Methods and systems for predicting onset and progression of medical conditions

Abstract

Provided herein are computer-implemented methods and systems for predicting the onset or progression of medical conditions using data from electronic health records (EHR) or electronic medical records (EMR), such as time-stamped blood-test values, in combination with trained predictive models. In some aspects, the predictive models comprise statistical or machine-learning algorithms trained using supervised learning on labeled clinical data. The disclosed methods and systems can generate individualized risk scores, classify subjects relative to operating thresholds, and optionally generate or record treatment recommendations in an EHR or EMR. The methods and systems can be applied to human or non-human subjects, enabling early detection of medical conditions, personalized risk assessment, and integration with healthcare information infrastructure.

Description

[0001] ALZAI-002-PCT

[0002] 2025-10-10

[0003] METHODS AND SYSTEMS FOR PREDICTING ONSET AND PROGRESSION OF MEDICAL CONDITIONS

[0004] INVENTORS: Amir Glik, Chen Hajaj, Orit Rephaeli, and Anat Goldstein

[0005] CROSS-REFERENCE

[0006]

[0001] Priority is claimed under PCT Art. 8(1) and Rule 4.10 to U.S. Provisional Application No. 63 / 706,020, filed October 10, 2024, and hereby incorporated by reference in its entirety for all purposes.

[0007] FIELD OF THE INVENTION

[0008]

[0002] The disclosure relates in some aspects to methods and systems for predicting the onset and / or progression of medical conditions using patient data, such as data from electronic medical records, for instance, blood test data. In some aspects, disclosed methods and systems enable early detection of medical conditions, personalized risk prediction, and integration with existing healthcare infrastructure.

[0009] BACKGROUND OF THE INVENTION

[0010]

[0003] Healthcare costs worldwide are on the rise, a trend attributed to aging populations, longer life expectancies, and the growing incidence of chronic and non-communicable diseases.

[0011]

[0004] These escalating costs place tremendous burdens on individuals, healthcare providers, and economies. Healthcare systems are struggling to balance rising demand with limited resources, leading to inefficiencies and disparities in access, especially to quality care.

[0012]

[0005] In this context, methods and systems capable of predicting the onset of medical conditions and generating models to identify individuals at high risk for such conditions would provide considerable benefits. Technologies to enable earlier detection and intervention, would for example reduce the need for costly late-stage treatments and hospitalizations, improve patient outcomes, and reduce the overall cost burdens.

[0013]

[0006] Provided herein are methods and systems to meet these needs and others and that have such advantages and improvements as will become known through the disclosure below.

[0014] INCORPORATION BY REFERENCE

[0015]

[0007] Each cited patent, publication, and non-patent literature is incorporated by reference in its entirety, as if each was incorporated by reference individually, and as if each is fully set forth herein. However, no such citation should be construed as an admission that a cited reference is from an area that is analogous or directly applicable to the invention, nor should any citation be construed as an admission that a document or underlying information, in any jurisdiction, is prior art or part of the common general knowledge in the art.

[0016] BRIEF SUMMARY OF THE INVENTION

[0017]

[0008] The following is a simplified summary of some embodiments of the invention in order to provide a basic understanding thereof. This summary is not an extensive overview of the invention, nor intended to ALZAI-002-PCT 2025-10-10 identify key or critical elements of the invention or to delineate its full scope. Its purpose is to present some embodiments and aspects of the invention in a simplified form as a prelude to the full description that follows.

[0009] In some aspects are provided computer-implemented methods for predicting the onset or progression of a medical condition in a subject, the methods comprising: (i) receiving, by a processor, a plurality of time-stamped blood-test values measured for the subject during at least one previous time period; (ii) normalizing each of the plurality of time-stamped blood-test values to a laboratory-specific reference interval to obtain age- and sex-adjusted standardized values; (iii) extracting, from the plurality of time-stamped blood-test values, a plurality of features comprising aggregations of values over time and change patterns in the values over time; (iv) applying, to the plurality of features, a trained predictive model configured and trained to compute a predicted risk score for the subject, wherein the predictive model is trained to predict a probability of onset or progression of the medical condition during a subsequent time period; (v) selecting, based on evaluation on a temporally separated validation set, an operating threshold that optimizes a performance metric balancing sensitivity and specificity, optionally subject to a false-positive-rate constraint, and classifying the subject relative to the threshold; (vi) generating a treatment recommendation for the subject by selecting a therapeutic intervention based on the predicted risk score from a plurality of therapeutic interventions; and (vii) writing the predicted risk score, an uncertainty interval, and the treatment recommendation to structured fields of an electronic medical record associated with the subject.

[0018]

[0010] In some embodiments, the predictive model is trained in a supervised training session using a plurality of labeled training samples each associating values of at least some of the plurality of blood test values measured for a respective subject during the at least one previous time period with a label indicative of whether or not onset of the medical condition was detected in the respective subject.

[0019]

[0011] In some embodiments, the trained predictive model comprises a statistical model. In some embodiments, the trained predictive model comprises a machine learning model. In some embodiments, selecting the operating threshold comprises maximizing Youden’s J index on the temporally separated validation set subject to a predefined false-positive-rate constraint. In some embodiments, the trained predictive model is further trained to predict a rate of exacerbation or a rate of progression of the medical condition for the subject. In some embodiments, the trained predictive model is further trained to use the rate of exacerbation or the rate of progression to classify the subject.

[0020]

[0012] In some embodiments, the plurality of blood test values are selected from the group consisting of absolute basophil count (Baso Abs), absolute eosinophil count (EOS Abs), hemoglobin (Hb), hematocrit (Het), absolute lymphocyte count (Lymp Abs), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), mean corpuscular volume (MOV), absolute monocyte count (MONO Abs), red blood cell count (RBC), procalcitonin (PCT), platelet count (PLT), white blood cell count (WBC), red cell distribution width (RDW), albumin, calcium, chloride, creatinine, globulin, glucose, magnesium, phosphorus, potassium, protein, sodium, urea, uric acid, aspartate aminotransferase (AST / GOT), gamma-glutamyl transferase (GGT), ALZAI-002-PCT 2025-10-10 alanine aminotransferase (ALT / GPT), total bilirubin, thyroid-stimulating hormone (TSH), vitamin B12, prothrombin time (PT), partial thromboplastin time (PTT), and international normalized ratio (INR).

[0021]

[0013] In some embodiments, the methods further comprise ranking the plurality of blood test values according to an impact of each of the blood test values to performance of the predictive model in computing the predicted risk score. In some embodiments, the plurality of blood test values used by the trained predictive model comprises a subset of highest ranking blood tests. In some embodiments, the trained predictive model is further trained to compute the predicted risk score based on a physiological parameter of the subject. In some embodiments, the trained predictive model is further trained to compute the predicted risk score based on a health-related risk, behavior, or susceptibility of the subject. In some embodiments, the trained predictive model is further trained to compute the predicted risk score based on a behavioral parameter of the subject. In some embodiments, the trained predictive model is further trained to compute the predicted risk score based on a sociodemographic parameter of the subject. In some embodiments, the trained predictive model is further trained to compute the predicted risk score based on a medical parameter of the subject.

[0022]

[0014] In some embodiments, the plurality of features extracted from the plurality of blood test values comprise: (i) one or more aggregations of values of each blood test over the at least one previous time period, the aggregation selected from the group consisting of an average value, a maximum value, a minimum value, and a standard deviation; and (ii) one or more change patterns detected in the values of at least one blood test over the at least one previous time period, the change pattern selected from a group comprising values increasing over time, values decreasing over time, values increasing then decreasing, and significant alternations between increases and decreases. In some embodiments, the at least one previous time period has a duration of 1 year, 2 years, 3 years, 5 years, or 10 years prior to the subsequent time period. In some embodiments, the subsequent time period has a duration of 1 year, 2 years, 3 years, 5 years, or 10 years following the at least one previous time period.

[0023]

[0015] In some embodiments, the medical condition is a cancer, a cardiovascular disease, a circulatory system disease, a chronic respiratory disorder, a musculoskeletal disorder, a connective tissue disease, a skin disease, a liver disease, a genitourinary disease, a metabolic disorder, an endocrine disorder, a neurological disorder, a chronic inflammatory disorder, a depressive disorder, a mental health disorder, a sleep-wake disorder, a sexual health disorder, an allergy, a periodontal disease, or a pediatric condition.

[0024]

[0016] In some embodiments, the medical condition is selected from the group consisting of: chronic obstructive pulmonary disease (COPD), chronic lung disease, congestive heart failure (CHF), chronic kidney disease (CKD), stage 4-5 chronic kidney disease or end-stage renal disease (ESRD), diabetes with complications, stroke, a neurological disorder, and HI V / AIDS.

[0025]

[0017] In some embodiments, the methods further comprise prescribing, recommending, performing, or administering to the subject any of pharmacotherapy, a gene therapy, an immunotherapy, a radiation therapy, a physical therapy, psychotherapy, counseling, or another mental health treatment, a stem cell therapy or other ALZAI-002-PCT 2025-10-10 regenerative medicine intervention, a non-pharmacological intervention, a surgical intervention, a medical device intervention, a behavioral intervention, a nutritional intervention, and a lifestyle modification.

[0026]

[0018] In some embodiments, the subject is a human. In some embodiments, the subject is a non-human animal. In some embodiments, the subject is a population of non-human animals. In some embodiments, the population of non-human animals is cattle, pigs, goats, sheep, poultry, horses, donkeys, mules, bees, dogs, cats, fish, or crustaceans.

[0027]

[0019] In some aspects are provided systems for predicting the onset of a medical condition in a subject, the systems comprising: (i) a processor; and (ii) a memory, storing instructions that, when executed by the processor, cause the system to: (a) receive, via the processor, a plurality of time-stamped blood-test values measured for the subject during at least one previous time period; (b) normalize each of the plurality of time-stamped blood-test values to a laboratory-specific reference interval to obtain age- and sex-adjusted standardized values; (c) extract, from the plurality of time-stamped blood-test values, a plurality of features comprising aggregations of values over time and change patterns in the values over time; (d) apply, to the plurality of features, a trained predictive model configured and trained to compute a predicted risk score for the subject, wherein the predictive model is trained to predict a probability of onset or progression of the medical condition during a subsequent time period; (e) select, based on evaluation on a temporally separated validation set, an operating threshold that optimizes a performance metric balancing sensitivity and specificity, optionally subject to a false-positive-rate constraint, and classifying the subject relative to the threshold; (f) generate a treatment recommendation for the subject by selecting a therapeutic intervention based on the predicted risk score from a plurality of therapeutic interventions; and (g) write the predicted risk score, an uncertainty interval, and the treatment recommendation to structured fields of an electronic medical record associated with the subject. In some embodiments, the disclosed systems comprise memory storing instructions that, when executed by the processor, cause the system to perform the method according to any of the disclosed embodiments.

[0028]

[0020] In some aspects are provided non-transitory computer-readable media storing instructions that, when executed by a processor, cause a system to perform a method for predicting the onset of a medical condition in a subject, the method comprising: (i) receive, via the processor, a plurality of time-stamped blood-test values measured for the subject during at least one previous time period; (ii) normalize each of the plurality of time-stamped blood-test values to a laboratory-specific reference interval to obtain age- and sex-adjusted standardized values; (iii) extract, from the plurality of time-stamped blood-test values, a plurality of features comprising aggregations of values over time and change patterns in the values over time; (iv) apply, to the plurality of features, a trained predictive model configured and trained to compute a predicted risk score for the subject, wherein the predictive model is trained to predict a probability of onset or progression of the medical condition during a subsequent time period; (v) select, based on evaluation on a temporally separated validation set, an operating threshold that optimizes a performance metric balancing sensitivity and specificity, ALZAI-002-PCT 2025-10-10 optionally subject to a false-positive-rate constraint, and classifying the subject relative to the threshold; (vi) generate a treatment recommendation for the subject by selecting a therapeutic intervention based on the predicted risk score from a plurality of therapeutic interventions; and (vii) write the predicted risk score, an uncertainty interval, and the treatment recommendation to structured fields of an electronic medical record associated with the subject. In some embodiments, the disclosed media store instructions that, when executed by the processor, cause the system to perform the method according to any of the disclosed embodiments.

[0029]

[0021] Also provided are further methods, systems, and media as will be described and enabled herein.

[0030]

[0022] The foregoing has outlined broadly and in summary certain pertinent features of the disclosure so that the detailed description that follows may be better understood, and so the present contribution to the art can be more fully appreciated. This summary is a brief and general synopsis of only some of the disclosed aspects and embodiments, is provided solely for the benefit and convenience of the reader, and is not intended to limit in any manner the scope, or range of equivalents, to which the claims are entitled.

[0031]

[0023] Additional aspects and embodiments are described below. It will be appreciated by one in the art that all disclosed methods and systems are only exemplary, and will readily serve as a basis to modify or design other methods and systems for carrying out the same purposes, which are also within the scope hereof.

[0032]

[0024] The headings in this document are provided only to expedite its review by a reader. They should not be construed as limiting the invention or any of its aspects, embodiments, or applications in any manner.

[0033] BRIEF SUMMARY OF THE FIGURES

[0034]

[0025] To further clarify various aspects of the invention, certain exemplary embodiments are illustrated in the figures. The figures depict only illustrated embodiments of the invention and should not be considered limiting of its scope. Certain aspects of the invention are thus further described and explained with additional specificity and detail, but still by way of example only, with reference to the accompanying figures in which:

[0035]

[0026] FIG. 1 is a schematic illustration of an exemplary system for predicting onset of medical conditions, according to some embodiments of the disclosure;

[0036]

[0027] FIG. 2 is a flowchart of an exemplary computer-implemented method of predicting onset of medical conditions, such as can be implemented on the exemplary system in FIG. 1;

[0037]

[0028] FIG. 3 is a flowchart of an exemplary method for training a predictive model to predict the onset of a medical condition in a subject, such as the model used by the disclosed embodiments;

[0038]

[0029] FIG. 4 is a flowchart illustrating an exemplary feature extraction process according to some disclosed embodiments;

[0039]

[0030] FIG. 5 is a schematic block diagram illustrating an exemplary system architecture for predicting the onset or progression of medical conditions, in accordance with some embodiments of the disclosure;

[0040]

[0031] FIGS. 6, 7, 8, and 9 are graph charts illustrating the importance of selected features for the prediction of onset of an exemplary condition and / or exemplary condition risk factor with respect to duration of the previous time period during which the blood test values (features) were measured and the subsequent time ALZAI-002-PCT 2025-10-10 period, i.e., horizon, during which the onset of the condition is predicted (estimated) to develop, according to the exemplary embodiment of Example 11 herein; and

[0041]

[0032] FIG. 10 is a graph depicting age distribution for various classes in the population at the end of the study period of the exemplary embodiment of Example 11 herein.

[0042] DETAILED DESCRIPTION OF THE INVENTION

[0043]

[0033] While various exemplary aspects and embodiments are summarized above, the following description illustrates several exemplary embodiments in further detail to enable one having ordinary skill in the art to which the invention pertains (also as shorthand herein, “one” or “those” “of skill” or “in the art”) to make and use the full scope of the invention as claimed.

[0044]

[0034] It will be understood that many modifications, substitutions, changes, and variations in the described aspects, embodiments, applications, examples, and details of the disclosed invention (“disclosure”) can be made by those in the art without departing from the spirit of the invention, or the scope of the invention as claimed, and the general principles described may be applied to a wide range of aspects and embodiments.

[0045]

[0035] Thus, the invention should not be limited to the aspects and embodiments presented, but should be accorded the widest scope consistent with the principles and features disclosed. The description will make such principles and features apparent to those of ordinary skill, in that the expressly described as well as further embodiments will be both readily cognizable and readily creatable without undue experimentation, solely using the teachings of this disclosure together with the general knowledge in the art.

[0046] I. General Definitions and Terms

[0047]

[0036] The singular “a,” “an,” and “the” include plural referents unless context clearly indicates otherwise. While the term “one or more” may be used, its absence (or its replacement by the singular “a” or “an”) does not signify the singular only; rather, “one or more” simply emphasizes the possibility of a plurality in particular embodiments. “Or” means, and is interchangeable with, “and / or” unless context clearly indicates otherwise.

[0048]

[0037] The terms “comprising,” “including,” “such as,” and “having” are inclusive and not exclusive (i.e., they do not limit lists to the recited elements), and all are interchangeable with the phrases “including but not limited to” and “including without limitation.” The term “exemplary” means the same as “an example,” and both are the same as the phrase “one of a plurality of non-limiting examples,” and mean that other such examples may be known to those of skill, in view of the disclosure and the general knowledge in the art. An “exemplary” embodiment, or “examples” of features of an embodiment, will be understood to be representative example(s), but not to imply that any such example is a preferred example, or is from a group of preferred examples, unless expressly stated as such, or context shows otherwise.

[0049]

[0038] Numerical parameters should be construed in light of the number of reported significant digits and by applying ordinary rounding techniques. Notwithstanding that the numerical ranges and parameters setting forth the broad scope of embodiments are approximations, the numerical values set forth in the examples are reported as precisely as practicable. Numerical values in some embodiments may contain certain errors ALZAI-002-PCT 2025-10-10 necessarily resulting from the standard deviation found in their respective testing measurements.

[0050]

[0039] Unless otherwise indicated, all numbers and terms expressing quantities should be understood as being modified in some instances by the term “about,” even where not so stated explicitly. In alternative embodiments, such numbers and terms should be understood as not being modified by the term “about.” In some embodiments, such numbers and terms are approximations that can vary depending upon the desired properties of or sought to be obtained by a particular embodiment. In some embodiments, “about” refers to plus or minus five percent (±5%) of the recited unit of measure. In some embodiments, “about” refers to plus or minus ten percent (±10%) of the recited unit of measure. Where “about” is used to modify one number in a series or range, it should be understood to modify all numbers in the series or range, including, for a range, both the upper and lower bounds of the range. Thus, the term “about 1 , 2, or 3” is understood to mean “about 1 , about 2, or about 3” and the term “about 1 to 10” means “about 1 to about 10.” The term “substantially,” where used to modify a feature or limitation, must be read in the context of the disclosure and in light of the knowledge in the art to provide the appropriate certainty, such as by using an art-recognized standard for measuring “substantially” as a term of degree, or by otherwise ascertaining the scope as would one of skill.

[0051]

[0040] Herein, “in embodiments” is equivalent to, and simply shorthand for, the term “in some embodiments.”

[0052]

[0041] “Embodiments” refers to disclosed embodiments and, unless otherwise indicated, their equivalents.

[0053]

[0042] A “disclosed method” refers to any method of the invention, such as described herein or claimed, and, unless otherwise indicated, further including such methods as will be understood as equivalents in view of this disclosure and the general knowledge of the art, to one of skill. Reference to a “disclosed method” also will refer to the systems for performing the disclosed method, such as a “disclosed system” (as similarly defined), and will additionally refer to non-transitory computer-readable media storing instructions that, when executed by one or more processors, cause such systems to perform the method, each unless context demands otherwise. Thus, unless context demands otherwise, a “disclosed method” may refer to a “disclosed method or system” and also may refer to a “disclosed method, system, or computer-readable medium.” Unless context indicates otherwise, a “method” may mean, and be shorthand for, a “disclosed method.” Where context does not prohibit it, “a method” also may mean, and be shorthand for, the term “a method, including one or more steps or limitations of the method.” Similarly, where context does not prohibit it, “a system” also may mean, and be shorthand for, the term “a system, including one or more aspects or limitations of the system.”

[0054]

[0043] Generally, the nomenclature used and procedures performed herein are those known in fields relating to one or more aspects of the invention, e.g., bioinformatics, machine learning, medicine, computational biology, data science, and the like; and that will be well known and commonly employed in such fields.

[0055]

[0044] Unless described otherwise, disclosed techniques and procedures will be those that are standard and known in the art, and will be those performed according to conventional methods in the art. If no techniques or procedures are expressly disclosed for any methods or steps herein, it may be inferred that one or more standard techniques or procedures is performed. Standard techniques and procedures are those generally ALZAI-002-PCT 2025-10-10 performed according to conventional methods in the art. Unless defined otherwise, all technical terms have the meaning that the term would have to a person of ordinary skill in the art in question at the time of the invention.

[0056]

[0045] Where a specific term is used for a parameter, feature, category, classification, function, step, embodiment, aspect, or the like, including terms expressly defined herein for such purposes, and the term does not have an ordinary and customary meaning in the art, its use will be understood to reflect a preference rather than a limitation, unless the term provides necessary clarity or contributes to a required technical effect. Except in such cases, or where context demands otherwise, the use of an alternative term will be considered equivalent and within the scope of the disclosure. In short, where the inventors have acted as their own lexicographers, such terms should be understood to represent merely one chosen means of identification and connotation, which may have one or more equivalent alternatives.

[0057]

[0046] Further definitions to assist a reader in understanding various aspects and embodiments of the disclosed methods and systems are throughout; however, such definitions are also not intended to limit the scope of the invention, which is properly interpreted and understood by reference to the full specification (and any plain meaning known to one of skill in the relevant art) in view of the language of the claims. The terminology used herein is for the purpose of describing particular embodiments, and is not limiting.

[0058] II. Methods and Systems for Predicting Onset and Progression of Medical Conditions

[0059]

[0047] In some aspects are provided methods and systems for predicting the onset and / or progression of medical conditions. In some embodiments, the disclosed methods and systems generate a prediction of disease onset, disease progression, or both, using trained predictive models configured to analyze heterogeneous biomedical, physiological, and behavioral data.

[0060]

[0048] In some embodiments, the disclosed methods and systems enable identification or mitigation of risk factors associated with medical conditions by non-invasive means, and thereby enable earlier clinical intervention or preventive treatment, delaying or preventing disease onset or progression.

[0061]

[0049] In some embodiments, a disclosed method comprises receiving a plurality of physiological values for a subject, wherein each value is associated with a respective timestamp. Herein, unless context indicates otherwise, a “subject” may be understood as equivalent to a “target subject” or “target individual.” The plurality of physiological values may include, without limitation, laboratory measurements, vital signs, or device-recorded physiological signals collected at one or more time points.

[0062]

[0050] In some embodiments, a method further comprises transforming the plurality of physiological values into one or more normalized or standardized feature vectors suitable for input to a predictive model. The model, which may comprise a statistical model, a machine learning (ML) model, or another computational predictive framework, is trained using longitudinal datasets to estimate the probability or confidence of onset or progression of a medical condition.

[0063]

[0051] In some embodiments, the method comprises computing, with respect to the subject, based on the plurality of physiological values, an output indicating a probability, likelihood, or confidence value representing ALZAI-002-PCT 2025-10-10 the predicted onset, progression, or exacerbation of a medical condition. In some embodiments, the output comprises a continuous score, a binary classification, or a categorical risk label (e.g., low, moderate, or high risk). In some embodiments, the method further comprises identifying one or more input features that contribute most strongly to the predicted outcome, thereby generating an interpretable attribution or feature-importance profile for the subject.

[0064]

[0052] In some embodiments, the predictive model receives as input a combination of features derived from blood-based laboratory tests, physiological signals, and other health indicators. Exemplary laboratory-based inputs include Complete Blood Count (CBC), blood chemistry profile (e.g., electrolytes, enzymes, proteins, and lipid levels), and hormonal panels (e.g., thyroid-stimulating hormone (TSH), cortisol, sex hormones, insulin). In some embodiments, the predictive model additionally receives anthropometric and physiological parameters such as blood pressure, heart rate, body weight, or body-mass index (BMI), and in some embodiments, derived temporal features such as rate of change, variability, or slope across repeated measurements.

[0065]

[0053] In some embodiments, additional or alternative parameter values are incorporated into the predictive analysis to increase robustness and predictive power. These include, by way of example, vascular and metabolic risk factors (e.g., hypertension, hyperlipidemia, diabetes mellitus), sociodemographic variables (e.g., age, sex, race, socioeconomic status), and behavioral parameters (e.g., tobacco or alcohol use, medication adherence, and physical activity). In some embodiments, medical-history parameters (e.g., comorbid conditions, background disease, current medication regimen) are included, alone or in combination with physiological and laboratory data, to generate individualized risk profiles.

[0066]

[0054] In some embodiments, such multimodal features are extracted or derived from structured and unstructured electronic medical record (EMR) or electronic health record (EHR) data. These may include demographic information, medical history, medication lists, allergies, vital signs, immunization records, laboratory test results, imaging and radiology reports, progress notes, treatment plans, surgical records, visit or encounter history, family history, and social and behavioral health information. In some embodiments, additional data sources include wearable or home-monitoring devices, telehealth platforms, or patient-generated health data (PGHD). In some embodiments, such multimodal features are extracted or derived from structured, semi-structured, or unstructured data modalities, including tabular time-series data, free-text clinical notes, and imaging-derived quantitative parameters.

[0067]

[0055] In some embodiments, a disclosed method comprises processing the plurality of features using an ML algorithm to identify latent patterns, trends, or anomalies predictive of disease onset or progression. The model may be trained on longitudinal case-control data to learn relationships between historical measurements and subsequent diagnosis, or on survival or time-to-event data to estimate hazard or progression rates. The predictive model may apply regression, classification, or sequence modeling architectures, such as gradient-boosted decision trees, recurrent neural networks, or transformer-based models, optimized using cross-validation and feature-importance regularization. ALZAI-002-PCT

[0068] 2025-10-10

[0069]

[0056] In some embodiments, the predictive model outputs a risk score that is compared to one or more predefined thresholds to categorize the subject’s predicted condition likelihood. Herein, “risk score” refers to a continuous or categorical model output quantifying the predicted likelihood of onset or progression of a medical condition. In some embodiments, a method further comprises ranking or prioritizing a plurality of risk factors contributing to the prediction and generating a set of recommended preventive or therapeutic interventions associated with the ranked features. In some embodiments, the model output comprises a scalar risk score or categorical probability class, which is compared to one or more clinically defined thresholds to trigger recommendations or alerts. Thresholds may be dynamically calibrated to maximize sensitivity for early-stage conditions or specificity for advanced disease management.

[0070]

[0057] In some embodiments, the system automatically generates personalized recommendations or alerts to clinicians, enabling early intervention, individualized monitoring, or adaptive treatment planning.

[0071]

[0058] In some embodiments, the method further comprises assessing the rate of exacerbation or rate of progression of a medical condition and characterizing subgroups of subjects according to the rate of progression. In some embodiments, the method includes longitudinal recalibration of the predictive model as additional data become available, thereby improving model accuracy and stability over time.

[0072]

[0059] In some embodiments, the disclosed methods and systems determine, based on routine clinical test results such as blood panels, an individualized probability of developing a medical condition within a defined future time window (e.g., one year, three years, or five years). Such predictive analytics can be used to inform personalized health recommendations, early interventions, and dynamic treatment plans, enhancing the timeliness and precision of preventive healthcare.

[0073] III. Exemplary Methods for Training a Predictive Model

[0074]

[0060] In some aspects are provided methods and systems for training a predictive model to predict the onset of a medical condition using subject data, such as electronic medical record (EMR) or electronic health record (EHR) data, blood test data, and blood tests included within EMR data. Unless otherwise specified or required by context, references to EMR data are understood to encompass EHR data, and vice versa.

[0075]

[0061] In some embodiments, a method comprises receiving a dataset of subject data that includes blood test values with timestamps and labels indicating the presence or absence of a medical condition.

[0076]

[0062] Blood Test Data and Biomarkers. In some embodiments, blood test data used for training or inference include values for a plurality of biomarkers or analytes relevant to predicting one or more medical conditions. Blood test data may be obtained from any suitable clinical assay (e.g., plasma or serum testing) and may include biomarkers indicative of physiological states or medical conditions. For ease of reference, exemplary biomarkers and analytes are grouped below by general physiological category; however, such groupings are not limiting, and any combination of biomarkers (within or across categories) may be used. Biomarkers used to predict a particular medical condition need not be conventionally associated with that condition. In some embodiments, predictive relationships are identified among diverse biomarkers, including those not ALZAI-002-PCT

[0077] 2025-10-10 traditionally considered indicative of the condition being predicted.

[0078]

[0063] Hematologic Parameters: Hemoglobin (Hb), hematocrit (Het), red blood cell count (RBC), white blood cell count (WBC), platelet count (PLT), mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), red cell distribution width (RDW), absolute neutrophil count, absolute lymphocyte count (Lymp abs), absolute eosinophil count (Eos abs), absolute basophil count (Baso abs), absolute monocyte count (Mono abs), reticulocyte count, erythropoietin (EPO), immature granulocyte percentage, mean platelet volume (MPV), and neutrophil-to-lymphocyte ratio (NLR).

[0079]

[0064] Metabolic and Renal Function Markers: Glucose, urea (blood urea nitr., BUN), creatinine, uric acid, phosphorus, calcium, magnesium, potassium, sodium, chloride, bicarbonate, serum osmolality, total protein, albumin, globulin, anion gap, estimated glomerular filtration rate (eGFR), and cystatin C.

[0080]

[0065] Hepatic Function Markers: Alanine aminotransferase (ALT / GPT), aspartate aminotransferase (AST / GOT), alkaline phosphatase (ALP), gamma-glutamyl transferase (GGT), bilirubin (total and direct), albumin, globulin, total bile acids, lactate dehydrogenase (LDH), prothrombin activity (PTA), total protein, and glutamate dehydrogenase (GLDH).

[0081]

[0066] Coagulation and Fibrinolysis Markers: Prothrombin time (PT), partial thromboplastin time (PTT), international normalized ratio (INR), fibrinogen, D-dimer, and von Willebrand factor (vWF).

[0082]

[0067] Inflammatory and Immunologic Markers: C-reactive protein (CRP), high-sensitivity C-reactive protein (hs-CRP), serum amyloid A (SAA), serum amyloid P (SAP), tumor necrosis factor-alpha (TNF-a), interleukin-1 beta (IL-1 P), interleukin-6 (IL-6), interleukin-10 (IL-10), transforming growth factor-beta (TGF- ), immunoglobulins (IgG, IgA, IgM, IgE), anti-nuclear antibodies (ANA), myeloperoxidase (MPO), galectin-3, matrix met- alloproteinase-9 (MMP-9), soluble urokinase plasminogen activator receptor (suPAR), neutrophil gelatinase- associated lipocalin (NGAL), fibroblast growth factor 23 (FGF23), and fibroblast growth factor 21 (FGF21).

[0083]

[0068] Endocrine and Hormonal Markers: Thyroid-stimulating hormone (TSH), triiodothyronine (T3), thyroxine (T4), cortisol, adrenocorticotropic hormone (ACTH), parathyroid hormone (PTH), prolactin, growth hormone (GH), insulin-like growth factor-1 (IGF-1), insulin, proinsulin, C-peptide, testosterone, estradiol (E2), follicle-stimulating hormone (FSH), luteinizing hormone (LH), aldosterone, renin, ghrelin, leptin, and vitamin D.

[0084]

[0069] Cardiac and Vascular Markers: Troponin, creatine kinase (CK), myoglobin, B-type natriuretic peptide (BNP), N-terminal pro-B-type natriuretic peptide (NT-proBNP), homocysteine, lactoferrin, angiopoietin-2 (Ang-2), placental growth factor (PIGF), and soluble fms-like tyrosine kinase- 1 (sFlt-1).

[0085]

[0070] Lipid and Nutritional Markers: Total cholesterol, high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, triglycerides, lipoprotein(a), apolipoprotein A1 , apolipoprotein B, ferritin, transferrin, total iron-binding capacity (TIBC), zinc, vitamin B9 (folate), vitamin B12, and vitamin K.

[0086]

[0071] Metabolic and Energy-Related Biomarkers: Beta-hydroxybutyrate, ketones, serum free fatty acids, acylcarnitines, serum ammonia, lactate, pyruvate, uric acid, and carnitine (total and free).

[0087]

[0072] Neurotrophic and Growth Factors: Brain-derived neurotrophic factor (BDNF), fibroblast growth factor ALZAI-002-PCT

[0088] 2025-10-10

[0089] 23 (FGF23), fibroblast growth factor 21 (FGF21), and insulin-like growth factor-1 (IGF-1).

[0090]

[0073] Oncologic and Tumor Markers: Alpha-fetoprotein (AFP), carcinoembryonic antigen (CEA), cancer antigen-125 (CA-125), cancer antigen-19-9 (CA-19-9), beta-2 microglobulin, neuron-specific enolase (NSE), squamous cell carcinoma antigen (SCC), and serum free light chains (kappa / lambda).

[0091]

[0074] Preprocessing and Feature Engineering. In some embodiments, blood test values used for prediction are extracted from EMR data spanning historical time periods of, for example, 6 months to 10 years or longer (e.g., 15, 20, 25, 30, 40, or 50 years), with each value associated with a timestamp. Preprocessing may include cleaning, normalizing, unit harmonization, reference-range alignment, encoding categorical variables (e.g., diagnosis codes, medication identifiers), handling missing or irregularly sampled measurements (e.g., imputation or interpolation), and vectorizing temporal information. Extracted features may include aggregated statistics (e.g., mean, median, maximum, minimum, standard deviation, range, variance) and temporal features (e.g., trends, slopes, periodicity, volatility) derived from time-series transformations.

[0092]

[0075] Temporal Change Patterns. In some embodiments, temporal change patterns are used as features. Exemplary patterns include stepwise or “footstep” trajectories such as values that increase over time (“upstairs”), decrease over time (“downstairs”), increase then decrease (“mountain”), or show repeated alternations (“fingers”), as well as other complex or irregular longitudinal profiles. Using such patterns may enhance predictive performance and interpretability.

[0093]

[0076] Prediction Horizons and Label Construction. In some embodiments, the model predicts the probability and / or confidence of onset of one or more medical conditions during a subsequent time period (prediction horizon) based on historical blood test values measured during one or more previous time periods. Herein, “prediction horizon” refers to the temporal interval between the latest available input data and the modeled occurrence of the predicted condition. The prediction horizon may span, for example, 6 months, 1 year, 2 years, 3 years, 4 years, 5 years, or 10 years, or any interval in between, and may exceed 10 years (e.g., 15, 20, 25, 30, 35, 40, 45, or 50 years). Labels may indicate whether onset of a medical condition was observed during the subsequent period. For non-human animals, time periods may be scaled according to species-specific lifespan or healthspan.

[0094]

[0077] Label Derivation and Cohort Assignment. In some embodiments, labels are derived from electronic medical records by verifying the absence of a diagnosis of the medical condition during a lookback period and the presence or absence of onset during a subsequent prediction horizon. In some embodiments, preprocessing further classifies subjects into active and control cohorts based on clinical indicators such as diagnosis codes, medication prescriptions, or treatment histories. Subjects exhibiting evidence of active treatment or confirmed diagnosis may be assigned to an active cohort, while subjects lacking such indicators are assigned to a control cohort. In some embodiments, data quality procedures include reconciling inconsistent diagnosis and medication entries, removing duplicate records, and validating temporal ordering of laboratory results and outcome events. ALZAI-002-PCT

[0095] 2025-10-10

[0096]

[0078] Model Families. In some embodiments, a trained predictive model comprises one or more statistical or ML models. Exemplary statistical models include linear regression, logistic regression, time-series models, and Bayesian models. Exemplary ML models include decision trees, random forests, gradient boosting machines (e.g., XGBoost), support vector machines (SVMs), and neural networks. Neural networks may include feed-forward networks, convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory (LSTM) or gated recurrent unit (GRU) networks, transformer-based architectures, or combinations thereof. The choice of architecture may depend on whether input data are sequential, spatial, or tabular, and on the specific predictive objective.

[0097]

[0079] Supervised Training and Data Partitioning. In some embodiments, the dataset is divided into nonoverlapping subsets (e.g., training, validation, and test sets). During supervised training, model parameters are iteratively adjusted to minimize a loss function based on labeled samples. The validation set may be used to tune hyperparameters (e.g., learning rate, regularization strength, tree depth, number of layers). The test set may be used to evaluate generalization on data not seen during training or validation. In some embodiments, techniques for addressing class imbalance (e.g., class weighting, re-sampling, or focal loss) are employed.

[0098]

[0080] Classification Outputs and Thresholding. In embodiments, the trained model outputs a risk score that may be thresholded to produce a categorical output (e.g., binary “positive / negative” or “YES / NO”). Threshold selection may consider sensitivity, specificity, disease prevalence, and the costs and benefits of misclassifications. In some embodiments, thresholds are selected using procedural methods, such as optimization of ROC-based criteria (e.g., Youden’s J), cost-sensitive utility functions, or other decision analytic techniques.

[0099]

[0081] Model Validation and Performance Evaluation. In some embodiments, model performance is characterized using accuracy, precision, recall (sensitivity), specificity, F1 score, and area under the receiver operating characteristic curve (AUC-ROC). Additional metrics may include area under the precision-recall curve (AUPRC), Matthews correlation coefficient (MCC), calibration error, or Brier score. External validation may be performed using temporally separated or site-held-out datasets to assess generalizability. Calibration methods (e.g., Platt scaling or temperature scaling) and optional explainability techniques (e.g., feature attributions) may be employed. A predictive model may be characterized by its accuracy (i.e., its ability to correctly identify subjects who will and who will not experience onset during a specified time period) and by its sensitivity (i.e., its ability to correctly identify subjects who will experience onset during the specified time period). A “specified time period” may be any time period described herein.

[0100]

[0082] Model Accuracy and Sensitivity Targets. In some embodiments, a predictive model trained according to any embodiment herein achieves an accuracy and / or sensitivity of at least 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 95%, or 99% for predicting the onset and / or progression of one or more medical conditions in a subject during a specified time period. In some embodiments, the foregoing levels are achieved for conditions selected from the group consisting of: cancer; cardiovascular disease; circulatory system disease; chronic respiratory disorder; musculoskeletal disorder; connective tissue disease; skin disease; liver disease; ALZAI-002-PCT 2025-10-10 genitourinary disease; metabolic disorder; endocrine disorder; neurological disorder; chronic inflammatory disorder; depressive disorder; mental health disorder; sleep-wake disorder; sexual health disorder; allergy; periodontal disease; pediatric condition; and any negative health outcome, or risk, behavior, or susceptibility, that contributes to or is associated with a negative health outcome. In some embodiments, the foregoing levels are achieved for diseases in non-human animals, with time periods adjusted by species.

[0101]

[0083] Feature Ranking and Selection. In some embodiments, blood tests and / or derived features are ranked or prioritized by impact on model performance (e.g., accuracy, reliability, consistency, or explanatory value). In some embodiments, only a top-ranked subset is used to train the model and compute the risk score, reducing complexity, improving accuracy, and lowering compute and storage requirements.

[0102]

[0084] Model Refinement with Additional Subject Parameters. In some embodiments, a predictive model is refined using additional subject parameters, such as physiological (e.g., blood pressure, ECG, heart rate, BMI, EEG); vascular (e.g., hypertension, hyperlipidemia, ischemic heart disease, myocardial infarction, diabetes mellitus); behavioral (e.g., smoking, alcohol consumption, drug use, medication adherence, physical activity); medical (e.g., comorbidities, current medications); and sociodemographic (e.g., gender, race, ethnicity, education, age, height, weight); however such categories should not be limiting. In some embodiments, previously predicted risk scores for a first medical condition are used as input features for predicting a second condition, enabling hierarchical or cascaded models. Model-specific refinements may include, for example, weighting or structure adjustments in decision trees or ensembles, boosted emphasis on misclassified cases in gradient boosting, and added layers or feature channels in neural networks.

[0103]

[0085] Hierarchical and Cascaded Prediction Frameworks. In embodiments, the system employs hierarchical or cascaded predictive architectures in which the output of a 1st trained model, such as a risk score for a metabolic or vascular condition, is provided as an input feature to a 2nd model configured to predict a related or downstream condition. In embodiments, multiple predictive models operate sequentially or in parallel, with later models refining or contextualizing the outputs of earlier ones. Such multi-stage frameworks can capture interdependencies among medical conditions and improve both predictive accuracy and interpretability.

[0104]

[0086] Model-Specific Refinement Examples. In some embodiments, individual model families are refined using architecture-appropriate adjustments. For example, a random forest may be refined by reweighting or pruning trees that contribute to misclassifications, while a gradient boosting model (e.g., XGBoost) may incorporate subject-specific covariates into its boosting iterations to emphasize difficult or previously misclassified samples. In some embodiments, a convolutional neural network (CNN) is refined by adding or retraining convolutional filters to capture contextual or auxiliary input channels, and a deep learning neural network (DLNN) is retrained with additional input nodes representing newly available subject-level parameters or longitudinal features. Such refinements can reduce false positives and false negatives and improve calibration across heterogeneous datasets.

[0105]

[0087] Clinical and Predictive Applications. In some embodiments, the trained model outputs a risk score ALZAI-002-PCT 2025-10-10 (numerical, categorical, or binary) that may be used to guide further diagnostic testing, preventative interventions, treatment planning, monitoring, or automated health-system actions. In some embodiments, the model predicts disease progression rate and / or assigns subjects to progression-profile classes based on predicted symptoms, rate of change, or expected functional decline.

[0106]

[0088] Exemplary Training Method. Reference is made to FIG. 3, which is a flowchart of an exemplary method for training a predictive model. At 810, a dataset containing blood test values for multiple subjects over a past time period is received, with each value timestamped and each subject labeled as having or not having a medical condition. At 820, features are extracted from the blood test values (e.g., aggregates and temporal change patterns). At 830, the predictive model is trained using the extracted features and labels. At 840, the trained predictive model is output for classifying the onset of a medical condition in a subject.

[0107]

[0089] FIG. 4 illustrates an exemplary feature extraction and preprocessing pipeline according to some embodiments. Time-stamped blood-test values 401 are first aggregated statistically 402 to compute summary measures such as mean, maximum, minimum, and standard deviation across a defined historical window. Temporal pattern detection 403 identifies dynamic changes in analyte values over time, including upward or downward trends and oscillating patterns. The resulting features are then normalized 404, for example by scaling to laboratory-specific reference intervals or z-scores. A feature selection and ranking stage 405 determines the relative contribution of each feature to model performance and retains the highest-impact features. The selected and normalized features are assembled into a feature vector 406 that serves as input to the trained predictive model for risk-score computation.

[0108]

[0090] As shown in FIG. 5, the system may include an Electronic Medical Record (EMR) system 501, a Predictive Analytics System 502, a Prediction Output 503, and a Clinical Decision Support System 504, which may communicate through one or more networks or interfaces. In the illustrated embodiment, the EMR system 501 comprises stored blood-test data, other clinical data, and updated risk scores that are optionally received from and may be additionally exchanged with (i.e., 2-way) the predictive analytics system. The Predictive Analytics System 502 includes an API interface module configured to retrieve time-stamped subject data from the EMR system 501, a data-processing engine configured to standardize, transform, and feature-encode such data, a trained predictive model configured to compute one or more predicted risk scores, and a risk-score calculator that generates, aggregates, and / or normalizes those scores for one or more medical conditions. The predictive analytics system 502 transmits the resulting prediction to a Prediction Output 503, such as a GUI or report generator, and optionally to a Clinical Decision Support System 504.

[0109]

[0091] In some embodiments, the Clinical Decision Support System 504 compares the received predicted risk score against predefined thresholds and, when appropriate, issues alerts, recommendations, or treatment suggestions to a clinician. The communication paths shown by solid arrows represent direct data exchanges, while dashed arrows indicate optional or asynchronous feedback flows, such as the return of updated risk scores from the predictive analytics system 502 to the EMR system 501 for longitudinal record-keeping. ALZAI-002-PCT

[0110] 2025-10-10

[0111] IV. Exemplary System Architecture

[0112]

[0092] In some aspects are provided systems for performing the disclosed methods, such as for predicting onset and / or progression of medical conditions.

[0113]

[0093] Any embodiment referring to a method will be understood to refer equivalently to a system, such as a computer system that is configured to perform the method. Likewise, any embodiment referring to a step of a method will be understood to refer equivalently to a system configured to perform the step of the method.

[0114]

[0094] Embodiments of the disclosed methods can be implemented in numerous ways and using numerous forms of systems and types of system architecture, and that unless clearly indicated as being directed to a particular implementation or type of implementation, the disclosed embodiments are not limited to any such specific implementations. The following system architecture features are provided solely to further elucidate certain exemplary aspects and embodiments of the disclosure, but are not limiting.

[0115]

[0095] System Overview and Components. Reference is made to FIG. 1, which is a schematic illustration of an exemplary system 100 for predicting the onset and / or progression of medical conditions. The system may comprise one or more processors 110 configured to execute code 120 stored on one or more storage units 109. The code 120 may include executable instructions that (i) receive temporally tagged laboratory values and other subject data; (ii) transform or preprocess such data (e.g., normalization, encoding, feature extraction); and (iii) apply a trained predictive model 160 to compute a predicted risk score for one or more medical conditions. In some embodiments, the system further comprises a display device 210 and / or communicates with a client computing device 450 that presents predicted risk scores and associated context (e.g., confidence measures, contributing features) to authorized users via a user interface. A network 410 may couple the foregoing components and allow secure communication among servers, storage, and client devices.

[0116]

[0096] Exemplary Inference Workflow. Reference is made to FIG. 2, which is a flowchart of a computer-implemented method 700 of prospectively predicting onset and / or progression of medical conditions using the system of FIG. 1. At 710, the system receives values of a plurality of blood tests (and optionally additional subject parameters) measured for one or more subjects during previous time period(s), each value associated with a timestamp. At 720, the system applies one or more trained predictive models 160 to compute a predicted risk score based on features extracted from the received values. At 730, the system outputs the predicted risk score and, in some embodiments, explanatory information (e.g., confidence interval, contributing risk factors) to a display device 210 or to the EMR 510, where the output may be stored as a timestamped, auditable record.

[0117]

[0097] User Interfaces and Explainability. In some embodiments, the system provides role-based user interfaces tailored for clinicians, administrators, and data science personnel. Clinician-facing views may present predicted risk scores, classifications, confidence intervals, and a concise list of contributing features or feature attributions, which may be generated using explainability techniques (e.g., permutation importance or attribution methods) without limiting the models employed. Administrator or operations views may include alert ALZAI-002-PCT 2025-10-10 queues and acknowledgments, while data science views may expose performance dashboards, calibration checks, and drift indicators. In some embodiments, explanation artifacts are stored alongside predictions in the EMR to facilitate clinical review and audit.

[0118]

[0098] Deployment Topologies. In some embodiments, the disclosed system is deployed as a standalone application on a local computing device (e.g., desktop, laptop, workstation, or on-premises server) that stores code 120 and the trained model 160 and accesses subject data from local or connected databases. In other embodiments, the system is deployed as a cloud-based service accessible over a network 410 via a web interface or dedicated client, allowing scalable compute and storage and centralized model management. In yet other embodiments, the system is implemented as a service integrated with or connected to an EMR system 510 of a healthcare provider, as further described below.

[0119]

[0099] EMR Integration and Data Interoperability. In some embodiments, the system is directly integrated with an EMR system 510 through secure application programming interfaces (APIs), for example, APIs compliant with Health Level Seven (HL7) and Fast Healthcare Interoperability Resources (FHIR) standards, or via a dedicated middleware layer. Such integration can enable bidirectional data exchange between the disclosed system and clinical databases, including automated, periodic ingestion of structured patient data (e.g., laboratory values with timestamps, diagnostic codes and results, medication records, demographic and clinical parameters) and automated write-back of predicted risk scores and explanatory metadata to designated EMR fields or structured data objects. Stored predictions may be timestamped and auditable, forming part of the subject's medical record and remaining visible to authorized clinicians.

[0120]

[0100] Clinical Decision Support and Alerts. In some embodiments, the system includes a clinical decision support (CDS) module that evaluates predicted risk scores against configurable thresholds and triggers alerts within the EMR when a score exceeds a predefined limit. Alerts may recommend confirmatory tests, specialist referrals, or preventative interventions. The CDS module may log alert acknowledgments and downstream actions, and may support longitudinal visualization of successive risk scores to assist clinicians in monitoring trends over time.

[0121]

[0101] Security and Compliance. In some embodiments, the system is configured to protect sensitive health information and comply with applicable data-privacy and security regulations, such as HI PAA and GDPR. Protections may include access control and authentication, encryption of data in transit and at rest, secure key management, role-based authorization, network isolation, audit logging, and integrity checks. In some embodiments, de-identification or pseudonymization is performed for secondary uses (e.g., model training, validation, or quality assurance) in accordance with institutional policies and applicable laws.

[0122]

[0102] Data Lineage and Provenance. In some embodiments, the system records data lineage for inputs used during training and inference, including source system identifiers, assay or laboratory identifiers, units of measure, reference-range normalization steps, and transformation logs. Such provenance metadata may be stored with predictions and model versions to enable traceability, quality assurance, and reproducibility across ALZAI-002-PCT

[0123] 2025-10-10 deployments and over time.

[0124]

[0103] Monitoring, Model Management, and Drift Detection. In some embodiments, the system includes facilities for model versioning, deployment, and rollback; periodic performance monitoring using prospectively collected ground truth; calibration checks; data and concept drift detection; and retraining or fine-tuning workflows. Administrative interfaces may expose telemetry (e.g., latency, throughput, error rates), performance metrics (e.g., AUC-ROC, AUPRC, calibration), and audit trails for predictions, alerts, and model updates. In some embodiments, monitoring includes temporal and site-level stratifications to assess generalizability and equity across cohorts. In some embodiments, prediction outcomes and subsequent real-world clinical results are logged and used as feedback signals for continuous model retraining, enabling adaptive improvement in prediction accuracy and treatment optimization over time.

[0125] A. Exemplary Software and Software / Hardware Implementations

[0126]

[0104] In some embodiments, the disclosed systems comprise one or more computing devices, servers, or distributed computing nodes configured to perform the data ingestion, model execution, and output generation steps described herein. The system architecture may include both hardware and software components, integrated to enable efficient, real-time prediction of medical-condition onset or progression. A disclosed method or one or more steps of a disclosed method may be implemented in an application, such as a computer application. Methods or steps of methods implemented in an application may be implemented as one or more modules, such as software modules, which for example may comprise code executable on a computer system, and such code may be stored on computer-readable media, such as non-transitory media.

[0127]

[0105] In some exemplary applications, some embodiments of the disclosed methods or steps of such methods therefore are implemented in software, such as in software instructions executable by a computer processor, for example as an application, which may be web- or browser-based, on a smartphone, or on any computer system disclosed or known in the art. In some exemplary applications, a computer application is provided that performs one or more steps of the disclosed methods. Some such exemplary applications, and certain aspects and features of those applications, are described below. These applications are exemplary only, and numerous other such applications, comprising these and other aspects and features, and which may comprise more or less such aspects or features, as well as may comprise one or more modifications to any such aspects or features, and may be combined with any one or more aspects or features of any other disclosed embodiments, will be readily appreciated in view of these teachings and the ordinary skill in the art.

[0128]

[0106] In some exemplary applications, one or more embodiments of the disclosed methods, or one or more steps of such methods, are implemented in a combination of software and hardware, such as described herein or known in the art. For example, one or more steps or stages of any method, process, or algorithm described or illustrated herein (which for convenience and as shorthand may be simply “steps” of a disclosed “method”) may be embodied directly in hardware, in software such as executed by a processor, or in a combination of both hardware and software and / or other components. ALZAI-002-PCT

[0129] 2025-10-10

[0130]

[0107] To help illustrate the interchangeability of hardware and software, certain components, blocks, modules, means, steps, and the like may be disclosed generally in terms of their functionality. How such functionality is implemented as hardware and / or software may depend upon the particular use and any design constraints, which will be understood by one in the art in view of the disclosure. Those in the art will understand how to implement such functionality in various ways. For example, computer code may comprise one or more modules executing one or more processes to provide a useful result, and the modules may communicate with one another via means known in the art.

[0131]

[0108] In some embodiments, one or more Application Programming Interfaces (APIs) are used, such as to allow an API-calling program code component or hardware component to access and use one or more functions, methods, procedures, data structures, classes, and / or other services provided by the API- implementing component. An API can define one or more parameters passed between the components. See generally, e.g., U.S. Pat. No. 10,852,912 (Apple Inc.), incorporated by reference as if fully set forth herein.

[0132]

[0109] In some embodiments, one or more steps of a disclosed method are implemented in a computer application, such as in a software application, including in one or more modules, such as in one or more software modules. In some embodiments, an exemplary disclosed system comprises multiple software modules or software / hardware modules. One will understand that, although software modules run on hardware, reference may be made solely to the software for purposes of its illustration, but its execution on hardware will be well known to those of skill. In some embodiments, software, hardware, and / or combinations thereof are configured to implement one or more functions associated with the disclosed methods, including executing ML models for training, inference, and updating.

[0133] B. Exemplary Data-Processing and other Functional System Modules

[0134]

[0110] The disclosed systems comprise one or more software, hardware, and / or software / hardware components configured to execute the steps of the disclosed methods, such as the predictive modeling, data-processing, and treatment-recommendation operations described herein, and implemented as described.

[0135]

[0111] In some embodiments, the system includes one or more of the following functional modules.

[0136]

[0112] In some embodiments, the disclosed systems comprise one or more computing devices, servers, or distributed computing nodes configured to perform the data ingestion, model execution, and output generation steps described herein. The system architecture may include both hardware and software components, integrated to enable efficient, real-time prediction of medical-condition onset or progression.

[0137]

[0113] In some embodiments, the system includes a data-ingestion module configured to receive, parse, and standardize biomedical data from multiple sources, such as electronic medical records (EMR), laboratory information systems, wearable devices, and remote monitoring sensors. The data-ingestion module may perform temporal alignment, deduplication, and validation of received records, and may write normalized data into a feature database or vector store optimized for low-latency retrieval.

[0138]

[0114] In some embodiments, a feature-processing module transforms the normalized data into standardized ALZAI-002-PCT 2025-10-10 feature vectors, computing statistical and temporal descriptors such as mean, variance, slope, extrema, and earliest-versus-latest difference for each parameter. The feature-processing module may include one or more submodules for data imputation, categorical encoding, and feature scaling. In embodiments, feature transformations and aggregation functions are implemented in deterministic pipelines to ensure reproducibility across retraining cycles.

[0139]

[0115] In some embodiments, a predictive model execution engine applies a trained machine-learning model to the feature vectors to compute one or more output scores, such as predicted risk, probability of onset, or rate of progression for one or more medical conditions. The execution engine may support multiple model architectures, including gradient-boosted ensembles, neural networks, or hybrid frameworks. In embodiments, inference computations are automatically distributed across local and cloud nodes, depending on network latency and processor availability.

[0140]

[0116] In some embodiments, a policy engine receives the model outputs and associated feature attributions, and executes a multi-objective optimization process balancing predicted therapeutic efficacy, toxicity, and resource utilization. The policy engine may generate treatment recommendations, triage priorities, or preventive-care schedules based on the predicted risk score and configurable clinical thresholds. In certain embodiments, the policy engine implements causal-inference or uncertainty-estimation modules that compute confidence bounds or alternative outcome scenarios.

[0141]

[0117] In some embodiments, an orchestration layer manages inter-module communication and task allocation across a distributed computing environment. The orchestration layer may monitor hardware utilization, manage job queues, and provide cryptographically signed logs of data access, model versioning, and decision outputs, thereby ensuring traceability and auditability. In some embodiments, the orchestration layer coordinates secure federated-learning sessions that update the predictive model using distributed datasets without centralizing raw patient data.

[0142]

[0118] In some embodiments, the system includes one or more user interfaces or visualization dashboards that render model outputs in interpretable form. Such interfaces may display feature importance rankings, counterfactual “what-if” scenarios, confidence intervals, and longitudinal biomarker trajectories. In embodiments, the interface is integrated with EMR systems via standardized APIs (e.g., HL7 FHIR) and is configured to update risk assessments dynamically as new data become available.

[0143]

[0119] In some embodiments, the system operates within a secured computing environment, optionally including encryption of data in transit and at rest, access control, and audit logging to ensure compliance with healthcare data privacy regulations (e.g., HIPAA, GDPR). The system may further include an application programming interface (API) enabling external systems, such as laboratory instruments, infusion pumps, or apheresis devices, to receive configuration or treatment parameters based on model output.

[0144]

[0120] In some embodiments, two or more such components collectively form a modular and extensible computational framework that enables automated ingestion of biomedical data, predictive risk modeling, and ALZAI-002-PCT 2025-10-10 generation of actionable treatment recommendations. The architecture provides a concrete technological improvement to computer-implemented healthcare systems by enabling interpretable, real-time clinical decision support based on dynamic data streams.

[0145] C. Exemplary Machine Learning (ML) Algorithms and ML Models

[0146]

[0121] In some embodiments, software, hardware, and / or combinations thereof are configured to execute ML algorithms and / or ML models used by the disclosed predictive system during training and inference.

[0147]

[0122] Herein, without being bound by theory, and with such terms also known by their context, the term “ML algorithm” refers to a method or process for training an ML model, and the term “ML model” refers to a specific instantiation of the ML algorithm based on the input data it has learned from; i.e., after an ML algorithm is trained on data, it becomes an ML model. In some embodiments, the functionality of one or more ML algorithms or ML models is implemented in software, hardware, or a hardware / software combination, including the training, inference, and updating of ML models.

[0148]

[0123] In general, types of ML algorithms and ML models suitable for use in the disclosed methods, such as in the disclosed predictive models, include tree-based models, linear models, support vector machines (SVMs), neural networks, probabilistic models, instance-based learning models, ensemble methods, clustering algorithms, dimensionality reduction techniques, and reinforcement learning models, as well as any combination thereof.

[0149]

[0124] In some embodiments, a method comprises training a predictive model using an ML algorithm. In some embodiments, a method comprises training an ML model using an ML algorithm.

[0150]

[0125] In some embodiments, the ML algorithm used to train and / or ML model used in the disclosed predictive models include, as examples, decision trees, random forests, gradient boosting (e.g., XGBoost) machines, support vector machines, neural networks (for instance, convolutional neural networks and deep learning neural networks), and combinations thereof.

[0151]

[0126] In some embodiments, the ML algorithm used to train and / or ML model used in the predictive model is a tree-based model. In some embodiments, the tree-based model is a decision tree, random forest, gradient boosting machine (e.g., XGBoost, LightGBM, CatBoost), or extra trees.

[0152]

[0127] In some embodiments, the ML algorithm used to train and / or ML model used in the predictive model is a linear model. In some embodiments, the linear model is a linear regression, logistic regression, ridge regression, lasso regression, or elastic net.

[0153]

[0128] In some embodiments, the ML algorithm used to train and / or ML model used in the predictive model is a support vector machine (SVM). In some embodiments, the support vector machine is a linear support vector machine or non-linear support vector machine (e.g., using kernel methods).

[0154]

[0129] In embodiments, the ML algorithm used to train and / or ML model used in the predictive model is a neural network. In embodiments, the neural network is a feedforward neural network (fully connected network), convolutional neural network (CNN), recurrent neural network (RNN), long short-term memory (LSTM) ALZAI-002-PCT 2025-10-10 network, deep learning neural network (e.g., multi-layer perceptrons), or transformer (e.g., BERT, GPT).

[0155]

[0130] In some embodiments, the ML algorithm used to train and / or ML model used in the predictive model is a probabilistic model. In some embodiments, the probabilistic model is a Naive Bayes classifier, hidden Markov model (HMM), Gaussian mixture model (GMM), or Bayesian network.

[0156]

[0131] In some embodiments, the ML algorithm used to train and / or ML model used in the predictive model is an instance-based learning model. In some embodiments, the instance-based learning model is a k-nearest neighbors (KNN) model or locally weighted learning (LWL) model.

[0157]

[0132] In some embodiments, the ML algorithm used to train and / or ML model used in the predictive model is an ensemble method. In some embodiments, the ensemble method is bagging (bootstrap aggregating), boosting (e.g., AdaBoost, gradient boosting, XGBoost), stacking (stacked generalization), or a voting classifier.

[0158]

[0133] In some embodiments, the ML algorithm used to train and / or ML model used in the predictive model is a clustering algorithm. In some embodiments, the clustering algorithm is k-means, DBSCAN (density-based spatial clustering of applications with noise), hierarchical clustering, or mean shift.

[0159]

[0134] In some embodiments, the ML algorithm used to train and / or ML model used in the predictive model is a dimensionality reduction technique. In some embodiments, the dimensionality reduction technique is principal component analysis (PCA), linear discriminant analysis (LDA), t-distributed stochastic neighbor embedding (t-SNE), or singular value decomposition (SVD).

[0160]

[0135] In some embodiments, the ML algorithm used to train and / or ML model used in the predictive model is a reinforcement learning model. In some embodiments, the reinforcement learning model is Q-learning, a deep Q-network (DQN), a policy gradient method, or an actor-critic model.

[0161]

[0136] Although some embodiments herein are described with reference to a specific ML algorithm and / or ML model, the same embodiments but using one or more other ML algorithms and / or ML models should be considered equivalents, where such other ML algorithm(s) and / or ML model(s) are recognized by those of skill as also suitable for use in the embodiments, in view of the teachings herein, and without undue experimentation. Using a combination of ML algorithms or ML models, in place of a single ML algorithm or ML model, also will be recognized in some disclosed embodiments as an equivalent. It will be appreciated that the choice of ML algorithms and ML models, in embodiments, is exemplary and not limiting.

[0162] D. Implementation Embodiments Supporting Technological Improvements

[0163]

[0137] In some embodiments, the disclosed system comprises a predictive model execution engine configured to operate within a distributed computing environment. The engine transforms heterogeneous biomedical data, such as laboratory values, diagnostic codes, and physiological time series, into normalized feature vectors using a multimodal preprocessing module. In some embodiments, multimodal feature fusion is performed by concatenating or attention-weighting representations derived from heterogeneous data sources, such as laboratory results, wearable device signals, and clinical notes embeddings, thereby enabling cross-domain learning across biomedical and behavioral modalities. ALZAI-002-PCT

[0164] 2025-10-10

[0165]

[0138] In embodiments, the system includes an adaptive data-ingestion pipeline that standardizes, indexes, and stores data in a vectorized schema optimized for low-latency retrieval. The predictive model execution engine may output a risk or progression score that is continuously updated as new data are received through the ingestion pipeline, thereby enabling real-time adjustment of predicted outcomes and corresponding treatment recommendations. A continuous learning configuration allows the system to provide up to date predictions without manual data entry or model retraining, improving accuracy and clinical responsiveness.

[0166]

[0139] In some embodiments, a policy engine receives risk predictions and associated feature attributions and generates an actionable treatment recommendation by executing a multi-objective optimization process balancing efficacy, toxicity, and resource availability. The policy engine may implement counterfactual simulation, uncertainty estimation, or causal-inference modules, each configured to generate confidence bounds or alternate-outcome predictions used to refine treatment recommendations. In some embodiments, the policy engine further integrates feedback from observed treatment outcomes, enabling dynamic re-optimization of therapeutic recommendations as new evidence accumulates. The system may simulate alternate outcomes under different pharmacological or procedural options, compute expected utility scores, and automatically update the recommended intervention when predicted benefit exceeds a threshold improvement over standard of care.

[0167]

[0140] In some embodiments, the system includes an orchestration layer that dynamically allocates computation between cloud and edge nodes, with inference performed locally when network latency or bandwidth constraints exceed a model-defined threshold. In some embodiments, the system includes a federated learning module enabling distributed model retraining across multiple data custodians using secure aggregation protocols without centralizing patient-level data. Such distributed and federated configurations can preserve patient data privacy while maintaining real-time inference capability in clinical environments, thereby improving the security, speed, and scalability of computer-implemented healthcare analytics compared with conventional centralized architectures.

[0168]

[0141] In some embodiments, the system interfaces with external devices such as apheresis machines, infusion pumps, or wearable biosensors through standards-compliant APIs. The system transmits configuration parameters, monitors sensor feedback, and adjusts device operation in real time based on predicted physiological tolerance and biomarker response. This closed-loop integration allows the disclosed system to implement predictive-treatment workflows, wherein model outputs directly inform device settings or treatment delivery parameters, improving safety and therapeutic precision through continuous monitoring and adjustment. The orchestration layer further maintains cryptographically signed logs of model version, input features, and decision outputs to ensure deterministic re-execution and auditability.

[0169]

[0142] In some embodiments, the user interface includes explainability widgets rendering top-ranked features, counterfactual “what-if’ scenarios, and dynamic confidence intervals derived from the predictive model output. In some embodiments, the user interface further allows clinicians to explore how changes in ALZAI-002-PCT 2025-10-10 patient-specific variables, such as laboratory values or medication adherence, would alter predicted outcomes, providing interactive feedback on potential interventions.

[0170]

[0143] In some embodiments, the user interface is integrated with EMR systems and clinical dashboards, displaying individualized risk trajectories, treatment recommendations, and uncertainty visualizations. The interface may generate clinician alerts or scheduling recommendations when the predicted risk exceeds a configurable threshold. In some embodiments, the interface is linked to the policy engine, enabling a user to accept, modify, or reject a recommended treatment and automatically update the model state or treatment schedule based on clinician input. In embodiments, this bidirectional interface enables dynamic collaboration between human and machine decision-making, reducing cognitive load while maintaining clinical oversight. These visualization modules can improve interpretability and enable clinicians to evaluate treatment recommendations with transparent model reasoning.

[0171] E. Exemplary Model Configurations and Training Examples

[0172]

[0144] In some embodiments, the disclosed system executes on a distributed computing infrastructure comprising one or more processors, memory devices, and communication interfaces configured to process biomedical data, perform model inference, and output results to clinical-decision-support modules. In some embodiments, the system architecture includes cloud-based inference servers, local edge nodes, or containerized microservices that orchestrate data ingestion, preprocessing, and model execution. The computing environment may further comprise secure storage subsystems and audit-logging modules configured to maintain reproducibility and traceability of each model prediction.

[0173]

[0145] To further illustrate some embodiments of the disclosed methods and systems, the following exemplary configurations demonstrate representative input features, training data structures, and hyperparameter ranges that can be employed to achieve the described predictive performance. Other such exemplary configurations will be readily appreciated by those of skill in view of the teachings herein.

[0174]

[0146] In some embodiments, the predictive model comprises a neural network architecture, such as a multilayer perceptron, convolutional network, transformer, or recurrent network trained on time-series data.

[0175]

[0147] In some embodiments, ensemble or hybrid architectures combine tree-based and deep-learning components to optimize interpretability and temporal sensitivity.

[0176]

[0148] In a first exemplary configuration, the disclosed predictive modeling framework is implemented using a gradient-boosted decision tree ensemble (e.g., XGBoost) trained on a longitudinal clinical dataset derived from de-identified electronic medical records (EMR) spanning at least 10 years of patient data.

[0177]

[0149] The dataset may include both case and control subjects, where cases are defined as individuals diagnosed with the target condition during the observation window, and controls are individuals without such diagnosis during the same or a longer follow-up period. In some embodiments, data are partitioned into training, validation, and test sets using stratified sampling to preserve class balance and temporal independence between cohorts. In some embodiments, control subjects are selected from individuals without ALZAI-002-PCT 2025-10-10 diagnosis of the target condition throughout the observation and prediction horizons. Historical data for controls are aligned temporally to diagnosis-matched anchors (e.g., last encounter date) to ensure equivalent feature distributions and temporal windows.

[0178]

[0150] In some embodiments, model inputs include numerical, categorical, and time-varying features such as standardized blood test values (e.g., hemoglobin A1c, fasting glucose, lipid profile, creatinine, liver enzymes), demographic and anthropometric variables (e.g., age, sex, body-mass index), and derived temporal features (e.g., mean, standard deviation, slope, and earliest-versus-latest delta over time). In some embodiments, additional features include medication history indicators, diagnostic code embeddings, and comorbidity indices derived from longitudinal health records. Missing values may be imputed using model-based or distributional imputation, and continuous features may be normalized or quantile-transformed prior to training.

[0179]

[0151] In embodiments, representative model hyperparameters include a maximum tree depth of between 4 and 8, a learning rate (eta) between 0.01 and 0.1 , a subsample ratio between 0.6 and 0.9, and L1 / L2 regularization coefficients between 0.1 and 1.0. The model may be trained for between 200 and 1000 boosting rounds with early stopping based on validation-set performance (e.g., area under the receiver operating characteristic curve, AUROC). In embodiments, hyperparameters are optimized using Bayesian or grid-based search procedures to maximize predictive accuracy subject to interpretability and runtime constraints.

[0180]

[0152] In some embodiments, feature importance values are computed using SHAP (SHapley Additive exPlanations) or gain-based ranking to generate interpretable attributions. Herein, “feature importance” refers to a quantitative measure of each input variable’s contribution to the model output, as determined by model-specific attribution algorithms. The resulting model may achieve predictive accuracy exceeding 0.85 AUROC on held-out validation data while using only routinely collected laboratory and demographic features, demonstrating that high-fidelity prediction can be achieved with minimal data dimensionality and without specialized imaging or genomic modalities.

[0181]

[0153] In some embodiments, the trained model is stored as a serialized artifact containing the learned parameters, feature schema, and preprocessing pipeline metadata, enabling deterministic re-execution and integration into the clinical-decision-support infrastructure described herein.

[0182]

[0154] In some embodiments, the predictive model is periodically retrained or recalibrated using newly acquired clinical data to preserve predictive performance over time and across populations. The retraining process may employ online learning, incremental parameter updates, or transfer-learning techniques to compensate for data drift or shifts in population characteristics. In some embodiments, the retraining pipeline is automated within the distributed computing environment, performing continuous performance monitoring and triggering model updates when validation metrics fall below defined thresholds.

[0183]

[0155] In a second exemplary configuration, the disclosed methods and systems are implemented as follows, which illustrates further representative data-processing and model-training workflows applicable to prediction of various medical conditions, with implementation details applicable across multiple medical domains. ALZAI-002-PCT

[0184] 2025-10-10

[0185]

[0156] In some embodiments, the system receives multimodal input data comprising structured clinical information (e.g., demographic variables, medication history, and laboratory results) and temporal biosignal data (e.g., audio recordings, movement traces, or wearable-sensor outputs) collected from a plurality of subjects. Each record is linked to a confirmed diagnostic label or longitudinal clinical outcome.

[0186]

[0157] In some embodiments, data preprocessing includes quality-control filtering, normalization, and feature transformation. In embodiments for time-series modalities, the system computes spectral and statistical representations such as short-time Fourier transforms, mean and variance coefficients, slope and delta features, and temporally binned aggregates. Structured clinical variables can be one-hot encoded or z-score normalized, and all features are concatenated into unified feature vectors stored in a vector database indexed by subject identifier and timestamp.

[0187]

[0158] In embodiments, a predictive model is trained using supervised learning on the labeled dataset. The architecture may include a feature-embedding layer, one or more attention or recurrent modules configured to capture temporal dependencies, and a classification head outputting a probability or confidence score of disease presence or progression. Model hyperparameters may be optimized via grid or Bayesian search, or equivalent parameter-tuning methods over ranges that may include a learning rate (e.g., 1 xi o4— 5x103), batch size (e.g., 16-128), regularization strength (e.g., dropout 0.2-0.5), and optimizer type (e.g., Adam, RMSprop).

[0188]

[0159] In embodiments, model performance is evaluated using stratified cross-validation and held-out test sets, with metrics such as area under the receiver-operating-characteristic curve (AUG), precision-recall AUG, accuracy, and F1 -score. In embodiments, feature-importance or attribution analyses (e.g., SHAP, integrated gradients) identify the input variables contributing most strongly to prediction, improving model interpretability.

[0189]

[0160] In some embodiments, the trained model is deployed within a clinical decision framework to generate, for each subject, a personalized risk or progression score together with ranked contributing features. This exemplary computational pipeline, including multimodal data normalization, temporal-feature extraction, and interpretable machine learning inference, will be applicable to prediction of medical conditions disclosed herein, such as metabolic, cardiovascular, inflammatory, endocrine, and other disorders.

[0190] F. Exemplary Computer-Readable Media

[0191]

[0161] In some aspects are provided systems comprising computer storage products such as computer-readable media. In embodiments, a software module resides on a computer-readable medium. Herein, reference to “a computer-readable medium” or to “computer-readable media” should be understood also to refer to the other, unless context demands otherwise, and computer code, such as a software module, generally will be appreciated to be interchangeably storable on “a medium” or “media,” and the terms are thus equivalents, unless reference to one or the other is necessary to give technical effect to a term or embodiment.

[0192]

[0162] In some embodiments, a computer-readable medium has computer code thereon for performing various computer-implemented operations, such as operations related to one or more functions or steps associated with the disclosed methods. The computer-readable medium and the computer code thereon may ALZAI-002-PCT 2025-10-10 be designed for the purposes of the disclosure, may be of any kind known to those in the art, or may be a combination thereof, such as depending on the embodiment.

[0193]

[0163] Examples of computer-readable media include solid-state drives (SSDs), magnetic media such as hard disk drives (HDDs); removable disk storage including optical media such as CD-ROMs, DVDs, and holographic devices; magneto-optical media; RAM memory; flash memory; ROM memory; EPROM or EEPROM memory; and registers. Other examples of computer-readable media include hardware devices configured to store and / or execute program code, such as programmable microcontrollers, ASICs, and programmable logic devices (PLDs).

[0194]

[0164] Generally speaking, computer-readable media includes any tangible or non-transitory storage media or memory media such as electronic, magnetic, or optical media, including any one or more of the exemplary media described herein or known in the art, and which may be coupled to a computer system via a bus.

[0195]

[0165] Herein, the terms “tangible” and “non-transitory” describe a computer-readable medium excluding propagating electromagnetic signals, but unless context dictates otherwise such terms are not intended to limit the type of physical medium. For instance, the terms “non-transitory computer-readable medium” and “tangible memory” are intended to encompass types of storage devices that do not necessarily store information permanently, such as random access memory (RAM).

[0196]

[0166] Program instructions and / or data stored in tangible memory in non-transitory form may be transmitted by transmission media or signals such as electrical, electromagnetic, or digital signals, and also may be conveyed via a communication medium such as a network and / or a wireless link. The terms “tangible” and “non-transitory” thus should be understood as a limitation of the medium itself (i.e., that it is tangible, and not merely a signal), but not a limitation on the volatility or persistency of data stored or storable on the medium (e.g., RAM vs. ROM), unless such limitation is expressly made or demanded by the context.

[0197]

[0167] In some embodiments, a computer-readable medium is coupled to a processor so the processor can read information therefrom, and write information thereto. In other embodiments, a computer-readable medium is integral to a processor, for example where the computer-readable medium and the processor reside in an ASIC, which may reside in a user terminal. A processor and a computer-readable medium also may reside as discrete components in a user device, such as a user terminal. A “user terminal” includes any user device suitable for use in disclosed methods and systems, such as the examples described herein.

[0198] G. Exemplary Executable Instructions

[0199]

[0168] In some aspects are provided systems comprising computer-readable media, such as non-transitory media, storing executable instructions. In some embodiments, executable instructions are computer-executable instructions, machine-executable instructions, and / or processor-executable instructions.

[0200]

[0169] In some embodiments, executable instructions comprise computer code. Examples of computer code include machine code, such as produced by a compiler or other machine code generation mechanisms, scripting programs, PostScript programs, and / or other code or files containing higher-level code that are ALZAI-002-PCT 2025-10-10 executed by a computer using an interpreter or other code execution mechanism. For example, embodiments may be implemented using assembly language, Java, C, C#, C++, scripting languages, or other programming languages known in the art. Embodiments may be implemented in hardwired circuitry in place of, or in combination with, machine-executable software instructions.

[0201]

[0170] Depending on the embodiment, the disclosed components, blocks, modules, means, steps, stages, and the like, where implemented as executable instructions, such as processor-executable instructions, may be executed by any suitable processor known to those in the art. In some embodiments, a suitable processor may include a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof that is designed to perform or capable of performing the function(s) embodied in the executable instructions.

[0202]

[0171] In some embodiments, a suitable processor may include a general purpose processor, such as a microprocessor or any conventional processor, controller, microcontroller, or state machine. A suitable processor also may be implemented as a combination of computing devices, e.g., as a combination of microprocessors, of one or more microprocessors and a DSP or DSP core, and other such configurations. A suitable processor also may be one or more communication processors or other such processors designed to implement functionality in communication devices or other mobile or portable devices such as smartphones.

[0203]

[0172] Besides being embodied in software such as processor-executable instructions, one or more steps or stages of any disclosed method, process, or algorithm also may be embodied directly in hardware, or in a combination of both hardware and software and / or other components.

[0204]

[0173] In some embodiments, one or more steps or stages of any disclosed method, process, predictive model, or algorithm is embodied in machine-executable instructions, which are performed by a machine. In some embodiments, the machine is a computer system. In embodiments, the computer system, or other machine, may have a set of instructions, such as computer-executable instructions, or other machine-executable instructions, for causing the computer or the machine to perform or execute any one or more of the steps of a disclosed method, according to one or more embodiments.

[0205] H. Exemplary Computer Systems

[0206]

[0174] In some aspects are provided systems comprising a computer system. A computer system for shorthand may be referred to as a “computer,” and in embodiments is a user device, such as a user terminal.

[0207]

[0175] In an example, a computer system may include a processor, memory, non-volatile memory, and an interface device. Other common components (e.g., cache memory) will be appreciated by one of skill. The components of the computer system may be coupled together via a bus or another such known device.

[0208]

[0176] The disclosed methods and systems can be implemented in a variety of different computer systems, including general purpose computer systems, special purpose computer systems, hybrids of general purpose and special purpose computer systems, and other suitable data processing systems and devices. ALZAI-002-PCT

[0209] 2025-10-10

[0210]

[0177] A computer system may take any suitable physical form. By way of example and not limitation, the computer system may be an embedded computer system, a system-on-chip (SOC), a single-board computer system (SBC) (e.g., a computer-on-module (COM) or system-on-module (SOM)), a desktop computer system, a laptop or notebook computer system, an interactive kiosk, a cloud-based system, a mainframe, a smart TV, a tablet, a mobile phone or smartphone, a VR, AR, XR, MR, or other headset or device, a personal digital assistant (PDA), a wearable device (e.g., a smartwatch, fitness tracker, “smart sunglasses,” or other “wearables”), a server, or a combination of two or more of these, such as a mesh of computer systems.

[0211]

[0178] In some embodiments, a “computer system” may comprise one or more computer systems; may be unitary or distributed; may span multiple locations; may span multiple machines; or may reside in a cloud, which may include one or more cloud components in one or more networks. In embodiments, one or more computer systems may perform without substantial spatial or temporal limitation one or more steps of one or more disclosed methods. In embodiments, one or more computer systems may perform in real time or in batch mode one or more steps of one or more disclosed methods. In embodiments, one or more computer systems may perform at different times or at different locations one or more steps of one or more disclosed methods.

[0212]

[0179] It will be readily appreciated that additional software and hardware modules not described herein may be necessary for implementing disclosed methods and systems; however, such omitted modules will be well known to those in the art. For example, every computer system discussed herein requires a power supply module, though such mention was omitted.

[0213]

[0180] In some embodiments, the system supports horizontal and / or vertical scaling to accommodate variable workload, including autoscaling of model-serving instances, load balancing across replicas, and queue-based ingestion for batch or streaming data. High-availability configurations may include redundant instances across failure domains, health checks, rolling updates, and automated failover procedures. In some embodiments, data persistence and fault-tolerance mechanisms, such as distributed storage replication or checkpointing of intermediate states, are implemented to maintain operational continuity during component failure. In further embodiments, the deployment architecture may be orchestrated within cloud, hybrid, or edge environments to ensure portability and resource efficiency.

[0214] / . Exemplary Networking Means

[0215]

[0181] In embodiments, information from one computer system is transmitted to another computer system.

[0216]

[0182] For example, information from a user computer system may be transmitted to another computer system, such as where a user provides input to a browser-based application on a user laptop, and the like, as will be readily appreciated. Where information is transmitted from one computer system to another computer system, any communications interface as will be known in the art may be used, such as wireless networks (e.g., WiFi), local area networks (e.g., a building LAN), wide area networks (e.g., the internet, a cellular network, etc.), and / or by direct communication (e.g., NFC, Bluetooth, etc.). Communications interfaces can be configured to conduct wired and / or wireless communications. ALZAI-002-PCT

[0217] 2025-10-10

[0218]

[0183] Where information is transmitted from a user computer system to another computer system, such information in some embodiments is encrypted, for example using end-to-end encryption. In some embodiments, at least a portion of such information is encrypted, depending on data type or security policy.

[0219] J. Exemplary Input and Output Means

[0220]

[0184] In some aspects are provided methods comprising one or more steps where a user provides input to a computer system. Where a user provides input to a computer system (which will be understood to also include the software thereon), such as in a step or stage of a disclosed method, it will be understood that such input may be provided using any suitable input device known in the art.

[0221]

[0185] Examples of input devices include a physical keyboard, a virtual (e.g., touchscreen-based) keyboard, a touchscreen (which also may be the output device, e.g., a smartphone or tablet screen, a laptop screen), a keypad, a touchpad, a mouse, a joystick, and any other such device performing a similar function.

[0222]

[0186] In some embodiments, input comprises voice input, which for example may be provided using a microphone or any other auditory input device, e.g., by voice commands interpreted by voice recognition software (e.g., resident on the computer), or a voice recognition module (e.g., as part of an application that also performs other steps of a disclosed method).

[0223]

[0187] In some embodiments, input provided by a user to a computer system is processed in real-time by the computer system, or by software on the computer system performing one or more steps of a disclosed method. In some embodiments, input provided by a user to a computer system is subject to asynchronous or subsequent processing by the computer system, such as where a text message is sent by a user and stored for a period of time before processing.

[0224]

[0188] From the perspective of an individual computer system, one will appreciate that input may be provided to that computer system (including software thereon) indirectly, e.g., a user may provide input which is first provided to one or more other computer systems or software programs before it is provided to the computer system(s) that perform the step(s) or stage(s) of the method that use such input. It will be appreciated that input may pass through multiple such computer systems or software programs before being provided.

[0225]

[0189] Where output is provided by a computer system (which will be understood to also include the software thereon), such as in a step of a disclosed method, for example to a user, it will be understood that such output may be provided using any suitable output device known in the art, such as a monitor, a TV, a smartphone screen, a tablet screen, or any other such display device, as well as any other output device, such as a printer, performing a similar function. Output may comprise auditory output, such as provided by a speaker or other auditory output device. Output also may be provided by a computer system to a user for asynchronous or subsequent use thereby, such as where a text message is sent to a user, and stored within an app, where it is not read by the user until the app is subsequently opened. In embodiments, a user may receive a notification that it received an output, but not immediately receive the output, such as by choosing not to open the app.

[0226]

[0190] In one exemplary embodiment, output comprises outcomes of the predictive model(s), in accordance ALZAI-002-PCT 2025-10-10 with some disclosed embodiments, and which is provided to an exemplary user interface (e.g., GUI) presenting such outcomes. For example, the GUI may present the risk scores outputted by one or more predictive model(s) for onset of a medical condition, for example, for different time frames (as shown, risk for years 1, 5, and 10) and / or for different stages of the medical condition (e.g., progression), as described herein. The GUI may present risk scores for the onset of one or more risk factors likely to trigger the onset of the medical condition, such as described herein. The risk scores may be converted into binary values, indicating whether there is a risk or no risk, such as whether the risk score is above a threshold or below a threshold. The GUI may present other data, e.g., any one or more of data associated with the subject, a confidence level associated with a risk score, a change from a previous prediction, and time since a previous prediction.

[0227]

[0191] It will be readily appreciated that, as with input, output also may be provided by a computer system (including software thereon) indirectly, e.g., the computer system may prepare one or more outputs which are provided to one or more other computer systems or software programs before they are provided to a user, according to the step(s) of the disclosed method that involve such output. It will be appreciated that output may pass through multiple such computer systems or software programs before being provided.

[0228] V. Exemplary Medical Conditions Predicted by the Disclosed Methods

[0229]

[0192] In some aspects, the disclosed methods and systems enable non-invasive, low-cost, and highly accurate prediction of the onset and / or progression of a medical condition.

[0230]

[0193] Herein, “condition” and “medical condition” are used interchangeably and refer broadly to any disease, disorder, dysfunction, syndrome, or pathological state for which prediction of onset, progression, or exacerbation using the disclosed methods and systems provides clinical or research value.

[0231]

[0194] Unless otherwise specified, reference to a medical condition should be understood to include conditions defined according to the International Classification of Diseases, 11th Revision (ICD-11), as well as any equivalent or substantially similar conditions described in the International Classification of Diseases, 10th Revision (ICD-10) or similar accepted clinical criteria, including for non-human animals.

[0232]

[0195] The disclosed methods can be implemented to predict the onset and / or progression of a broad range of medical conditions, including but not limited to those exemplified in the categories below. It will be appreciated that individual conditions may fit into multiple categories; however, they are grouped here by predominant physiological or etiological characteristics for convenience and clarity of description.

[0233]

[0196] Cancers: In some embodiments, the disclosed methods and systems predict the onset and / or progression of cancer or neoplastic disease. Exemplary cancers include carcinomas (e.g., breast, lung, prostate, and colorectal cancers), sarcomas (e.g., osteosarcoma, soft-tissue sarcoma), leukemias (e.g., acute myeloid leukemia, chronic lymphocytic leukemia), lymphomas (e.g., Hodgkin and non-Hodgkin lymphomas), melanomas (e.g., cutaneous melanoma), and central nervous system malignancies (e.g., glioblastoma, astrocytoma). In some embodiments, the predictive model evaluates longitudinal laboratory, imaging, and EMR-derived data to identify early signatures of oncogenesis, tumor progression, or recurrence risk. ALZAI-002-PCT

[0234] 2025-10-10

[0235]

[0197] Cardiovascular Diseases In some embodiments, the disclosed methods and systems predict the onset and / or progression of a cardiovascular disease. Exemplary cardiovascular diseases include ischemic heart disease (IHD), myocardial infarction (Ml), angina, coronary artery disease (CAD), hypertension, congestive heart failure (CHF), arrhythmias (e.g., atrial fibrillation), peripheral artery disease (PAD), cerebrovascular disease (e.g., stroke, transient ischemic attack (TIA)), aortic aneurysm, and deep vein thrombosis (DVT). In some embodiments, predictive features include combinations of laboratory, physiological, and behavioral data associated with vascular integrity, cardiac strain, and systemic metabolic state.

[0236]

[0198] Circulatory System Diseases: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a circulatory system disease. Exemplary circulatory system diseases include varicose veins, chronic venous insufficiency, Raynaud’s disease, lymphedema, vasculitis, and thromboangiitis obliterans (Buerger’s disease), as well as other circulatory conditions such as arteriovenous malformations and vascular dementia. In some embodiments, predictive features are derived from longitudinal hemodynamic, vascular-imaging, or peripheral sensor data capturing perfusion dynamics, vessel compliance, and microcirculatory integrity.

[0237]

[0199] Chronic Respiratory Disorders: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a chronic respiratory disorder. Exemplary chronic respiratory disorders include asthma, chronic obstructive pulmonary disease (COPD), chronic bronchitis, emphysema, bronchiectasis, and interstitial lung disease (e.g., idiopathic pulmonary fibrosis), as well as other chronic respiratory conditions such as pulmonary hypertension. In some embodiments, predictive modeling incorporates physiological, environmental, and behavioral inputs to identify early risk signatures of functional decline or exacerbation.

[0238]

[0200] Musculoskeletal Disorders: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a musculoskeletal disorder. Exemplary musculoskeletal disorders include osteoarthritis, degenerative disc disease, osteoporosis, tendinitis, and bursitis, as well as other conditions such as fibromyalgia. In some embodiments, predictive features include mechanical-load parameters, activity metrics, and inflammatory or metabolic markers associated with tissue degeneration and recovery potential.

[0239]

[0201] Connective Tissue Diseases: In embodiments, the disclosed methods and systems predict the onset and / or progression of a connective tissue disease. Exemplary connective tissue diseases include systemic lupus erythematosus (SLE), antiphospholipid antibody syndrome (APLA), Sjogren’s disease, scleroderma, mixed connective tissue disease (MCTD), and Ehlers-Danlos syndrome, as well as other autoimmune-related conditions affecting connective tissues. In embodiments, predictive modeling integrates longitudinal laboratory and physiological data to detect immunologic or fibrotic trends preceding clinical manifestation.

[0240]

[0202] Skin Diseases: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a skin disease. Exemplary skin diseases include psoriasis, eczema (atopic dermatitis), contact dermatitis, seborrheic dermatitis, and acne, as well as allergic or inflammatory dermatoses. In some embodiments, predictive models identify dynamic relationships between inflammatory activity, environmental ALZAI-002-PCT

[0241] 2025-10-10 exposure, and physiological stress indicators to forecast flare onset or treatment response.

[0242]

[0203] Liver Diseases: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a liver disease. Exemplary liver diseases include non-alcoholic fatty liver disease (NAFLD), alcoholic fatty liver disease, viral hepatitis (e.g., A, B, C, D, E), and cirrhosis, as well as other hepatic conditions such as hepatocellular carcinoma and cholestasis. In some embodiments, predictive features include longitudinal metabolic and biochemical indicators reflecting hepatic function, systemic inflammation, and treatment response.

[0243]

[0204] Genitourinary Diseases: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a genitourinary disease. Exemplary genitourinary diseases include chronic kidney disease (CKD), glomerulonephritis, nephrotic syndrome, polycystic kidney disease (PKD), urinary incontinence, and urinary tract infections (UTIs), as well as other genitourinary conditions such as benign prostatic hyperplasia (BPH) and interstitial cystitis. In some embodiments, predictive features include renal function indices, electrolyte-balance parameters, and derived physiological markers capturing filtration efficiency and systemic hydration dynamics.

[0244]

[0205] Metabolic Disorders: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a metabolic disorder. Exemplary metabolic disorders include type 2 diabetes mellitus, lipid-metabolism disorders (e.g., hyperlipidemia), insulin resistance, and metabolic syndrome, as well as complex conditions such as chronic kidney disease (CKD) associated with metabolic dysfunction and other metabolism-related disorders such as obesity. In some embodiments, predictive features are derived from longitudinal biochemical, anthropometric, and behavioral data representing metabolic regulation, insulin sensitivity, and energy-balance dynamics.

[0245]

[0206] Endocrine Disorders: In some embodiments, the disclosed methods and systems predict the onset and / or progression of an endocrine disorder. Exemplary endocrine disorders include thyroid disorders (e.g., hyperthyroidism, hypothyroidism), hypogonadism, adrenal-gland disorders (e.g., Cushing’s syndrome, Addison’s disease), pituitary-gland disorders (e.g., acromegaly, pituitary adenoma), and polycystic ovary syndrome (PCOS), as well as endocrine-related conditions such as osteoporosis and obesity. In some embodiments, predictive modeling integrates hormonal, metabolic, and physiological parameters to assess functional dysregulation across endocrine axes.

[0246]

[0207] Neurological Disorders: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a neurological disorder. Exemplary neurological disorders include epilepsy, Parkinson’s disease, multiple sclerosis, migraine, and amyotrophic lateral sclerosis (ALS), as well as other conditions such as Huntington’s disease and peripheral neuropathy. In some embodiments, predictive models analyze electrophysiological, behavioral, and biochemical time-series data to identify signatures of neural degeneration, excitability, or network imbalance preceding clinical presentation.

[0247]

[0208] Chronic Inflammatory Disorders: In some embodiments, the disclosed methods and systems predict ALZAI-002-PCT 2025-10-10 the onset and / or progression of a chronic inflammatory disorder. Exemplary chronic inflammatory disorders include rheumatoid arthritis, inflammatory bowel disease (IBD) (e.g., Crohn’s disease, ulcerative colitis), psoriasis, systemic lupus erythematosus (SLE), and ankylosing spondylitis, as well as related conditions such as endometriosis and sarcoidosis. In some embodiments, predictive features include immune, metabolic, and physiological markers indicative of sustained inflammatory activation and tissue remodeling.

[0248]

[0209] Depressive Disorders: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a depressive disorder. Exemplary depressive disorders include major depressive disorder (MDD), persistent depressive disorder (dysthymia), seasonal affective disorder (SAD), premenstrual dysphoric disorder (PMDD), postpartum depression, and treatment-resistant depression (TRD), as well as depressive states secondary to chronic illness or systemic disease. In some embodiments, predictive features include physiological, behavioral, and digital-interaction data reflecting circadian rhythm, neurovegetative balance, and affective stability.

[0249]

[0210] Mental Health Disorders (Other than Depressive Disorders): In some embodiments, the disclosed methods and systems predict the onset and / or progression of a mental-health disorder other than a depressive disorder. Exemplary such disorders include neurodevelopmental disorders (e.g., attention-deficit / hyperactivity disorder (ADHD)), psychotic disorders (e.g., schizophrenia, schizoaffective disorder), catatonia, mood disorders (e.g., bipolar disorder), anxiety or fear-related disorders (e.g., generalized anxiety disorder (GAD), panic disorder, social anxiety disorder), obsessive-compulsive or related disorders (OCD), stress-related and dissociative disorders, feeding and eating disorders (e.g., anorexia nervosa, bulimia nervosa, binge-eating disorder), bodily-distress or experience disorders, substance-use or addictive-behavior disorders (e.g., alcohol-use disorder, opioid-use disorder), impulse-control and disruptive-behavior disorders, personality disorders (e.g., borderline personality disorder (BPD)), paraphilic disorders, factitious disorders, and neurocognitive disorders. In some embodiments, predictive models incorporate multimodal data from physiological sensors, speech or behavioral analytics, and self-reported metrics to forecast relapse, symptom worsening, or treatment response.

[0250]

[0211] Sleep-Wake Disorders: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a sleep-wake disorder. Exemplary sleep-wake disorders include insomnia, hypersomnolence disorder, narcolepsy, obstructive sleep apnea, and circadian rhythm sleep-wake disorders (e.g., delayed sleep phase disorder), as well as sleep-related movement disorders (e.g., restless legs syndrome) and sleep-related breathing disorders such as primary snoring, obstructive sleep apnea, or upper-airway resistance syndrome (UARS). In embodiments, predictive models analyze longitudinal sleep architecture, respiration, and autonomic-activity data to identify early indicators of dysregulation and comorbid risk.

[0251]

[0212] Sexual Health Disorders: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a sexual-health disorder. Exemplary sexual-health disorders include erectile dysfunction (ED), premature ejaculation, female sexual-arousal disorder, sexual-pain disorders (e.g., vaginismus, ALZAI-002-PCT 2025-10-10 dyspareunia), and low libido (hypoactive sexual-desire disorder), as well as conditions such as paraphilic disorders and gender dysphoria. In some embodiments, predictive features include hormonal, vascular, psychological, and behavioral indicators reflecting sexual function and endocrine or neurophysiological state.

[0252]

[0213] Allergies: In some embodiments, the disclosed methods and systems predict the onset and / or progression of an allergy. Exemplary allergies include food allergies (e.g., peanuts, shellfish), seasonal allergies (e.g., hay fever, pollen allergy), drug allergies (e.g., penicillin), and insect-sting allergies (e.g., bee sting allergy), as well as other allergic reactions (e.g., allergic rhinitis, allergic asthma). In some embodiments, predictive features capture immune-response trends, environmental exposures, and longitudinal sensitization profiles to anticipate flare onset and treatment needs.

[0253]

[0214] Periodontal Diseases: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a periodontal disease. Exemplary periodontal diseases include gingivitis, chronic periodontitis, aggressive periodontitis, and necrotizing periodontal disease. In some embodiments, predictive modeling integrates systemic inflammatory markers, oral-microbiome metrics, and behavioral data (e.g., smoking, oral-care habits) to estimate risk and progression rate.

[0254]

[0215] Pediatric Conditions: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a pediatric condition. Exemplary pediatric conditions include asthma, neonatal jaundice, cystic fibrosis, childhood obesity, and allergic conditions (e.g., food allergies, eczema), as well as conditions such as sudden infant death syndrome (SIDS) and developmental delays. In some embodiments, predictive models adapt feature selection and weighting to pediatric reference ranges and growth-adjusted trajectories, improving model accuracy across developmental stages.

[0255]

[0216] Hematologic Disorders: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a hematologic disorder. Exemplary hematologic disorders include anemia (e.g., iron-deficiency anemia, hemolytic anemia), thalassemia, sickle cell disease, myelodysplastic syndromes, leukemias, lymphomas, and coagulopathies (e.g., hemophilia, von Willebrand disease). In some embodiments, predictive features include longitudinal hematologic indices, iron and erythropoiesis biomarkers, and clotting-factor activity profiles to detect evolving hematologic dysfunction or treatment response.

[0256]

[0217] Gastrointestinal Disorders: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a gastrointestinal disorder. Exemplary gastrointestinal disorders include gastroesophageal reflux disease (GERD), peptic ulcer disease, irritable bowel syndrome (IBS), inflammatory bowel disease (IBD) (e.g., Crohn’s disease, ulcerative colitis), and hepatic or pancreatic disorders with gastrointestinal involvement, such as pancreatitis or cholestatic liver disease. In some embodiments, predictive models integrate biochemical, inflammatory, and motility-related data to assess mucosal injury, dysbiosis, or motility abnormalities preceding symptomatic onset.

[0257]

[0218] Reproductive Disorders: In some embodiments, the disclosed methods and systems predict the onset and / or progression of a reproductive disorder. Exemplary reproductive disorders include polycystic ovary ALZAI-002-PCT 2025-10-10 syndrome (PCOS), endometriosis, premature ovarian insufficiency, male and female infertility, erectile dysfunction, and reproductive cancers (e.g., ovarian, testicular, endometrial, and prostate cancer). In some embodiments, predictive features include hormonal, metabolic, and anatomical parameters reflecting reproductive endocrine function, gametogenesis, and tissue homeostasis.

[0258]

[0219] Immune or Immunodeficiency Disorders: In some embodiments, the disclosed methods and systems predict the onset and / or progression of an immune or immunodeficiency disorder. Exemplary immune or immunodeficiency disorders include primary immunodeficiency syndromes (e.g., common variable immunodeficiency (CVID), severe combined immunodeficiency (SCID)), secondary immunodeficiencies (e.g., HIV / AIDS, chemotherapy-induced immunosuppression), and autoimmune or autoinflammatory disorders (e.g., multiple sclerosis, rheumatoid arthritis, systemic lupus erythematosus (SLE)). In some embodiments, predictive modeling analyzes immune-cell subsets, cytokine expression, and infection or inflammation histories to identify latent immune dysregulation before clinical manifestation.

[0259] VI. Exemplary Risks, Behaviors, and Susceptibilities Predicted by the Disclosed Methods

[0260]

[0220] In some aspects, the disclosed methods and systems predict not only the onset or progression of medical conditions but also health-related risks, behaviors, susceptibilities, and other non-disease outcomes associated with increased morbidity, mortality, or adverse health events.

[0261]

[0221] Such predictions may include, for example, susceptibility to infection, behavioral risk factors, or physiological vulnerabilities that contribute to negative health outcomes. In some embodiments, these predictions are used or are optionally used in conjunction with medical-condition predictions, such as those derived from blood test values or other clinical data, to generate composite or conditional risk assessments.

[0262]

[0222] Infectious Disease Susceptibility. In some embodiments, the disclosed methods and systems predict susceptibility to infectious diseases, including the likelihood of contracting or developing complications from such infections. Exemplary infectious diseases include bacterial infections (e.g., pneumonia, tuberculosis, streptococcal pharyngitis, bacterial meningitis, urinary tract infections), viral infections (e.g., influenza, HIV, hepatitis, COVID-19, dengue fever, mononucleosis), fungal infections (e.g., candidiasis, aspergillosis), and parasitic infections (e.g., malaria, toxoplasmosis). In some embodiments, predictive models analyze immune, environmental, and behavioral data to estimate infection likelihood, severity, or complication risk.

[0263]

[0223] STD Susceptibility. In some embodiments, the disclosed methods and systems predict susceptibility to sexually transmitted diseases (STDs), including the likelihood of infection and the likelihood of developing complications therefrom. Exemplary STDs include bacterial infections (e.g., chlamydia, gonorrhea, syphilis), viral infections (e.g., HIV, human papillomavirus (HPV), herpes simplex virus (HSV), hepatitis B), and parasitic infections (e.g., trichomoniasis). In some embodiments, predictive features include behavioral, hormonal, and immunological parameters correlated with infection susceptibility and progression risk.

[0264]

[0224] Drug Resistance: In some embodiments, the disclosed methods and systems predict drug resistance, including the likelihood of developing resistance, the rate of progression, or the severity of resistance to ALZAI-002-PCT 2025-10-10 pharmacologic treatment. Exemplary drug resistance includes clopidogrel (Plavix) resistance, statin resistance, insulin resistance in type 2 diabetes, antibiotic resistance, antiviral resistance, and anticoagulant resistance (e.g., warfarin resistance), as well as resistance to medications used for chronic disease management (e.g., anti hypertensives, proton pump inhibitors). In some embodiments, predictive models integrate pharmacogenomic, metabolic, and adherence-related features to identify early signatures of diminished therapeutic response.

[0265]

[0225] Amenorrhea Age: In some embodiments, the disclosed methods and systems predict age of amenorrhea, including early or late onset of natural or pathological cessation of menstruation. Predictive features may include hormonal levels, genetic predispositions, reproductive health history, body mass index (BMI), stress level, and underlying conditions such as polycystic ovary syndrome (PCOS) or thyroid dysfunction. In some embodiments, amenorrhea age prediction is combined with reproductive health modeling to guide fertility counseling or menopausal management.

[0266]

[0226] Infertility Risk: In embodiments, the disclosed methods and systems predict infertility risk, including the likelihood of experiencing reduced fertility, negative reproductive outcomes, or infertility-related conditions such as PCOS or low sperm count. Predictive features include menstrual-cycle regularity, hormonal balance, reproductive history, genetic predispositions, and lifestyle or environmental exposures. In embodiments, predictive models assess temporal changes in reproductive biomarkers to estimate evolving fertility potential.

[0267]

[0227] Trauma Risk: In some embodiments, the disclosed methods and systems predict trauma risk, including the likelihood of exposure to traumatic events, the likelihood of negative psychological or physiological outcomes following trauma, and the likelihood of developing trauma-related disorders thereafter. Exemplary contributing factors include history of exposure to violence, childhood adversity, occupational or environmental hazards, substance use, and pre-existing mental health conditions. In some embodiments, predictive models integrate behavioral, environmental, and psychophysiological data to estimate susceptibility to trauma-related outcomes.

[0268]

[0228] Fall Risk: In some embodiments, the disclosed methods and systems predict fall risk, including the likelihood of experiencing a fall, sustaining injury, or requiring prolonged recovery. Predictive features include balance and gait parameters, muscle strength, medication side effects, visual acuity, environmental context, and underlying chronic conditions (e.g., Parkinson’s disease, diabetes, osteoporosis). In some embodiments, predictive models leverage continuous sensor data and clinical history to anticipate fall risk and recommend preventive interventions.

[0269]

[0229] Negative Health Habits: In embodiments, the disclosed methods and systems predict engagement in health-related behaviors associated with adverse outcomes, including initiation, persistence, escalation, and difficulty in cessation. Exemplary behaviors include tobacco use, alcohol consumption, recreational drug use, sedentary lifestyle, risky sexual behavior, and unhealthy eating patterns, as well as other detrimental habits such as excessive screen time or irregular sleep. In embodiments, predictive models incorporate behavioral, ALZAI-002-PCT 2025-10-10 environmental, and psychological data to identify latent patterns of habit formation and persistence.

[0270]

[0230] Bleeding or Cloting Risk-. In some embodiments, the disclosed methods and systems predict bleeding or thrombotic risk, including likelihood of hemorrhage, thrombosis, or coagulopathy due to genetic, physiological, or pharmacologic factors. Exemplary predictors include platelet count, clotting-factor activity, medication exposure (e.g., anticoagulants, antiplatelet agents), hepatic function, and inflammatory state. In some embodiments, predictive models analyze hematologic and pharmacodynamic data to optimize therapeutic safety and guide dosing strategies.

[0271]

[0231] Gastrointestinal Complication Risk-. In embodiments, the disclosed methods and systems predict gastrointestinal (Gl) complication risk, including susceptibility to inflammatory, ulcerative, or motility-related disorders. Exemplary risks include likelihood of Gl bleeding, ulcer recurrence, inflammatory bowel flare, or hepatic-gastrointestinal metabolic disruption. In embodiments, predictive models combine biochemical, microbiome, and behavioral data to anticipate Gl complications and guide preventive management.

[0272]

[0232] Those skilled in the art will recognize that the foregoing examples illustrate a broad range of predictive targets, encompassing not only diagnosable medical conditions but also underlying risks, behaviors, and susceptibilities that contribute to disease onset, exacerbation, and overall health outcomes. In embodiments, these predictive targets are modeled jointly, allowing the system to infer causal or correlative pathways between behavioral, physiological, and clinical domains, enhancing both predictive accuracy and clinical utility.

[0273] VII. Exemplary Diseases of Non-Human Animals Predicted by the Disclosed Methods

[0274]

[0233] In some embodiments, the disclosed methods and systems predict the onset and / or progression of diseases in non-human animals. Exemplary non-human animals include farm animals, agricultural pollinators, aquaculture species, domestic pets, and other animals of agricultural, ecological, or commercial importance. Examples include livestock (e.g., cattle, pigs, sheep, and goats), poultry (e.g., chickens, turkeys, ducks, and geese), aquaculture species (e.g., tilapia, salmon, shrimp, and oysters), fur-bearing animals (e.g., mink and fox), game animals (e.g., deer and wild boar), animals kept for conservation, education, or entertainment (e.g., elephants, lions, giraffes, pandas, tigers, gorillas, zebras, rhinoceroses, polar bears, and cheetahs), and fiber-producing animals (e.g., alpacas and llamas), as well as insects such as silkworms and honeybees.

[0275]

[0234] In some embodiments, the disclosed methods and systems predict disease onset or progression for an individual animal, while in other embodiments, they predict the onset or progression of disease for a population, herd, flock, colony, or aquaculture group. For example, a model may predict the emergence of an infection across a cattle herd, the spread of avian influenza in a poultry flock, or colony collapse disorder in a commercial apiary. In some embodiments, predictive features are derived from physiological, behavioral, environmental, or genomic data collected over time, enabling early detection of disease or stress conditions and improving outcomes in animal health and production systems.

[0276]

[0235] Cattle: In some embodiments, the disclosed methods and systems predict the onset and / or progression of disease in cattle. Exemplary such diseases include aflatoxicosis, anthrax, bluetongue, botulism, ALZAI-002-PCT 2025-10-10 bovine ephemeral fever, bovine viral diarrhea virus, brucellosis, cattle tick infestation, copper deficiency, enzootic bovine leucosis, foot-and-mouth disease, hydatid disease, intestinal torsion, Johne’s disease, lumpy jaw, lumpy skin disease, melioidosis, neosporosis, salmonellosis, screw-worm fly infestation, tetanus, transport tetany, vibriosis, and papillomavirus-associated warts. In some embodiments, predictive features include herd-level sensor data, feed composition, environmental exposure, and biomarker trends related to metabolic stress and immune response.

[0277]

[0236] Pigs: In some embodiments, the disclosed methods and systems predict the onset and / or progression of disease in pigs. Exemplary diseases include aflatoxicosis, African swine fever, anthrax, encephalitis, leptospirosis, copper deficiency, influenza A, intestinal torsion, tetanus, brucellosis, melioidosis, piglet anemia, piglet scours, salmonellosis, swine brucellosis, and lumpy jaw. In some embodiments, predictive models analyze growth curves, feed intake, behavioral activity, and environmental metrics to identify early indicators of infection or nutritional deficiency.

[0278]

[0237] Goats: In some embodiments, the disclosed methods and systems predict the onset and / or progression of disease in goats. Exemplary diseases include anthrax, akabane disease, brucellosis, caprine arthritis encephalitis, cattle tick infestation, copper deficiency, Johne’s disease, leptospirosis, lumpy jaw, melioidosis, neosporosis, salmonellosis, and tetanus. In embodiments, predictive models integrate milk yield, body-condition trends, and herd-exposure data to anticipate subclinical disease or vector-borne infection.

[0279]

[0238] Sheep: In some embodiments, the disclosed methods and systems predict the onset and / or progression of disease in sheep. Exemplary diseases include anthrax, akabane disease, bluetongue, brucellosis, cattle tick infestation, copper deficiency, enzootic bovine leucosis, foot-and-mouth disease, hydatid disease, Johne’s disease, lumpy jaw, melioidosis, neosporosis, ovine brucellosis, salmonellosis, screw worm fly infestation, tetanus, transport tetany, and vibriosis. In some embodiments, predictive features include lambing patterns, feed efficiency, and weather-related stress markers contributing to disease risk.

[0280]

[0239] Poultry: In some embodiments, the disclosed methods and systems predict the onset and / or progression of disease in poultry, including chickens, turkeys, ducks, and geese. Exemplary diseases include avian influenza, avian paramyxovirus, blackhead, botulism, external parasitism, fowl cholera, fowl pox, infectious laryngotracheitis, Marek’s disease, Newcastle disease, salmonella enteritidis, screw-worm fly infestation, spotty liver, and helminthic infection. In some embodiments, predictive models analyze environmental humidity, feed contamination, and flock-level sensor data to detect early deviations in respiratory or digestive health.

[0281]

[0240] Horses: In some embodiments, the disclosed methods and systems predict the onset and / or progression of disease in horses. Exemplary diseases include African horse sickness, anthrax, Australian bat lyssavirus, botulism, brucellosis, cattle tick infestation, equine herpesvirus type 1 (EHV-1), equine infectious anemia, equine influenza, equine viral arteritis, Hendra virus, Japanese encephalitis, leptospirosis, melioidosis, neosporosis, nipah virus, rabies, strangles, tetanus, and neurologic syndromes of infectious or inflammatory ALZAI-002-PCT 2025-10-10 origin. In some embodiments, predictive features include gait analysis, cardiovascular and respiratory monitoring, and exposure to vector populations.

[0282]

[0241] Donkeys and Mules: In some embodiments, the disclosed methods and systems predict the onset and / or progression of disease in donkeys and mules. Exemplary diseases include African horse sickness, anthrax, cattle tick infestation, EHV-1 , equine infectious anemia, equine influenza, equine viral arteritis, Japanese encephalitis, melioidosis, neosporosis, nipah virus, rabies, strangles, and tetanus. In some embodiments, predictive models combine physiological data with climatic and transport metrics to anticipate infectious or stress-related outbreaks.

[0283]

[0242] Bees: In some embodiments, the disclosed methods and systems predict the onset and / or progression of disease in bees, including honeybees. Exemplary diseases include American foulbrood, European foulbrood, chalkbrood virus, sacbrood virus, Nosema spp. infection, Varroa destructor parasitism, deformed wing virus, colony collapse disorder, Israeli acute bee paralysis virus, chronic bee paralysis virus, Kashmir bee virus, black queen cell virus, and Apis iridescent virus. In some embodiments, predictive features include hive temperature, acoustic activity, foraging behavior, and microclimate data, enabling early detection of colony stress and pathogen load.

[0284]

[0243] Dogs: In some embodiments, the disclosed methods and systems predict the onset and / or progression of disease in dogs. Exemplary chronic diseases include osteoarthritis, chronic kidney disease (CKD), diabetes mellitus, hypothyroidism, Cushing’s disease (hyperadrenocorticism), dilated cardiomyopathy (DCM), chronic bronchitis, mitral valve disease, epilepsy, and inflammatory bowel disease (IBD). In some embodiments, predictive features include activity tracking, heart-rate variability, and metabolic profiles for early identification of chronic or degenerative disease.

[0285]

[0244] Cats: In some embodiments, the disclosed methods and systems predict the onset and / or progression of disease in cats. Exemplary diseases include Australian bat lyssavirus infection, Nipah virus infection, feline immunodeficiency virus (FIV), feline leukemia virus (FeLV), chronic kidney disease (CKD), diabetes mellitus, hypertrophic cardiomyopathy, hyperthyroidism, and feline asthma. In some embodiments, predictive models use behavioral, respiratory, and biochemical data to assess susceptibility to chronic or infectious conditions and parasite exposure.

[0286]

[0245] Fish: In some embodiments, the disclosed methods and systems predict the onset and / or progression of disease in fish, including farmed species such as salmon and tilapia. Exemplary diseases include infectious salmon anemia, bacterial kidney disease, sea lice infestations, and viral hemorrhagic septicemia. In some embodiments, predictive features include water temperature, salinity, oxygen levels, and feeding behavior to forecast outbreaks or stress-induced disease in aquaculture systems.

[0287]

[0246] Crustaceans: In some embodiments, the disclosed methods and systems predict the onset and / or progression of disease in crustaceans, such as prawns, lobsters, and crabs. Exemplary diseases include white spot disease and other viral or bacterial infections that can significantly affect aquaculture and commercial ALZAI-002-PCT 2025-10-10 seafood production. In some embodiments, predictive models analyze water quality, microbial load, and population dynamics to detect environmental conditions conducive to pathogen proliferation.

[0288]

[0247] Those skilled in the art will recognize that the disclosed methods and systems can be adapted across animal species and population scales to predict both individual and collective disease dynamics. In some embodiments, such predictions inform interventions in veterinary care, livestock management, aquaculture biosecurity, pollinator preservation, or wildlife conservation. In embodiments, by modeling temporal and environmental data in conjunction with biological measurements, the disclosed systems improve early detection, optimize treatment strategies, and enhance overall health management in non-human animal populations.

[0289] VIII. Exemplary Methods Further Comprising Treatment and Prevention

[0290]

[0248] In some aspects are provided methods of predicting the onset and / or progression of a condition, wherein the method further comprises a treatment being selected, prescribed, and / or administered to a subject based on the output of a predictive model.

[0291]

[0249] In some embodiments, the predictive model provides a quantitative or categorical prediction that informs a treatment plan by identifying a therapy or intervention most likely to prevent, delay, or mitigate the predicted condition, or to optimize a therapeutic response while minimizing adverse effects.

[0292]

[0250] In some embodiments, the disclosed system transmits risk predictions, confidence metrics, and explanatory outputs to downstream modules that recommend, prioritize, or schedule diagnostic tests, pharmacotherapies, or preventive interventions. Such modules may be integrated with electronic medical record (EMR) systems or clinical workflow engines to generate alerts, treatment plans, or resource-allocation recommendations in real time based on the predicted onset or progression probability.

[0293]

[0251] In some embodiments, a prediction according to a disclosed method informs a decision by a clinician or caregiver (e.g., a physician, nurse practitioner, or other healthcare provider) to prescribe a pharmacotherapy or other treatment to a subject. In some embodiments, the prediction informs a decision to recommend a preventive or prophylactic measure to postpone the onset of a medical condition or to reduce the likelihood or severity of a symptom thereof. In some embodiments, a prediction informs the selection or modification of a non-pharmacological intervention or a lifestyle modification, such as adjusting dietary, exercise, or behavioral habits. In embodiments, the predictive model identifies not only the probability of a condition’s occurrence but also one or more predicted treatment responses, enabling adaptive, individualized intervention strategies.

[0294]

[0252] Herein, “treating” or “treatment” refers to causing a desired biological, pharmacological, or measurable clinical effect in an animal, including a human or non-human animal. Treatment includes, unless specified or indicated otherwise: (a) preventing a condition from occurring in a predisposed subject; (b) inhibiting or arresting disease development; (c) causing regression or symptom reduction; (d) protecting from or alleviating a pathology or symptom; (e) reducing, decreasing, inhibiting, ameliorating, or preventing the onset, severity, duration, progression, frequency, or probability of a condition or symptom; and (f) preventing or inhibiting ALZAI-002-PCT 2025-10-10 worsening or comorbid progression. In some embodiments, “treatment” includes prophylaxis and prevention. In other embodiments, “treatment” does not include prophylaxis or prevention. When used as a noun, a “treatment” includes any pharmacological or non-pharmacological therapy or intervention that may be prescribed, recommended, administered, performed, or the like, for the purpose of treating a subject.

[0295]

[0253] A treatment may be directly administered or self-administered, including under clinical guidance.

[0296]

[0254] Pharmacotherapy. In some embodiments, the method comprises predicting the onset and / or progression of a medical condition and further includes administering or recommending a pharmacotherapy determined to have a predicted therapeutic efficacy above a defined model threshold for the subject. The predictive model may identify an optimal drug class, agent, dosage regimen, or combination based on the subject’s biomarker profile, comorbidities, pharmacogenomic markers, predicted metabolism, or interaction risk. In some embodiments, the model output includes predicted probabilities of therapeutic response, adverse events, and drug-drug interactions, enabling automatic generation of a ranked treatment list or dose-adjustment recommendation. In some embodiments, the predictive model uses prior medication response data or digital twin simulations to estimate treatment trajectories and identify early nonresponders. For example, a model predicting early-stage type 2 diabetes may further predict enhanced metformin responsiveness when administered at a low dose, extended-release schedule combined with increased physical activity, whereas a model identifying hepatic impairment may predict reduced benefit and suggest alternative agents, such as SGLT2 inhibitors, with a titration plan calibrated to renal function. In some embodiments, the model may recommend dynamic titration intervals or alternate-day dosing based on predicted hepatic enzyme activity, thereby minimizing adverse reactions while maintaining glycemic control. In some embodiments, the model further predicts optimal timing, formulation, or combination therapy to achieve synergistic benefit, for example, co-administration of an anti-inflammatory agent when a pro-inflammatory cytokine signature is detected. In other embodiments, the model identifies a predicted responder subgroup to a precision therapeutic, guiding clinical trial stratification or post-market dosing optimization.

[0297]

[0255] Gene Therapy In some embodiments, the predictive model identifies subjects with a high predicted likelihood of developing a gene-associated disorder or a strong predicted therapeutic response to a gene-based intervention. The method may further comprise administering or recommending a gene therapy, such as CRISPR-mediated gene correction, antisense oligonucleotide modulation, or viral-vector-based gene supplementation, selected according to model-predicted tissue tropism, immune tolerance, or safety profile. In some embodiments, the predictive model evaluates vector-host compatibility parameters including serotype preference, promoter activity, or neutralizing antibody titers to predict transduction efficiency and immune reactivity. The model output may recommend an optimal viral vector system (e.g., AAV, lentivirus, adenovirus) or delivery route (e.g., intrathecal, intramuscular, intravenous) predicted to achieve target gene expression with minimal off-target activity. In some embodiments, the model predicts the most effective therapeutic window for gene therapy, such as early pre-symptomatic administration in high risk genotypes, or postponement until ALZAI-002-PCT 2025-10-10 inflammatory biomarkers subside to minimize vector clearance. The system may also simulate dose-response curves based on predicted vector copy number and cell-type turnover, enabling individualized treatment scheduling and long-term expression stability modeling.

[0298]

[0256] Immunotherapy. In some embodiments, the predictive model determines a probability distribution for immune response or immune-related adverse events, including cytokine release risk, checkpoint sensitivity, and autoimmunity propensity. Based on this output, the method may further include administering or adjusting an immunotherapy, such as checkpoint inhibition, monoclonal antibody therapy, or cytokine modulation, at a dosage or schedule predicted to optimize efficacy-to-toxicity ratio. In some embodiments, the predictive model also identifies the optimal therapeutic window for combination immunotherapy, including adjuvant vaccination timing or alternating cytokine blockade cycles, based on model-predicted immune exhaustion kinetics or antigen presentation dynamics. For example, a model predicting low PD-L1 expression and high T-cell exhaustion may recommend combination PD-1 / CTLA-4 inhibition with targeted cytokine therapy, whereas a model detecting elevated autoantibody titers may predict increased autoimmune risk and recommend delayed checkpoint reactivation or dose fractionation. In some embodiments, the model predicts patient-specific dosing curves, immune cell expansion rates, or serum biomarker kinetics to dynamically adapt therapy in near real time. In other embodiments, predicted immune activation trajectories inform prophylactic administration of immunomodulatory agents (e.g., corticosteroids or IL-6 inhibitors) to preempt cytokine-release syndromes, improving patient safety and reducing hospitalization.

[0299]

[0257] Radiation Therapy In some embodiments, the predictive model determines radiosensitivity, optimal dose fractionation, or predicted radiation-induced toxicity. The method may further comprise performing a radiation therapy according to model-recommended dose or field configuration, or combining radiotherapy with concurrent pharmacotherapy when the model predicts synergistic benefit.

[0300]

[0258] Physical Therapy In some embodiments, the predictive model estimates neuromuscular recovery potential, injury recurrence risk, or biomechanical compensation patterns. The method may further comprise performing or recommending a physical therapy protocol tailored to the predicted rehabilitation trajectory, intensity, or muscle group responsiveness, thereby improving mobility and reducing reinjury risk.

[0301]

[0259] Psychotherapy In some embodiments, the predictive model identifies the likelihood of positive response to specific psychotherapeutic modalities based on behavioral, linguistic, or physiological data. The method may further comprise performing or recommending a psychotherapy, counseling, or cognitive behavioral intervention most likely to reduce predicted recurrence or symptom severity for the subject’s mental health profile.

[0302]

[0260] Regenerative Medicine: In some embodiments, the predictive model determines suitability or response likelihood to a regenerative medicine therapy, such as stem-cell transplantation, platelet-rich plasma (PRP), exosome therapy, or engineered tissue grafts. The method may further comprise administering or recommending such therapy at a predicted time point or anatomical site determined by model-derived ALZAI-002-PCT 2025-10-10 regeneration probability. In some embodiments, the predictive model predicts the local microenvironmental conditions most favorable for regenerative success, such as inflammatory cytokine ratios, extracellular matrix remodeling, or angiogenic signaling levels, thereby identifying an optimal intervention window. The model may recommend, for example, autologous mesenchymal stem-cell (MSC) injection in a joint predicted to exhibit maximal responsiveness based on cartilage biomarker trajectories, or PRP treatment when fibrinogen levels fall within a model-defined therapeutic range. In some embodiments, the system further predicts which regenerative modality, such as cell-based, scaffold-based, or secretome-based, offers the highest probability of tissue integration and functional restoration, and may recommend combination with a supportive intervention (e.g., mechanical offloading or microcurrent stimulation) to enhance engraftment probability.

[0303]

[0261] Blood Filtration and Extracorporeal Therapies: In some embodiments, the disclosed methods further comprise performing or recommending a blood filtration or extracorporeal therapy in response to a model-predicted indication, risk profile, or response likelihood. The predictive model may determine that the subject exhibits a predicted benefit from selective removal of circulating toxins, immune complexes, lipoproteins, or inflammatory mediators based on laboratory, hemodynamic, or molecular biomarkers. In some embodiments, the blood filtration intervention comprises plasmapheresis, therapeutic plasma exchange (TPE), apheresis, or double-filtration plasmapheresis (DFPP). In some embodiments, the predictive model identifies the likely efficacy of one filtration modality over another by evaluating biomarker persistence, plasma protein concentration, or immune activation profiles. The system may predict, for example, that a subject with elevated cytokine levels or autoantibody titers will respond favorably to plasma exchange, whereas a subject with hyperlipidemia or refractory inflammation will benefit more from selective DFPP. In some embodiments, the predictive model further estimates optimal treatment timing, filtration volume, or replacement fluid composition based on predicted hemodynamic tolerance and biomarker kinetics. For instance, for a subject at predicted high risk of cytokine storm or autoimmune flare, the model may recommend initiating TPE within a defined pre-symptomatic window or adjusting exchange volume according to predicted plasma viscosity or fibrinogen depletion rate. In some embodiments, extracorporeal therapies may further include hemofiltration, hemodialysis, hemoadsorption, immunoadsorption, or lipid apheresis, and the predictive model may identify subjects for whom these modalities are predicted to optimize metabolic clearance, immune modulation, or toxicant removal efficiency. In some embodiments, the system automatically prioritizes a filtration modality when predicted multi-organ stress or metabolic dysfunction exceeds a threshold, providing real-time, data-driven recommendations to clinical systems or care teams.

[0304]

[0262] Non-Pharmacological Intervention: In some embodiments, the predictive model predicts benefit from a non-pharmacological intervention, such as neuromodulation, transcranial magnetic stimulation, phototherapy, or biofeedback. The method may further comprise performing or recommending the intervention with parameters (e.g., frequency, intensity, duration) optimized by the model output for the subject’s predicted neurophysiological responsiveness. ALZAI-002-PCT

[0305] 2025-10-10

[0306]

[0263] Surgical Intervention-. In some embodiments, the predictive model identifies subjects likely to experience favorable surgical outcomes or reduced postoperative complications. The method may further comprise performing or recommending a surgical procedure when the predicted benefit-to-risk ratio exceeds a defined threshold, or delaying surgery where predicted recovery potential is low. The predictive output may also guide selection of surgical technique or anesthesia type.

[0307]

[0264] Medical Device Intervention-. In some embodiments, the predictive model identifies subjects predicted to respond favorably to a medical device-based therapy, such as implantable cardiac defibrillators, insulin pumps, neurostimulators, or continuous glucose monitors. The method may further comprise selecting device parameters, implantation timing, or programming schedules according to predicted performance metrics or adherence likelihood. In some embodiments, the model predicts physiological tolerance thresholds, device calibration intervals, or sensor-drift correction schedules, automatically generating parameter recommendations for device configuration. For example, the system may predict that a neurostimulator’s pulse width and amplitude should be reduced during certain sleep phases to minimize desensitization, or that an insulin pump’s basal rate should be modulated in anticipation of circadian metabolic variation predicted from prior glucose activity correlation data. In some embodiments, the predictive model interfaces with the device’s telemetry feed to adapt control parameters in real time, maintaining closed-loop optimization of therapeutic efficacy and safety. The model output can further inform pre-implantation device selection, predicting comparative benefit between alternative device types or stimulation targets based on historical response data.

[0308]

[0265] Behavioral Intervention-. In some embodiments, the predictive model identifies behavioral factors most strongly correlated with the predicted condition. The method may further comprise performing or recommending a behavioral intervention targeting those specific factors, such as structured adherence programs, digital coaching, or stress-reduction training, predicted to yield the greatest reduction in model-estimated risk.

[0309]

[0266] Nutritional Intervention-. In some embodiments, the predictive model determines nutritional factors most relevant to the predicted condition, such as macronutrient imbalance, micronutrient deficiency, or metabolic sensitivity. The method may further comprise prescribing or recommending a nutritional intervention, including a personalized diet plan, macronutrient adjustment, or supplementation schedule. The model may predict both expected benefit and potential contraindications based on genetic or metabolic data, producing a treatment plan optimized for the individual’s predicted nutrient utilization and disease risk.

[0310]

[0267] Lifestyle Modification-. In some embodiments, the predictive model identifies lifestyle parameters most influential on the predicted condition, such as sleep quality, physical activity, dietary pattern, substance use, or stress level. The method may further comprise prescribing or recommending lifestyle modifications, including structured exercise regimens, stress-management practices, or sleep schedule optimization, prioritized according to the model’s predicted effect size on disease prevention or symptom reduction. In some embodiments, the model produces an individualized “lifestyle impact vector,” quantifying marginal predicted ALZAI-002-PCT 2025-10-10 risk reduction per modifiable parameter, which is then used to rank interventions by relative efficacy. For example, the system may predict that improving sleep efficiency from 70% to 85% yields a 12% reduction in cardiovascular risk, whereas reducing sodium intake below a modeled threshold yields only marginal benefit for that specific subject. In some embodiments, the model further predicts the subject’s adherence likelihood, motivation trajectory, or behavioral reinforcement profile, allowing the system to recommend complementary digital or behavioral interventions to sustain engagement. The system may, for example, combine predictive behavioral modeling with continuous wearable sensor feedback to adjust exercise intensity or recommend rest periods dynamically, based on physiological response data. In some embodiments, the predictive model evaluates lifestyle interactions with pharmacologic or nutritional interventions, automatically generating a combined risk-reduction plan that maximizes synergistic effects while mitigating counterproductive overlap (e.g., predicting that aggressive caloric restriction during immunotherapy may elevate fatigue risk, and automatically recommending moderate adjustment instead).

[0311]

[0268] Combination Therapies: In some embodiments, any of the above treatments or interventions are combined according to predicted synergistic benefit, reduced toxicity, and / or improved adherence likelihood derived from the predictive model. The model may generate multi-objective treatment policies that balance competing goals, such as maximizing therapeutic gain while minimizing cumulative toxicity, using individualized reward functions or counterfactual simulations. In some embodiments, the predictive system employs policy-gradient or reinforcement-learning frameworks to recommend sequential or alternating intervention schedules. For example, the model may predict that alternating DFPP and immunotherapy maximizes autoantibody clearance while avoiding immune rebound, or that staggered initiation of pharmacotherapy and behavioral modification optimizes long-term adherence. In some embodiments, the model predicts the interaction term between two interventions (e.g., nutrition-drug synergy or radiation-chemotherapy sensitization) and recommends personalized timing, dose ratio, or sequence to achieve maximal predicted outcome. These recommendations may be generated automatically as part of an adaptive decision support loop that continuously refines treatment policy based on longitudinal response data. In some embodiments, the system outputs a ranked intervention plan specifying both the combination and the predicted net-benefit score, enabling clinicians to select from transparent, model-derived options rather than heuristic or guideline-based defaults.

[0312]

[0269] Technological and Clinical Integration: Those skilled in the art will recognize that these treatment-related embodiments demonstrate that the disclosed methods are not limited to passive data analysis but instead produce actionable, technologically enabled outputs that directly inform or optimize clinical decision making. By coupling predictive modeling with intervention selection, dosing, timing, or combination strategy, the disclosed systems improve therapeutic precision, reduce unnecessary treatment, and enable adaptive management based on continuously updated model predictions. This integrated approach yields a concrete improvement in computer-implemented healthcare decision support and patient outcomes, ALZAI-002-PCT 2025-10-10 distinguishing the disclosed methods from abstract mental processes or conventional diagnostic reasoning.

[0313]

[0270] Any one or more of the disclosed treatments can be combined with any other such treatment(s).

[0314] IX. Examples

[0315]

[0271] The following examples are included for illustrative purposes only and are not intended to be limiting. Unless expressly indicated, each example is a prophetic example that describes prospective or simulated evaluations consistent with the disclosed methods and systems and is not based on actual patient data.

[0316] Example 1 : Predicting the Onset and Progression of a Medical Condition Using Blood Test Data

[0317]

[0272] In this example, predictive models are evaluated for determining a predicted risk score representing the likelihood, onset risk, or progression probability of a medical condition. The medical condition may be any of the medical conditions described herein. A predictive model may comprise any of a decision tree, a random forest ensemble, a gradient-boosting model (e.g., XGBoost), a support vector machine (SVM), or a neural network such as a convolutional or deep-learning neural network.

[0318]

[0273] The evaluation uses a retrospective-style study design with simulated cohorts comprising several thousand control subjects (also, “patients”) and several thousand subjects diagnosed with a medical condition.

[0319]

[0274] Multiple configurations of training data from both groups (controls and subjects) are evaluated to assess the sufficiency and comparative performance of various feature sets:

[0320]

[0275] Configuration 1: A heterogeneous dataset comprising multimodal data sources, including but not limited to: EMR data, laboratory panels, derived longitudinal statistics, physiological measurements, sociodemographic attributes, and behavioral signals; as well as imaging data (e.g., computed tomography (CT), magnetic resonance imaging (MRI), functional MRI (fMRI), positron-emission tomography (PET), ultrasound, echocardiography, or retinal imaging); genomic, transcriptomic, proteomic, and metabolomic data; wearable- or sensor-derived digital health data (e.g., continuous glucose monitoring, photoplethysmography (PPG), accelerometer data, sleep-stage monitoring, and activity tracking); and unstructured clinical or patient-reported data such as physician notes, symptom diaries, or questionnaire responses.

[0321]

[0276] Configuration 2: A dataset solely comprising EMR data, including blood test data and one or more additional or alternative parameter values, including any of physiological parameters (e.g., blood pressure, electrocardiography (ECG), heart rate, heart-rate variability (HRV), weight, body-mass index (BMI)), vascular risk factors (e.g., hypertension, hyperlipidemia, ischemic heart disease, myocardial infarction, diabetes mellitus), sociodemographic parameters (e.g., age, gender, race, education, socioeconomic status), behavioral parameters (e.g., smoking, alcohol consumption, medication use, and frequency and type of physical activity), and medical parameters (e.g., a medical condition, a background disease, and an administered medication).

[0322]

[0277] Configuration 3 A dataset solely comprising blood test data stored as part of EMR data.

[0323]

[0278] Blood test data stored as part of EMR data may comprise, consist essentially of, or consist of any combination or subset of the following analytes, panels, or derived laboratory values. Different embodiments may include different combinations, subsets, or panels of these tests, such as depending on availability, ALZAI-002-PCT 2025-10-10 predictive contribution, or model configuration, as will be appreciated by one of skill in view of these teachings.

[0324]

[0279] Complete blood count and standard chemistry panels: absolute basophil count (Baso abs), absolute eosinophil count (EOS abs), hemoglobin (Hb), hematocrit (Het), absolute lymphocyte count (Lymp abs), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), mean corpuscular volume (MOV), absolute monocyte count (MONO abs), red blood cell count (RBC), platelet count (PLT), white blood cell count (WBC), red cell distribution width (RDW), albumin, calcium, chloride, creatinine, globulin, glucose, magnesium, phosphorus, potassium, total protein, sodium, urea, uric acid, aspartate aminotransferase (AST / GOT), alanine aminotransferase (ALT / GPT), gamma-glutamyl transferase (GGT), total bilirubin, thyroid-stimulating hormone (TSH), vitamin B12, prothrombin time (PT), partial thromboplastin time (PTT), and international normalized ratio (INR).

[0325]

[0280] Extended metabolic, endocrine, and inflammatory markers: high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, total cholesterol, triglycerides, tumor necrosis factor alpha (TNF-a), interleukin-6 (IL-6), C-reactive protein (CRP), high-sensitivity C-reactive protein (hs-CRP), transforming growth factor (TGF), creatine kinase (CK), cortisol, adrenocorticotropic hormone (ACTH), parathyroid hormone (PTH), prolactin, growth hormone (GH), insulin-like growth factor 1 (IGF-1), testosterone, estradiol (E2), follicle-stimulating hormone (FSH), luteinizing hormone (LH), aldosterone, renin, vitamin D, ferritin, total iron binding capacity (TIBC), transferrin, vitamin B9 (folate), zinc, HbA1c, ketones, ghrelin, and leptin.

[0326]

[0281] Specialized, cardiovascular, and immunologic markers: D-dimer, B-type natriuretic peptide (BNP), NT-proBNP, troponin, fibrinogen, erythrocyte sedimentation rate (ESR), lactate dehydrogenase (LDH), homocysteine, immunoglobulins (IgG, IgA, IgM, IgE), vitamin K, anti-nuclear antibodies (ANA), apolipoprotein A1 , apolipoprotein B, fibroblast growth factor 23 (FGF23), fibroblast growth factor 21 (FGF21), erythropoietin (EPO), calcitonin, alkaline phosphatase (ALP), neutrophil-to-lymphocyte ratio (NLR), galectin-3, brain-derived neurotrophic factor (BDNF), neutrophil gelatinase-associated lipocalin (NGAL), soluble urokinase plasminogen activator receptor (suPAR), von Willebrand factor (vWF), placental growth factor (PIGF), soluble fms-like tyrosine kinase-1 (sFlt-1), matrix metalloproteinase-9 (MMP-9), angiopoietin-2 (Ang-2), interleukin-1 beta and -10 (IL-1 p, IL-10), glutamate dehydrogenase (GLDH), myeloperoxidase (MPO), and serum amyloid A (SAA).

[0327]

[0282] Tumor, metabolic, and other specialized markers: alpha-fetoprotein (AFP), carci noembryonic antigen (CEA), cancer antigen 125 (CA-125), cancer antigen 19-9 (CA19-9), neuron-specific enolase (NSE), beta-2 microglobulin, squamous cell carcinoma antigen (SCC), total bile acids, lactoferrin, serum ammonia, beta-hydroxybutyrate, serum osmolality, serum free fatty acids, acylcarnitines, serum free light chains (kappa / lambda), C-peptide, proinsulin, serum amyloid P (SAP), and serum amyloid A (SAA).

[0328]

[0283] Historical data reflect measurements taken up to 10 years prior to diagnosis of the medical condition for subjects in the patient cohort, and up to 10 years prior to a corresponding reference time point for subjects in the control cohort. The reference time point for control subjects may be randomly assigned or matched to the diagnosis dates of corresponding patients, ensuring comparable temporal distributions across cohorts. ALZAI-002-PCT

[0329] 2025-10-10

[0330]

[0284] In each configuration, subject data may be structured as time-series sequences, static snapshots, or derived statistical features capturing longitudinal patterns, such as averages, extrema, variances, ranges, slopes, or temporal-change descriptors or temporal deltas (e.g., differences between early and late windows or between the earliest and latest available values). For each blood test for example, multiple derived features may be computed, such as mean, maximum, minimum, standard deviation, and temporal deltas. Feature importance is then quantified by the model for each configuration and each time horizon.

[0331]

[0285] The predictive models are further trained and evaluated under both knowledge-weighted and knowledge-excluded conditions. In the knowledge-weighted configuration, features conventionally associated with the target medical condition based on clinical expertise are assigned higher feature weights or selection priority. In the knowledge-excluded configuration, such features are intentionally omitted or down-weighted to test whether the model autonomously identifies alternative or counterintuitive predictors. In these configurations, feature selection is not constrained to variables conventionally associated with the target condition, and the system autonomously identifies, ranks, and refines features (including counterintuitive or cross-domain signals) based on their contribution to predictive performance. Comparative evaluation between conditions enables determination of whether clinical priors improve or constrain predictive performance, and whether non-traditional features can meaningfully enhance accuracy, sensitivity, or interpretability.

[0332]

[0286] In prophetic examples comparing predictive model performance across three data configurations (i.e., (1) all relevant multimodal data, (2) all EMR data, and (3) blood test data only) models utilizing Configuration 3 will demonstrate performance that approaches or is equivalent to the accuracy of models utilizing Configuration 1 . Models using Configuration 2 will provide only marginal improvement over Configuration 3.

[0333]

[0287] This surprising result demonstrates that the structured, time-stamped blood test data alone contains substantially all predictive signals required for accurate prediction, while providing a critical technological improvement over systems reliant on complex, unstructured multimodal or EMR data. This improvement includes reduced computational overhead, faster inference times, simplified system architecture, and avoidance of data integration challenges inherent in heterogeneous data sources.

[0334]

[0288] In one illustrative evaluation, an XGBoost predictive model is trained using the foregoing datasets, and in each configuration the model outputs a predicted risk score and a class label determined by comparison to a configurable threshold (e.g., 0.5). Equivalent training and evaluation procedures may be applied to alternative predictive models (e.g., random forest, gradient-boosting variants, SVMs, and neural networks) to facilitate comparative benchmarking and cross-validation across architectures.

[0335]

[0289] Comparisons are also made by training an XGBoost predictive model and the alternative predictive models using subject data that solely comprises EMR data and blood test data, including blood tests as part of EMR data. Historical blood test values reflect blood tests taken up to 10 years prior to diagnosis. Blood tests with the highest risk for a medical condition are included in the model based on clinical knowledge.

[0336]

[0290] Based on an analysis of blood test values and the relative contribution or impact of those values on ALZAI-002-PCT 2025-10-10 model performance (e.g., influence on accuracy, precision, or feature importance ranking), each model configuration may autonomously or semi-autonomously select a subset of the highest-ranking features as the active input set for training and prediction. In some embodiments, this feature-selection process is performed separately for each configuration (heterogeneous, EMR + blood tests, or blood tests only), allowing direct comparison of which variables most strongly contribute to predictive accuracy under different data conditions.

[0337]

[0291] Feature rankings may differ across time horizons and prior measurement periods. For instance, certain blood test values may exhibit stronger predictive contribution when derived as longitudinal changes (e.g., multi-year slopes or deltas) rather than static measurements. The predictive importance of a given analyte may also vary depending on the length of the prior data window (e.g., one, five, or 10 years) and the forecast horizon (e.g., one, two, or five years after the observation window).

[0338]

[0292] Results and Comparative Evaluation. The comparative evaluation across Configurations 1-3 will demonstrate that models trained solely on blood test data (Configuration 3) achieve predictive accuracy approaching or equaling that of models trained on multimodal heterogeneous datasets (Configuration 1).

[0339]

[0293] Surprisingly, the blood test-only configuration will be sufficient to identify subjects at elevated risk for developing or progressing in a medical condition, indicating that a compact feature space derived from standard laboratory data captures latent physiological signatures of disease onset. Inclusion of heterogeneous multimodal data (Configuration 1) will substantially increase computational burden and data-acquisition requirements without commensurate improvement in predictive performance.

[0340]

[0294] Only marginal improvement will be observed using broader EMR data (Configuration 2), where additional EMR-derived or non-laboratory modalities are included, and it will only be observed for certain medical conditions, whereas it will provide no improvement for other medical conditions. Notably, the modest gains yielded in interpretability and calibration incurs substantially greater data-preprocessing complexity.

[0341]

[0295] These findings collectively demonstrate that accurate, resource-efficient risk prediction can be achieved using routinely collected blood test data alone, constituting a significant and unexpected simplification relative to approaches requiring multi-source or high-dimensional data integration.

[0342]

[0296] Performance Metrics. In one example, a gradient-boosting model trained under Configuration 3 correctly identifies greater than about 70% of subjects who later experience onset or progression of a medical condition during a specified future time period based on blood test data collected up to 10 years prior.

[0343]

[0297] In some embodiments, predictive models trained according to Configuration 3 of the present example achieve an accuracy or sensitivity of at least 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 95%, or 99% for predicting the onset or progression of a medical condition within a specified future period, based on data from prior observation windows. In some embodiments, predictive models trained according to Configuration 3 of the present example achieve such an accuracy or sensitivity during a future time period of 1 year, 2 years, 3 years, 4 years, 5 years; at least 1 year, at least 2 years, at least 3 years, at least 4 years, at least 5 years; or greater than 1 year, greater than 2 years, greater than 3 years, greater than 4 years, or greater than 5 years. ALZAI-002-PCT

[0344] 2025-10-10

[0345]

[0298] Technological Advantages. These results will demonstrate multiple technological improvements, including reduced data-processing requirements, minimized dependence on heterogeneous data integration pipelines, and lower computational cost during both training and inference. By showing that predictive accuracy can be maintained using standardized laboratory data already stored in EMR systems, the disclosure enables easier clinical deployment, improved reproducibility, and broader applicability across healthcare systems lacking multimodal datasets. These findings will also support that the disclosed methods and systems provide a technical improvement in data efficiency and predictive modeling, enabling earlier disease-risk detection with reduced infrastructure complexity compared to existing multimodal diagnostic approaches.

[0346] Example 2: Predicting a Health-Related Risk, Behavior, or Susceptibility Using Blood Test Data

[0347]

[0299] In another example, predictive models are trained and evaluated to determine a predicted risk score representing the likelihood, behavioral propensity, or susceptibility of a subject to experience a health-related risk, behavior, or negative outcome. Such risks, behaviors, or susceptibilities may include predisposition to unhealthy behaviors (e.g., smoking relapse, alcohol misuse, sedentary behavior), elevated physiological risk (e.g., metabolic syndrome, cardiovascular risk), or other negative health-related outcomes described herein.

[0348]

[0300] The modeling framework, training configurations, and comparative evaluation procedures correspond generally to those described in Example 1 , with a principal distinction that the output variable here represents a risk or behavioral susceptibility rather than the onset or progression of a diagnosed medical condition. A predictive model may comprise any of a decision tree, random forest, gradient-boosting model (e.g., XGBoost), support vector machine (SVM), or neural network such as a convolutional or deep-learning model.

[0349]

[0301] As in Example 1 , simulated retrospective cohorts are constructed comprising several thousand control subjects and several thousand subjects identified as having exhibited, or later exhibiting, a health-related risk, behavior, or susceptibility (collectively, “affected subjects”). Multiple configurations of training data are also likewise evaluated:

[0350]

[0302] Configuration T. A heterogeneous multimodal dataset including EMR data, laboratory panels, physiological and sociodemographic data, behavioral and lifestyle information, imaging data, and optionally omics or wearable-derived data;

[0351]

[0303] Configuration 2: EMR data including blood test data and additional parameter values (e.g., physiological, behavioral, sociodemographic, and medical factors) as described in Example 1 ; and

[0352]

[0304] Configuration 3 Blood test data stored as part of EMR data, comprising any subset or combination of the analytes and laboratory panels set forth above.

[0353]

[0305] Historical data reflect measurements taken up to 10 years before identification of the health-related risk or behavior in affected subjects, and up to 10 years before a matched reference time point in control subjects. As described in Example 1, features may be structured as time-series sequences, static snapshots, or derived longitudinal statistics (e.g., averages, extrema, variances, slopes, and temporal deltas).

[0354]

[0306] Predictive models are trained under both knowledge-weighted and knowledge-excluded conditions. In ALZAI-002-PCT 2025-10-10 the knowledge-weighted configuration, features conventionally associated with the risk or behavior are given higher priority based on prior clinical knowledge (for instance, HDL / LDL ratio for cardiovascular risk or HbA1c for metabolic susceptibility). In the knowledge-excluded configuration, such conventional variables are down-weighted or omitted to evaluate whether the model autonomously identifies counterintuitive predictors.

[0355]

[0307] Comparative Results. Across all configurations, models trained solely on blood test data (Configuration 3) will achieve predictive accuracy closely approaching or for some medical conditions equivalent to that of the heterogeneous multimodal dataset (Configuration 1), again demonstrating that routinely collected laboratory data can encode sufficient latent information to forecast not only clinical outcomes but also behavioral or susceptibility-related risks. The inclusion of additional EMR variables (Configuration 2) may for some conditions provide modest improvement in calibration and interpretability but will not materially alter predictive performance. Inclusion of broad heterogeneous data (Configuration 1) increases computational and data-integration complexity without commensurate gain in accuracy.

[0356]

[0308] Feature Importance and Temporal Dynamics. For each configuration, feature importance rankings are computed to quantify the relative contribution of individual blood test values and derived temporal features. The most informative features may vary across different prior observation windows (e.g., one-, five-, or ten-year periods) and across prediction horizons (e.g., one- to five-year forecasts). For example, longitudinal variability in markers such as HbA1c, triglycerides, or C-reactive protein may carry stronger predictive weight than single time-point measurements, reflecting evolving physiological risk states.

[0357]

[0309] Performance and Technical Advantage. In one representative implementation, a gradient-boosting model trained under Configuration 3 will correctly classify greater than about 70% of subjects who later manifest a targeted risk or behavioral susceptibility within a one-year horizon, based solely on blood test data from prior years. In some embodiments, predictive accuracy or sensitivity ranges from at least 20% to 99%, depending on condition type, feature subset, and observation period. These findings confirm that high-accuracy risk and behavior prediction can be achieved without reliance on multimodal or high-dimensional data integration, providing a technical improvement in data efficiency, computational scalability, and deployability across healthcare and consumer-health contexts.

[0358]

[0310] Together, Examples 1 and 2 illustrate that the disclosed framework and the methods and systems of its implementation and use will be broadly applicable to both clinical and preclinical contexts: it will predict not only the onset or progression of defined medical conditions but also latent health-related risks, behavioral tendencies, or physiological susceptibilities preceding overt disease. Across these domains, the results collectively will demonstrate that high-accuracy prediction is achievable using standard blood test data alone, providing a scalable, interpretable, and computationally efficient alternative to existing multimodal approaches.

[0359] Example 3: Predicting the Onset and Progression of a Disease in a Non-Human Animal

[0360]

[0311] In a further example, predictive models are trained and evaluated to determine a predicted risk score representing the likelihood, onset risk, or progression probability of a disease in a non-human animal. ALZAI-002-PCT

[0361] 2025-10-10

[0362]

[0312] The non-human animal may be any mammal, avian, or other vertebrate species for which biological samples, veterinary records, or health-monitoring data are available, including, for example, companion animals (e.g., dogs, cats, horses), livestock (e.g., cattle, swine, sheep, goats, poultry), animals raised for food (e.g., fish, crustaceans), and research animals (e.g., rodents, non-human primates).

[0363]

[0313] Each predictive model may comprise any of a decision tree, random forest ensemble, gradient-boosting model (e.g., XGBoost), support vector machine (SVM), or neural network such as a convolutional or deep-learning model. The modeling framework, data-processing structure, and comparative evaluation correspond generally to those described in Example 1 , except that veterinary and species-specific data are used in place of human clinical data.

[0364]

[0314] Simulated retrospective cohorts are constructed comprising several thousand control animals and several thousand animals diagnosed with or later developing a disease (“affected animals”). Multiple training configurations are evaluated:

[0365]

[0315] Configuration 1: A heterogeneous dataset comprising multimodal veterinary data, including structured records (e.g., veterinary EMR analogs), laboratory panels, imaging data (e.g., radiographs, ultrasound, MRI, CT, PET), physiological monitoring data, behavioral observations, dietary intake logs, and activity-tracking or wearable sensor data.

[0366]

[0316] Configuration 2: Veterinary records including blood test data and one or more additional or alternative parameter values, including physiological parameters (e.g., heart rate, weight, body-mass index (BMI) or equivalent species-specific indices), vascular risk factors or metabolic indicators (e.g., hypertension, hyperlipidemia, diabetes mellitus, metabolic syndrome), demographic parameters (e.g., age, sex, breed), behavioral parameters (e.g., activity frequency, feeding patterns, stress or aggression scores), and medical parameters (e.g., background disease, vaccination history, administered medication).

[0367]

[0317] Configuration 3 Blood test data stored as part of veterinary records, comprising any subset or combination of analytes and laboratory values as set forth in Example 1 .

[0368]

[0318] Historical data reflect measurements taken up to 10 years before diagnosis of a disease in affected animals, and up to 10 years before a corresponding reference time point for control animals. The reference time point may be matched to diagnosis dates of affected animals to ensure comparable temporal distributions across cohorts. “10 years” also will be understood to be adjusted in view of the lifespan of the animal(s). As described in Example 1 , features may be structured as time-series sequences, static snapshots, or derived longitudinal statistics (e.g., means, extrema, variances, slopes, or temporal deltas).

[0369]

[0319] Predictive models are trained under both knowledge-weighted and knowledge-excluded conditions. In the knowledge-weighted configuration, features conventionally associated with the disease under veterinary understanding (e.g., elevated liver enzymes for hepatic disorders, altered hematocrit for anemia, or abnormal renal markers for kidney disease) are given higher priority or initial weighting. In the knowledge-excluded configuration, such variables are down-weighted or omitted to test whether the model autonomously identifies ALZAI-002-PCT

[0370] 2025-10-10 unconventional or cross-system predictors.

[0371]

[0320] Comparative Results. Across all configurations, models trained solely on blood test data (Configuration 3) will achieve predictive accuracy approaching that of or equivalent to the full heterogeneous dataset (Configuration 1), with only marginal gains when multimodal or behavioral data are incorporated. These findings will indicate that standard blood test data alone can capture latent physiological signatures sufficient to predict onset or progression of a disease in non-human animals, even across diverse species. The inclusion of broader veterinary record data (Configuration 2) will only modestly improve calibration and interpretability but will substantially increase data-collection complexity.

[0372]

[0321] Feature Importance and Model Output. For each configuration, feature-importance scores quantify the contribution of individual laboratory values to predictive performance. Temporal dynamics (e.g., multi-year changes in albumin, ALT, or creatinine) may show stronger predictive power than single-time-point measurements. The model output includes (i) a predicted risk score representing probability of disease onset or progression, and (ii) a class label derived from a configurable threshold (e.g., 0.5). In some embodiments, a progression index may also be generated to estimate the rate of disease advancement.

[0373]

[0322] Performance and Technical Advantages. In one representative implementation, a gradient-boosting model trained under Configuration 3 correctly identifies greater than about 70% of animals that subsequently experience disease onset or progression within a one-year period, based on blood test data collected up to 10 years prior. In embodiments, predictive accuracy or sensitivity may range from at least 20% to 99%, depending on species, disease type, and temporal window. These findings demonstrate that the disclosed methods and systems are not limited to human clinical data and can be generalized to veterinary or experimental settings, offering improvements in data efficiency, early detection, and cross-species disease modeling.

[0374]

[0323] Together with Examples 1 and 2, this example will demonstrate that the disclosed predictive framework operates effectively across humans and non-human animals alike, capturing shared physiological principles reflected in standard laboratory measurements. The ability to achieve accurate prediction using only routine blood test data across species represents a significant and unexpected simplification over existing approaches that require species-specific multimodal data integration or behavioral monitoring.

[0375] Example 4: Integration with EMR (or EHR) Systems to Predict the Onset of Conditions

[0376]

[0324] In another example, the disclosed methods and systems are implemented as a software service integrated with an electronic medical record (EMR) or electronic health record (EHR) system. Integration may occur directly within a medical facility’s computing environment (e.g., clinic, hospital, or health network) or through a secure application programming interface (API) linking the predictive system to the EMR database.

[0377]

[0325] The integrated system comprises code configured to access, extract, and process patient data stored within the EMR system, including but not limited to laboratory results, blood test values, vital signs, demographic data, diagnostic codes, medication history, and clinical encounter summaries. The system may operate in conjunction with one or more predictive models trained according to the embodiments described ALZAI-002-PCT

[0378] 2025-10-10 herein to predict the onset and / or progression of one or more medical conditions.

[0379]

[0326] Each predictive model may draw upon the same underlying EMR data but apply model-specific feature sets optimized for different conditions. For example, separate models may be trained to predict the onset or progression of hypertension, diabetes mellitus, ischemic heart disease, or myocardial infarction, each leveraging condition-specific subsets of biomarkers and clinical attributes. The system can execute multiple predictive models concurrently or sequentially, aggregating their outputs to produce individualized predicted risk scores and condition-specific progression indices for each subject.

[0380]

[0327] In some embodiments, the system operates passively, continuously monitoring the EMR system for new or updated data entries without requiring active physician input. Upon ingestion of new data (e.g., a recently added laboratory result or updated vital sign), the system automatically retrains or recalculates a subject's predicted risk score. Updated predictions are stored back into the EMR database, linked to the corresponding patient record, and optionally surfaced within the physician’s EMR dashboard or alert interface.

[0381]

[0328] Computed risk scores and corresponding explanatory outputs (e.g., most influential features or time-series trends) are made available for review by authorized medical professionals, enabling rapid identification of subjects at elevated risk for specific conditions. In some embodiments, the system generates automated summaries or notifications highlighting newly elevated risk categories or statistically significant changes in predictive trajectories.

[0382]

[0329] The system may also maintain longitudinal patient models, updating predictions over time as additional data become available, thereby allowing dynamic monitoring of risk trajectories and treatment responses. For instance, following the initiation of a prescribed treatment or lifestyle intervention, the system can track changes in the predicted risk score, quantify improvements, or flag persistent or worsening indicators for further clinical evaluation. In some embodiments, the system further communicates results to auxiliary clinical systems (e.g., scheduling, triage, or care-management software) to prompt proactive interventions, follow-up testing, or treatment adjustments.

[0383]

[0330] From an implementation standpoint, integration with EMR or EHR systems, as well as with other auxiliary clinical systems, will provide several technological improvements, including: (1) automated data synchronization, eliminating manual data entry and reducing clinician workload; (2) continuous longitudinal modeling, enabling time-aware prediction updates as new data accumulate; (3) resource efficiency, as the models can operate on existing EMR data without additional diagnostic instrumentation; and (4) enhanced decision support, by embedding interpretable predictive outputs directly within the clinician’s workflow. These and other improvements will collectively enable predictive analytics to function as a seamless extension of existing EMR infrastructures, enhancing clinical decision-making, operational efficiency, and preventive care.

[0384]

[0331] Thus are provided not only predictive models capable of identifying early disease risk from minimal data inputs, but also system architectures designed for real-world deployment in healthcare environments, ensuring continual, automated, and interpretable risk assessment within the clinical record ecosystem. ALZAI-002-PCT 2025-10-10

[0385] Example 5: Predicting the Onset, Progression, or Exacerbation of Chronic Lung Disease (e.g., COPD)

[0386]

[0332] In this example, predictive models are trained to determine a predicted risk score representing the likelihood, onset risk, or exacerbation probability of a chronic lung disease, such as chronic obstructive pulmonary disease (COPD), emphysema, or chronic bronchitis. The framework parallels that described in Example 1 but is specialized for pulmonary conditions. Multiple configurations of training data are evaluated to assess the sufficiency and comparative performance of different feature sets:

[0387]

[0333] Configuration 1: A heterogeneous dataset comprising multimodal inputs such as EMR data, laboratory panels, spirometry results, imaging data (e.g., CT or chest X-ray studies), wearable sensor data (e.g., oxygen saturation, respiration rate, accelerometry), behavioral and environmental exposure data (e.g., smoking history, air quality index (AQI), occupational hazards), and unstructured clinical notes.

[0388]

[0334] Configuration 2: EMR data including blood test data and other structured parameters such as physiological measurements (e.g., heart rate, oxygen saturation, respiratory rate, blood pressure), demographic and behavioral data (e.g., age, gender, smoking status, physical activity level), and medical parameters (e.g., comorbid cardiovascular disease or medication use).

[0389]

[0335] Configuration 3 Blood test data stored within EMR data, optionally combined with demographic variables such as age and sex and others such as disclosed herein and appreciated by those of skill.

[0390]

[0336] Historical data reflect measurements taken up to 10 years before diagnosis or documented exacerbation of COPD in affected subjects, and up to 10 years before a matched reference point in control subjects. Data are structured as time-series sequences, static snapshots, or derived longitudinal features (e.g., slopes, deltas, or multi-year averages).

[0391]

[0337] Predictive models include gradient-boosting frameworks (e.g., XGBoost), random forests, supportvector machines, and deep-learning architectures. Each model outputs a continuous predicted risk score and a binary or multiclass label (e.g., no disease, stable disease, progressing, or at-risk for exacerbation).

[0392]

[0338] The models are trained under knowledge-weighted and knowledge-excluded conditions. Under the former, clinically recognized pulmonary indicators (e.g., pulmonary-function metrics or prior respiratory diagnoses) are given higher weight; under the latter, such features are excluded or down-weighted to allow autonomous discovery of counterintuitive or cross-domain predictors.

[0393]

[0339] Comparative evaluation will reveal that, as in earlier examples, blood test-only datasets (Configuration 3) achieve predictive accuracy approaching that of multimodal datasets, with minimal loss of performance but substantially reduced computational and data-collection overhead.

[0394]

[0340] The models further identify emergent relationships among heterogeneous parameters, such as interactions between systemic inflammatory markers, longitudinal oxygen-saturation variability, and physical activity trends, that contribute disproportionately to predictive power despite lacking conventional pulmonary interpretation. Feature-attribution modules highlight these latent variables for expert review, facilitating discovery of novel correlates of lung-function decline. ALZAI-002-PCT

[0395] 2025-10-10

[0396]

[0341] Clinical Application and Treatment Selection. The system may operate in conjunction with an EMR or EHR interface as described in Example 4. When the predicted risk score for COPD onset or exacerbation exceeds a configurable threshold, the system may automatically generate a risk-stratification report within the subject's medical record. This report may recommend review by a clinician and suggest evidence-based interventions tailored to disease stage and risk profile.

[0397]

[0342] In some embodiments, the clinician may initiate or modify any one or more of: (1) pharmacologic interventions, such as inhaled bronchodilators, corticosteroids, or phosphodiesterase inhibitors; (2) non-pharmacologic interventions, such as pulmonary rehabilitation, smoking-cessation counseling, air-quality control, or structured exercise programs; and / or (3) follow-up diagnostics, such as repeat spirometry, chest imaging, or home oxygen monitoring.

[0398]

[0343] Following intervention, the system may continue longitudinal surveillance, which may be performed passively, such as by recalculating the predicted risk score as new EMR data or laboratory results become available. Downward trends in the predicted risk score following intervention may be stored as indicators of treatment response, and may be used to provide feedback for therapy optimization.

[0399]

[0344] Results and Advantages. Across the evaluated configurations, the blood test-only model will achieve predictive accuracy comparable or equivalent to that of models using full heterogeneous inputs, revealing that systemic biochemical trends alone encode sufficient information to anticipate pulmonary decline or exacerbation risk. The ability to perform continuous, automated COPD risk monitoring using only standard EMR-available data provides a significant improvement in scalability, interpretability, and cost efficiency over traditional models requiring specialized pulmonary instrumentation or clinician-driven data entry.

[0400] Example 6: Predicting the Onset, Progression, or Decompensation of Congestive Heart Failure (CHF)

[0401]

[0345] In this example, predictive models are trained to determine a predicted risk score representing the likelihood, onset risk, or decompensation probability of congestive heart failure (CHF) or related cardiac dysfunction. The modeling framework parallels that described in Example 1 and Example 5, adapted to cardiac pathophysiology and longitudinal cardiovascular monitoring. Multiple configurations of training data are evaluated to assess the contribution of heterogeneous versus minimal input features:

[0402]

[0346] Configuration 1 A heterogeneous dataset comprising EMR data, laboratory panels, cardiac imaging data (e.g., echocardiography, MRI, CT), electrocardiographic data, hemodynamic parameters (e.g., blood pressure, pulse pressure, heart rate, heart-rate variability), wearable sensor or telemonitoring data (e.g., thoracic impedance, daily weight, physical-activity level), and structured or unstructured clinical notes reflecting symptoms such as orthopnea or dyspnea.

[0403]

[0347] Configuration 2: EMR data including blood test data and structured physiological measurements (e.g., heart rate, blood pressure, weight, oxygen saturation), demographic and behavioral attributes (e.g., age, sex, exercise level, medication adherence), and relevant comorbidities or medical treatments.

[0404]

[0348] Configuration 3 Blood test data stored within EMR data, optionally combined with demographic ALZAI-002-PCT 2025-10-10 variables such as age and sex and others such as disclosed herein and appreciated by those of skill.

[0405]

[0349] Historical data reflect measurements taken up to 10 years prior to the first recorded CHF diagnosis or decompensation event for affected subjects, and up to 10 years prior to a matched reference time point for control subjects. Features may be structured as time-series sequences, static snapshots, or derived longitudinal statistics (e.g., means, extrema, slopes, or temporal deltas).

[0406]

[0350] Predictive models may include gradient-boosting frameworks (e.g., XGBoost), random forests, support-vector machines, and deep-learning architectures. Each model outputs a continuous predicted risk score and a binary or multi-class label (e.g., no disease, stable CHF, decompensating).

[0407]

[0351] Training will be performed under both knowledge-weighted and knowledge-excluded conditions. Under the former, variables historically linked to cardiac decompensation are prioritized; under the latter, such features are excluded or down-weighted to enable autonomous discovery of unconventional predictors. Across configurations, models trained solely on blood test data (Configuration 3) demonstrate predictive accuracy approaching that of multimodal datasets, showing that systemic biochemical and metabolic patterns alone encode substantial information relevant to cardiac dysfunction.

[0408]

[0352] The models autonomously identify emergent cross-domain relationships, such as covariation between physiological stability indicators and derived behavioral or environmental metrics, which together exert a stronger effect on the predicted risk score than any single input. Feature-attribution analyses highlight latent temporal patterns and inter-parameter interactions that improve calibration and discrimination beyond conventionally recognized markers.

[0409]

[0353] Clinical Application and Treatment Selection. When deployed within an EMR or EHR environment as described in Example 4, the system automatically updates and records a subject's predicted CHF risk each time new data become available. If the predicted risk score exceeds a configurable threshold, the system generates a clinical alert or report summarizing the underlying contributing factors and trends. The treating clinician may then select or adjust one or more therapeutic interventions based on the predicted risk profile.

[0410]

[0354] In some embodiments, the clinician may initiate or modify any one or more of: (1) pharmacologic treatments, such as diuretics, angiotensin-converting-enzyme inhibitors, angiotensin-receptor blockers, beta-blockers, or mineralocorticoid-receptor antagonists; (2) device-based or non-pharmacologic interventions, such as cardiac resynchronization therapy, implantable defibrillators, fluid-management protocols, or dietary-sodium restriction; and / or (3) lifestyle or monitoring strategies, such as weight tracking, exercise regimens, or telehealth follow-up scheduling.

[0411]

[0355] Following intervention, the system may continue longitudinal surveillance, automatically recalculating the predicted risk score and trending trajectories to quantify treatment response. Reductions in predicted risk over successive evaluations may be logged as indicators of effective therapy, whereas sustained or worsening trajectories can trigger escalation or follow-up recommendations.

[0412]

[0356] Results and Advantages. Across all configurations, the blood test-only model yields predictive ALZAI-002-PCT 2025-10-10 accuracy comparable to multimodal input configurations, illustrating that early signals of cardiac stress and fluid imbalance are encoded in systemic laboratory and metabolic data accessible from standard EMR systems. The integration of such predictive modeling within routine clinical infrastructure enables continuous, automated, and interpretable monitoring of heart failure risk without requiring specialized cardiac instrumentation or manual data collection, providing measurable improvements in scalability, preventive care, and clinical workflow efficiency.

[0413] Example 7: Predicting the Onset, Progression, or Adverse-Event Probability of CKD and ESRD

[0414]

[0357] In this example, predictive models are trained to determine a predicted risk score representing the likelihood, rate of progression, or probability of an adverse clinical event associated with chronic kidney disease (CKD) or end-stage renal disease (ESRD). The modeling framework parallels that of Examples 1 through 6 but is adapted for renal physiology and longitudinal metabolic decline. Multiple configurations of training data are evaluated to compare predictive performance across progressively simplified input sets:

[0415]

[0358] Configuration 1: A heterogeneous dataset comprising EMR data, laboratory data, physiological measurements, renal imaging or ultrasonography results, wearable or sensor-based measurements of hydration status or blood pressure, and unstructured clinical notes capturing symptom descriptions or treatment adherence indicators.

[0416]

[0359] Configuration 2: EMR data including blood test data and structured physiological or behavioral parameters such as blood pressure, heart rate, fluid intake, weight, medication use, demographic characteristics, and comorbid conditions (e.g., diabetes mellitus, hypertension).

[0417]

[0360] Configuration 3 Blood test data stored within EMR data, optionally combined with demographic variables such as age and sex and others such as disclosed herein and appreciated by those of skill.

[0418]

[0361] Historical data reflect laboratory and physiological measurements collected up to 10 years prior to the first diagnosis of CKD or ESRD in affected subjects, and up to 10 years prior to a matched reference time point in control subjects. Features may be represented as time-series sequences, static snapshots, or derived longitudinal descriptors (e.g., averages, extrema, rates of change, or multi-year deltas).

[0419]

[0362] Predictive models may include gradient-boosting frameworks (e.g., XGBoost), random-forest ensembles, support-vector machines, or neural networks. Each model outputs a continuous predicted risk score and a categorical or binary label (e.g., no disease, stable CKD, progressing, end-stage).

[0420]

[0363] Training will be performed under knowledge-weighted and knowledge-excluded conditions. In the knowledge-weighted configuration, features conventionally associated with renal function are prioritized; in the knowledge-excluded configuration, such variables are removed or down-weighted, allowing the model to autonomously identify alternative predictors of renal decline. Across all configurations, models trained solely on blood test data (Configuration 3) achieve predictive accuracy approaching that of the multimodal dataset (Configuration 1), demonstrating that systemic biochemical and metabolic patterns alone contain sufficient latent information to forecast renal deterioration or dialysis dependence. ALZAI-002-PCT

[0421] 2025-10-10

[0422]

[0364] The models also reveal emergent feature interactions, such as temporal patterns linking electrolyte balance, nutritional status, and physiologic variability, that exert greater influence on the predicted risk score than any individual input. Feature-attribution modules highlight these multi-factor relationships, enabling the identification of novel, previously unrecognized predictors of CKD progression or decompensation.

[0423]

[0365] Clinical Application and Treatment Selection. When integrated with an EMR or EHR system, the predictive system may continuously update each subject's predicted CKD or ESRD risk as new laboratory or clinical data become available. Upon exceeding a configurable threshold, the system may generate an automated report summarizing risk trends and key contributing factors. A treating clinician may select or adjust one or more therapeutic interventions tailored to the subject’s disease stage and predicted progression rate.

[0424]

[0366] In some embodiments, the clinician may initiate or modify any one or more of: (1) pharmacologic interventions, such as renin-angiotensin-aldosterone-system modulators, phosphate binders, erythropoiesisstimulating agents, or bicarbonate supplementation; (2) non-pharmacologic interventions, such as dietary protein or sodium restriction, optimization of hydration and blood pressure control, and referral to a nephrology specialist; and / or (3) preparatory or monitoring measures, such as planning for renal replacement therapy, adjusting dialysis schedules, or deploying home-based monitoring devices for fluid balance.

[0425]

[0367] Following intervention, the system may continue longitudinal surveillance, recalculating the predicted risk score at each data update and recording trends that reflect treatment response or stabilization. Declining risk scores may indicate therapeutic efficacy, while persistently elevated or increasing scores can trigger automated recommendations for earlier intervention or escalated follow-up.

[0426]

[0368] Results and Advantages. Comparative evaluation demonstrates that the blood test-only configuration yields predictive accuracy comparable to multimodal datasets, revealing that systemic laboratory parameters alone can effectively anticipate renal decline and adverse-event risk. Integrating such predictive analytics within routine clinical workflows enables automated, continuous, and interpretable monitoring of kidney function, providing a scalable and resource-efficient framework for early detection, risk stratification, and individualized treatment optimization.

[0427] Example 8: Predicting the Onset, Progression, or Complication Risk of Diabetes Mellitus

[0428]

[0369] In this example, predictive models are trained to determine a predicted risk score representing the likelihood, progression rate, or complication probability of diabetes mellitus, including microvascular and macrovascular complications such as nephropathy, neuropathy, and cardiovascular disease. The modeling framework parallels that of Examples 1 through 7 but is adapted to glycemic control and metabolic homeostasis. Multiple configurations of training data are evaluated to compare performance across heterogeneous and minimal input sets:

[0429]

[0370] Configuration 1: A heterogeneous dataset comprising EMR data, laboratory data, continuous glucose monitoring (CGM) data, wearable sensor metrics (e.g., physical activity, sleep quality, heart-rate variability), medication, and unstructured clinical notes describing symptoms, lifestyle factors, or treatment adherence. ALZAI-002-PCT

[0430] 2025-10-10

[0431]

[0371] Configuration 2: EMR data including blood test data and structured physiological and behavioral parameters such as fasting glucose, weight, blood pressure, age, sex, physical activity level, and medications.

[0432]

[0372] Configuration 3 Blood test data stored within EMR data, optionally combined with demographic variables such as age and sex and others such as disclosed herein and appreciated by those of skill.

[0433]

[0373] Historical data reflect measurements collected up to 10 years before the first recorded diabetes diagnosis or complication event in affected subjects and up to 10 years before a matched reference time point for control subjects. Data may be structured as time-series sequences, static snapshots, or derived longitudinal descriptors (e.g., multi-year means, slopes, deltas, or measures of intra-individual variability).

[0434]

[0374] Predictive models may include gradient-boosting frameworks (e.g., XGBoost), random-forest ensembles, support-vector machines, or neural-network architectures. Each model outputs a continuous predicted risk score and a categorical or binary label (e.g., no disease, stable diabetes, progressing, complication likely).

[0435]

[0375] Training will be conducted under knowledge-weighted and knowledge-excluded conditions. In the knowledge-weighted configuration, features conventionally associated with diabetic control (e.g., glucose or lipid metrics) are prioritized; in the knowledge-excluded configuration, these are removed or down-weighted to allow the system to autonomously identify non-traditional predictors. Comparative evaluation demonstrates that models trained solely on blood test data (Configuration 3) achieve predictive performance approaching that of multimodal datasets (Configuration 1), confirming that systemic biochemical patterns alone capture sufficient latent information to forecast glycemic instability and complication risk.

[0436]

[0376] The models further identify emergent relationships among heterogeneous domains, such as correlations between metabolic variability, inflammation indices, and physical-activity fluctuations, which together exert a greater collective influence on the predicted risk score than any single factor. Feature-attribution modules highlight such latent predictors, enabling discovery of novel or cross-domain signatures of disease progression.

[0437]

[0377] Clinical Application and Treatment Selection. When integrated with an EMR or EHR system, the predictive system may continuously update each subject's risk score as new laboratory or sensor data become available. When the predicted risk score exceeds a configurable threshold, an automated risk-stratification report may be generated and stored within the patient record. The treating clinician may then select or modify one or more interventions responsive to the identified risk profile and predicted complication trajectory.

[0438]

[0378] In some embodiments, the clinician may initiate or modify any one or more of: (1) pharmacologic interventions, such as initiation or titration of glucose-lowering agents, lipid-lowering therapies, or antihypertensive medications; (2) non-pharmacologic interventions, such as medical nutrition therapy, structured exercise or weight-management programs, and patient education on lifestyle modification; and / or (3) monitoring strategies, such as implementation of continuous glucose monitoring, foot-care programs, or ophthalmologic screening to detect early complications. ALZAI-002-PCT

[0439] 2025-10-10

[0440]

[0379] Following intervention, the system may continue longitudinal surveillance, recalculating predicted risk scores and generating trend reports that reflect improvement, stability, or deterioration. Downward trends in predicted risk following intervention are recorded as indicators of therapeutic efficacy, while persistent or increasing scores may trigger automated alerts recommending clinician review or escalation of care.

[0441]

[0380] Results and Advantages. Across configurations, the blood test-only predictive model achieves accuracy and sensitivity comparable to that of multimodal models, highlighting that routine laboratory data suffice to anticipate diabetic progression and complication risk. Integration of the disclosed predictive framework within clinical infrastructure enables continuous, data-driven monitoring of disease control, enhances individualized treatment planning, and supports proactive prevention of adverse outcomes through early, model-guided intervention.

[0442] Example 9: Predicting the Onset, Recurrence, or Progression of Stroke and Neurological Disorders

[0443]

[0381] In this example, predictive models are trained to determine a predicted risk score representing the likelihood, onset risk, recurrence probability, or progression rate of stroke or major neurological disorders, including cerebrovascular and neurodegenerative conditions. The modeling framework parallels that of Examples 1 through 8 but is adapted for vascular, metabolic, and neurophysiological data streams relevant to brain health and neural resilience. Multiple configurations of training data are evaluated to assess the contribution of multimodal versus minimal feature sets:

[0444]

[0382] Configuration 1: A heterogeneous dataset comprising EMR data, laboratory data, neuroimaging data (e.g., MRI, CT, PET, or ultrasound imaging of cerebral vasculature), electrophysiologic or biosignal data (e.g., EEG, ECG, or HRV), wearable or sensor-derived metrics (e.g., gait, sleep, speech, or movement asymmetry), and unstructured clinical or behavioral notes.

[0445]

[0383] Configuration 2: EMR data including blood test data and structured physiological, behavioral, and demographic variables such as blood pressure, heart rate, body-mass index, smoking status, physical-activity level, medication use, and prior cardiovascular or neurological diagnoses.

[0446]

[0384] Configuration 3 Blood test data stored within EMR data, optionally combined with demographic variables such as age and sex and others such as disclosed herein and appreciated by those of skill.

[0447]

[0385] Historical data reflect measurements obtained up to 10 years prior to the first stroke or neurological diagnosis in affected subjects and up to 10 years prior to a matched reference point in control subjects. Data may be represented as time-series sequences, static snapshots, or derived longitudinal descriptors (e.g., multi-year means, extrema, slopes, or temporal deltas).

[0448]

[0386] Predictive models may include gradient-boosting frameworks (e.g., XGBoost), random-forest ensembles, support-vector machines, or neural-network architectures. Each model outputs a continuous predicted risk score and a categorical or binary label (e.g., no event, at risk, recurrent event predicted).

[0449]

[0387] Training will be performed under knowledge-weighted and knowledge-excluded conditions. In the knowledge-weighted configuration, variables traditionally associated with cerebrovascular or neurological ALZAI-002-PCT 2025-10-10 disease are prioritized. In the knowledge-excluded configuration, those same features are down-weighted or excluded, allowing the system to autonomously identify unconventional or cross-domain predictors. Across configurations, models trained solely on blood test data (Configuration 3) achieve predictive accuracy approaching that of multimodal input sets, suggesting that systemic biochemical and metabolic patterns alone encode latent indicators of neurovascular vulnerability.

[0450]

[0388] The models will autonomously identify emergent feature interactions linking vascular, metabolic, and behavioral domains (for example, relationships between autonomic variability, sleep irregularity, and inflammatory load) that exert greater combined influence on the predicted risk score than any single factor. Feature-attribution and interpretability analyses highlight these latent temporal and cross-modal relationships, revealing previously unrecognized predictors of cerebrovascular events or neurodegenerative progression.

[0451]

[0389] Clinical Application and Treatment Selection. When integrated within an EMR or EHR environment as described in Example 4, the predictive system continuously updates each subject's neurological-risk profile as new laboratory, imaging, or wearable data are ingested. When a predicted risk score exceeds a configurable threshold, the system automatically generates a report summarizing contributing features, longitudinal trends, and recommended follow-up intervals. The treating clinician may then select or modify one or more interventions to mitigate stroke or neurological-disease risk and enhance recovery or prevention outcomes.

[0452]

[0390] In some embodiments, the clinician may initiate or modify any one or more of: pharmacologic interventions, such as antiplatelet or anticoagulant therapy, lipid-lowering or antihypertensive agents, or neuroprotective medications; (2) non-pharmacologic interventions, such as dietary modification, smoking cessation, exercise programs, cognitive training regimens, or sleep-apnea management; and / or (3) monitoring and rehabilitation strategies, such as speech therapy scheduling, gait-analysis follow-up, or integration with remote neurological monitoring tools.

[0453]

[0391] Following intervention, the system may continue longitudinal surveillance, recalculating the predicted risk score upon acquisition of new data and plotting changes as indicators of therapeutic efficacy or deteriorating status. Declining risk scores are recorded as markers of successful intervention, whereas increasing scores may automatically prompt alerts recommending further evaluation or escalation of care.

[0454]

[0392] Results and Advantages. Across all configurations, the blood test-only model demonstrates predictive accuracy and sensitivity comparable to that of multimodal datasets, illustrating that early neurovascular risk signatures can be inferred from routinely collected laboratory data. Integration of the disclosed predictive modeling within existing clinical infrastructure enables continuous, interpretable, and cost-efficient neurological risk monitoring, providing a technological improvement over episodic or symptom-triggered assessment methods by facilitating proactive prevention and individualized care management.

[0455] Example 10: Predicting the Onset, Progression, or Complication Risk of HIV and AIDS

[0456]

[0393] In this example, predictive models are trained to determine a predicted risk score representing the likelihood of infection, progression rate, or complication probability associated with human immunodeficiency ALZAI-002-PCT 2025-10-10 virus (HIV) infection or acquired immunodeficiency syndrome (AIDS). The modeling framework parallels that of Examples 1 through 9 but is adapted to immunologic, virologic, and behavioral data relevant to disease control and treatment outcomes. Multiple configurations of training data are evaluated to assess the performance and sufficiency of multimodal versus simplified feature sets:

[0457]

[0394] Configuration 1: A heterogeneous dataset comprising EMR data, laboratory data, immunologic profiles, medication-adherence telemetry, wearable or digital health data (e.g., HRV, activity levels, sleep indices), and unstructured clinical notes describing symptoms, adherence history, or psychosocial factors.

[0458]

[0395] Configuration 2: EMR data including blood test data and structured physiological, behavioral, and demographic parameters such as medications, adherence frequency, age, sex, and comorbid infection status.

[0459]

[0396] Configuration 3: Blood test data stored within EMR data, optionally combined with demographic variables such as age and sex and others such as disclosed herein and appreciated by those of skill.

[0460]

[0397] Historical data reflect laboratory and behavioral measurements obtained up to 10 years prior to the first recorded clinical progression event or complication, and up to 10 years prior to a matched reference point in control subjects. Data may be structured as time-series sequences, static snapshots, or derived longitudinal descriptors (e.g., multi-year means, extrema, slopes, or temporal deltas).

[0461]

[0398] Predictive models may include gradient-boosting frameworks (e.g., XGBoost), random-forest ensembles, support-vector machines, or neural-network architectures.

[0462]

[0399] Each model may output a continuous predicted risk score and a categorical or binary label (e.g., no infection, controlled infection, progressing, at risk of complication).

[0463]

[0400] Training will be performed under knowledge-weighted and knowledge-excluded conditions. In the knowledge-weighted configuration, variables conventionally associated with viral replication or immune decline are prioritized; in the knowledge-excluded configuration, such variables are removed or down-weighted, allowing the system to autonomously identify non-traditional predictors. Comparative evaluation demonstrates that models trained solely on blood test data (Configuration 3) achieve predictive performance approaching that of multimodal datasets (Configuration 1), revealing that systemic biochemical and immunologic patterns alone can encode sufficient latent information to forecast disease progression or treatment failure.

[0464]

[0401] The models uncover emergent feature interactions linking immune function, inflammatory activity, and behavioral engagement metrics. Deviations in immune-cell ratios, inflammatory load, and adherence telemetry may for example together exert greater combined influence on the predicted risk score than any single factor. Feature-attribution and interpretability modules highlight these complex relationships, enabling identification of novel, cross-domain signatures predictive of disease progression, drug resistance, or adverse outcomes.

[0465]

[0402] Clinical Application and Treatment Selection. When integrated with an EMR or EHR system as described in Example 4, the predictive system continuously updates each subject's risk score as new laboratory, behavioral, or adherence data are received. When the predicted risk score exceeds a configurable threshold, the system automatically generates a risk-stratification summary identifying contributing factors and ALZAI-002-PCT 2025-10-10 trends. The treating clinician may then select or adjust one or more therapeutic or supportive interventions tailored to the subject's current immune status, viral dynamics, and behavioral context.

[0466]

[0403] In some embodiments, the clinician may initiate or modify any one or more of: (1) pharmacologic interventions, such as initiation, substitution, or dosage adjustment of antiretroviral therapy (ART), or addition of prophylactic treatments for opportunistic infections; (2) non-pharmacologic interventions, such as adherence counseling, nutritional optimization, or behavioral-health support to mitigate stress or depression that impact treatment maintenance; and / or (3) monitoring and preventive measures, such as increased laboratory-testing frequency, early toxicity screening, vaccination scheduling, or linkage to community-based care programs.

[0467]

[0404] Following intervention, the system may continue longitudinal surveillance, recalculating the predicted risk score at each data update and displaying trend trajectories indicative of stabilization or progression. Decreases in predicted risk following adherence or regimen improvement are recorded as evidence of therapeutic success, whereas increasing or fluctuating scores may automatically trigger clinician alerts recommending further diagnostic review or medication-regimen optimization.

[0468]

[0405] Results and Advantages. Comparative evaluation across configurations reveals that predictive models trained solely on blood test data achieve accuracy comparable to that of multimodal datasets, demonstrating that immunologic and systemic laboratory parameters alone can reliably forecast disease progression, treatment efficacy, and complication risk. Integration of the disclosed predictive modeling within clinical infrastructure provides continuous, interpretable, and resource-efficient monitoring of patient health, enabling proactive adjustment of therapy and early identification of adverse outcomes to improve long-term treatment success and quality of life.

[0469]

[0406] The foregoing examples illustrate the versatility and generalizability of the disclosed methods and systems for predictive modeling across a wide spectrum of diseases, disorders, and health-related risks. Each example will demonstrate that robust prediction of onset, progression, or complication risk can be achieved using systematically comparable configurations of multimodal, EMR-based, and blood test-only datasets, revealing that complex biomedical phenomena can be accurately modeled using even routine clinical inputs.

[0470]

[0407] The examples further show that models trained under knowledge-weighted and knowledge-excluded conditions can autonomously identify both intuitive and counterintuitive predictors, including cross-domain or emergent feature relationships that extend beyond established medical understanding.

[0471]

[0408] Those of skill in the art will appreciate that the principles, architectures, and evaluation frameworks described herein are not limited to the specific medical conditions exemplified above. The disclosed approaches may be applied, without undue experimentation, to additional diseases, disorders, or physiological conditions, including those affecting neurological, cardiovascular, metabolic, autoimmune, infectious, oncologic, or psychiatric systems, and as otherwise disclosed herein, by retraining or adapting the predictive models to relevant datasets. In all such embodiments, the like configuration logic, comparative methodology, and longitudinal monitoring framework as taught herein may be employed to enable early detection, ALZAI-002-PCT 2025-10-10 progression tracking, and treatment optimization across a diverse range of clinical or research applications.

[0472] Example 11 : Exemplary Working Implementation and Experimental Evaluation

[0473]

[0409] To illustrate some embodiments of the disclosed methods and systems, reference is made to the following working implementation, which demonstrates representative data-processing, model-training, and evaluation procedures suitable for prediction of medical conditions based on clinical laboratory data. The following experiments were performed using a dataset involving subjects diagnosed with a representative condition and control subjects, and those of skill in the art will appreciate that the same methodology and computational architecture are applicable to prediction of any of the conditions described herein.

[0474]

[0410] This example and related disclosure are set forth in co-pending International Patent Application No. PCT / IL2024 / 050342, filed April 3, 2024 and published October 10, 2024 as WO2024209468A1, and U.S. Patent Application No. 18 / 910,178, filed October 9, 2024 and published January 30, 2025 as US 2025 / 0037877 A1 , as well as U.S. Provisional Application No. 63 / 456,555, filed April 3, 2023, each of which is incorporated herein by reference in its entirety. Figures referenced below correspond to representative charts, tables, and feature-importance visualizations in the above co-pending applications illustrating embodiments of the disclosed predictive framework, derived from analogous experimental datasets such as described below.

[0475]

[0411] Exemplary experiments were conducted to demonstrate the performance of trained predictive models in forecasting onset of a neurodegenerative condition and to evaluate their improvement over existing and / or traditional prediction methods. Several predictive models (classification models) were evaluated, including a decision tree, a random-forest ensemble comprising a plurality of decision trees, and an XGBoost ML model.

[0476]

[0412] The experiments were based on retrospective analysis of several thousand control subjects and several thousand diagnosed subjects. The diagnosed population was defined according to recognized diagnostic guidelines for dementia due to Alzheimer's disease (AD), as an exemplary neurodegenerative disease (the 2011 National Institute of Aging (NIA) criteria for the the diagnosis of dementia due to AD).

[0477]

[0413] For both diagnosed and control subjects, historical blood-test values represented up to 10 years of longitudinal laboratory data preceding diagnosis. Blood tests with the highest clinical relevance for neurodegenerative risk were selected for inclusion.

[0478]

[0414] In one exemplary experiment, an XGBoost predictive model was trained using blood-test data from 14,249 control subjects and 9,232 diagnosed subjects. The model was adapted and trained to perform binary classification, indicating whether a respective subject was estimated to develop the target condition. Blood tests evaluated included complete blood count (CBC) and standard chemistry-panel parameters (not all subjects had all values), comprising: BASO Abs, EOS Abs, HB, HOT, LYMPH Abs, MCH, MCHC, MOV, MONO Abs, RBC, PCT, PLT, WBC, RDW, ALBUMIN, CALCIUM, CHLORIDE, CREATININE, GLOBULIN, GLUCOSE, MAGNESIUM, PHOSPHORUS, POTASSIUM, PROTEIN, SODIUM, UREA, URIC ACID, AST (GOT), GGT, ALT (GPT), BILIRUBIN TOTAL, TSH, VITAMIN B12, PT, PTT, and INR.

[0479]

[0415] Feature-importance analysis identified a subset of the highest-ranking laboratory values as most ALZAI-002-PCT 2025-10-10 predictive, including calcium, vitamin B12, folic acid, RBC, albumin, creatinine, globulin, glucose, and WBC. A probability threshold of 0.5 was defined for the binary classification to differentiate between subjects predicted to develop the target condition and those predicted not to.

[0480]

[0416] Results summarized in Table 1 illustrate performance of the trained XGBoost model for predicting onset risk at several previous-period durations (e.g., 1 , 5, 10 years) and prediction horizons (e.g., 1-10 years). TABLE 1. Representative performance metrics for disclosed predictive models, illustrating variation in accuracy, precision, recall, and F1 -score as a function of historical medical data length and prediction horizon:

[0481]

[0417] FIGS. 6-9 show feature-importance rankings across varying combinations of history and prediction windows. For example, in a one-year-horizon prediction based on one year of prior data, albumin exhibited the highest importance, whereas glucose ranked lowest (FIG. 6). In a one-year-horizon prediction based on five years of prior data, globulin ranked highest and vitamin B12 lowest (FIG. 8).

[0482]

[0418] For each laboratory test, multiple temporal features were derived, such as the mean, maximum, minimum, standard deviation, range, change between first and last observation, and change between early and late periods. FIG. 9 shows the relative contribution of each input feature to the model’s output (predicted risk score) for a one-year-horizon prediction based on 10 years of history. In that example, the “GLOBULIN_final_mark_diff” feature had the greatest impact, suggesting that longitudinal changes in globulin levels were most strongly associated with disease onset. Other highly influential features included “RBC_final_mark_diff,” “GLOBULIN_final_mark_min,” and “ALBUMIN Jnaljriarkjriin.” Features such as “WBC_final_mark_std” and “HB_final_mark_max” showed a lesser contribution.

[0483]

[0419] The trained model correctly identified 76% of subjects who were later diagnosed within one year, using data collected up to 10 years before diagnosis. A model trained on five years of blood-test data correctly ALZAI-002-PCT

[0484] 2025-10-10 predicted diagnoses five years in advance with 78% sensitivity (accuracy) and 81 % precision.

[0485]

[0420] Extended Evaluation on a Larger Cohort: A subsequent set of experiments was conducted using an expanded longitudinal dataset encompassing 504,219 participants from a defined healthcare provider network. All individuals aged 47 years or older as of January 1 , 2000 were included. Subjects with alternative neurodegenerative diagnoses or non-Alzheimer etiologies (e.g., Parkinson’s disease, stroke, frontotemporal dementia, or Lewy body disease) were excluded. Diagnosed cases were defined by clinical criteria and / or pharmaceutical treatment history with agents such as donepezil, galantamine, rivastigmine, or memantine.

[0486]

[0421] The longitudinal cohort study spanned from 2000 to 2022, and collected data included demographic parameters (birth date, gender, socioeconomic status), recorded vital signs, smoking habits, medication, and laboratory findings. Data preprocessing required consolidation of demographic, laboratory, diagnosis, and prescription datasets into a unified record. Subjects were categorized into four classes: cognitive-healthy controls, subjects with the target neurodegenerative condition (i.e., AD), subjects with cognitive decline not due to that condition, and subjects with mixed or uncertain etiology. Out of the 504,219 total participants, 11.8% were classified as affected AD subjects (59,441), comparable to 10.8% prevalence reported in medical literature; 381 ,754 were classified as cognitive healthy controls; 35,640 were classified as having cognitive decline not due to AD; and 27,384 were classified as AD subjects with prior non-AD diagnosis which may explain their cognitive decline; the latter two classifications were excluded.

[0487]

[0422] To prepare data for ML analysis, one blood-test record per subject per year was generated. Missing data were imputed, categorical variables one-hot encoded, and continuous features scaled. Derived statistics (e.g., means, deltas, standard deviations) were added as engineered features.

[0488]

[0423] Gender distribution was 53.8% female / 46.2% male for cognitive healthy controls and AD subjects together, 52.2% female / 42.8% male for cognitive health controls alone, and 64.2% female / 35.8% male for AD subjects alone. The male to female gender ratio of 1 :1.8 among the AD subjects is consistent with the ratios in the literature. The age distribution of cognitive healthy controls subjects was similar to that of the entire cohort, while the age distribution of AD subjects resembled a normal distribution (see FIG. 10).

[0489]

[0424] Biomarker Subsets and Model Training: Processed laboratory results were separated into two datasets: (i) Common Biomarkers, available for > 90% of the cohort, and (ii) Uncommon Biomarkers, available for « 50%. Classification models were trained to forecast the probability of diagnosis across nine combinations of historical observation periods (1 , 5, 10 years) and prediction horizons (1, 5, 10 years). For example, a 10-year-history / 1 -year-horizon model used 2006-2016 data for a 2017 diagnosis.

[0490]

[0425] Each dataset was split 80:20 into training and test partitions using stratified shuffling (employing the Stratified ShuffleSplit cross-validator technique) to maintain class balance. Confidence thresholds for classification were selected using Youden’s J statistic to optimize sensitivity versus false-positive rate.

[0491]

[0426] 63 total models were developed for each dataset (common and uncommon biomarkers) comprising 7 models x 3 history ranges x 3 horizons. Table 2 provides a summary of the Common Biomarkers dataset, ALZAI-002-PCT 2025-10-10 presenting a proportion of the representative condition diagnosed (AD) for each training and test dataset as part of the exemplary embodiments of this Example.

[0492] TABLE 2. Table of Common Biomarkers Training and Test Dataset Size and Proportion:

[0493]

[0427] TABLE 3 presents performance metrics of the selected models for the Common Biomarkers, in accordance with this Example. The rows of the table show various metrics (accuracy, AUG, precision, recall, and F1 score) for different models across 1, 5, and 10-year horizons. Notably, a model based on 5 years of historical data and a 10-year horizon achieved an accuracy of 0.81 , an AUG of 0.88, a precision of 0.26, a recall of 0.82, F1 scores of 0.4, and a false positive rate of 20%.

[0494] TABLE 3. Table of Common Biomarkers Dataset - Model Performance: ALZAI-002-PCT

[0495] 2025-10-10

[0496]

[0428] These results demonstrate that the disclosed predictive-modeling architecture can achieve high predictive accuracy using only standardized laboratory data collected over routine clinical timeframes, without requiring imaging or genomic modalities. The working implementation confirms that the disclosed methods and systems are technically operable and capable of generating reliable onset and progression predictions for diverse medical conditions. Those of skill in the art, in view of the teachings herein, will appreciate that analogous predictive frameworks, such as disclosed embodiments, when trained on appropriately structured datasets (e.g., hematologic, endocrine, cardiovascular, inflammatory, or metabolic profiles), can achieve similar performance and can inform treatment selection or preventive interventions as described herein.

[0497]

[0429] Those skilled in the art will appreciate that the computational framework, data preprocessing architecture, and predictive modeling methods illustrated in the foregoing working example are not limited to any specific condition or conditions, such as AD or neurodegenerative conditions, but are applicable to other disease classes disclosed herein, including endocrine, metabolic, cardiovascular, infectious, oncologic, and inflammatory disorders, with adaptations made according to the teachings of this disclosure, and the general knowledge in the art available to one of skill. The methodology of feature extraction, longitudinal aggregation, and gradient-boosted modeling can be adapted to alternate sets of biomarkers, physiological parameters, and outcome definitions, without departing from the disclosed principles. All described examples are illustrative of a generalizable predictive-modeling framework rather than a condition- or class-specific implementation.

[0498] I. Exemplary Technological Improvements

[0499]

[0430] The disclosed methods and systems provide multiple technological improvements over conventional approaches to medical data analysis, predictive modeling, and electronic record integration.

[0500]

[0431] In some embodiments for example, disclosed methods and systems transform heterogeneous, longitudinal biomedical data into compact, interpretable feature representations suitable for automated risk modeling, thereby improving computational efficiency, model generalizability, and interpretability. ALZAI-002-PCT

[0501] 2025-10-10

[0502]

[0432] In some embodiments, an improvement arises from the discovery that high-fidelity predictive performance can be achieved using only a subset of routinely available clinical inputs, such as standardized blood test data, rather than requiring high-dimensional multimodal datasets. In such embodiments, this reduction in input dimensionality yields measurable gains in computational efficiency, including decreased data-acquisition requirements, reduced storage and network-transfer demands, faster model training and inference, and lower processor and energy utilization across distributed computing environments.

[0503]

[0433] In some embodiments, predictive performance of the disclosed models exceeds benchmark statistical baselines, such as logistic regression or proportional-hazard models, by at least 10-20% in AUROC or equivalent performance metrics. Such measurable gains in predictive accuracy, sensitivity, and precision demonstrate concrete technological improvements over conventional computational approaches to disease-risk modeling.

[0504]

[0434] In some embodiments, the disclosed architecture improves data-processing efficiency through the integration of multimodal data normalization, temporal feature engineering, and model explainability within a unified computational pipeline. Such enhancements enable the disclosed systems to operate continuously or asynchronously with EMR or EHR infrastructures, automatically updating predictions in response to new data streams without requiring manual preprocessing or clinician intervention. In some embodiments, adaptive resource allocation and distributed-execution architectures dynamically balance computational workloads across local and remote nodes, routing inference to the most efficient processor based on latency, network bandwidth, and hardware utilization.

[0505]

[0435] In some embodiments, a policy engine transforms calibrated risk outputs into actionable therapeutic recommendations by executing multi-objective optimization that accounts for predicted clinical benefit, adverse-event likelihood, cost, and adherence probability. Such policy generation yields deterministic and auditable treatment-selection pathways that reduce manual computation and enable re-execution and traceability across devices and clinical environments.

[0506]

[0436] In some embodiments, the disclosed systems implement counterfactual or causal simulation modules that estimate outcomes under alternative interventions (e.g., pharmacotherapy versus combined lifestyle modification), enabling individualized treatment-effect estimation and dynamic therapy switching. Constrained learners may be employed to ensure clinically valid causal directions and to improve model interpretability and stability relative to unconstrained black-box models.

[0507]

[0437] In some embodiments, uncertainty-aware inference mechanisms compute calibrated confidence or conformal bounds alongside each risk prediction, allowing automatic safety gating, alert suppression, or re-routing of uncertain cases to higher-fidelity inference pipelines. Such uncertainty quantification improves computational efficiency by minimizing unnecessary downstream computations and enhances clinical safety by ensuring predictable and bounded system behavior.

[0508]

[0438] In some embodiments, the disclosed systems support incremental and streaming learning through ALZAI-002-PCT 2025-10-10 bounded-memory state representations and change-data capture from clinical data sources. Shadow-deployment modes evaluate updated models on live data in parallel with production models, automatically rolling back to prior versions if drift or degradation is detected, thereby improving reliability and uptime relative to static batch retraining paradigms.

[0509]

[0439] In some embodiments, federated learning and privacy-preserving training techniques allow distributed model updates across multiple institutions without centralizing protected health information. Secure aggregation, differential-privacy noise injection, and per-feature gradient clipping reduce privacy risk and network-transfer volume while maintaining or improving predictive performance.

[0510]

[0440] In some embodiments, model compression techniques such as pruning, quantization, and distillation enable hybrid cloud-edge execution and on-device inference using limited clinical inputs. The runtime automatically selects an execution target (edge versus cloud) based on device resources and latency requirements, reducing bandwidth and energy consumption while preserving model fidelity.

[0511]

[0441] In some embodiments, optimized data schemas and standards-compliant integration (e.g., HL7-FHIR APIs) enable direct ingestion, transformation, and indexing of structured laboratory and clinical data into EMR-compatible databases. Dynamic caching and incremental-update mechanisms reduce redundant computations and accelerate longitudinal risk recalculation across large-scale record sets, achieving significant performance improvements relative to known static or batch-based methods.

[0512]

[0442] In some embodiments, the disclosed methods and systems also introduce enhancements to graphical user interfaces and data-presentation modules, including dynamically updated biomarker trajectories, interactive feature-importance plots, counterfactual “what-if’ visualizations, and risk-trend dashboards rendered with compressed data payloads. These features enable responsive, data-rich user interactions while maintaining low latency and efficient client-server communication. The user interface further improves interpretability by rendering causal explanations, uncertainty intervals, and intervention rationales derived from the policy engine, allowing clinicians to understand system reasoning without exporting data to external tools.

[0513]

[0443] In some embodiments, the system includes an explainability engine configured to compute and render feature-importance scores, SHAP values, or gain-based rankings. These visualizations enable clinical users to interpret which input features most influenced the model’s prediction and to evaluate counterfactual “what-if’ scenarios by interactively modifying feature values and observing resultant risk changes. Such explainability interfaces improve transparency, reduce cognitive load, and enhance the interpretability of ML-based recommendations within clinical settings.

[0514]

[0444] In some embodiments, the disclosed systems further interface with therapeutic or diagnostic devices, such as extracorporeal blood-filtration systems, neuromodulation hardware, or infusion pumps, in a closed- loop configuration. Model outputs can parameterize therapy modalities, timing, or dosing (e.g., initiation of plasmapheresis, DFPP, or hemoadsorption based on predicted cytokine kinetics or hemodynamic tolerance). Safety constraints are enforced in real time prior to command execution, providing a computer-controlled ALZAI-002-PCT 2025-10-10 feedback mechanism that dynamically adapts treatment protocols based on physiological data.

[0515]

[0445] In some embodiments, adaptive thresholding and resource-aware alerting modules automatically tune alert levels according to institutional constraints, such as available personnel or device capacity, thereby optimizing workload distribution and minimizing false alarms. In some embodiments, the system employs real-time drift detection, fairness monitoring, and cryptographically signed model-version logs to enable reproducibility, governance, and compliance auditing.

[0516]

[0446] The disclosed systems achieve a technical improvement in computational efficiency by reducing feature dimensionality through automated relevance scoring, enabling real-time inference on standard clinical hardware. The integration of interpretable ML outputs into actionable treatment recommendations constitutes a further technological improvement by transforming abstract data correlations into operational clinical guidance.

[0517]

[0447] Across the exemplary embodiments described herein, disclosed methods and systems demonstrate that accurate and computationally efficient prediction of disease onset, progression, and health-related risks can be achieved using minimal standardized data. Whether applied to human or non-human subjects, these systems generalize across species, data modalities, and predictive objectives, revealing that routine clinical laboratory values may encode sufficient latent information for robust early prediction. This discovery enables cross-domain translation, reduced data-acquisition burden, and substantial improvements in scalability, accessibility, and clinical utility compared with prior high-dimensional or multimodal approaches.

[0518]

[0448] The demonstrated working implementation of Example 11 and the inventors’ co-pending applications further provides empirical validation of the disclosed methods and systems and the technological advantages thereof, including efficient feature-ranking, scalable temporal data processing, and model interpretability.

[0519]

[0449] Collectively, these and other disclosed features constitute concrete improvements to the functioning of computers and computer networks, rather than mere automation of abstract analytical tasks. The disclosed predictive models achieve improved accuracy, sensitivity, and computational efficiency through novel data representations, causal reasoning, and adaptive deployment strategies, while the surrounding system architecture provides enhanced interoperability, interpretability, and responsiveness relative to conventional predictive-analytics platforms.

[0520]

[0450] Advantages over prior systems, including those described in U.S. Pub. Nos. 2022 / 189637A1 (Cerner Innovation Inc.), 2023 / 082019A1 (Early Signal LLC), 2022 / 122253A1 (Fujifilm Corp.), 2022 / 351371 A1 (Macquarie University), and 2020 / 166525A1 (Nipro Corp.), include improved feature-ranking algorithms that reduce processing time and memory requirements, adaptive data-partitioning and policy-generation techniques that enhance treatment selection efficiency, and user-interface modules that provide real-time interpretability and causal transparency of model outputs. Without conceding that any such disclosures constitute prior art, the presently disclosed methods and systems achieve demonstrably superior technological performance across multiple computational layers and clinical integration pathways.

[0521]

[0451] Those skilled in the art will further recognize that additional technological improvements are inherent ALZAI-002-PCT 2025-10-10 in and facilitated by the disclosed methods and systems, as illustrated in the figures and claims.

[0522] II. Additional Exemplary Aspects and Embodiments

[0523]

[0452] Notwithstanding the claims, and among the other various exemplary and non-limiting aspects and embodiments, are also the following additional exemplary and non-limiting aspects and embodiments.

[0524]

[0453] In some such exemplary aspects are provided computer-implemented methods for predicting the onset of a medical condition in a subject, the methods comprising: (i) receiving, via a processor, a plurality of blood test values for the subject, wherein the blood test values were measured during at least one previous time period, and each of the blood test values is associated with a respective time stamp; (ii) extracting, from the blood test values, a plurality of features; (iii) applying, to the plurality of features, a trained predictive model to compute a predicted risk score for the subject based on a plurality of features, wherein the trained predictive model is trained to predict a probability of onset a medical condition during a subsequent time period, based on the plurality of features; and (iv) outputting a predicted risk score, wherein the predicted risk score is indicative of the probability of onset of the medical condition in the subject during the subsequent time period.

[0525]

[0454] In some further exemplary aspects are disclosed systems for predicting the onset of a medical condition in a subject, the systems comprising: (i) a processor; and (ii) a memory, storing instructions that, when executed by the processor, cause the system to: (a) receive, via a processor, a plurality of blood test values for the subject, wherein the blood test values were measured during at least one previous time period, and each of the blood test values is associated with a respective time stamp; (b) extract, from the blood test values, a plurality of features; (c) apply, to the plurality of features, a trained predictive model to compute a predicted risk score for the subject based on a plurality of features, wherein the trained predictive model is trained to predict a probability of onset a medical condition during a subsequent time period, based on the plurality of features; and (d) output a predicted risk score, wherein the predicted risk score is indicative of the probability of onset of the medical condition in the subject during the subsequent time period.

[0526]

[0455] In yet further exemplary aspects are disclosed non-transitory computer-readable media storing instructions that, when executed by a processor, cause a system to perform a method for predicting the onset of a medical condition in a subject, the method comprising: (i) receiving, via a processor, a plurality of blood test values for the subject, wherein the blood test values were measured during at least one previous time period, and each of the blood test values is associated with a respective time stamp; (ii) extracting, from the blood test values, a plurality of features; (iii) applying, to the plurality of features, a trained predictive model to compute a predicted risk score for the subject based on a plurality of features, wherein the trained predictive model is trained to predict a probability of onset a medical condition during a subsequent time period, based on the plurality of features; and (iv) outputting a predicted risk score, wherein the predicted risk score is indicative of the probability of onset of the medical condition in the subject during the subsequent time period.

[0527]

[0456] In some exemplary embodiments of the methods, systems, and media, the predictive model is trained in at least one supervised training session using a plurality of labeled training samples each associating values ALZAI-002-PCT 2025-10-10 of at least some of the plurality of blood test values measured for a respective subject during the at least one previous time period with a label indicative of whether or not onset of the medical condition was detected in the respective subject. In some embodiments of the methods, systems, and media, the trained predictive model comprises a statistical model. In some embodiments of the methods, systems, and media, the trained predictive model comprises a machine learning model. Some embodiments of the methods, systems, and media further comprise classifying the probability of onset of the medical condition in the subject according to a binary classification based on a comparison of the predicted risk score to a predetermined threshold.

[0528]

[0457] In some exemplary embodiments of the methods, systems, and media, the trained predictive model is further trained to predict a rate of exacerbation or a rate of progression of the medical condition for the subject.

[0529]

[0458] In some exemplary embodiments of the methods, systems, and media, the trained predictive model is further trained to use the rate of exacerbation or the rate of progression to classify the subject.

[0530]

[0459] In some embodiments of the methods, systems, and media, the plurality of blood test values are any of absolute basophil count (Baso abs), absolute eosinophil count (EOS abs), hemoglobin (Hb), hematocrit (Het), absolute lymphocyte count (Lymp abs), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), mean corpuscular volume (MOV), absolute mononucleosis (MONO abs), red blood cell count (RBC), procalcitonin (PCT), platelet count (PLT), white blood cells count (WBC), red cell distribution width (RDW), albumin, calcium, chloride, creatinine, globulin, glucose, magnesium, phosphorus, potassium, protein, sodium, urea, uric acid, aspartate aminotransferase (AST / GOT), gamma-glutamyl transferase (GGT), alanine aminotransferase (ALT / GPT), bilirubin total, thyroid stimulating hormone (TSH), vitamin b12, prothrombin time (PT), Partial thromboplastin time (PTT), and international normalized ratio (INR).

[0531]

[0460] Some embodiments of the methods, systems, and media further comprise ranking the plurality of blood test values according to an impact of each of the blood test values to performance of the predictive model in computing the predicted risk score. In some embodiments of the methods, systems, and media, the plurality of blood test values used by the trained predictive model comprises a subset of highest ranking blood tests. In embodiments of the methods, systems, and media, the trained predictive model is further trained to compute the predicted risk score based on a physiological parameter of the subject. In embodiments of the methods, systems, and media, the trained predictive model is further trained to compute the predicted risk score based on a health-related risk, behavior, or susceptibility of the subject. In embodiments of the methods, systems, and media, the trained predictive model is further trained to compute the predicted risk score based on a behavioral parameter of the subject. In embodiments of the methods, systems, and media, the trained predictive model is further trained to compute the predicted risk score based on a sociodemographic parameter of the subject. In embodiments of the methods, systems, and media, the trained predictive model is further trained to compute the predicted risk score based on a medical parameter of the subject.

[0532]

[0461] In some embodiments of the methods, systems, and media, the plurality of features extracted from the plurality of blood test values include at least one of: (i) an aggregation of values of each blood test over the at ALZAI-002-PCT 2025-10-10 least one previous time period, the aggregation selected from a group comprising an average value, a maximum value, a minimum value, and a standard deviation; and (ii) a change pattern detected in the values of at least one blood test over the at least one previous time period, the change pattern selected from a group comprising values increasing over time, values decreasing over time, values increasing then decreasing, and significant alternations between increases and decreases. In some embodiments of the methods, systems, and media, the at least one previous time period has a duration of 1 year, 2 years, 3 years, 5 years, or 10 years prior to the subsequent time period. In some embodiments of the methods, systems, and media, the subsequent time period has a duration of 1 , 2, 3, 5, or 10 years following the at least one previous time period.

[0533]

[0462] In some embodiments of the methods, systems, and media, the medical condition is a cancer, a cardiovascular disease, a circulatory system disease, a chronic respiratory disorder, a musculoskeletal disorder, a connective tissue disease, a skin disease, a liver disease, a genitourinary disease, a metabolic disorder, an endocrine disorder, a neurological disorder, a chronic inflammatory disorder, a depressive disorder, a mental health disorder, a sleep-wake disorder, a sexual health disorder, an allergy, a periodontal disease, or a pediatric condition. Some embodiments of the methods, systems, and media further comprise prescribing, recommending, performing, or administering to the subject a pharmacotherapy, a gene therapy, an immunotherapy, a radiation therapy, a physical therapy, psychotherapy, counseling, or another mental health treatment, a stem cell therapy or other regenerative medicine intervention, a non-pharmacological intervention, a surgical intervention, a medical device intervention, a behavioral intervention, a nutritional intervention, or a lifestyle modification. In some embodiments of the methods, systems, and media, the subject is a human. In some embodiments of the methods, systems, and media, the subject is a non-human animal. In some embodiments of the methods, systems, and media, the subject is a population of non-human animals. In some embodiments of the methods, systems, and media, the non-human animal or the population of non-human animals is a non-human animal from, or is a population of, cattle, pigs, goats, sheep, poultry, horses, donkeys, mules, bees, dogs, cats, fish, or crustaceans.

[0534]

[0463] Also provided are such further methods, systems, and media as described and enabled herein.

[0535]

[0464] The foregoing description, for purposes of explanation, uses specific nomenclature to provide a thorough understanding of the invention. However, it will be apparent to one of skill that specific details are not required in order to practice the invention. Thus, the foregoing description of specific embodiments of the invention is presented for purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise forms disclosed, and many modifications and variations are possible in view of the above teachings. The embodiments were chosen and described in order to best explain the principles of the invention and its practical applications, through the elucidation of specific examples, and to thereby enable others skilled in the art to best utilize the invention and various embodiments with various modifications as are suited to the particular use contemplated, when such uses are beyond the specific examples disclosed.

[0536]

[0465] Accordingly, the scope of the invention shall be defined solely by the claims and their equivalents.

Claims

ALZAI-002-PCT2025-10-10CLAIMSThe invention claimed is:1 . A computer-implemented method for predicting the onset or progression of a medical condition in a subject, the method comprising: i. receiving, by a processor, a plurality of time-stamped blood-test values measured for the subject during at least one previous time period; ii. normalizing each of the plurality of time-stamped blood-test values to a laboratory-specific reference interval to obtain age- and sex-adjusted standardized values; ill. extracting, from the plurality of time-stamped blood-test values, a plurality of features comprising aggregations of values over time and change patterns in the values over time; iv. applying, to the plurality of features, a trained predictive model configured and trained to compute a predicted risk score for the subject, wherein the predictive model is trained to predict a probability of onset or progression of the medical condition during a subsequent time period; v. selecting, based on evaluation on a temporally separated validation set, an operating threshold that optimizes a performance metric balancing sensitivity and specificity, optionally subject to a false-positive-rate constraint, and classifying the subject relative to the threshold; vi. generating a treatment recommendation for the subject by selecting a therapeutic intervention based on the predicted risk score from a plurality of therapeutic interventions; and vii. writing the predicted risk score, an uncertainty interval, and the treatment recommendation to structured fields of an electronic medical record associated with the subject.2 The method of claim 1 , wherein the predictive model is trained in a supervised training session using a plurality of labeled training samples each associating values of at least some of the plurality of blood test values measured for a respective subject during the at least one previous time period with a label indicative of whether or not onset of the medical condition was detected in the respective subject.3 The method of claim 1 , wherein the trained predictive model comprises a statistical model.4 The method of claim 1 , wherein the trained predictive model comprises a machine learning model.5 The method of claim 1 , wherein selecting the operating threshold comprises maximizing Youden’s J index on the temporally separated validation set subject to a predefined false-positive-rate constraint.ALZAI-002-PCT 2025-10-106. The method of claim 1 , wherein the trained predictive model is further trained to predict a rate of exacerbation or a rate of progression of the medical condition for the subject.

7. The method of claim 6, wherein the trained predictive model is further trained to use the rate of exacerbation or the rate of progression to classify the subject.

8. The method of claim 1 , wherein the plurality of blood test values are selected from the group consisting of absolute basophil count (Baso Abs), absolute eosinophil count (EOS Abs), hemoglobin (Hb), hematocrit (Het), absolute lymphocyte count (Lymp Abs), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), mean corpuscular volume (MOV), absolute monocyte count (MONO Abs), red blood cell count (RBC), procalcitonin (PCT), platelet count (PLT), white blood cell count (WBC), red cell distribution width (RDW), albumin, calcium, chloride, creatinine, globulin, glucose, magnesium, phosphorus, potassium, protein, sodium, urea, uric acid, aspartate aminotransferase (AST / GOT), gamma-glutamyl transferase (GGT), alanine aminotransferase (ALT / GPT), total bilirubin, thyroid-stimulating hormone (TSH), vitamin B12, prothrombin time (PT), partial thromboplastin time (PTT), and international normalized ratio (I NR).9 The method of claim 1 , further comprising ranking the plurality of blood test values according to an impact of each of the blood test values to performance of the predictive model in computing the predicted risk score.10 The method of claim 9, wherein the plurality of blood test values used by the trained predictive model comprises a subset of highest ranking blood tests.11 The method of claim 1 , wherein the trained predictive model is further trained to compute the predicted risk score based on a physiological parameter of the subject.12 The method of claim 1 , wherein the trained predictive model is further trained to compute the predicted risk score based on a health-related risk, behavior, or susceptibility of the subject.13 The method of claim 1 , wherein the trained predictive model is further trained to compute the predicted risk score based on a behavioral parameter of the subject.14 The method of claim 1 , wherein the trained predictive model is further trained to compute the predicted risk score based on a sociodemographic parameter of the subject.15 The method of claim 1 , wherein the trained predictive model is further trained to compute the predicted risk score based on a medical parameter of the subject.ALZAI-002-PCT 2025-10-1016. The method of claim 1 , wherein the plurality of features extracted from the plurality of blood test values comprise: i. one or more aggregations of values of each blood test over the at least one previous time period, the aggregation selected from the group consisting of an average value, a maximum value, a minimum value, and a standard deviation; and ii. one or more change patterns detected in the values of at least one blood test over the at least one previous time period, the change pattern selected from a group comprising values increasing over time, values decreasing over time, values increasing then decreasing, and significant alternations between increases and decreases.

17. The method of claim 1 , wherein the at least one previous time period has a duration of 1 year, 2 years, 3 years, 5 years, or 10 years prior to the subsequent time period.

18. The method of claim 1 , wherein the subsequent time period has a duration of 1 year, 2 years, 3 years, 5 years, or 10 years following the at least one previous time period.

19. The method of claim 1 , wherein the medical condition is a cancer, a cardiovascular disease, a circulatory system disease, a chronic respiratory disorder, a musculoskeletal disorder, a connective tissue disease, a skin disease, a liver disease, a genitourinary disease, a metabolic disorder, an endocrine disorder, a neurological disorder, a chronic inflammatory disorder, a depressive disorder, a mental health disorder, a sleep-wake disorder, a sexual health disorder, an allergy, a periodontal disease, or a pediatric condition.

20. The method of claim 1 , wherein the medical condition is selected from the group consisting of: chronic obstructive pulmonary disease (COPD), chronic lung disease, congestive heart failure (CHF), chronic kidney disease (CKD), stage 4-5 chronic kidney disease or end-stage renal disease (ESRD), diabetes with complications, stroke, a neurological disorder, and HIV / AIDS.21 . The method of claim 1 , further comprising prescribing, recommending, performing, or administering to the subject any one or more of pharmacotherapy, a gene therapy, an immunotherapy, a radiation therapy, a physical therapy, psychotherapy, counseling, or another mental health treatment, a stem cell therapy or other regenerative medicine intervention, a non-pharmacological intervention, a surgical intervention, a medical device intervention, a behavioral intervention, a nutritional intervention, and a lifestyle modification.

22. The method of claim 1 , wherein the subject is a human.ALZAI-002-PCT2025-10-1023. The method of claim 1 , wherein the subject is a non-human animal.

24. The method of claim 23, wherein the subject is a population of non-human animals.

25. The method of claim 24, wherein the population of non-human animals is cattle, pigs, goats, sheep, poultry, horses, donkeys, mules, bees, dogs, cats, fish, or crustaceans.

26. A system for predicting the onset of a medical condition in a subject, the system comprising: i. a processor; and ii. a memory, storing instructions that, when executed by the processor, cause the system to: a. receive, via the processor, a plurality of time-stamped blood-test values measured for the subject during at least one previous time period; b. normalize each of the plurality of time-stamped blood-test values to a laboratory-specific reference interval to obtain age- and sex-adjusted standardized values; c. extract, from the plurality of time-stamped blood-test values, a plurality of features comprising aggregations of values over time and change patterns in the values over time; d. apply, to the plurality of features, a trained predictive model configured and trained to compute a predicted risk score for the subject, wherein the predictive model is trained to predict a probability of onset or progression of the medical condition during a subsequent time period; e. select, based on evaluation on a temporally separated validation set, an operating threshold that optimizes a performance metric balancing sensitivity and specificity, optionally subject to a false-positive-rate constraint, and classifying the subject relative to the threshold; f. generate a treatment recommendation for the subject by selecting a therapeutic intervention based on the predicted risk score from a plurality of therapeutic interventions; and g. write the predicted risk score, an uncertainty interval, and the treatment recommendation to structured fields of an electronic medical record associated with the subject.

27. The system of claim 26, comprising the memory storing instructions that, when executed by the processor, cause the system to perform the method according to any of claims 1-25.ALZAI-002-PCT 2025-10-1028. A non-transitory computer-readable medium storing instructions that, when executed by a processor, cause a system to perform a method for predicting the onset of a medical condition in a subject, the method comprising: i. receive, via the processor, a plurality of time-stamped blood-test values measured for the subject during at least one previous time period; ii. normalize each of the plurality of time-stamped blood-test values to a laboratory-specific reference interval to obtain age- and sex-adjusted standardized values; ill. extract, from the plurality of time-stamped blood-test values, a plurality of features comprising aggregations of values over time and change patterns in the values over time; iv. apply, to the plurality of features, a trained predictive model configured and trained to compute a predicted risk score for the subject, wherein the predictive model is trained to predict a probability of onset or progression of the medical condition during a subsequent time period; v. select, based on evaluation on a temporally separated validation set, an operating threshold that optimizes a performance metric balancing sensitivity and specificity, optionally subject to a false-positive-rate constraint, and classifying the subject relative to the threshold; vi. generate a treatment recommendation for the subject by selecting a therapeutic intervention based on the predicted risk score from a plurality of therapeutic interventions; and vii. write the predicted risk score, an uncertainty interval, and the treatment recommendation to structured fields of an electronic medical record associated with the subject.

29. The computer-readable medium of claim 28, storing instructions that, when executed by the processor, cause the system to perform the method according to any of claims 1-25.

30. Any further method, system, or computer-readable medium as described and enabled herein.

Images