Screening, diagnostic, and treatment tools for cognitive and other medical conditions

Abstract

The disclosed systems and methods identify undiagnosed patients within a patient population that are suffering from or at risk of developing a target medical condition that is clinically diagnosed in a group of diagnosed patients in the same population using a machine learning model trained on data of the diagnosed patients. Data of the undiagnosed patients is sourced from insurance company records, patient medical records, prescription history, treatment history, biographic, and other available data. The machine learning model extracts data about the undiagnosed patient from these sources that, alone or in combination, correlate to the target medical condition clinically diagnosed in the diagnosed patients. The undiagnosed patients that are at risk of or should be diagnosed with the target medical condition are identified by this correlation to the diagnosed patient population. The system and method can then communicate with the undiagnosed patient to ensure they are properly diagnosed and treated.

Description

PATENT SCREENING, DIAGNOSTIC, AND TREATMENT TOOLS FOR COGNITIVE AND OTHER MEDICAL CONDITIONS CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority and benefit from the U.S. Provisional Patent Application 63 / 660,915, filed June 17, 2024, and titled, “SCREENING, DIAGNOSING, AND TREATING PATIENTS FOR COGNITIVE AND OTHER DISEASES AND CONDITIONS,” which is incorporated herein by reference in its entirety for all purposes. BACKGROUND

[0002] Many medical conditions go undiagnosed or underdiagnosed because patients lack the ability, education, or access to specialized medical providers that excel at identifying and treating these patients. For example, patients suffering from such cognitive conditions that visit their primary care provider often present with subtle or undetectable symptoms or a combination of symptoms that, only in combination, indicate further investigation or a diagnosis. Even when a primary care provider suspects a patient suffers from a cognitive or other subspecialty condition, they are often not trained to adequately treat or support the patient. Specialist providers that diagnose and treat cognitive conditions tend to have long wait lists to evaluate and treat patients, which oftentimes causes patients to suffer progression of the disease and frustration while awaiting treatment because most primary care or non- specialist providers cannot provide them adequate care. Many more patients go undiagnosed entirely for various reasons including lack of time or training by their primary care or non- specialist provider or that their condition is simply in an early stage that is difficult to detect or diagnose. Fortunately, good treatments exist, particularly for certain cognitive conditions like cognitive impairment due to Alzheimer’s disease, if a patient is diagnosed and treated at an early stage (mild cognitive impairment or mild dementia) rather than a more advanced stage (moderate or severe dementia). However, early-stage diagnostic services and treatments remain difficult for patients to access due to the nature of cognitive conditions being challenging for patients to self-identify and difficult for primary care or non-specialist providers to identify due to lack of training, time available for clinical diagnostics, or expertise with ordering and interpreting advanced diagnostic studies.

[0003] With the gold standard of specialist care unavailable or delayed for many patients suffering from cognitive and other medical conditions, most patients do not gain access to thePATENT best treatment options available for their condition. Patients and their families suffer delayed medical treatment, increased cost, emotional trauma, lower quality of life, and other consequences from patients being undiagnosed or having a delay in diagnosis. Most communities across the globe have a patient population suffering from undiagnosed and / or untreated cognitive conditions that far exceeds the capacity of the available cognitive specialists to diagnose and treat them.

[0004] The industry needs to develop new techniques for screening, diagnosing, and treating cognitive and other medical conditions, particularly through upskilling primary or non- specialist care providers to reduce waiting times for patients to be diagnosed and receive appropriate and timely treatment and to improve overall patient outcomes. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] Non-limiting and non-exhaustive embodiments of the invention are described with reference to the following drawings. In the drawings, like reference numerals refer to like parts throughout the various figures, unless otherwise specified, wherein:

[0006] FIG.1 is an example of phases of progression of a cognitive medical condition.

[0007] FIG.2 is an example screening, diagnostic, and treatment system.

[0008] FIG. 3 is an example of a method for screening for a target medical condition using the system shown in FIG.2.

[0009] FIGS. 4A and 4B show a screening algorithm with an AI model that is both trained and then used to screen patients for a target medical condition.

[0010] FIGS. 5A is an example flowchart of diagnosing patients with a target medical condition.

[0011] FIG.5B is an example flowchart of diagnosing a patient with a cognitive condition.

[0012] FIG.6 is an example of a flowchart of a diagnosis method for a cognitive condition.

[0013] FIG. 7 is an example flowchart of treatment and management for a cognitive condition.

[0014] FIG. 8 is another aspect of an example flowchart of treatment and management for a cognitive condition.

[0015] FIG. 9 is an example flowchart of treatment and management for a cognitive condition when the patient suffers from agitation.

[0016] FIG. 10 is an example flowchart of treatment and management for a cognitive condition when the patient suffers from visual hallucinations.PATENT

[0017] FIG.11 is an example diagnostic flowchart for a cognitive condition when the patient suffers from late-onset behavioral changes. DETAILED DESCRIPTION

[0018] This disclosure describes systems and methods for a self-learning computational engine that transforms a conventionally static clinical evaluation of a patient for a target medical condition into a dynamic, probabilistic guidance system. The disclosed systems and methods computationally weigh data from an undiagnosed patient to help guide primary care providers (PCPs) or other non-specialist providers or clinicians through the most probable diagnostic pathway to screen, diagnose, and treat patients for target medical conditions. These techniques use unique artificial intelligence (AI) models trained on data from already diagnosed patients for a specific target medical condition along with other dynamically updated diagnostic and treatment algorithms. The AI models are iteratively and dynamically updated as data from newly diagnosed patients continues to be used to train the AI model. The dynamically updated diagnostic and treatment algorithms are continuously updated with data about the patient as it becomes available and about newly available diagnostic and treatment options. Because these AI models and diagnostic and treatment algorithms are interconnected, the new patient data can be used to continuously and dynamically update the patient’s diagnosis and treatment with each new piece of data. The disclosed systems and methods use the AI model to guide non-specialist providers, such as PCPs, to an early and accurate differentiation of target medical conditions that are conventionally only diagnosed by specialists due to the required specialty training and skill to perform conventionally required evaluations to diagnose and treat such conditions.

[0019] For example, the disclosure helps PCPs diagnose neurodegenerative diseases, such as cognitive conditions like dementia due to Alzheimer’s Disease (AD) when conventionally only specialists like a neurologist specifically trained in cognitive conditions (Behavioral Neurologists) diagnoses and treats such patients. Generally, the disclosed systems and methods allow for early, accurate, and consistent differentiation of subspecialty conditions in a way that is not possible using conventional medical screening, diagnostic, and treatment techniques, especially when the screening, diagnosis, and treatment are being performed by a non-specialist provider, such as a PCP. Non-specialist providers as disclosed here are those medical providers that do not have specialized training in a particular medical condition but instead have generalized or primary care medical training.PATENT

[0020] For sake of ease, the disclosed systems and methods are described below as a “platform.” The platform can include integrated modules that work together in a single computer architecture system or can be separate modules that operate independently in other examples. However, the comprehensive platform that provides patients and providers with complete support to screen for, diagnose, and treat the patient for a target condition are fully integrated in a single platform in most of the examples discussed herein. The fully integrated platform has four interconnected processes that includes a computational engine that converts raw data into weighted probabilities; a guidance system that helps providers avoid common diagnostic errors for non-specialists; a dynamically changing patient state engine that evolves to continually recalculate a downstream diagnosis over time as the patient state evolves; and a self-learning system that refines logic at a population level. Although the disclosed platforms are intended to help non-specialists like PCPs screen for, diagnose, and treat patients for various medical conditions, they can also be used by any provider including specialist providers that may be specially trained for the target medical condition or may be a sub- specialty provider for the target medical condition.

[0021] The process begins before the provider clinically evaluates the patient. The platform's computational engine ingests and analyzes hundreds to thousands of variables from disparate, longitudinal sources (e.g., health information exchange data, insurance claims, prescription data, procedural claims, existing lab and other diagnostic results, family history, and patient biographic data). The computational engine does not simply flag keywords; it performs feature engineering to convert raw data points into computationally significant inputs. For example, a history of “recurrent depression” from insurance claims data is converted into a specific numerical feature. This numerical feature then directly influences the weighting of a “Yes” branch at a decision point of “Does patient have longstanding psychiatric history, questionable mild cognitive impairment (MCI), and recurrent depression / anxiety?” decision point. This automated, multi-variable analysis by the platform generates a precise, quantified pre-test probability for various clinical pathways—a task that is computationally impossible for a human expert to perform with the same scale, speed, and objectivity.

[0022] The platform uses these calculated probabilities to actively guide the clinician, especially non-specialists, through the complex diagnostic decision tree, which helps prevent common diagnostic pitfalls. For example, a PCP might see the word “atrophy” on a magnetic resonance imaging (MRI) report and default to a preliminary AD diagnosis (or, less accurate, a generic “dementia” diagnosis without any consideration of the actual neuropathologic process driving the symptoms or impairment). The disclosed platform prevents this errorPATENT from happening because it already processed risk factors for cerebrovascular disease, which means the system would interpret “Atrophy > expected for age” in an MRI brain scan without contrast result and immediately highlight the subsequent branching point: “Regional or diffuse atrophy?” Instead of the generic “dementia” or unsubstantiated AD diagnosis. The platform would emphasize findings like moderate small vessel degeneration (“moderate+ small vessel degen.”) from the radiologist’s report on the patient’s MRI, which an unaided, non-specialist might otherwise overlook or not be trained to detect.

[0023] The platform then presents a ranked list of differential diagnoses with explicit probabilities (e.g., 1. Alzheimer's Disease - 65%, 2. Vascular Dementia - 30%, 3. FTD - 5%). It can synthesize this into a nuanced summary with explanations and evidentiary support, such as: “Mild dementia (CDR 1.5) likely due to a mixed major neurocognitive disorder— probable late-onset Alzheimer's disease and possible Vascular dementia (small vessel subtype).” This level of quantified guidance at pivotal junctures, especially early in a patient’s diagnosis and treatment suffering from a cognitive condition, is a core difference from conventional, intuition-based medical practice and is what is needed to address the 92% of mild cognitive impairment cases that currently go undiagnosed. Specifically for dementia, the platform helps guide non-specialist practitioners to find an accurate, early diagnosis possibly during the critical window when certain pharmacological treatments can be most effective if started early, such as amyloid-targeting monoclonal antibody therapies (like lecanemab and donenamab) for early stage AD.

[0024] The platform also treats the diagnostic journey as an evolving stochastic process, not as a single, static event that places the patient on a static treatment path. This is achieved through a methodology of dynamic longitudinal modeling, which uses an engine that maintains a unique computational “state” for each patient. This computational state can be understood as a time-updated, multi-dimensional vector representing the patient’s health status, derived from the synthesis of all ingested clinical, claims, and biographic data. Each new piece of data about a patient acts as an input that causes the system to re-calculate all downstream probabilities for that patient. For example, if an initial assessment leads to an MRI that is abnormal with or without atrophy (“Abnml w / wo atrophy”), the platform's recommendation might be an MRI brain with contrast or an FDG-PET CT brain scan. If that scan then returns results that the patient only has mild bilateral anterior temporal hypometabolism without posterior cingulate or precuneus hypometabolism (interpreted as “Not abnml / equivocal”) the patient's state is updated. The engine automatically recalculates the diagnostic downstream probabilities and guides the non-specialist provider to the nextPATENT logical step in the flowchart, such as “Consider CSF biomarkers or amyloid PET,” while de- prioritizing other paths. This ensures the diagnostic momentum is maintained and logically sound.

[0025] Building upon this dynamic longitudinal model, the platform further includes a digital twin simulation capability to forecast patient-specific trajectories and compare counterfactual outcomes. For a given patient, a personalized “digital twin” is constructed by calibrating the population-level model with the patient’s specific longitudinal data (e.g., clinical, imaging, genetic, and other data). The engine then runs “what-if” simulations to forecast how different clinical decisions are likely to impact future outcomes (such as initiating an amyloid-targeting therapy like lecanemab versus donanemab versus continuing only with conservative symptomatic management, or proceeding with an amyloid PET scan now versus waiting).

[0026] The output is presented to a medical provider as a set of comparative trajectory graphs, showing the probable course and uncertainty bands for a range of key clinical and functional outcomes. These include forecasts for cognitive measures (such as Clinical Dementia Rating or Mini-Mental State Examination scores), functional status (such as ability to perform basic and instrumental activities of daily living), behavioral and neuropsychiatric symptoms, and the probabilistic risk of future relevant adverse events (e.g., falls, episodes of agitation, delirium / encephalopathy, seizures, visual hallucinations, delusions, or hospitalizations). These trajectory forecasts can be generated for various time horizons, from the upcoming weeks and months to many years, with the confidence bands representing the range of possible outcomes widening as the prediction extends further into the future, accurately reflecting the increasing uncertainty over longer periods.

[0027] This simulation capability uniquely empowers non-specialist medical providers to help patients and families navigate the high degree of uncertainty inherent in cognitive and other subspecialty conditions. It transforms the clinical encounter with a patient into a proactive, data-driven conversation by presenting probable outcomes as visual forecasts with clear probability estimates and confidence bands, a level of experiential foresight typically reserved for seasoned specialists. This enables the provider to articulate the potential impactPATENT of different complex care decisions, facilitating a deeper, shared decision-making process that allows patients and their caregivers to become more engaged and make fully informed choices about their future care pathway. The disclosed platform’s logic is not static. Rather, it is a learning system (both supervised and unsupervised) that improves its own diagnostic accuracy over time by analyzing population-level data for patients with the target medical condition. The platform creates a feedback loop. When a patient’s journey results in a final, confirmed diagnosis (e.g., “Probable DLB / PDD”), the platform analyzes the entire data pathway that led to that outcome. By doing this across thousands of patients, it can identify subtle, early data patterns that are highly predictive of specific pathologies. The platform’s model then adjusts its internal algorithms to give these patterns greater weight in future cases. This means the system’s ability to accurately calculate the pre-test probabilities for screening, diagnostic, and treatment options improves with every patient it processes.

[0028] Generally, the disclosed techniques, platforms, systems, and methods provide comprehensive patient care to help non-specialist medical providers screen, diagnose, and treat patients for health conditions and diseases. Specialists and sub-specialists can also use the system to help in their screening, diagnosis, and treatment of patients. However, most patients do not have access to specialists and instead seek care through their PCPs or other non-specialist providers. Most PCPs are not sufficiently trained in specialty care to screen, diagnose, and treat patients for complex or specialty conditions. Because of this lack of training or simply lack of time the non-specialists are able to spend with a patient; patients often go undiagnosed or underdiagnosed with target medical conditions. Even when PCPs are proactive and suspect the patient suffers from a target medical condition, the PCP may over- or under- order diagnostics to confirm their clinical evaluation and may over- or under- treat the patient. Patients in need of specialty care who have been diagnosed or are at risk for a target medical condition often must wait long periods of time to engage with a specialist. Unfortunately, global healthcare systems cannot provide the specialty care most patients need in a timely manner. Often, patients receive delayed treatment, sometimes substantially delayed treatment, because they are waiting to see a specialist. Other patients are never diagnosed because their target medical condition requires a specialist to screen for, diagnose, and treat it. Patients and their caregivers and families suffer and healthcare costs increase substantially when patients are not diagnosed and treated in a timely manner early in the disease stages. The disclosed techniques and platforms help non-specialists, often PCPs,PATENT screen for, diagnose, and treat these patients using gold standards typically employed only in a specialist’s clinical setting but adapted to a population based screening module and a non- specialist’s diagnosis and treatment module.

[0029] In most of the examples explained here, the target medical condition is a cognitive condition, such as dementia. The same or similar techniques can be used to screen, diagnose, and treat neurocognitive and neurological conditions as well as many diseases and conditions, including cancer; stroke; and pulmonary, cardiovascular, and immune system disease and conditions. Each disease, condition, or category of the same has a unique AI model and set of dynamically adaptable diagnostic and treatment algorithms to help screen patients in a patient population to identify those patients that at risk for or likely diagnosable with the target medical condition. Likewise, each disease, condition, or category requires different diagnoses and treatment steps to best help patients.

[0030] The comprehensive platform includes integrated modules to screen, diagnose, and treat patients at risk for and / or suffering from the target medical condition. This integrated platform could be linked modules that perform the screening, diagnosis, and treatment techniques or it could be a central platform that communicates with separate but electronically connected screening, diagnosis, and treatment steps. Each of the screening, diagnosis, and treatment modules could be used alone or in any combination. For example, an undiagnosed patient could be screened as being at risk for the target medical condition or likely should be diagnosed with the target medical condition using the screening module, then diagnosed for the target medical condition using the diagnosis module, then treated for the target medical condition using the treatment module.

[0031] To help screen for patients in a population that are undiagnosed but are at risk and in need of a diagnosis and / or treatment, an AI model is used to extract portions of ingested data about undiagnosed patients that correlate to similar or the same data of patients that have been diagnosed with the same target medical condition. Using a population based approach, the AI model is trained on the data of diagnosed patients who have been clinically diagnosed, likely by a specialist provider using a gold standard, with the target medical condition. The platform identifies patterns for the diagnosed patient from multiple ingested data sources like insurance claims, prescriptions, treatment, co-morbidities, patient biographic data, and medical records. That data about the diagnosed patient is used to train the AI model to identify other patients that are undiagnosed and either at risk of or are likely diagnosable with the target medical condition. The technique can work well when the same or a similar set of data sources are used for both the diagnosed and the undiagnosed patients. Not all ingestedPATENT data sources for the undiagnosed patients need to exactly match those of the diagnosed patients. Some data sources for the undiagnosed patients are unavailable, not yet existing, or are underdeveloped / pre-mature because they are undiagnosed and have not yet received certain treatments and prescriptions that would later appear in their medical and prescriptions records after a diagnosis, for example.

[0032] The screening module includes a control algorithm for population based screening for the target medical condition that offers or helps guide non-specialist medical providers through a sequence of steps to help non-specialist providers screen undiagnosed patients for the target medical condition. As mentioned above, many examples explained in this application relate to screening for cognitive conditions like dementia although the population based control algorithm can be used to screen for many other diseases and / or conditions, as needed. FIG. 1 shows a progression of phases of dementia 100 as an example of a target medical condition that the disclosed systems can help screen for, diagnose, and treat. However, the disclosed platform can also help screen for, diagnose, and treat other medical conditions like cancer, stroke, pulmonary, cardiovascular, endocrine, immune, and other neurological system conditions and diseases. Specific to dementia, there are preclinical 102, mild cognitive / behavioral 104, mild dementia 106, moderate dementia 108, and severe dementia 110 phases of the disease progression. Helping to screen for, diagnose, and treat dementia as early in the disease progression as possible produces the best patient outcomes. For example, patients diagnosed with dementia in the preclinical and mild cognitive / behavioral impairment phases of the disease progression are identified by the platform’s AI screening model at a stage where a timely diagnosis is possible, overcoming common barriers that lead to a high percentage of dementia cases in primary care that otherwise go undiagnosed. This early, AI-enabled diagnosis places patients in the critical window to begin treatment with emerging disease-modifying therapies, such as amyloid- targeting monoclonal antibodies like lecanemab and donanemab, which are indicated for these early stages but are less effective when these therapies are started when the patient’s disease state is progressed to more advanced stages. The platform further leverages this early identification of patients by actively flagging patients who are potential candidates forPATENT clinical trials, thereby serving as a tool to accelerate research for next-generation treatments. Simultaneously, the system empowers patients and families by delivering disease stage- appropriate educational resources and planning checklists, allowing them to proactively address future legal, financial, and care needs while the patient still possesses full capacity. Furthermore, the platform’s control algorithm or “guidance engine” is the mechanism that enables significant downstream economic and systemic benefits by recommending evidence- based pharmacological and non-pharmacological interventions, such as comprehensive care coordination and caregiver support programs. This systemic approach to treating patients in cognitive decline, particularly during the MCI or mild dementia stages, can delay progression to more advanced disease, thereby postponing entry into long-term care or nursing home settings. This delay translates into meaningful reductions in both societal costs and caregiver burden. For example, modeling based on published outcomes from U.S. patient populations has estimated that a 20 - 30% slowing of cognitive decline over three years could yield per- person savings of approximately $1,700 - $2,600 and spare 210 - 315 hours of unpaid caregiver time for individuals with MCI. For those already in early-stage dementia, the estimated benefit rises to $2,500 - $4,000 in savings and 386 - 579 hours of caregiver support. When scaled to a population level, treating one million individuals early could result in aggregate societal savings of $3 - 4 billion and avert nearly half a billion hours of unpaid caregiving within just three years. These projections underscore the potential for early detection and timely intervention to not only preserve individual function but also alleviate downstream healthcare utilization, including emergency visits and prolonged hospitalizations.

[0033] FIG. 2 illustrates a system diagram of the disclosed platform 200. The platform 200 has memory 202 and a processor 204 or series of processors. The platform 200 also has input / output devices 206 that receive and display input and output, respectively, received into and sent out of the platform 200. The platform 200 also includes communications 208 that communicate with electronically connected external devices, such as provider or patient devices (not shown) and external servers, processors, and other platforms or services. Still further, the platform 200 includes a memory device 210 or devices that can include variousPATENT memory data stores for storing patient data, control algorithms, AI models, and the like. Any suitable memory devices can be included in the platform 200.

[0034] The platform 200 shown in FIG. 2 includes a screening module 212, a diagnostic module 214, and a treatment module 216. As discussed above, the screening module 212, diagnostic module 214, and treatment module 216 can be integrated into a single memory 202, as shown in FIG. 2. Alternatively, one or more of the screening module 212, diagnostic module 214, and treatment module 216 can be stored on a separate memory (not shown) and electronically coupled to the platform memory 202 to support seamless integration between the modules.

[0035] The screening module 212 of the platform 200 includes external data 218, electronic medical records (EMR) 220 for patients, population data 222, and, in some example embodiments, a condition-specific secondary screening tool 224. The external data 218 can be ingested from various sources such as health information exchange, insurance company claims, prescription data, EMRs for undiagnosed patients, procedure or surgical claims, existing lab, imaging, and other diagnostic results, family history, and patient biographic data, for example. The sources of this data may each be different or some categories of data may be sourced from the same data source. The external data is continually updated about patients. Of particular interest, the undiagnosed patients that are screened for a target medical condition have their data continuously updated as new data is created about them from the various sources from which the platform ingests data. The screening module then can continuously update its screening of these undiagnosed patients. When a change occurs in an undiagnosed patient that places them at risk of or considered for a diagnosis of a target medical condition, then the undiagnosed patient is identified as such and a risk profile is created about the undiagnosed patient.

[0036] The screening module 212 also includes population data 222 that is from a population of patients that includes the undiagnosed patients and also includes diagnosed patients. The diagnosed patients in the population are diagnosed with the target medical condition for which the undiagnosed patients are screened. The screening module 212 uses an AI algorithm trained on the data of the diagnosed patients that relates to the target medicalPATENT condition to recognize patterns and data in the undiagnosed patients that indicate that the undiagnosed patients are at risk for or considered for a diagnosis of the target medical condition. Before the data for the undiagnosed patient is ingested into the AI screening model, the raw ingested data about the undiagnosed patient is converted into computationally significant inputs that include a specific numerical feature. These features directly influence the weighting of a branch of the screening criteria to identify a patient as at risk for or considered for a diagnosis of the target medical condition.

[0037] The diagnostic module 214 helps patients that are either identified as at risk or considered to be likely diagnosable with the target medical condition using the screening module 212 or any other newly input patient that is undergoing a diagnosis for the target medical condition. A patient going through the diagnostic module 214 of the platform 200 is not required to be a patient identified as at risk or being considered for a diagnosis by the screening module 212. A patient can use any one or more of the modules, and in some examples, an undiagnosed patient uses all of the modules of the platform from screening through treatment.

[0038] The diagnostic module 214 evaluates undiagnosed patients for the target medical condition using the external data 218 and the EMR 220 from the same sources as the screening module 212. The external data 218 and EMR 220 for the undiagnosed patient are continuously updated as the screening and / or diagnosis of the undiagnosed patient evolves over time as the patient is diagnosed. The dynamic updating of patient data likewise updates the weights applied to each viable diagnostic and / or treatment option. The diagnostic module 214 actively guides a provider, often a non-specialist provider, using non-specialist ordered diagnostic support 226 that is based on the pathological process that is actually driving the symptoms the patient is suffering. As discussed above, the examples herein mostly describe that the disclosed platforms are used by non-specialist providers; however, any provider can use the disclosed platforms, as needed, including specialists and sub-specialist providers. The non-specialist ordered diagnostic support 226 of the diagnostic module 214 outputs a ranked list of differential diagnoses with explicit probabilities for each one based on a predetermined pathological process for the target medical condition.

[0039] The non-specialist ordered diagnostic support 226 helps support non-specialists to identify the most likely differential diagnoses for the patients. The properly identified differential diagnoses help the non-specialist provider order the proper tests and performPATENT necessary additional diagnostic actions to help determine the actual diagnosis of the patients. In conventional systems, the non-specialist must diagnose the patient without help, which leads to incorrect, incomplete, or non-existent diagnoses of the patients. Other times, the conventional systems without diagnostic support tools result in non-specialist providers over- testing or under-testing the patient to reach inconclusive, ineffective, or incorrect diagnoses because they are too general or focus on the incorrect pathology driving the patient’s symptoms or impairment. Weighting of the various branches of likely diagnostics and / or treatments guides the non-specialist through the differential diagnosis process and helps them avoid misdiagnoses or failure to diagnose and uses a process that is based on the objective patient data.

[0040] Similar concepts are applied to the patient’s treatment using the treatment module 216. The treatment module also has access to the patient’s external data and EMR. The treatment module 216 also can access the diagnostic module data 228 for the patient if that patient was diagnosed in the platform. Alternatively, the patient is diagnosed in another system and the patient is then treated using the treatment module 216. The diagnostic module data 228 is the data generated about the patient’s diagnosis by the diagnostic module 214. Similar in concept to the way the diagnostic module 214 guides the non-specialist provider through a diagnosis, the treatment decision support for the patient’s care team 230 guides the non-specialist provider through the treatment of the patient by providing early stage treatment suggestions and options based on a pathological process that identifies treatment options early and accurately for the patient’s diagnosis and is based on the continuously updated data about the patient from the various external data 218 sources, the EMR 220, and the diagnostic module data 228.

[0041] The platform 200 also includes additional control logic 232 in its memory 202 that globally controls the ingestion, processing, and control of all data ingestion, data analysis, and control instructions within the platform 200. The additional control logic 232 is a specific set of instructions for controlling the ingestion of data from the external sources, the ingestion of EMRs of the patient, and the continuous updating of each category of ingested data. The additional control logic 232 also controls the instructions between the screening, diagnostic, and treatment modules and all input / output, communication, and memory.

[0042] FIG. 3 shows a flowchart of an example screening module 300 in the disclosed platforms. The screening module 300 ingests insurance claim data 302, medical record (EMR and otherwise) data 304, patient biographic data 306, and data from any other external source 308 that is relevant to an undiagnosed patient’s screening, diagnosis, or treatment.PATENT This data is ingested into a screening machine learning model 310. The screening machine learning model 310 is trained on data from diagnosed patients 312 within a patient population that also includes the undiagnosed patient. The raw data for the diagnosed patients is converted to computationally significant inputs that are then used to generate a precise, quantified pre-test probably for various clinical pathways. When the patient is identified as being at risk or likely considered for diagnosis with a target medical condition, that training data is then sent to a specialist clinician or provider that evaluates the suggestion from the AI model and performs a clinical or other evaluation of the patient using the gold standard for diagnosis 314 to confirm that the patient should be diagnosed with the target medical condition. Once confirmed by the specialist provider 314, the diagnosed patient’s data is used to train the AI model that later screens undiagnosed patients for the target medical condition. The screening module analyzes all data of the diagnosed patient to help identify patterns and combinations of data that indicate a target medical condition.

[0043] When using the screening module 300 to screen the undiagnosed patient, the same ingested insurance claim data 302, medical record data 304, biographic data 306, and other external source data 308 is ingested to or otherwise accessed by the screening machine learning model 310. That data about the undiagnosed patient is compared to the data about the diagnosed patients within the patient population to identify the undiagnosed patient as being at risk for or likely diagnosable with the target medical condition. In some examples, although not required, the output of the screening machine learning model 310 that identifies the undiagnosed patient as being at risk for or likely considered for diagnosis with the target medical condition is sent to a medical provider for evaluation 316. In this screening process, the medical provider evaluation can be a gold standard clinical evaluation by a specialist, like the step in the data used to train the model, or can be a non-specialist or other provider that evaluates the patient data 316. That data is output to the undiagnosed patient 318 in the form or a message or communication relaying the outcome that the patient is at risk of or can be likely considered to be diagnosed with the target medical condition. The output to the undiagnosed patient can include a risk profile for the undiagnosed patient based on the differential diagnoses options the AI model identifies for the undiagnosed patient along with optional explanations and likelihood of each diagnosis.

[0044] Optionally, the screening process 300 can include a secondary, medical condition specific screening tool 320 that is used with the undiagnosed patient to help screen for the target medical condition. For example, in neurocognitive conditions, cognitive condition specific questionnaires for the patient and / or the patient’s caregiver are sent to the patientPATENT and / or caregiver respectively, the answers to the questionnaires are used only as secondary input to the output from the AI model to help validate a diagnosis. With patients being screened for cognitive conditions, the questionnaire is the AD8 questionnaire in some example embodiments. However, any other cognitive condition specific questionnaire could be used or a questionnaire could be used that is based on but includes additional targeted questions beyond the AD8 questionnaire. While helpful, the AD8 questionnaire is subjective because it is completed by the patient who may be suffering from the target medical condition and either does not wish to comply with the process or may be suffering symptoms that prevent them from completing the questionnaire with useable data. In conventional systems, the AD8 questionnaire, for example, is used as a primary rather than a secondary screening tool. Relying on the AD8 questionnaire as the primary source of data relating to the undiagnosed patient’s symptoms or pathological state often produces inaccurate results and missed diagnoses. Conventionally, non-specialists rely on only the subjective AD8 questionnaire for target conditions like all-cause dementia, which too often misses diagnoses. A patient with dementia or AD inherently has limited cognitive capacity to complete such a questionnaire. While the AD8 questionnaire is used as the primary example of a secondary, cognitive condition specific questionnaire for cognitive conditions, other questionnaires that are condition specific are used as secondary screening tools for other diseases and medical conditions.

[0045] FIGS. 4A and 4B show a screening module flowchart 400A, 400B to both train a screening model and then screen undiagnosed patients for a cognitive condition. In FIG. 4A, the training 402 portion of the screening module trains the machine learning or AI model based on diagnosed patients within a patient population. The training data includes population patient insurance claims and / or health information exchange data 406, clinical population patient data 408, and biographic data about the population of patients 410. The insurance claims and / or health information exchange data 406 includes prescription data 412, diagnoses and procedures 414, and other treatments 416 that the diagnosed patients received for the cognitive condition. As discussed above, the patient population includes both undiagnosed and diagnosed patients for the target medical condition. The diagnosed patients’ data is used to train the AI model to help screen for undiagnosed patients that are at risk of or likely diagnosable with the target medical condition. The population insurance claim and / or health information exchange data 406 along with the clinical population patient data 408 and the biographic data of the population of diagnosed patients 410 is used to identify computationally significant inputs to the AI model. Raw data related to prescriptions, newPATENT diagnoses and procedures, or other treatments for the diagnosed patients helps identify data that links the patient profile to the diagnosis of a cognitive condition. Raw data on the clinical population patient data includes data from Admission-Discharge-Transfer (ADT) feeds; pharmacy data; laboratory results (e.g., HbA1c, lipid panels, hormone and vitamin levels); vital signs (e.g., blood pressure, Body Mass Index (BMI)); structured data such as problem lists and allergy lists; and text from clinical notes like diagnostic study reports (e.g., electroencephalogram and imaging studies) and discharge summaries. The data also includes scores from various clinical and neuropsychological scales such as the Clinical Dementia Rating (CDR) scale score, the Functional Assessment Staging Test (FAST), Neuropsychiatric Inventory Questionnaire (NPI-Q), Patient Health Questionnaire-9 (PHQ-9), the Mini-Mental State Examination (MMSE), the Montreal Cognitive Assessment (MoCA), among others. The data further includes genetic data such as Apolipoprotein E (APOE) genotype, particularly e4 carrier status. This also includes quantitative data from the electroencephalogram studies and, importantly, diagnostic imaging studies themselves, such as volumetric data from structural brain MRI scans (e.g., T1 sagittal, T2-FLAIR axial, T2- FLAIR coronal MPRAGE, SWI, HEMOFLASH, or T2*GRE) and quantitative measures from FDG PET and amyloid beta PET scans (e.g., standardized uptake value ratio (SUVr) and centiloid values). Furthermore, the system captures quantitative digital phenotyping data, including user interaction data with the platform, and voice recordings for speech analysis to derive acoustic features, such as mean pause duration, jitter, shimmer, Mel-Frequency Cepstral Coefficients (MFCCs), and change in other speech-related measures that link to neuroanatomically and clinically significant cognitive impairments, such as syntactic variation, vocabulary frequency, prosody, and articulation. Raw data on the biographic data for the diagnosed patients includes age, sex, race, ethnicity, educational achievement, insurance type, and geocoded census tract data derived from zip and Federal Information Processing Standards (FIPS) codes to provide socioeconomic context. This collective raw data is converted to computationally significant inputs by a multi-stage data processing pipeline. This pipeline standardizes or normalizes continuous numerical data (e.g., lab results), and converts categorical data (e.g., specific diagnosis codes) into a numerical format using methods like one-hot encoding. For unstructured data, such as text from clinical notes, Natural Language Processing (NLP) is employed to extract key concepts and sentiment. Importantly, the pipeline generates dynamic temporal features by analyzing trends and frequencies of events (e.g., falls, hospitalizations) over various time windows and calculates measures of within-individual variance to detect subtle but significant changes from aPATENT patient’s own baseline. In some embodiments, neural network architectures, such as Long Short-Term Memory (LSTM) networks, may be used to automatically learn feature representations from such sequential data. A computationally significant input is a well- defined, machine-readable feature that has been processed from one or more raw data points to represent a specific, informative aspect of a patient’s state, history, or a measure of the variability within their own data over time.

[0046] For example, the raw data includes a complex combination of clinical and behavioral observations that are difficult to differentiate manually. The system is designed to distinguish between conditions with overlapping symptoms, such as sporadic behavioral variant Frontotemporal Dementia (bvFTD), late-onset primary psychiatric disorders (PPD), and functional cognitive disorder (FCD). The screening algorithm performs feature engineering on the collective patient data 418 by identifying predictive patterns in the raw data, such as the presence of apathy and low letter fluency scores which are more indicative of bvFTD, versus the presence of significant depressive symptoms which strongly favor a PPD diagnosis and transforming these patterns into distinct, weighted features for the model. For instance, the model learns to differentiate the “internal inconsistency” characteristic of FCD from the progressive decline of neurodegeneration by analyzing patterns in task performance over time. The features engineering performed on the patient data 418 prepares the data as a population patient training data set specific to the cognitive condition 420. This data set includes the diagnosed patients with their relevant data that relates to their cognitive condition. That data set is used to train an explainable boost algorithm that screens for the cognitive condition 422. The explainable boost algorithm is used here because it provides three key advantages: (1) high model intelligibility, as it is a type of Generalized Additive Model (GAM) that learns a distinct, interpretable function for each feature, allowing the contribution of each variable to the final risk prediction to be visualized and quantified individually, which is essential for clinician trust and validation; (2) high predictive accuracy that is competitive with more complex, opaque “black box” models, achieved through a modern boosting process where the model cyclically fits residuals to learn each feature’s contribution; and, (3) the inherent ability to automatically detect and model pairwise interaction terms, allowing the system to capture mediating and modifying effects between variables (e.g., how the predictive power of a given comorbidity is amplified by other combined factors, such as a patient’s age, educational achievement, and environment-related geocoding data derived from zip and FIPS codes) without sacrificing the model’s overall transparency.PATENT

[0047] Machine learning (ML) is being used more frequently to enable the analysis of data and assist in making decisions in multiple industries, such as the medical screening, diagnostics, and treatment platforms disclosed herein. In order to benefit from using machine learning, a machine learning algorithm is applied to a set of training data and labels to generate a “model” which represents what the application of the algorithm has “learned” from the training data. Each element (or example, in the form of one or more parameters, variables, characteristics or “features”) of the set of training data is associated with a label or annotation that defines how the element should be classified by the trained model. A machine learning model is a set of layers of connected neurons that operate to make a decision (such as a classification) regarding a sample of input data. When trained (i.e., the weights connecting neurons have converged and become stable or within an acceptable amount of variation), the model operates on a new element of input data to generate the correct label or classification as an output.

[0048] In general, certain of the methods, models or functions described herein, such as the screening module, non-specialist ordered diagnostic support of the diagnostic module, and the treatment decision support for the care team of the treatment module, may be embodied in the form of a trained GAM, where the model is implemented by the execution of a set of computer-executable instructions or representation of a data structure. The instructions may be stored in (or on) a non-transitory computer-readable medium and executed by a programmed processor or processing element. The set of instructions may be conveyed to a user through a transfer of instructions or an application that executes a set of instructions (such as over a network, e.g., the Internet). The set of instructions or an application may be used by an end-user through access to a SaaS platform or a service provided through such a platform. A trained GAM, trained machine learning model, or other form of decision or classification process may be used to implement one or more of the methods, functions, processes, or operations described herein. A GAM or deep learning model may be characterized in the form of a data structure in which are stored data representing a set of layers containing nodes, and connections between nodes in different layers are created (or formed) that operate on an input to provide a decision or value as an output.

[0049] The explainable boost model 422 shown in FIG. 4A is a type of GAM or machine learning model that receives the diagnosed patient data set 420 for training purposes. Once trained, the explainable boost model 422 is able to receive data sets from undiagnosed patients 424 to provide screening for the cognitive condition. The undiagnosed patient test data set includes target condition-specific data for the undiagnosed patient, which is aPATENT collection of data on the undiagnosed patient. The undiagnosed patient is selected from the patient population 426 for screening based on any criteria, which could be that they were randomly selected, are new to the patient population, have a certain combination of biographic data (e.g., age, weight, etc.), or the like. After the undiagnosed patient is selected, the patient data is collected. The same or similar categories of data are collected for the undiagnosed patient as for the diagnosed patients used to train the explainable boost algorithm.

[0050] Here, in FIG. 4A, the same data is ingested for the undiagnosed patient as for the diagnosed patient – insurance claims and / or health exchange data 428, clinical undiagnosed patient data 430, and biographic data 432. The undiagnosed insurance claims and / or health exchange data mirrors the diagnosed patients in that the platform 400A ingests prescription data 434, diagnoses and procedures 436, and other treatment data 438. Data in all these noted categories may not exist about the undiagnosed patient at the time of analysis, but the platform attempts to ingest it if it is available or ingest as data from as many of these categories of data as is available. To prepare the undiagnosed patient test data set 424, the platform 400A performs feature engineering on the undiagnosed patient raw data 440 using the same technique as the platform 400A uses to perform feature engineering on the diagnosed patient data set 418. The prepared undiagnosed patient data set 424 includes a target condition-specific data for the undiagnosed patient 442. This is data that relates to the cognitive condition for which the patient is being screened. The undiagnosed patient test data set is ingested into the explainable boost algorithm for screening the patient for the cognitive condition.

[0051] FIG. 4B shows the output from and training input to the explainable boost algorithm 422. The explainable boost algorithm 422 identifies a patient as being at risk for or likely diagnosable with the cognitive condition. The platform creates a risk profile for the undiagnosed patient based on the received the output of the explainable boost algorithm 444. The risk profile includes data about the differential diagnosis that the explainable boost algorithm generated for the undiagnosed patient and can include identification of various likely diagnoses, percentages of likelihood, and explanations for pathological reasons that support each possible diagnosis. The risk profile is used by the patient’s medical provider and can also be transmitted to the patient in some examples.

[0052] In the embodiment shown in FIG. 4B, the platform 400B transmits a secondary, cognitive condition-specific screening tool to the undiagnosed patient 446 and optionally the patient’s caregiver. The secondary, cognitive condition-specific screening tool is a set ofPATENT targeted questions that assess the patient’s cognitive state. However, each answer is subjective and the patient may be unavailable, unwilling, or inaccurate in their responses due to the inherent nature of their cognitive condition affecting their ability to engage in answering the questionnaire. In some examples, the answers to the cognitive condition specific questionnaire must be above a certain threshold value, such as a level 2 on the AD8 questionnaire for dementia. This example embodiment means that the platform outputs the risk profile and an offer or initiation of a comprehensive diagnostic evaluation with a specialist medical provider when the patient is determined to be at risk for or likely diagnosable with the cognitive condition and when their answers (and / or their caregiver’s answers) exceed the predetermined threshold. The offer or initiation of a comprehensive diagnostic evaluation with a specialist medical provider identifies the patient as a patient in need to care for the cognitive condition and begins the process of having them clinically evaluated by a specialist to diagnose then treat the patient 450.

[0053] In this same example, if the answers on the secondary, cognitive condition-specific screening tool are below a threshold, then the platform does not determine whether the patient is at risk of or likely to be diagnosable with the cognitive condition. In that case, no communication is transmitted to the undiagnosed patient 448. However, as the undiagnosed patient’s future condition changes, this same analysis continues to iteratively occur based on new patient data, such as changes in prescriptions, procedures, symptoms, diagnoses of co- morbidities, and the like, which could later identify the undiagnosed patient as at risk or likely to be diagnosable with the cognitive condition. In alternative examples, the platform 400B transmits the risk profile to the undiagnosed patient even when their answers to the secondary, cognitive condition-specific screening tool are below a threshold value. If this occurs, the undiagnosed patient’s risk profile includes a note of the below threshold score on the questionnaire for consideration by a specialist provider during the undiagnosed patient’s comprehensive diagnostic evaluation.

[0054] FIGS. 5A and 5B illustrate an example of diagnosing an undiagnosed patient in this disclosed platform 500A. In FIG. 5A, the non-specialist provider clinically examines the undiagnosed patient for a target condition 502, evaluates existing diagnostics available of the undiagnosed patient that are relevant to the target condition 504, and analyzes undiagnosed patient data for adjacent or compounding conditions to the target medical condition 506. This includes the conventional clinical examination of the patient and all existing diagnostics and patient data related to the target medical condition. Using a non-specialist ordered diagnostic support tool, the non-specialist provider identifies possible or probable targetedPATENT diagnoses of one or more target medical conditions for the undiagnosed patient 508. The non-specialist provider assigns a stage of the target medical condition to the patient based on the clinical examination, existing diagnostics, and the adjacent / compounding conditions 510. The platform does not require the non-specialist to assign a condition stage to the patient.

[0055] FIG. 5B illustrates another example of diagnosing an undiagnosed patient in the disclosed platform 500B. Here, the non-specialist provider evaluates whether non- pharmacological treatments are effective 514, determines if the patient is experiencing late onset behavioral changes 516, determines if psychosis symptoms are present 520, and evaluates whether the patient’s caregiver is experiencing difficulty caring for the patient or is experiencing burnout 522. These analyses in FIG. 5B can be in addition to performing a clinical evaluation and / or evaluating the existing diagnostics and / or the patient data for adjacent or compounding conditions related to the target medical condition, as shown in FIG. 5A. The non-specialist ordered diagnostic support tool helps the non-specialist identify objective changes in diagnoses, prescriptions, clinical evaluation, behavior, and psychosis of the patient 524 to ensure the patient receives diagnoses as early as possible so they can access treatment and obtain optimal patient outcomes. When these objective data changes occur, the platform 500B dynamically updates the patient’s clinical evaluation to determine if changes in treatment or additional imaging and diagnostic testing is recommended or required 526. The dynamic updating of the patient’s clinical evaluation includes reassessing the diagnostics and treatment of the patient based on the new objective patient data. The most current patient data is continuously evaluated to update diagnoses and treatment for the patient so the patient receives diagnoses and treatment as early as possible in their condition. Oftentimes, diseases, such as cognitive conditions, progress to more severe stages over time and updated evaluations of the patient’s objective data is important to ensure adequate and timely diagnoses and treatments to maintain a high quality of life and to produce the best patient outcomes.

[0056] FIG. 5B shows the platform 500B shares the patient’s most recent, dynamically updated clinical evaluation, diagnostics, and treatment with all of the patient’s care providers, which often includes various specialty providers, sub-specialty providers, and therapy (speech language pathology, physical therapy, occupational therapy, psychotherapy) providers. The coordination of care with the most recent and dynamically updated objective patient data increases the quality of care across the team of medical providers caring for the patient.PATENT

[0057] FIG.6 shows an example diagnostic algorithm used by the platform 600 to help non- specialists diagnose patients using ordered diagnostic support. The ordered diagnostic support is a specific order of steps to evaluate a patient for dementia in this example. The order of the evaluation steps is important to efficiently and effectively reach a diagnosis and / or updated treatment and to help weight various diagnostic and treatment paths. The non-specialist ordered diagnostic support starts by assessing objective cognitive and / or behavioral impairment during the patient’s clinical examination 602. If the provider determines that objective cognitive and / or behavioral impairment is present on the examination, then the diagnostic algorithm prompts the provider to determine whether the impairment is reversible as indicated by lab studies 604. If the cause is reversible, then it is treated. If the non-specialist provider determines that no clinically objective cognitive and / or behavioral impairment is present during the clinical evaluation of the patient, which correlates to a score of less than 0.5 on the CDR 608, then the platform designates the patient as cognitively unimpaired. The algorithm then prompts the provider to consider if the patient may have Subjective Cognitive Decline (SCD) or a Functional Cognitive Disorder (FCD), particularly if symptoms are reported despite the lack of objective findings. This step is clinically significant because SCD can be an early manifestation of an underlying neurodegenerative process like Alzheimer’s disease, especially in individuals with high cognitive reserve where standard tests may have ceiling effects related to high premorbid intellectual functioning. Identifying a potential FCD is also important, as this condition may be reversible with targeted treatments for underlying mood disorders (such as depression or anxiety) with psychopharmacologic or psychotherapeutic approaches (e.g., cognitive behavioral therapy, CBT). The platform therefore recommends ongoing longitudinal surveillance, prompting a reassessment in 6 to 12 months, or sooner if new symptoms or findings emerge, to monitor the patient’s status and ensure timely access to appropriate interventions. Emergence of objective neurologic findings or cognitive deficits might merit advanced diagnostic studies (e.g., a specific type of PET brain scan) or may indicate that an individual has become eligible for amyloid-targeting therapy; longitudinal surveillance to detect the earliest objective changes is clinically significant because interventions like amyloid-targeting therapies are more likely to be most effective the earlier the neuropathologic stage of disease.

[0058] The ordered diagnostic support then helps the non-specialist provider evaluate whether the cause of the objective cognitive and / or behavioral impairment is reversible 604. If it is reversible, then the cause is treated. If the cause is not reversible, then the orderedPATENT diagnostic support helps the non-specialist determine if the patient has a longstanding psychiatric history, questionable mild cognitive impairment (MCI), and recurrent depression and / or anxiety. This specific combination is important to help the non-specialist identify whether a patient is in need of treatment specifically for depression and / or anxiety as the primary condition or is in need of further evaluation for cognitive impairment. By first differentiating whether the patient primarily suffers from reversible causes of any objective cognitive and / or behavioral impairment and / or longstanding psychiatric history, questionable MCI, and recurrent depression and / or anxiety, the patients who suffer from these conditions can be properly treated, and the patients who do not can receive more targeted diagnostic evaluation for a possible cognitive condition.

[0059] If the patient is not suffering from a reversible cause of their symptoms and does not have the longstanding psychiatric history, questionable MCI, and recurrent depression and / or anxiety, the platform 600 helps the non-specialist provider determine the patient’s CDR score 610. The CDR score is the disease progression shown in FIG. 1 with the preclinical stage receiving a score of “0”; the MCI scoring a 0.5; mild dementia scoring a 1.0; moderate dementia scoring a 2.0; and severe dementia scoring a 3.0. Further, the platform 600 helps the non-specialist provider provide a working diagnosis based on clinical criteria for a possible or probable neurodegenerative diagnosis 612. The platform also prompts the non- specialist provider to evaluate the patient’s MRI or CT scan 614. If the patient’s MRI and / or CT scan are not abnormal, then the ordered diagnostic algorithm prompts the non-specialist provider to determine that a neurodegenerative MCI or dementia is not excluded as the proper diagnosis and to consider empirical treatment, additional imaging and / or diagnostic tests, reassessment, and / or surveillance 616. If the non-specialist provider determines that the patient is determined to be cognitively unimpaired with no objective cognitive or behavioral impairment identified 608; the objective cognitive and / or behavioral impairment is reversible 604; and / or the patient has a longstanding psychiatric history, questionable MCI, and recurrent depression and / or anxiety that needs to be treated; or that the patient’s MRI or CT scan is not abnormal, and neurodegenerative MCI or dementia cannot be excluded, then the ordered diagnostic algorithm prompts the non-specialist provider to assess the patient for changes in 6-12 months or sooner as determined by the provider 618.

[0060] When the non-specialist provider reviews the MRI and / or CT scan of the patient, the ordered diagnostic support algorithm can prompt them to do triage to determine if there is a tumor, inflammation, or the like indicated on the patient’s imaging 618. If the MRI and / or CT scan show neurodegenerative MCI or dementia that cannot be excluded, then the orderedPATENT diagnostic algorithm prompts the non-specialist provider to order additional imaging for the patient, such as volumetric MRI and FDG PET 620. The ordered diagnostic algorithm 600 can also prompt the non-specialist provider to consider testing for the cerebrospinal fluid (CSF) biomarker or amyloid PET 622. If those results meet or exceed a certain predetermined threshold, then the ordered diagnostic algorithm 400 can output a suggested diagnosis of “probable AD” 624.

[0061] The ordered diagnostic algorithm 600 also prompts the non-specialist provider to consider Normal Pressure Hydrocephalus (NPH) imaging markers based on the MRI and / or CT scan results. If the NPH marker is present, the ordered diagnostic support algorithm 600 can output a suggested diagnosis of “possible NPH.” If the NPH diagnosis is considered typical and there is no likely other diagnosis (another type of dementia or mixed dementia), then the ordered diagnostic algorithm 600 can output a suggested diagnosis of “probable NPH” to the non-specialist provider.

[0062] The ordered diagnostic algorithm 600 can also prompt the non-specialist provider to evaluate the MRI and / or CT scan for cerebrovascular injury and high vascular risk factors 630. If the cerebrovascular injury or high vascular risk factors are present in the imaging, then the ordered diagnostic algorithm outputs a recommended diagnosis of “possible VaD.” If the imaging indicates a typical VaD diagnosis and no other likely diagnosis or mixed dementia exists, then the ordered diagnostic algorithm outputs a recommended diagnosis of “probable VaD” 634. If the MRI and / or CT scan results do not indicate cerebrovascular injury or high vascular risk, then the ordered diagnostic support tool prompts the non- specialist provider to consider whether the patient meets dementia with Lewy bodies and Parkinson’s Disease (DLB / PDD) diagnosis criteria 636. If the patient meets these diagnostic criteria, then the ordered diagnostic algorithm 600 outputs a recommended diagnosis of “possible DLB / PDD” 638. If the imagine indicates the DLB / PDD diagnosis is typical and there is no other likely diagnosis, then the ordered diagnostic algorithm 600 outputs the recommended diagnosis is “probable DLB / PDD” 640.

[0063] If the MRI and / CT scan results show atrophy in the patient’s brain, then the ordered diagnostic algorithm 600 prompts the non-specialist provider to determine whether the atrophy is diffuse or regional 642. If it is diffuse and / or symmetric on both sides of the brain, then the ordered diagnostic algorithm 600 prompts the non-specialist provider to consider whether there is cerebrovascular injury or high vascular risk factors 630. However, if the atrophy on the patient’s MRI and / or CT scan results is regional, then the ordered diagnostic algorithm 600 prompts the non-specialist provider to determine if there is an atrophy patternPATENT 644. If the atrophy pattern is asymmetric temporal or anterior predominant, then the ordered diagnostic algorithm 600 outputs a recommended diagnosis of “Possible FTD” 646. If the patient’s symptoms are consistent with the diagnostic criteria for the behavioral variant of FTD (bvFTD, such as presenting with three of the following features that are persistent or recurrent: early behavioral disinhibition; early apathy or inertia; early loss of sympathy or empathy; early perseverative, stereotyped, or compulsive / ritualistic behavior; hyperorality and dietary changes; neuropsychological profile with executive / generation deficits with relative sparing of memory and visuospatial functions), then the ordered diagnostic algorithm outputs a recommended diagnosis of “Possible bvFTD” 648. If the atrophy pattern is posterior or hippocampal, then the ordered diagnostic algorithm 600 outputs a recommended diagnosis of “possible AD” 650. If the MRI and / or CT scan results show the presence of cerebral amyloid angiopathy (CAA) and there is no other likely diagnoses, then the ordered diagnostic algorithm 600 outputs a recommended diagnosis of “probable AD” 652. If more than one diagnosis is likely based on the clinical evaluation and the biomarker criteria discussed above, then the ordered diagnostic algorithm 600 outputs a diagnosis of mixed dementia 654.

[0064] FIG. 7 shows a chart 700 matching the stage of cognitive condition 702 with the possible or probable cognitive condition diagnosis 704 and an all-cause vulnerability for cognitive decline category 706. The combination of the stage of cognitive decline and the possible / probable diagnosis (e.g., from the ordered diagnostic algorithm), matches the proposed treatment options for gold standard care. For example, if the patient’s diagnosis is possibly or probably AD and the cognitive decline stage if stage 3 or above (MCI or greater), the treatment algorithm can suggest offering the patient eligibility for disease modifying therapy and consider offering off-label treatments or clinical trials 708; if not contraindicated, the patient can try donepezil, a rivastigmine patch, or galantamine 710, non-pharmacologic intervention (such as cognitive rehabilitation therapy or supportive collaborative care models), or a clinical trial that is actively recruiting and that the patient meets criteria for eligibility (e.g., inclusion and exclusion criteria, geographic proximity) to enroll into the clinical trial 714. If the diagnosis is possibly or probably VaD and the stage of cognitive decline is at least MCI, then the treatment algorithm suggests offering an off-label trial of acetylcholinesterase inhibitor (AChEI) 714 or trial of aggressive vascular risk factor management, which includes intensive control of hypertension, hyperlipidemia, and diabetes. If the diagnosis is possibly or probably DLB / PDD and the stage of cognitive decline is at least MCI, then the treatment algorithm suggests offering AChEI trial or a trial ofPATENT carbidopa / levodopa if Parkinson’s Disease is present 716. Alternatively, if REM Sleep Behavioral Disorder (RBD) is present, the treatment algorithm can suggest a trial of melatonin follow by a low dose BZD QHS 716. If visual hallucinations are present the treatment algorithm suggests an anti-psychotic 716. If the diagnosis is possibly or probably FTD / bvFTD and the stage of cognitive decline is at least MCI, then the treatment algorithm suggests educating the patient regarding limiting medications, considering participating in an off-label AChEI trial although the patient is cautioned of possible behavior symptom changes and a lower likelihood of treatment success in the trial 718. If the possibly or probable diagnosis is NPH and the stage of cognitive decline is MCI or higher, then the treatment algorithm suggests to the non-specialist provider to evaluate the patient for whether a VP shunt is needed and consider offering the patient participation in an off-label AChEI trial.

[0065] For all cause vulnerability for cognitive conditions 706 at any stage of cognitive decline up to severe dementia, the treatment algorithm suggests to the non-specialist provider to counsel the patient for reducing risk factors and increasing protective factors; evaluate the patient for mood disorders and neuropsychological and behavioral symptoms, and sleep disturbances; evaluates the patient for geriatric and frailty risks; evaluate all medication dosing; counsels the patient to increase the frequency of medication before increasing the dosing; and ensure that the patient receives the maximum benefit of a medication before considering additional or alternative medication options 722. For severe dementia patients, the treatment algorithm suggests to the non-specialist provider to consider palliative and hospice care 724. For all cause vulnerability for cognitive conditions 706 and at stages of cognitive decline from MCI through severe, the treatment algorithm suggests treatment that includes epilepsy management and other condition management medication 726 and / or education and support programs, including home care and estate planning 728. For all cause vulnerability for cognitive conditions 706 at preclinical stages of cognitive decline, the treatment algorithm suggests to the non-specialist provider to have the patient engage is cognitive skill and performance training 730. For all cause vulnerability for cognitive conditions 706 at cognitive decline though mild dementia, the treatment algorithm suggests to the non-specialist provider that cognitive therapy and social and creative therapies are helpful to the patient. For all cause vulnerability for cognitive conditions 706 with mild to moderate dementia, the treatment algorithm suggests to the non-specialist provider that cognitive therapy and counseling on a safe home environment are important along with social and creative therapies 734.PATENT

[0066] The ordered treatment steps align with the diagnosis and stage of cognitive decline the patient suffers. The treatment options are targeted to this combination of patient data and are based on the ordered diagnostic algorithm results that help weigh the most accurate and precise diagnostic and treatment paths. With such precise and accurate diagnostic techniques provided by the ordered diagnostic algorithm and the level of cognitive decline, the patient can then receive targeted treatment that improves their quality and length of life. The specific examples above that include offering the patient an opportunity to participate in a clinical trial for a new or off-label medication will change over time. The available prescriptions and therapy options will also change over time with advancements in medicine. However, the concept of a targeted diagnosis—particularly for complex classes of diseases and medical conditions that conventionally require specialist diagnosis and care like cognitive conditions—combined with evaluation of the patient for their cognitive decline stage provides patients with early and the most effective treatments available.

[0067] Specific to treatment of cognitive conditions with neuropsychiatric or neurobehavioral symptoms 800, the treatment algorithm 800 considers non-pharmacological considerations like the onset of a new illness, change in medications, change in caregiver support or caregiver feedback, and ensuring that the patient’s basic needs are met 802. This data is ingested through various external sources described above like insurance claims data, health exchanges, prescription data, EMRs, and patient biographic data that is continuously and dynamically updated with each data entry or change the patient experiences. The treatment algorithm 800 prompts the non-specialist provider to determine whether the patient is responding to non-pharmacological treatment 804. If the patient is responding to non- pharmacological treatment, then the treatment algorithm suggests to the non-specialist provider to re-assess the patient as needed or at a regular interval 806. If the patient is not responding to non-pharmacological treatment, but has neuropsychiatric and / or neurobehavioral symptoms, then the treatment algorithm prompts the non-specialist provider to consider medication options based on the patient’s symptoms 808. If the patient’s symptoms can include depression 810, anxiety 812, mania 814, psychosis 816, insomnia 818, or sexual disinhibition 820. Some patients present with multiple neuropsychiatric and / or neurobehavioral symptoms. In that case, combination treatments can be used. In the situation with a patient presenting with a single neuropsychiatric and / or neurobehavioral symptom, the treatment algorithm 800 presents various targeted prescription options for patients that are specific to the patient’s neuropsychiatric and / or neurobehavioral symptom(s). For the combination symptoms (not shown), the treatment algorithm suggestsPATENT single or multiple medication options to treat all symptoms or to prioritize treatment of more intense or primary symptoms over less severe or secondary symptoms.

[0068] FIG. 9 shows a treatment algorithm 900 specific to patients suffering cognitive decline that have agitation as a symptom. The treatment algorithm 900 prompts the non- specialist provider to assess whether the patient suffers from psychosis and to identify the agitation symptoms the has 902. The treatment algorithm 900 prompts the non-specialist provider to assess the patient for causes and risks of the agitation and / or psychosis 904. The treatment algorithm 900 then prompts the non-specialist provider to determine if there are treatable causes of the agitation, such as delirium, sensory deficit, or a previous psychotic diagnosis like schizophrenia 906. If yes, then the treatment algorithm 900 prompts the non- specialist provider to treat those causes, such as treating the delirium, maximizing the patient’s hearing and vision capabilities, ensuring the patient has optimal treatment of prior psychiatric disorders, and that the patient is counseled to increase their social, physical, sensory, and / or mental stimulation 908. The treatment algorithm 900 also prompts the non- specialist provider to determine if the patient suffering agitation is at significant risk to themselves or other 910, and if yes, recommends and helps the patient and their caregiver implement safety strategies, ensure adequate support for the caregiver to help the patient, and provide short-term prescription drug treatment if the patient is a severe risk to harm themselves or others 912.

[0069] The treatment algorithm 900 for patients suffering with agitation symptoms prompts the non-specialist provider to determine if the patient is distressed by their situation 914. Sometimes, this includes asking probing questions to the patient and their caregiver to determine the distress the patient is experiencing. If the patient is not distressed but still suffers from agitation symptoms, then the treatment algorithm 900 prompts the non-specialist provider to monitor the patient’s psychosis, explain the psychosis and risk to the patient and their caregiver, and determine a time for re-evaluation of the patient that is situationally appropriate or if the risk worsens 916. If the patient experiences distress caused by their agitation or other symptoms, the treatment algorithm 900 prompts the non-specialist provider to consider starting anti-psychotic medication for the patient 918. The treatment algorithm 900 can output recommendations on how to discuss the risks and benefits of starting an anti- psychotic with the patient and caregiver and recommendations on dosing appropriately for the patient. The treatment algorithm 900 can prompt the non-specialist provider to reassess the patient a predetermined time period after starting medication to help control agitation symptoms 920d, such as 4-6 weeks after starting an anti-psychotic medication for the patient.PATENT

[0070] If the patient is responding well to the anti-psychotic medication, then the treatment algorithm 900 can suggest to the non-specialist provider to consider withdrawing the patient from the anti-psychotic medication after a period of time, such as 12 weeks, and reassessing for recurrence of agitation symptoms 922. If the patient is not responding well to the anti- psychotic medication after a trial period of 4-6 weeks, then the treatment algorithm 900 prompts the non-specialist provider to consider increasing the does of the anti-psychotic medication, changing the medication, or withdrawing the patient from the medication 924. The treatment algorithm 900 also outputs recommendations to the non-specialist provider treating patients with agitation symptoms on how to consider the needs of the patient’s caregiver 926. The treatment algorithm 900 recommends that the non-specialist provider assess whether the caregiver is distressed or overburdened; gives careful and detailed explanations of the patient’s symptoms and psychosis; considers enhanced levels of practical support and caregiving for the patient and / or the caregiver; and considers the presence of anxiety or depression and offers a formal treatment plan to the caregiver.

[0071] FIG. 10 shows a treatment algorithm specific to patients suffering cognitive decline that have visual hallucinations as a symptom 1000. The treatment algorithm 1000 prompts a non-specialist provider to determine whether the patient has developed new-onset visual hallucinations 1002, and if not, then to verify if the visual hallucinations are longstanding 1004. If the patient suffers chronic visual hallucinations 1004, the treatment algorithm 1000 prompts the non-specialist provider to determine whether the patient has gradually increased intensity or impaired insight from the visual hallucinations 1006. If there is an increase in intensity or impaired insight from the visual hallucinations, then the treatment algorithm 1000 prompts the non-specialist provider to provide the patient with education around the visual hallucinations, address the patient’s risk factors for the visual hallucinations, consider medication changes to lower the frequency or intensity of or better control the visual hallucinations, or consider another specific treatment for the visual hallucinations 1008. Also, if the patient has chronic visual hallucinations, then the treatment algorithm 1000 prompts the non-specialist provider to determine whether they are mild, non-distressing, and have preserved insight 1010 and, if yes, the treatment algorithm 1000 also outputs a recommendations to the non-specialist provider to monitor the patient, provide education about the visual hallucinations, and address the patient’s risk factors for the visual hallucinations 1012.

[0072] The treatment algorithm 1000 prompts the non-specialist provider to determine if the patient has acute new-onset visual hallucinations or if they have a marked increase inPATENT frequency or intensity of the visual hallucinations 1014 rather than relatively stable, longstanding, chronic visual hallucinations 1004. If the visual hallucinations are acute or the patient has a marked increase in them, then the treatment algorithm 1000 prompts the non- specialist provider to screen the patient for encephalopathy; review their medications, especially any new medications and their respective changed dosage, if applicable; and treatment any underlying infection, metabolic derangement, or acute illness associated with or that may be causing the patient's visual hallucinations 1016. The treatment algorithm 1000 provides prompts and recommendations to the non-specialist provider to help educate the patient 1018, determine their risk factors for visual hallucinations 1020, determine whether the patient needs medication changes to control the visual hallucinations 1022, and treat the patient’s psychosis 1024. For example, the treatment algorithm 1000 prompts the non- specialist provider to educate the patient that the visual hallucinations are non-reality phenomena and also educate the patient about contributing factors to the visual hallucinations and what the patient can do to improve their symptoms and to create a plan if the symptoms worsen. The treatment algorithm 1000 also can determine the patient’s risk factors by monitoring their vision and correct any vision impairments and assess for and treat depression, anxiety, and RBD, for example. The treatment algorithm 1000 can also evaluate the patient’s visual hallucinations risk factors by regularly reviewing the medication list and educating the patient on medications to avoid that would likely increase the risk of visual hallucinations or increase their frequency or intensity.

[0073] The treatment algorithm 1000 prompts the non-specialist provider to consider, in order, the patients medication changes. First, the treatment algorithm 1000 recommends to the non-specialist provider to stop the patients’ anti-cholinergics if the patient is taking them; then to reduce or decrease the patient’s amantadine if the patient is taking it. Next, the treatment algorithm 1000 recommends to the non-specialist provider to reduce the patient’s dopamine agonists, if the patient is taking them, then to reduce or decrease the COMT inhibitors if the patient is taking it. Finally, the treatment algorithm 1000 recommends to the non-specialist provider to reduce the patient’s levodopa if the patient is taking it. The treatment algorithm 1000 helps the non-specialist provider treat the patient’s psychosis with various medication options, such as cholinesterase inhibitors, second-generation anti- psychotics, clozapine, and ECT when there are difficulties with other medication options or if the non-specialist provider has difficulty finding the proper dosing for the patient with other medications.PATENT

[0074] FIG. 11 illustrates a treatment algorithm specific to patients suffering cognitive decline that have late onset behavioral changes as a symptom 1100. The treatment algorithm 1100 prompts the non-specialist provider to clinically evaluate the patient for hallucinations with collateral information and assess them for bvFTD symptoms with hallucinations. The treatment algorithm 1100 also prompts the non-specialist provider to screen the patient for major psychosis symptoms using various techniques, such as evaluating the patient for DSM- 5 criteria, and for social cognition based on an informant-based history the non-specialist takes for the patient. The treatment algorithm 1100 prompts the non-specialist provider to incorporate a detailed neurological examination, mental status examination, and score from various diagnostics like MoCA, ACE-III, and DCQ into their clinical evaluation of the patient.

[0075] If bvFTD is part of the differential diagnosis with late onset behavioral changes 1104, the treatment algorithm 1100 outputs a recommendation to the non-specialist provider to refer the patient to a psychiatry specialist, and possible speech language pathology (SLP) for a multi-disciplinary evaluation that include neurologic and psychiatric inputs 1106. The treatment algorithm 1100 also output a recommendation to the non-specialist provider to gather at least one behavioral clinical scale evaluation for the patient (e.g., FBI, SRI) and a neuro psychological evaluation inputs to consider a potential referral for a formal neuro psychological evaluation 1106. The treatment algorithm 1100 also prompts the non-specialist provider to perform one or more structured tests for social cognition, such as the Ekman 60 faces or SEA and obtain a brain MRI with at least 3-D T1 sequences and FLAIR 1106. The treatment algorithm 1100 also prompts the non-specialist provider to perform a standardized review rating scales and visual qualification of regional cortical atrophy for the patient or to use automated volumetry if it is available 1106. The treatment algorithm 1100 prompts the non-specialist provider to evaluate the patient for an unexplained upper or lower motor neurological symptom, and refer them for an electromyography (EMG) if they have one 1106.

[0076] The treatment algorithm 1100 prompts the non-specialist provider to evaluate whether the primary diagnosis for the patient is psychiatric 1108. If it is, then the treatment algorithm 1100 prompts the non-specialist provider to further evaluate the patient to reach a specific psychiatric diagnosis, treat the patient for the psychiatric diagnosis, and consider genetic testing for C9orf72 if one or more first degree relatives of the patient have FTD, amyotrophic lateral sclerosis (ALS), or other early-onset neurodegenerative disease 1110. If the diagnosis is uncertain after the non-specialist provider evaluates the patient 1112, then thePATENT treatment algorithm 1100 prompts the non-specialist provider to evaluate the patient using a CSF analysis for A-beta 41, tau, and p-tau, or amyloid-beta PET to assess the patient for AD 1114. The treatment algorithm 1100 also prompts the non-specialist provider to consider serum or CSF Nfl if reference values are available for the patient and to perform genetic testing of all FTD mutations if the patient has at least one first degree relative with bvFTD or ALS 1114. The treatment algorithm 1100 also prompts the non-specialist provider to screen the patient for C9orf72 if they are considering a diagnosis of bvFTD even if the patient has no family history of bvFTd 1114. Further, the treatment algorithm 1100 prompts the non- specialist provider to consider C9orf72 screening if they suspect the patient has bvFTD but the patient is not meeting the full criteria for a diagnosis and there are prominent psychiatric symptoms or a family history of late-onset primary psychiatric diagnoses 1114. If, after the treatment algorithm prompts the non-specialist provider to evaluate the patient for bvFTD, the non-specialist provider still is uncertain of the patient’s diagnosis, then the treatment algorithm 1100 prompts the non-specialist provider to reconsider the psychiatric origin of the late onset behavioral change symptoms and to treat the symptoms empirically 1118. If the non-specialist provider determines the patient has a diagnosis of probable bvFTD, then the treatment algorithm 1100 prompts the non-specialist provider to screen the patient for C9orf72 and genetic testing for all FTD mutations if the patient has at least one first degree relative with bvFTD, late onset primary psychosis, ALS or other early onset neurodegenerative diagnoses 1122.

[0077] Each of the diagnosis and treatment algorithms shown in FIGS. 6-11 include a dynamically updated and continuously changing set of decisions. New therapies and medications become available as medical advancements are made. New data about an individual patient is continuously updated based on their continuously changing medical state, new diagnoses, aging, and the like. These algorithms ingest dynamically updated data and change decision flows in an adaptable manner. Conventional diagnosis and treatment approaches require human medical providers that do not ingest new data nor do they adapt to the most current gold standard in care for available diagnostic and treatment options. Typically, a team of medical providers treat a patient with a complex medical condition, such as a cognitive condition, and none has access to all insurance claims data, treatments and procedures, prescriptions, medical records, and caregiver data for their common patient. The disclosed systems and methods overcome this barrier and allow teams of providers to seamlessly care for patients with the best and most complete patient data available and with the gold standard diagnostics and treatments available.PATENT

[0078] The foregoing description, for purposes of explanation, used specific nomenclature to provide a thorough understanding of the disclosure. However, it will be apparent to one skilled in the art that the specific details are not required in order to practice the systems and methods described herein. The foregoing descriptions of specific embodiments or examples are presented by way of examples for purposes of illustration and description. They are not intended to be exhaustive of or to limit this disclosure to the precise forms described. Many modifications and variations are possible in view of the above teachings. The embodiments or examples are shown and described in order to best explain the principles of this disclosure and practical applications, to thereby enable others skilled in the art to best utilize this disclosure and various embodiments or examples with various modifications as are suited to the particular use contemplated. It is intended that the scope of this disclosure be defined by the following claims and their equivalents.

Claims

PATENT What is claimed is:

1. A method of identifying an undiagnosed patient that is suffering from or at risk for a target medical condition, the undiagnosed patient part of a patient population that also includes diagnosed patients clinically diagnosed with the target medical condition, the method comprising: receiving one or more of insurance claims records, medical record, or biographic data of the undiagnosed patient from one or more patient data sources; using a machine learning model trained on data from the diagnosed patients in the patient population, extracting data from the one or more of the insurance claims, medical record, or biographic data of the undiagnosed patient that, alone or in combination, correlates to target medical condition data of the diagnosed patients; identifying the undiagnosed patient as being diagnosed with or at risk for the target medical condition based on the correlation between the extracted data of the undiagnosed patient and the target medical condition data relating to the diagnosed patients in the patient population; and transmitting a communication to the undiagnosed patient that includes the risk profile and an offer of or initiation of a comprehensive diagnostic evaluation by a medical provider of the undiagnosed patient for the target medical condition.

2. The method of claim 1, wherein the target medical condition includes one or both of a cognitive condition and a neurological condition.

3. The method of clam 1, wherein the one or more patient data sources are independent of each other.

4. The method of claim 1, wherein the undiagnosed patient has not been examined for the target medical condition before identifying the undiagnosed patient as being diagnosed with or at risk for the target medical condition.

5. The method of claim 1, wherein the biographic data includes age, sex, and the insurance claims data with or without pharmacy data.PATENT 6. The method of claim 1, wherein the insurance claims data includes data about co- morbidities, family or genetic risk, healthcare services utilization related to the target medical condition or adjacent conditions, and prescriptions for drugs or treatments that indicate the target medical condition or an adjacent condition to the target medical condition.

7. The method of claim 1, wherein the machine learning model includes an explainable boosting machine algorithm.

8. The method of claim 1, wherein the identified, undiagnosed patient and the risk profile are used to train the machine learning model.

9. The method of claim 1, wherein, before transmitting the communication to the undiagnosed patient that includes the risk profile, transmitting a secondary, target medical condition-specific screening tool to the undiagnosed patient that includes questions to be completed by the undiagnosed patient or a caregiver of the undiagnosed patient, the questions relating to the target medical condition of the undiagnosed patient.

10. The method of claim 9, further comprising determining that responses from the undiagnosed patient to the secondary, target medical condition-specific screening tool meet a threshold value, and transmitting the communication to the undiagnosed patient that includes the risk profile and the offer of or initiation of the comprehensive diagnostic evaluation by the medical provider of the undiagnosed patient for the target medical condition is based on the responses from the undiagnosed patient to the secondary, target medical condition- specific screening tool meeting the threshold value.

11. The method of claim 10, wherein the secondary screening tool for cognitive impairment is the AD8 questionnaire.

12. The method of claim 12, wherein the threshold value for increased risk of a target medical condition causing cognitive decline is 2 or greater on the AD8 questionnaire.

13. The method of claim 9, further comprising determining that responses from the undiagnosed patient to the secondary, target medical condition-specific screening tool do not meet a threshold value, and canceling transmission of the communication to the undiagnosedPATENT patient that includes the risk profile and the offer of or initiation of the comprehensive diagnostic evaluation by the medical provider of the undiagnosed patient for the target medical condition based on the responses from the undiagnosed patient not meeting the threshold value.

14. The method of claim 9, further comprising determining that responses from the undiagnosed patient to the secondary, target medical condition-specific screening tool do not meet a threshold value, and confirming an instruction to transmit the communication to the undiagnosed patient that includes the risk profile and the offer of or initiation of the comprehensive diagnostic evaluation by the medical provider of the undiagnosed patient for the target medical condition.

15. A system of identifying an undiagnosed patient that is suffering from or at risk for a target medical condition, the undiagnosed patient part of a patient population that also includes diagnosed patient clinically diagnosed with the target medical condition, the system comprising: an input configured to receive one or more of insurance claims records, medical record, or biographic data of the undiagnosed patient from one or more patient data sources; a processor configured to: using a machine learning model trained on data from the diagnosed patients in the patient population, extract data from the one or more of the insurance claims, medical record, or biographic data of the undiagnosed patient that, alone or in combination, correlates to target medical condition data of the diagnosed patients; and identify the undiagnosed patient as being diagnosed with or at risk for the target medical condition based on the correlation between the extracted data of the undiagnosed patient and the target medical condition data relating to the diagnosed patients in the patient population; and an output configured to transmit a communication to the undiagnosed patient that includes the risk profile and an offer of or initiation of a comprehensive diagnostic evaluation by a medical provider of the undiagnosed patient for the target medical condition.

16. The system of clam 15, wherein the one or more patient data sources are independent of each other.PATENT 17. The system of claim 15, wherein the processor is further configured to determine that the undiagnosed patient has not been examined for the target medical condition before identifying the undiagnosed patient as being diagnosed with or at risk for the target medical condition.

18. The system of claim 15, wherein the insurance claims data includes data about co- morbidities, family or genetic risk, healthcare services utilization related to the target medical condition or adjacent conditions, and prescriptions for drugs or treatments that indicate the target medical condition or an adjacent condition to the target medical condition.

19. The system of claim 15, wherein the machine learning model includes an explainable boosting machine algorithm.

20. The system of claim 15, wherein the processor is further configured to train the machine learning model on the identified, undiagnosed patient and the risk profile.

21. The system of claim 15, wherein, the processor is further configured to, before the output transmits the communication to the undiagnosed patient that includes the risk profile, transmit a secondary, target medical condition-specific screening tool to the undiagnosed patient that includes questions to be completed by the undiagnosed patient or a caregiver of the undiagnosed patient, the questions relating to the target medical condition of the undiagnosed patient.

22. The system of claim 21, wherein the processor is further configured to determine that responses from the undiagnosed patient or the caregiver to the secondary, target medical condition-specific screening tool meet a threshold value, and transmit the communication to the undiagnosed patient that includes the risk profile and the offer of or initiation of the comprehensive diagnostic evaluation by the medical provider of the undiagnosed patient for the target medical condition based on the responses from the undiagnosed patient to the secondary, target medical condition-specific screening tool meeting the threshold value.

Images