Predicting mild cognitive impairment based on patient co-morbidities

Abstract

A platform is disclosed for the training and deployment of statistical or predictive models for predicting risk data and etiologies for cognitive impairment. Patient medical data may be obtained for a patient. The patient medical data may include comorbidity data. The predictive models may predict a mild cognitive impairment risk for the patient by applying a first portion of the patient medical data to a first predictive model trained to predict mild cognitive impairment risk using first medical record training data for first subjects. Based on the predicted risk of mild cognitive impairment, the predictive models may predict a mild cognitive impairment etiology for the patient by applying a second portion of the patient medical data to a second predictive model trained to predict mild cognitive impairment etiologies using second medical record training data for second subjects that satisfy a mild cognitive impairment condition.

Description

Attorney Docket No.: 08061.0066-00304 CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of priority to U.S. Provisional Application No. 63 / 607,067, filed December 6, 2023, and of U.S. Provisional Application No.63 / 561,285, filed March 4, 2024. Each of these applications is hereby incorporated by reference in its entirety. TECHNICAL FIELD

[0002] The present disclosure relates to machine learning and statistical models, as well as training machine learning or statistical models to predict cognitive impairment risks or etiologies. The trained models can be used to generate treatment recommendations for patients. BACKGROUND

[0003] Individuals with mild cognitive impairment (MCI) have memory loss and cognitive deficits beyond those expected as a result of normal aging. As compared to individuals without MCI, individuals with MCI have an increased likelihood of developing dementia (e.g., Alzheimer’s disease, Lewy body dementia, or vascular dementia). In particular, MCI can be a transitional stage before developing Alzheimer’s disease (AD). In such instances, early-stage Alzheimer’s disease may present as MCI. Emerging treatments can delay the progression of Alzheimer’s disease. Such treatments may provide greater benefits when started earlier in the progression of Alzheimer’s disease.

[0004] However, MCI can be difficult to detect. Clinicians may have limited time with patients during patient visits to observe signs and symptoms of MCI. Patients may not disclose symptoms to clinicians, and family members may assume that cognitive impairment symptoms are a natural part of aging. Accordingly, cognitive impairment may only be identified when a patient has progressed to later stages of Alzheimer’s disease. Thus, thereAttorney Docket No.: 08061.0066-00304 remains a need to identify earlier stages of cognitive impairment, e.g., to facilitate earlier therapeutic intervention.

[0005] Furthermore, the appropriate treatment for a patient with MCI can depend on the etiology of MCI for that patient. MCI can arise from one or a combination of different causes, and treatments appropriate for MCI having one etiology may be ineffective for MCI having another etiology. There thus is a need to better identify MCI of particular etiology, for instance MCI associated with Alzheimer’s disease. SUMMARY

[0006] Systems and methods are disclosed herein for predicting cognitive impairment risk data, etiology data, or treatments for cognitive impairment. In accordance with disclosed embodiments, cognitive impairment risk data, etiology data, or treatments for cognitive impairment can be predicted, at least in part, by prediction models disclosed herein using patient data.

[0007] Disclosed embodiments include a system including at least one processor and at least one computer-readable, non-transitory medium containing instructions. The instructions, when executed by the at least one processor, can cause the system to perform operations. The operations can include obtaining patient medical data for a patient, the patient medical data including comorbidity data. The operations can further include predicting a mild cognitive impairment risk for the patient by applying a first portion of the patient medical data to a first predictive model trained to predict mild cognitive impairment risk using first medical record training data for first subjects. The operations can further include, based on the predicted risk of mild cognitive impairment, predicting a mild cognitive impairment etiology for the patient by applying a second portion of the patient medical data to a second predictive model trained to predict mild cognitive impairment etiologies using second medical record training data for second subjects, potentially different from the first subjects, that satisfy a mild cognitiveAttorney Docket No.: 08061.0066-00304 impairment condition. The second subjects may be associated with mild cognitive impairment etiology labels.

[0008] In some embodiments, at least one of the first predictive model or the second predictive model can include at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest. In some embodiments, a first portion of the first subjects may be associated with a mild cognitive impairment label, and a second portion of the first subjects may be associated with a control label. In some embodiments, the first and second subjects satisfy an age criterion. In some embodiments, the patient medical data further includes at least one of socio-demographic data or health status data. In some embodiments, the first medical record training data includes at least one of socio-demographic data or health status data. In some embodiments, the second medical record training data includes at least one of socio-demographic data or health status data. In some embodiments, the comorbidity data includes at least one of cardiovascular condition data, metabolic condition data, neurological condition data, psychiatric condition data, respiratory condition data, sleep condition data, or substance abuse data. In some embodiments, the comorbidity data includes data for at least two comorbidities. In some embodiments, the first portion of the patient medical data may be the same as the second portion of the patient medical data. In some embodiments, the first portion of the patient medical data may be different from the second portion of the patient medical data.

[0009] Disclosed embodiments include a method of treatment. Some disclosed embodiments involve obtaining patient medical data for a patient. The patient medical data may include comorbidity data. Some disclosed embodiments include predicting a mild cognitive impairment risk for the patient by applying a first portion of the patient medical data to a first predictive model trained to predict mild cognitive impairment risk using first medical record training data for first subjects. Some disclosed embodiments include, based on the predictedAttorney Docket No.: 08061.0066-00304 risk of mild cognitive impairment, predicting a mild cognitive impairment etiology for the patient by applying a second portion of the patient medical data to a second predictive model trained to predict mild cognitive impairment etiologies using second medical record training data for second subjects that satisfy a mild cognitive impairment condition. The second subjects may be associated with mild cognitive impairment etiology labels. Some disclosed embodiments include, based at least in part on the prediction of the mild cognitive impairment etiology, applying a therapeutic agent to the patient, e.g., lecanemab, e.g., where the patient has been identified as having mild cognitive impairment associated with Alzheimer’s disease.

[0010] In some embodiments, at least one of the first predictive model or the second predictive model includes at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest. In some embodiments, a first portion of the first subjects may be associated with a mild cognitive impairment label, and a second portion of the first subjects may be associated with a control label. In some embodiments, the first and second subjects satisfy an age criterion. In some embodiments, the patient medical data further includes at least one of socio-demographic data or health status data. In some embodiments, the first medical record training data includes at least one of socio-demographic data or health status data. In some embodiments, the second medical record training data includes at least one of socio-demographic data or health status data.

[0011] Some disclosed embodiments include, based on a determination that the predicted etiology is Alzheimer’s disease, applying a first therapeutic agent; or based on a determination that the predicted etiology is Lewy body dementia, applying a second therapeutic agent; or based on a determination that the predicted etiology is Vascular dementia, applying a third therapeutic agent. In some embodiments, the therapeutic agentAttorney Docket No.: 08061.0066-00304 may be an anti-amyloid beta protofibril antibody. In some embodiments, the antibody may be lecanemab. In some embodiments, the comorbidity data includes at least one of cardiovascular condition data, metabolic condition data, neurological condition data, psychiatric condition data, respiratory condition data, sleep condition data, or substance abuse data.

[0012] Some disclosed embodiments include a system including at least one processor and at least one computer-readable, non-transitory medium containing instructions. The instructions, when executed by the at least one processor, can cause the system to perform operations. The operations can include obtaining a training dataset including patient medical data. The training dataset may include a first portion including subjects satisfying a mild cognitive impairment condition and a second portion including control subjects. The operations may include training a predictive model, using the training dataset, to predict a risk of mild cognitive impairment using the training dataset. The operations may include providing the trained predictive model. In some embodiments, the patient medical data may include at least one of comorbidity data, socio-demographic data, or health status data. In some embodiments, the predictive model may include at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest. In some embodiments, the comorbidity data includes data for at least 2 comorbidities. In some embodiments, the comorbidity data includes data for 2-30 comorbidities. In some embodiments, the comorbidity data includes medical comorbidity data or behavioral comorbidity data. In some embodiments, the medical comorbidity data includes data on one or more of stroke, transient ischemic attack, kidney disease, aphasia, atherosclerosis, gait abnormalities, concussion, cerebral palsy, incontinence, traumatic brain injury, or rapid eye movement (REM) disorder. In some embodiments, the medical comorbidity data does not include data on one or more of cancer, polyneuropathies, demyelinating disease, or COPD. InAttorney Docket No.: 08061.0066-00304 some embodiments, the behavioral comorbidity data includes data on one or more of depression, bipolar disorder, psychosis, hallucination, schizophrenia, substance abuse, anxiety, delusion, inability to follow a command, irritability, confusion, inability to communicate, falls, feeling or living alone, poor mental state, or impulsivity. In some embodiments, the socio-demographic data includes data on age or race. In some embodiments, subjects satisfying the mild cognitive impairment condition and the control subjects satisfy an age criterion. In some embodiments, the patient medical data for each subject satisfying the mild cognitive impairment condition includes an initial diagnosis of MCI and at least one subsequent diagnosis of MCI or AD. In some embodiments, the patient medical data for each control subject does not include a record of MCI, AD, a cognitive problem, or a dementia medication.

[0013] Some disclosed embodiments include a system including at least one processor and at least one computer-readable, non-transitory medium containing instructions. The instructions, when executed by the at least one processor, can cause the system to perform operations. The operations may include obtaining a training dataset including patient medical data. The training dataset may include subjects satisfying a mild cognitive impairment condition and having different etiologies for the mild cognitive impairment condition. The operations may include training a predictive model, using the training dataset, to predict an etiology for the subjects. The operations may include providing the trained predictive model.

[0014] In some embodiments, the patient medical data includes at least one of comorbidity data, socio-demographic data, or health status data, the subjects satisfy a sole-etiology dementia condition. In some embodiments, the comorbidity data includes at least one of cardiovascular condition data, metabolic condition data, neurological condition data, psychiatric condition data, respiratory condition data, sleep condition data, or substance abuse data. In some embodiments, the comorbidity data and health status data are obtained beforeAttorney Docket No.: 08061.0066-00304 or within a year of diagnosis of the mild cognitive impairment condition. In some embodiments, the predictive model includes at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest. In some embodiments, the predicted etiology includes at least one of Alzheimer’s disease, Lewy body dementia, or Vascular dementia. In some embodiments, the predicted etiology may be the sole etiology. Some disclosed embodiments involve training the predictive model to predict a treatment based on the predicted etiology.

[0015] Some disclosed embodiments include a system including at least one processor and at least one computer-readable, non-transitory medium containing instructions. The instructions, when executed by the at least one processor, can cause the system to perform operations. The operations may include obtaining training data for first subjects. The training data may include patient medical data for each subject. The patient medical data may include comorbidity data and socio-demographic data. In some embodiments, each subject satisfies an age criterion. In some embodiments, at least a portion of the first subjects satisfy a mild cognitive impairment condition. The operations may include training a predictive model, using the training data, to predict cognitive impairment risk data for a first subject. The operations may include obtaining patient medical data for a second subject. The operations may include predicting cognitive impairment risk data for the second subject by applying the patient medical data for the second subject to the trained predictive model. The operations may include providing the predicted cognitive impairment risk data. In some embodiments, the comorbidity data includes at least one of medical comorbidity data or behavioral comorbidity data. In some embodiments, the medical comorbidity data includes data on one or more of stroke, transient ischemic attack, kidney disease, aphasia, atherosclerosis, gait abnormalities, concussion, cerebral palsy, incontinence, traumatic brain injury, or rapid eye movement (REM) disorder. In some embodiments, the medical comorbidity data does notAttorney Docket No.: 08061.0066-00304 include data on one or more of cancer, polyneuropathies, demyelinating disease, or COPD. In some embodiments, the behavioral comorbidity data includes data on one or more of depression, bipolar disorder, psychosis, hallucination, schizophrenia, substance abuse, anxiety, delusion, inability to follow a command, irritability, confusion, inability to communicate, falls, feeling or living alone, poor mental state, or impulsivity. In some embodiments, the patient medical data includes at least one diagnosis or record abstracted using natural language processing from clinical notes. In some embodiments, the predictive model includes at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest. In some embodiments, the patient medical data for each of the first subjects satisfying the mild cognitive impairment condition includes an initial diagnosis of MCI and at least one subsequent diagnosis of MCI or AD.

[0016] Some disclosed embodiments include a system including at least one processor and at least one computer-readable, non-transitory medium containing instructions. The instructions, when executed by the at least one processor, can cause the system to perform operations. The operations may include obtaining training data for first subjects satisfying a mild cognitive impairment condition. The training data may include patient medical data for each subject. The patient medical data may include comorbidity data and socio-demographic data. Each subject may satisfy an age criterion. The operations may include training a predictive model, using the training data, to predict MCI etiology data for a first subject. The operations may include obtaining patient medical data for a second subject. The operations may include predicting MCI etiology data for the second subject by applying the patient medical data for the second subject to the trained predictive model. The operations may include providing the predicted MCI etiology data. In some embodiments, the socio-demographic data includes one or more of gender, age at MCI diagnosis, race, education, or marital status. In someAttorney Docket No.: 08061.0066-00304 embodiments, the patient medical data includes at least one diagnosis or record abstracted using natural language processing from clinical notes. In some embodiments, the at least one diagnosis or record includes at least one of stroke, ITA, diabetes, hypercholesterolemia, hypertension, atrial fibrillation, insomnia, depression, traumatic brain injury, or a sleep disorder. In some embodiments, the predictive model includes at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest. In some embodiments, the comorbidity data may include at least one of medical comorbidity data or behavioral comorbidity data.

[0017] The disclosed embodiments further include computer-readable, non-transitory media containing instructions for configuring systems to perform the above-recited operations, and methods corresponding to the above-recited operations.

[0018] The foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings, which are incorporated in and constitute part of this disclosure, together with the description, illustrate and serve to explain the principles of various example embodiments.

[0020] FIG.1 depicts an exemplary platform for developing, validating, and deploying predictive models for predicting cognitive impairment risk or etiology, consistent with disclosed embodiments.

[0021] FIG.2 depicts an exemplary process of predicting cognitive impairment risk for a subject, consistent with disclosed embodiments.

[0022] FIG.3 depicts an exemplary process for predicting cognitive impairment etiology for a subject, consistent with disclosed embodiments.Attorney Docket No.: 08061.0066-00304

[0023] FIG.4 depicts an exemplary process for treating cognitive impairment, consistent with disclosed embodiments.

[0024] FIGs.5A-5E depict the results of an investigation into development of a model for the prediction of cognitive impairment etiologies using electronic health record (EHR) data.

[0025] FIGs.6A-6I depict results of an investigation into developing a risk prediction model for mild cognitive impairment (MCI) using EHR data.

[0026] FIGs.7A-7D depict results of another investigation into developing a risk prediction model for MCI using EHR data.

[0027] FIGs.8A-8D depict results of another investigation into developing a risk prediction model for MCI using EHR data.

[0028] FIGs.9A-9H depict results of another investigation into developing a risk prediction model for MCI using EHR data. DETAILED DESCRIPTION

[0029] The following detailed description refers to the accompanying drawings. Wherever possible, the same reference numbers are used in the drawings and the following description to refer to the same or similar parts. While several illustrative embodiments are described herein, modifications, adaptations, and other implementations are possible. For example, substitutions, additions, or modifications may be made to the components illustrated in the drawings, and the illustrative methods described herein may be modified by substituting, reordering, removing, or adding steps to the disclosed methods. Accordingly, the following detailed description is not limited to the disclosed embodiments and examples. Instead, the proper scope is defined by the appended claims.

[0030] Emerging treatments can delay the progression of Alzheimer’s disease (AD). Such treatments may provide greater benefits when started earlier in the progression of Alzheimer’s disease. Mild cognitive impairment (MCI) can indicate incipient Alzheimer’sAttorney Docket No.: 08061.0066-00304 disease. But detection of MCI can be difficult, and the memory loss and cognitive deficits characteristic of MCI can have other etiologies beyond, or in addition to, Alzheimer’s disease. Treatments intended for Alzheimer’s disease may be ineffective or harmful to patients having MCI arising from such other etiologies. Conversely, treatments provided early for Alzheimer’s disease patients or those at risk of developing Alzheimer’s disease may provide improved efficacy and superior therapeutic outcomes.

[0031] One such therapeutic treatment for early Alzheimer’s disease is to target amyloid beta(A ) protofibrils for clearance from the brain, e.g., using an anti-A protofibril antibody suchas lecanemab (also known as BAN2401). The terms “BAN2401” and “lecanemab” are used interchangeably and refer to a humanized IgG1 monoclonal version of mAb158, which is a murine monoclonal antibody raised to target protofibrils and disclosed in WO 2007 / 108756 and Journal of Alzheimer’s Disease 43: 575-588 (2015). BAN2401 comprises three heavy chain complementarity determining regions (HCDR1, HCDR2, and HCDR3) comprising amino acid sequences of SEQ ID NO: 1 (HCDR1), SEQ ID NO: 2 (HCDR2), and SEQ ID NO: 3 (HCDR3); and three light chain complementarity determining regions (LCDR1, LCDR2, and LCDR3) comprising amino acid sequences of SEQ ID NO: 4 (LCDR1), SEQ ID NO: 5 (LCDR2), and SEQ ID NO: 6 (LCDR3) and is described in WO 2007 / 108756 and in Journal of Alzheimer’s Disease 43:575-588 (2015). BAN2401 comprises (i) a heavy chain variable region comprising the amino acid sequence of SEQ ID NO: 7 and (ii) a light chain variable region comprising the amino acid sequence of SEQ ID NO: 8. The full length heavy chain and light chain sequences of BAN2401 are set forth in SEQ ID Nos: 9 and 10 and are also described in WO 2007 / 108756 and in Journal of Alzheimer’s Disease 43:575-588 (2015).

[0032] While anti-A protofibril treatment can promote amyloid clearance in the brain, itremains a challenge to identify patients with mild cognitive impairment due to Alzheimer’sAttorney Docket No.: 08061.0066-00304 Disease, early Alzheimer’s Disease, or those at risk of developing Alzheimer’s Disease, who would benefit from early therapeutic intervention.

[0033] Existing predictive models trained to detect Alzheimer’s disease may be unsuitable for early detection of MCI. Such models may be trained using patients having a diagnosis of Alzheimer’s disease. But the timeline of progression from MCI to Alzheimer’s disease can vary greatly between patients. A patient may therefore exhibit detectable signs of MCI for many years before being diagnosed with Alzheimer’s disease. A model trained to use a diagnosis of Alzheimer’s disease as a proxy for a diagnosis of MCI may therefore be unreliable.

[0034] Without being bound by theory, to improve on these challenges, the disclosed embodiments include predictive models suitable for predicting MCI risk and predictive models suitable for predicting MCI etiology. As described herein, such predictive models can be used in a method of treating patients with MCI, or separately. The disclosed embodiments can enable early interventions, including emerging treatments such as monoclonal antibody treatments that are indicated for early stages of disease. Such interventions can reduce clinical and economic burdens of disease progression to AD. Furthermore, such predictive models can be used to screen patients for clinical studies, thereby excluding patients unlikely to benefit from treatment and improving the predictive power of such studies.

[0035] Predictive models consistent with disclosed embodiments can include statistical and machine learning models suitable for identifying relationships between input data and output results. Such models can include regression models (e.g., logistic regression models; ridge, lasso, or elastic net regression models; time series regression models; or the like), support vector machines, gradient boosting, Bayes classifiers, neural networks, decision trees, random forests, ensemble models, or other suitable statistical and machine learning models.Attorney Docket No.: 08061.0066-00304

[0036] In some embodiments, suitable predictive models can include regularized logistic regression models and ensemble tree-based models. Regularized logistic regression models can include Bayesian elastic net models. Such models can use a suitable prior (e.g., a mixture double-exponential prior) selected to reduce the complexity of the model, thus preventing overfitting and increasing model robustness. Ensemble tree-based models can include Stochastic Gradient Boosting models, which can combine predictions from multiple decision trees to generate the final predictions. The nodes in each of the multiple decision trees can be trained using different random subsets of input features. The individual decision trees can therefore differ and potentially capture different signals from the data.

[0037] Predictive models consistent with disclosed embodiments can be configured to predict one or more of MCI risk, MCI etiology (e.g., AD, Lewy body dementia, vascular dementia, or the like), recommended confirmatory tests, or potential or recommended treatments for a subject.

[0038] MCI risk data can be or include a score. When the value of the score satisfies a condition (e.g., is greater than or less than a suitable threshold), the score can support a diagnosis of MCI for the subject. MCI risk data can be or include a likelihood. The likelihood can indicate a probability that the subject has MCI. MCI risk data can be or include a label or category (e.g., MCI or non-MCI) for the subject. The MCI risk data can indicate a current risk of MCI, or a risk of being diagnosed with MCI (or satisfying conditions for a diagnosis of MCI) within a time interval (e.g., within the next 6 months, the next 12 months, the next 18 months, the next 24 months, or more).

[0039] In some embodiments, MCI risk data may be Boolean (e.g., returning a yes or no indication of risk for MCI) or may indicate a position of the patient on a scale (e.g., a “low,” “medium,” or “elevated” risk of MCI). In some embodiments, MCI risk data may indicate a relative risk, such as a patient being more likely or less likely to develop MCI in comparisonAttorney Docket No.: 08061.0066-00304 to subjects in the training data. In another example, MCI risk data may express a risk relative to a certain category or characteristic of the individual (e.g., a 69-year-old individual may be predicted to have a higher risk of MCI than similar individuals in a 65-79 age category). In some embodiments, MCI risk data may include a prediction of likelihood or timeline for progressing from MCI to AD.

[0040] In some embodiments, MCI etiology data can be used in a prediction of a dementia etiology, such as Alzheimer’s disease, Lewy body dementia, or vascular dementia. In some embodiments, the predicted MCI etiology may be a sole etiology, as described herein. MCI etiology data can be or include one or more scores. A score can correspond to an MCI etiology. When a score satisfies a condition (e.g., is greater than or less than a suitable threshold), the score can support a diagnosis of the corresponding MCI etiology for the subject. MCI etiology data can be or include one or more likelihoods. A likelihood can correspond to an MCI etiology. The likelihood can indicate a probability of the corresponding MCI etiology for the subject. MCI etiology data can be or include one or more labels or categories for the subject. In some embodiments, only one label or category can be provided (e.g., the label corresponding to the most likely etiology). In some embodiments, multiple labels or categories can be provided.

[0041] Recommended confirmatory tests can depend on predicted risk or predicted etiologies. Such tests can include cognitive assessment tests, biomarker-based tests (e.g., tests using blood, plasma, cerebrospinal fluid, or the like), imaging tests (e.g., magnetic resonance imaging (MRI) scans, positron emission tomography (PET) scans, or the like), or other suitable tests.

[0042] Recommended treatments can depend on predicted risk, predicted etiologies, or the results of any confirmatory tests. Recommended treatments can include the use of therapeuticAttorney Docket No.: 08061.0066-00304 agents. Such therapeutic agents can include, for patients with MCI caused by AD, monoclonal antibody treatments (e.g., lecanemab, aducanumab, donanemab, or the like).

[0043] Predictive models consistent with disclosed embodiments can be configured to generate predictions using inputs including patient medical data. In some embodiments, patient medical data can include at least one of socio-demographic data, comorbidity data (e.g., medical comorbidity data, behavioral comorbidity data, or the like), health status data, or other suitable baseline data. In some embodiments, patient medical data may additionally include medication data (e.g., data relating to specific medications or classes of medications). In some embodiments, patient medical data may additionally include cognitive test scores, behavioral questionnaires, caregiver surveys, and / or genetic or other biomarkers (e.g., such as ApoE 4 mutation status, a level of phosphorylated tau, a ratio of phosphorylated tau to non-phosphorylated tau, or the ratio of A 42 / A 40 in a blood sample). Amyloid 1-42 (A 42)refers to an amyloid beta monomer from amino acid 1 to 42 of the full-length protein.Amyloid 1-40 (A 1-40) refers to an amyloid beta monomer from amino acid 1 to 40 of thefull-length protein. Phosphorylated tau may refer to phosphorylated Tau217 (pTau217), phosphorylated Tau181 (pTau181), or the like. In some cases, the ratio of phosphorylated tau to non-phosphorylated tau may be a ratio of pTau217 to non-phosphorylated Tau217.

[0044] In some embodiments, socio-demographic data may describe social and / or demographic characteristics of the subject. As an example, socio-demographic data may indicate one or more of age, sex, geographic region (e.g., US Census region), income, education, marital status, occupation, race, national origin, or ethnicity. In certain embodiments, socio-demographic data may comprise two or more of age, sex, geographic region (e.g., US Census region), income, education, marital status, occupation, race, national origin, or ethnicity.Attorney Docket No.: 08061.0066-00304

[0045] In some embodiments, comorbidity data can indicate the presence and / or severity or one or more diseases or conditions in a subject. In some embodiments, comorbidity data can include medical comorbidity data and / or behavioral comorbidity data. In some embodiments, the comorbidity data may include metrics or standard measures of severity of comorbidities, such as a comorbidity index. For example, comorbidity data may include the Charlson Comorbidity Index (CCI) or the Elixhauser Comorbidity Index (ECI). In some examples, the comorbidity data may comprise data for at least 2, at least 5, at least 10, at least 20, at least 30, at least 40, or at least 50 comorbidities. In some embodiments, the comorbidity data may comprise data for 2-5 comorbidities, 2-10 comorbidities, 2-20 comorbidities, 2-30 comorbidities, 2-40 comorbidities, 2-50 comorbidities, 5-10 comorbidities, 5-20 comorbidities, 5-30 comorbidities, 5-40 comorbidities, 5-50 comorbidities, 10-20 comorbidities, 10-30 comorbidities, 10-40 comorbidities, 10-50 comorbidities, 20-30 comorbidities, 20-50 comorbidities, or 30-50 comorbidities.

[0046] In some embodiments, medical comorbidity data may include data indicating the presence and / or severity of one or more cardiovascular conditions (e.g., stroke, transient ischemic attack (TIA), congestive heart failure, heart attack, cardiac arrest, atrial fibrillation, hypercholesterolemia, ischemic heart disease, atherosclerosis, hypertension), metabolic conditions (e.g., hyperlipidemia, obesity, metabolic syndrome, thyroid disease, kidney disease, weight loss, diabetes), cancer, neurological conditions (e.g., cognitive conditions or neurological diseases, injuries, and dysfunctions, including concussion, traumatic brain injury, concussion, and the like), respiratory conditions (e.g., chronic obstructive pulmonary disorder (COPD)), sensory conditions (e.g., hearing loss, vision loss, smell or taste disturbances, or the like), sleep conditions (e.g., insomnia, sleep apnea, rapid eye movement (REM) disorder), movement disorders (e.g., cerebral palsy and other paralytic syndromes, gait abnormalities), urinary disorders (e.g., incontinence), or bowel disorders (e.g.,Attorney Docket No.: 08061.0066-00304 constipation). Additionally, or alternatively, medical comorbidity data may include data indicating the presence and / or severity of one or more endocrine, musculoskeletal, soft tissue, blood (e.g., anemia), or infectious diseases or conditions. In some embodiments, medical comorbidity data may include data indicating the presence and / or severity of one or more of stroke / TIA, atherosclerosis, kidney disease, aphasia, concussion, traumatic brain injury (TBI), gait abnormalities, cerebral palsy, incontinence, or rapid eye movement (REM) disorder. In some embodiments, medical comorbidity data may include data indicating the presence and / or severity of one or more of stroke / TIA, gait abnormalities, constipation, urinary incontinence, kidney disease, or sleep apnea. In some embodiments, medical comorbidity data may exclude data relating to certain diseases or conditions. For example, in some embodiments, medical comorbidity data may exclude data relating to one or more of cancer-related conditions, polyneuropathies and other disorders of the peripheral nervous system, demyelinating disease (CNS), or COPD.

[0047] In some embodiments, behavioral comorbidity data may include data indicating the presence and / or severity of substance abuse (e.g., drug or alcohol abuse), psychiatric disorders (e.g., depression, hallucination, bipolar, schizophrenia, psychosis, anxiety, or delusion), emotional symptoms (e.g., combativeness, loneliness, agitation, apathy, or irritability), or the like.

[0048] In some embodiments, health status data may describe the health of a subject. For example, health status data may include biometric data (e.g., body mass index (BMI) or blood pressure), smoking status (e.g., smoker or previous smoker), vision impairments, hearing impairments, or laboratory results (e.g., lipids, hemoglobin A1c, high-density lipoproteins, low-density lipoproteins, total cholesterol, triglycerides, white blood cell count, estimated glomerular filtration rate (EGFR)).Attorney Docket No.: 08061.0066-00304

[0049] In some embodiments, the patient medical data can be obtained at least in part from medical record(s) for the subject, such as electronic health record (EHR) data. In some embodiments, at least a portion (and, in some cases, all) of the patient medical data comprises data that is routinely collected by primary care physicians (PCPs). In some embodiments, the patient medical data may not include MRI scans, PET scans, or cerebrospinal fluid (CSF) biomarker data. In some embodiments, the patient medical data can be obtained at least in part from insurance record(s) for the subject, such as insurance claims. In some embodiments, at least a portion of comorbidity data (e.g., medical comorbidity data, behavioral comorbidity data) may be obtained from diagnosis codes (e.g., International Classification of Diseases (ICD) diagnosis codes) in the patient medical data. In some embodiments, at least a portion of comorbidity data (e.g., medical comorbidity data, behavioral comorbidity data) may be obtained by extracting signs of comorbidity symptoms from clinical notes in the patient medical data using natural language processing (NLP). In some embodiments, the comorbidity data may be obtained using a combination of diagnosis codes (e.g., ICD diagnosis codes) and NLP-derived records.

[0050] A predictive model consistent with disclosed embodiments can be trained using a training dataset. The training dataset can include patient medical data for patients that satisfy a screening criterion. As described herein, in some embodiments a training dataset for training a predictive model to detect MCI can include patient medical data for a cohort of subjects diagnosed with MCI and a cohort of subjects without a diagnosis of MCI (or dementia, such as AD). As described herein, in some embodiments, a training dataset for training a predictive model to detect MCI etiologies can include subjects having a diagnosis of MCI with a known etiology (or combination of known etiologies in some embodiments). By restricting the training cohort to similarly situated subjects, the performance of the predictive model can be improved.Attorney Docket No.: 08061.0066-00304

[0051] While described herein with respect to predicting MCI risk and MCI etiology, systems and methods consistent with disclosed embodiments can be used to predict risk and etiology of neurological disease, dysfunction, or injury more generally. Such neurological disease, dysfunction, or injury can include any condition that affects the central nervous system, resulting in impaired movement, cognition, or behavior. Neurological disease, dysfunction, or injury can include central nervous system diseases (e.g., Alzheimer’s disease, or the like), disorders (e.g., mild cognitive impairment (MCI), or the like), or injuries (e.g., strokes, traumatic brain injury, or the like).

[0052] Based on a diagnosis using a training set or system disclosed herein, a patient may betreated, e.g., by receiving an anti-A protofibril antibody, e.g., lecanemab. Exemplarytreatment regimens for lecanemab, as well as methods of identifying patients for treatment based on biomarkers that may be used in conjunction with the methods disclosed herein, include those provided in PCT / US2022 / 073576, the disclosure of which is incorporated herein by reference in its entirety.

[0053] Models consistent with disclosed embodiments could provide a rapid, adaptable, and low-cost tool for clinicians (e.g., primary care physicians) to detect potential MCI in patients. Such models could alert clinicians to an elevated risk of undetected cognitive impairment in a patient, providing an entry point for triage when a patient is flagged for elevated risk for cognitive impairment. Such models could be integrated into EHR systems and passively recognize the MCI risk. For example, a model integrated into an EHR system could generate a report indicating an elevated risk of undetected cognitive impairment during or after a physical exam based in part on the results of the physical exam. Such a report could be generated during or after the physical exam. For example, the EHR system could be configured with action rules (e.g., flag patient for future evaluation, recommend particular test, or the like) based on predicted MCI risk levels.Attorney Docket No.: 08061.0066-00304

[0054] FIG.1 depicts an exemplary platform 100 for developing, validating, and deploying predictive models for predicting cognitive impairment data, consistent with disclosed embodiments. Platform 100 can be configured to obtain input data from other systems (e.g., such as electronic health record (EHR) systems, insurance systems, imaging systems, or medical laboratory systems, not shown in FIG.1) or record(s) 101.

[0055] Platform 100 can be configured to generate datasets suitable for training predictive models using components such as extract transform load (ETL) engine 110 and dataset creation engine 115. Platform 100 can be configured to train models using training engine 120. Trained models can be used in the prediction phase by prediction engine 130. A user can interact with user device 199 to control and configure platform 100. The user can also provide data concerning subjects to, and receive predictions from, platform 100 by interacting with user device 199.

[0056] As may be appreciated, the particular arrangement of components depicted in FIG.1 is not intended to be limiting. Platform 100 can include additional components (e.g., additional databases, data sources, processing systems, or the like) or fewer components (e.g., by combining databases or processing systems). The functionality of the existing components can be combined or distributed among additional systems, without departing from the envisioned embodiments.

[0057] Components of FIG.1 can be implemented using one or more computing systems (e.g., a laptop, desktop, workstation, computing cluster, on-premises or off-premises cloud computing platform, or the like). For example, a computing cluster or workstation can implement ETL engine 110, dataset creation engine 115, or training engine 120. As an additional example, a desktop or laptop (e.g., user device 199 or another device) can implement prediction engine 130. As an additional example, the components of platform 100Attorney Docket No.: 08061.0066-00304 (e.g., apart from user device 199) can be implemented using containerized services on a cloud computing platform. As may be appreciated, these examples are not intended to be limiting.

[0058] Consistent with disclosed embodiments, record(s) 101 can include one or more storage locations for data usable by platform 100 to predict MCI risk or MCI etiology. In various embodiments such data can include medical record information for the subjects. Such medical record information can include insurance records, medical records, case notes, clinical trial records, requisition information (e.g., pertaining to biomarker testing), imaging data, or the results of laboratory tests. In some embodiments, record(s) 101 may include socio-demographic data. In some embodiments, records(s) 101 may include comorbidity data (e.g., medical comorbidity data, behavioral comorbidity data). In some embodiments, record(s) 101 may include health status data. In some embodiments, record(s) 101 may include medication data. In some embodiments, the information contained in record(s) 101 can include (or be useable to generate) patient medical data, as described herein.

[0059] Consistent with disclosed embodiments, ETL engine 110 can be configured to obtain data in varying formats from one or more sources (e.g., record(s) 101, or the like). The disclosed embodiments are not limited to any particular format of the obtained data, or method for obtaining this data. For example, the obtained data can be or include structured data or unstructured data. ETL engine 110 can interact with the various data sources to receive or retrieve the data.

[0060] ETL engine 110 can transform the data into suitable format(s) and load the transformed data into a target component or database of platform 100. In some embodiments, transforming the data can include performing quality control processing on obtained data. Such quality control processing can include confirming that data is usable (e.g., that the subject satisfies inclusion criteria for the model to be trained, that required input data for a subject is complete, or the like). In some embodiments, transforming the data can includeAttorney Docket No.: 08061.0066-00304 processing the data into a standard format or structure. As may be appreciated, the input data obtained from record(s) 101 may not be in a suitable format for training a predictive model. Similarly, input data obtained from different ones of record(s) 101 may have different formats. Accordingly, ETL engine 110 can clean the obtained input data such that the input data, although originating from a variety of different sources, has a consistent format.

[0061] In some embodiments, ETL engine 110 can enrich medical record information by generating additional data using the medical record information. For example, ETL engine 110 can normalize data (e.g., converting data to a standard format, such as a standard format for dates), or the like. In another example, ETL engine 110 may normalize data by converting data with equivalent meanings to a standard term (e.g., for a dataset including instances of both “heart attack” and “myocardial infarction,” ETL engine 110 may change instances of “heart attack” to “myocardial infarction”). In some embodiments, ETL engine 110 can remove unnecessary or unwanted variables or data from the input dataset. For example, when a medical record contains information unrelated to predicting an MCI risk or MCI etiology, ETL engine 110 can create a version of the medical record that contains only the information related to predicting MCI risk or MCI etiology. In some embodiments, ETL engine 110 may exclude data relating to certain one or more comorbidities. For example, in some embodiments, ETL engine 110 may exclude data relating to one or more of cancer-related conditions, polyneuropathies and other peripheral nervous system disorders, demyelinating disease, and COPD.

[0062] Consistent with disclosed embodiments, ETL engine 110 can load the transformed data into another component of platform 100, such as dataset creation engine 115 (or into a suitable data storage, from which dataset creation engine 115 can retrieve the data).

[0063] In some embodiments, dataset creation engine 115 can be configured to generate training samples or inference samples from data received from ETL engine 110. DatasetAttorney Docket No.: 08061.0066-00304 creation engine 115 can be configured to extract any necessary input data features from the transformed data received from ETL engine 110. In some embodiments, dataset creation engine 115 may extract comorbidity data (e.g., medical comorbidity data, behavioral comorbidity data) from input data for a subject. For example, in some embodiments, dataset creation engine 115 may identify the presence of a comorbidity (e.g., a medical comorbidity, a behavioral comorbidity) based on a diagnosis code (e.g., an ICD diagnosis code) in the medical record information and / or by using a natural language processing algorithm to identify signs or symptoms of the comorbidity in the medical record information. In some embodiments, dataset creation engine 115 can generate features based on combinations of subject data and patient population data (e.g., biomarker values for a subject normalized by biomarker values for a population).

[0064] In some embodiments, dataset creation engine 115 can be configured to accept label information provided by a user through user device 199. For example, dataset creation engine 115 can be configured to provide data (or metadata concerning the data) received from ETL engine 110 to user device 199 for display. In response, dataset creation engine 115 can receive label information (e.g., identification of a subject as having mild cognitive impairment, comorbidity data for a patient obtained during an assessment, or the like).

[0065] In some embodiments, dataset creation engine 115 can be configured to associate labels with training samples. As an example, when predicting cognitive impairment etiology, a diagnosis of vascular dementia (e.g., as indicated by a diagnosis code, such as an ICD diagnosis code) can be noted in a medical record of a subject. Dataset creation engine 115 can then include in a training example the patient medical data for the subject and a vascular dementia label (e.g., a “ground truth” class label, or the like). As an additional example, findings supporting a diagnosis of vascular dementia can be noted in a medical record of a subject. In some embodiments, dataset creation engine 115 may use a natural languageAttorney Docket No.: 08061.0066-00304 processing algorithm to identify the findings as signs or symptoms of vascular dementia. Dataset creation engine 115 can then include in a training example the patient medical data for the subject and a vascular dementia label. Dataset creation engine 115 can be configured to store training samples in data storage 105.

[0066] Consistent with disclosed embodiments, model storage 103 can be a storage location for models usable by components of platform 100 (e.g., training engine 120, or prediction engine 130). The disclosed embodiments are not limited to any particular implementation of model storage 103. Consistent with disclosed embodiments, model storage 103 can be implemented using one or more relational databases, object-oriented or document-oriented databases, tabular data stores, graph databases, distributed file systems, or other suitable data storage options.

[0067] Consistent with disclosed embodiments, data storage 105 can be a storage location for prepared datasets usable by training engine 120 or prediction engine 130. The disclosed embodiments are not limited to any particular implementation of data storage 105. Consistent with disclosed embodiments, data storage 105 can be implemented using one or more relational databases, object-oriented or document-oriented databases, tabular data stores, graph databases, distributed file systems, or other suitable data storage options.

[0068] Consistent with disclosed embodiments, training engine 120 can be configured to train, or create and train, models. Training engine 120 can be configured to create models (e.g., in response to a command to create a trained model of a particular type using an input dataset) or obtain existing models from model storage 103. Training engine 120 can be configured to create or train models using training datasets obtained from data storage 105. In some embodiments, training engine 120 can be configured to store trained models in model storage 103.Attorney Docket No.: 08061.0066-00304

[0069] Consistent with disclosed embodiments, training engine 120 can include model training and model evaluation components. Training engine 120 can be configured to train a model using a model training component and then determine performance measure values for the model using a model evaluation component.

[0070] In some embodiments, training engine 120 can provide a model and a cross-validation or holdout portion of a training dataset to the model evaluation component. In some embodiments, training engine 120 can specify one or more performance measures. Additionally, or alternatively, the model evaluation component can be configured with a predetermined or default set of performance measures. In some embodiments, the performance measures can include confusion matrices, mean-squared-error, mean-absolute- error, sensitivity or selectivity, receiver operating characteristic curves or area under such curves, precision and recall, F-measure, or any other suitable performance measure. In some embodiments, performance measure values can be displayed to a user through user device 199. The user may then interact through user device 199 with training engine 120 to update the model.

[0071] In some embodiments, training engine 120 can automatically update the model being trained based on the performance measure values. In various embodiments, training engine 120 can update the model being trained in response to user input provided through user device 199. Updating the model can include one or more of performing additional training (e.g., using the existing training dataset or another training dataset), modifying the model (e.g., changing the input features used by the model, changing the architecture of the model, or the like), or changing the training environment (e.g., changing training hyperparameters, changing a division of the training dataset into training, cross-validation, and holdout portions, or the like).Attorney Docket No.: 08061.0066-00304

[0072] Consistent with disclosed embodiments, prediction engine 130 can be configured to predict MCI risk, MCI etiology, or the like for a subject using a prediction model. In some embodiments, prediction engine 130 can obtain the trained prediction model from model storage 103. In some embodiments, prediction engine 130 can obtain input data for the subject from data storage 105. In some embodiments, prediction engine 130 can obtain the input data from another data storage location. This alternative data storage location can be associated with another entity or user. For example, prediction engine 130 can receive or retrieve the subject data from a healthcare system controlled by an entity distinct from the entity that controls prediction engine 130. In some embodiments, components of platform 100 (e.g., ETL engine 110, dataset creation engine 115, or the like) can be configured to process the retrieved subject data into suitable data for inputting to the prediction model.

[0073] Consistent with disclosed embodiments, the subject data can include patient medical data or data usable by platform 100 to generate patient medical data. The prediction engine 130 can apply the patient medical data to the trained prediction model to provide, for example, output data indicating MCI risk or MCI etiology for the subject. The output can be provided by prediction engine 130 to user device 199. In some embodiments, the output can be stored on a computing device associated with platform 100 or provided to another system.

[0074] Consistent with disclosed embodiments, user device 199 can provide a user interface for interacting with other components of platform 100. The user interface can be a graphical user interface. The user interface can enable a user to configure ETL engine 110 to extract, transform, and load data according to user specifications. The user interface can enable the user to specify how the transformed data received by dataset creation engine 115 is converted into labeled training data (or suitable patient data). In some embodiments, the user interface can enable the user to interact with dataset creation engine 115 to manually or semi-manually label or annotate the training data. In some embodiments, the user interface can enable a userAttorney Docket No.: 08061.0066-00304 to provide data or models to training engine 120 for training, or to prediction engine 130 for identification and classification.

[0075] In some embodiments, the user interface can enable a user to interact with training engine 120 to create or select a model for training, create or select a dataset for use in training the model, or select training parameters or hyperparameters. In some embodiments, the user interface can enable a user to interact with training engine 120 to display information related to training of the model (e.g., performance measure values, loss function values during training or functions thereof, or other training information). In some embodiments, the user interface can enable a user to interact with prediction engine 130 to select a training model and patient data. In some embodiments, the user interface can enable a user to interact with prediction engine 130 to display prediction data, store prediction data on a computing device, or transmit the prediction data to another system.

[0076] Components of platform 100 can be implemented using one or more computing devices. Such computing devices can include tablets, laptops, desktops, workstations, computing clusters, or cloud computing platforms. In some embodiments, components of platform 100 can be implemented using cloud computing platforms. For example, one or more of ETL engine 110, dataset creation engine 115, training engine 120, and prediction engine 130 can be implemented on a cloud computing platform. In some embodiments, components of platform 100 can be implemented using on-premises systems. For example, record(s) 101 or user device 199 can be, or be hosted on, on-premises systems. As an additional example, model storage 103 or data storage 105 can be, or be hosted on, on- premises systems.

[0077] Components of platform 100 can communicate using any suitable method. In some embodiments, two or more components of platform 100 can be implemented as microservices or web services. Such components can communicate using messages transmitted on aAttorney Docket No.: 08061.0066-00304 computer network. The messages can be implemented using SOAP, XML, HTTP, JSON, RCP, or any other suitable format. In some embodiments, two or more components of platform 100 can be implemented as software, hardware, or combined software / hardware modules. Such components can communicate using data or instructions written to or read from a memory (e.g., a shared memory), function calls, or any other suitable communication method.

[0078] As may be appreciated, the particular structure of platform 100 is not intended to be limiting. Consistent with disclosed embodiments, any two or more of record(s) 101, model storage 103, or data storage 105 can be combined, or hosted on the same computing device. Consistent with disclosed embodiments, ETL engine 110 and dataset creation engine 115 can be omitted from platform 100. In such embodiments, datasets formatted and configured for use by training engine 120 or prediction engine 130 can be deposited in data storage 105 by another system or using another method. Consistent with disclosed embodiments, ETL engine 110 and dataset creation engine 115 can be combined. In such embodiments, data extraction, transformation, and loading can be combined with feature extraction, labeling, and sample creation. Consistent with disclosed embodiments, training engine 120 and prediction engine 130 can be combined.

[0079] Though shown with one user device 199, platform 100 could have multiple user devices. Different user devices could be associated with different entities or different users having different roles. For example, user device 199 could be associated with a software engineer or data scientist who is developing the prediction model, while another user device could be associated with a clinician who is using the prediction model.

[0080] User device 199 can be combined with one or more other components of platform 100. In some embodiments, user device 199 and at least one of ETL engine 110, dataset creation engine 115, training engine 120, or prediction engine 130 can be implemented by theAttorney Docket No.: 08061.0066-00304 same computing device. In various embodiments, user device 199 and at least one of model storage 103 or data storage 105 can be implemented by the same computing device. As may be appreciated, platform 100 can be integrated into a method for treating subjects or for conducting clinical trials. Prediction engine 130 can use a trained prediction model and input data for the subject to predict data indicating MCI risk or MCI etiology. The predicted MCI risk or MCI etiology can be used to determine a patient treatment plan for the patient. In some embodiments, the MCI risk or MCI etiology for the subject can be used as a screening tool for a clinical trial, or the like.

[0081] FIG.2 depicts an exemplary process 200 of predicting the risk of cognitive impairment for a subject, consistent with disclosed embodiments. Process 200 can be performed as part of process 400, described herein, or separately. For convenience of description, process 200 is described as being performed using platform 100. However, process 200 can also be performed at least in part using another computing system. Process 200 can include a dataset creation phase, a training phase, and a prediction phase. In the dataset creation phase, components of platform 100 (e.g., ETL engine 110, dataset creation engine 115, or the like) can obtain training data from databases (e.g., record(s) 101, or the like). In the training phase, components of platform 100 (e.g., training engine 120, or the like) can create or refine predictive models for predicting MCI risk data from patient medical data. In the prediction phase, components of platform 100 (e.g., prediction engine 130, or the like) can use prediction model(s) generated in the training phase to predict MCI risk data from patient medical data for a subject. In some instances, the predicted MCI risk data can be used manage treatment for the subject. In some embodiments, based on the prediction that the subject has MCI, an etiology of the MCI can be investigated, as described herein.

[0082] In step 210 of process 200, components of platform 100 (e.g., ETL engine 110, dataset creation engine 115, or the like) can obtain training data, consistent with disclosedAttorney Docket No.: 08061.0066-00304 embodiments. The training data may include patient medical data for subjects, as described herein. In some embodiments, the patient medical data may include at least one of socio- demographic data, comorbidity data, health status data, medication data, or other suitable baseline data. In some embodiments, any included socio-demographic data can describe at least one of social and / or demographic characteristics of the subject. In some embodiments, any included comorbidity data can include at least one of medical comorbidity data or behavioral comorbidity data. In some embodiments, any included health status data may describe at least one health characteristic of the subject. In some embodiments, any included medication data may describe at least one medication or class of medications taken by the subject. Patient medical data may be assessed as-of an index date for the subject or within a certain time period of the index date. In some embodiments, subsequent changes in the medical history of the subject may not be included in the training data for that subject if they occurred after the index date.

[0083] In some embodiments, the obtained training data in step 210 may concern subjects satisfying an inclusion criterion. The inclusion criterion may concern the age or cognitive status of the subject. For example, the obtained training data may be limited to subjects satisfying a given age criterion, such as having an age of at least 35, 40, 50, 55, 60, 65, 70, or 80 years old in the year of the index date. In some embodiments, the obtained training data may be limited to subjects falling within one or more given age groups, such as 40-49 years old, 50-64 years old, 65-79 years old, or at least 80 years old. Additionally, or alternatively, the obtained training data may include patient medical data for subjects satisfying a cognitive impairment condition. In some embodiments, a subject may satisfy the cognitive impairment condition when the subject has a diagnosis of MCI (e.g., as indicated by a diagnosis code, such as an ICD diagnosis code), or exhibits signs or symptoms indicative of MCI (e.g., clinician notes indicating signs of cognitive impairment, such as confusion or inability toAttorney Docket No.: 08061.0066-00304 respond to instructions; test results, such as amyloid positivity on a PET scan; a biomarkerscore, such as a plasma, serum, or cerebrospinal fluid p-Tau181, p-Tau217, A 1-42 score, orA 1-40 score; or the like). In some embodiments, a subject may satisfy the cognitiveimpairment condition if the subject has an initial diagnosis of MCI and one or more subsequent diagnoses of MCI, AD, or a related dementia. In certain embodiments, a subject may not satisfy the cognitive impairment condition if the subject has only one MCI diagnosis and does not have a subsequent MCI, AD, or related dementia diagnosis.

[0084] In some embodiments, a first portion of the subjects may satisfy the cognitive impairment condition. A second portion of subjects may satisfy a control condition. Such patients may not have a diagnosis of MCI or exhibit signs or symptoms indicative of MCI. In some embodiments, such patients may lack a diagnosis of (or signs or symptoms indicative of) another neurological disease, disorder, or condition. In some embodiments, a subject may satisfy the control condition if the subject does not have a diagnosis for (or signs or symptoms indicative of) MCI, AD, or a related dementia as of an index date and for at least two years after the index date. In some embodiments, the control condition may further require that a subject not have a diagnosis for (or signs or symptoms indicative of) a condition associated with cognitive impairment. Non-limiting examples of conditions associated with cognitive impairment include delirium; disorientation; attention deficit; impairment in comprehension, concentration, speech, executive function, language, or naming; inability to find the right word, lack of judgment, and lack of insight. In some embodiments, the control condition may further require that a subject not have any records of taking a dementia-related medication (e.g., lecanemab, aducanumab, donanemab, donepezil, memantine, memantine / donepezil, galantamine, rivastigmine).

[0085] Platform 100 may be used to determine the obtained training data. For example, the components of platform 100 can obtain at least a portion of the training data from a databaseAttorney Docket No.: 08061.0066-00304 (e.g., record(s) 101 or the like) or another system. In some embodiments, the components of platform 100 can generate at least a portion of the training data. For example, dataset creation engine 115 can identify subjects having a diagnosis of MCI or exhibiting signs or symptoms indicative of MCI (e.g., for inclusion in the first cohort of subjects) or subjects lacking such diagnoses or indications (e.g., for inclusion in the control subjects). In some embodiments, dataset creation engine 115 can exclude certain subjects from inclusion in one or more cohorts. For example, in some embodiments, patients with diagnosis records for Parkinson’s disease or parkinsonism may be excluded from the first cohort of subjects and the control subjects. In some embodiments, dataset creation engine 115 may also stratify or code subjects based on patient medical data. For example, dataset creation engine 115 may code subjects into age groups (e.g., a first age category for 50- to 64-year-old subjects, a second age category for 65- to 79-year-old subjects, a third age category for greater than 80-year-old subjects, and the like).

[0086] In step 220 of process 200, components of platform 100 (e.g., training engine 120, or the like) can train a predictive model to predict MCI risk, consistent with disclosed embodiments. For example, a predictive model may be trained to predict MCI risk for a subject given patient medical data for that subject. In some embodiments, training engine 120 can create the predictive model and then store the predictive model in model storage 103. In some embodiments, training engine 120 can obtain a predictive model from model storage 103, or another database or system, and then refine the model. As described herein, the trained predictive model may include any suitable predictive model, such as a Bayesian logistic lasso regression, a stochastic gradient boosting machine, an extreme gradient boosting, or a random forest model.

[0087] In some embodiments, training engine 120 can obtain hyperparameters for training the predictive model. The particular hyperparameters obtained can depend on the type ofAttorney Docket No.: 08061.0066-00304 predictive model and the disclosed embodiments are not limited to any particular set of hyperparameters. For example, a neural network model may have hyperparameters governing layer arrangement and configuration, batch size, dropout, or the like. As an additional example, a gradient boosted model may have hyperparameters governing learning rate, number of trees, bagging fraction, tree depth, or the like.

[0088] In some embodiments, a user can interact with user device 199 to provide hyperparameters to training engine 120. In some embodiments, training engine 120 can receive or retrieve hyperparameters from another component of platform 100. In some embodiments, training engine 120 can generate suitable hyperparameters. For example, training engine 120 can be configured to conduct an iterative or adaptive search of a predetermined hyperparameter space (e.g., through training predictive models, evaluating the performance of the models, and updating the selected hyperparameters based on the performance of the models).

[0089] In some embodiments, training engine 120 can train the predictive model using the hyperparameters and the training data obtained in step 210. For example, the obtained training data may include a training dataset of patient medical data. In some embodiments, the training dataset may have a first portion including subjects satisfying a mild cognitive impairment condition, as described herein. For example, the first portion of the training dataset may include subjects having an MCI diagnosis. In some embodiments, subjects in the first portion of the training dataset may have at least one MCI diagnosis and at least one subsequent diagnosis for MCI, AD, or a related dementia. In some embodiments, the training dataset may have a second portion including control subjects. For example, the control subjects may refer to subjects without an MCI diagnosis. In some embodiments, control subjects may not have a diagnosis (or signs or symptoms) of MCI, AD or a related dementia. In some embodiments, control subjects may not have a diagnosis (or signs or symptoms) ofAttorney Docket No.: 08061.0066-00304 cognitive problems associated with cognitive impairment (e.g., delirium; disorientation; attention deficit; impairment in comprehension, concentration, speech, executive function, language, or naming; inability to find the right word, lack of judgment, or lack of insight). In some embodiments, control subjects may not have any records of taking a dementia-related medication (e.g., lecanemab, aducanumab, donanemab, donepezil, memantine, memantine / donepezil, galantamine, rivastigmine). The dataset creation engine 115 may randomly select training data for subjects from record(s) 101 such that the ratio of subjects with MCI to control (e.g., non-MCI) subjects may be, as an example, in the range of 30:1 to 3:1.

[0090] In some embodiments, training engine 120 can be configured to evaluate the performance of multiple model designs using the same training dataset. The performance of a model design can be determined using k-fold cross validation. In some embodiments, training engine 120 can be configured to evaluate the performance of the best-performing model design by dividing the training dataset into training and validation subsets. Training engine 120 can evaluate model designs by performing k-fold cross validation using the training subset. Training engine 120 can select a model design and evaluate the performance of that model design using the validation subset.

[0091] Some disclosed embodiments may involve providing the trained predictive model. For example, platform 100 may provide the trained predictive model to model storage 103. The trained predictive model may be stored together with other predictive models in model storage 103. The predictive model can be used as-is or trained further using training engine 120. Additionally, or alternatively, the trained predictive model may be provided to generate predictions. For example, the trained predictive model may be provided to predict MCI risk using patient medical data stored in data storage 105, or received from user device 199.Attorney Docket No.: 08061.0066-00304

[0092] In step 230 of process 200, components of platform 100 (e.g., ETL engine 110, dataset creation engine 115, prediction engine 130, or the like) can obtain patient medical data for a patient. The patient medical data may be obtained from medical record(s), insurance record(s), and / or other sources. In some embodiments, the components of platform 100 can obtain at least a portion of the patient data from a database (e.g., record(s) 101 or the like) or from another system. For example, platform 100 can be configured to accept prediction requests from other systems. Additionally, or alternatively, individual data may be obtained from user device 199. In some embodiments, the components of platform 100 can generate at least a portion of the patient data.

[0093] In some embodiments, the patient and the subjects from which the training data was obtained can satisfy similar inclusion criteria, such as an age criterion (e.g., the patient may have an age corresponding to an age category of the training data). In some embodiments, the patient may differ from the subjects from which the training data was obtained.

[0094] In some embodiments, the patient medical data for the patient can be the same as the patient medical data included in the training data. For example, when the training patient medical data includes a combination of a certain socio-demographic data, health status data, and comorbidity data, the prediction patient medical data can include the same socio- demographic data, health status data, and comorbidity data. As may be appreciated, obtaining the prediction patient medical data can include reformatting or arranging the prediction patient medical data to match the format or arrangement of the training patient medical data. Similarly, obtaining the prediction patient medical data can include handling missing values or erroneous values in the prediction patient medical data. Furthermore, when obtaining the training patient medical data includes generating certain values (e.g., coding age groups, conditions, or the like), obtaining the prediction patient medical data can similarly include generating these values.Attorney Docket No.: 08061.0066-00304

[0095] In step 240 of process 200, components of platform 100 (e.g., prediction engine 130, or the like) can predict the MCI risk for a patient, consistent with disclosed embodiments. Patient medical data for the patient can be input (e.g., by prediction engine 130, or the like) to a trained prediction model to predict the MCI risk.

[0096] In some embodiments, platform 100 can provide the predicted MCI risk data. For example, platform 100 can provide the predicted MCI risk data to a user of platform 100 (e.g., by providing the predicted risk data to user device 199 for display), store the predicted MCI risk data in a component of platform 100, provide the predicted MCI risk data to another system (e.g., a system that provided a prediction request), or the like. For example, a user (e.g., a clinician, the subject, a relative, family member, or caregiver of the subject, or the like) may interact with user device 199 to generate predicted MCI risk data for a subject based on patient medical data for the subject.

[0097] As may be appreciated, a trained predictive model can be used to screen or select patients for inclusion in a clinical trial. MCI risk data can be predicted for candidate patients and can be used to enrich the clinical trial population with patients likely to benefit from a treatment. As may be appreciated, treatment effect, study size, and study power can be related. By selecting patients with MCI or at risk of MCI in a study of MCI treatments (or treatments for MCI of a particular etiology), fewer patients can be enrolled, or study power can be increased, or detectable treatment effect size reduced, or some combination of the foregoing.

[0098] FIG.3 depicts a process 300 for predicting MCI etiology for subjects, consistent with disclosed embodiments. Process 300 can be used as part of process 400, as described herein, or as part of another process. For convenience of description, process 300 is described as being performed using platform 100. However, process 300 can also be performed at least in part using another computing system. Process 300 can include a dataset creation phase, aAttorney Docket No.: 08061.0066-00304 training phase, and a prediction phase. In the dataset creation phase, components of platform 100 (e.g., ETL engine 110, dataset creation engine 115, or the like) can obtain training data from databases (e.g., record(s) 101, or the like). In the training phase, components of platform 100 (e.g., training engine 120, or the like) can create or refine predictive models for predicting MCI etiology data from patient medical data. In the prediction phase, components of platform 100 (e.g., prediction engine 130, or the like) can use prediction model(s) generated in the training phase to predict MCI etiology data from patient medical data for a subject. The subject may be selected for such prediction based on MCI risk data generated using another predictive model. For example, the MCI risk data can include a score, likelihood, category, or the like that indicates that the subject has MCI or has an elevated risk of MCI. In some instances, the predicted MCI etiology data can be used to manage treatment for the subject.

[0099] In some embodiments, the trained predictive model can be used with patients having MCI. In a prior step, the patients can be identified as having MCI. As may be appreciated, a predictive model trained to predict MCI etiology may be simpler, easier to train, or exhibit superior performance as compared to a predictive model trained to both detect MCI and predict MCI etiology.

[0100] In step 310 of process 300, components of platform 100 (e.g., ETL engine 110, dataset creation engine 115, or the like) can obtain training data, consistent with disclosed embodiments. The training data may include patient medical data for subjects. In some embodiments, the patient medical data can be included in, or be generated using data included in, medical records of the subjects. In some embodiments, the patient medical data may include comorbidity data. For example, the comorbidity data may include medical comorbidity data and / or behavioral comorbidity data, as described herein. In some examples, the comorbidity data may comprise data for at least 2, at least 5, at least 10, at least 20, atAttorney Docket No.: 08061.0066-00304 least 30, at least 40, or at least 50 comorbidities. In some embodiments, the comorbidity data may comprise data for 2-5 comorbidities, 2-10 comorbidities, 2-20 comorbidities, 2-30 comorbidities, 2-40 comorbidities, 2-50 comorbidities, 5-10 comorbidities, 5-20 comorbidities, 5-30 comorbidities, 5-40 comorbidities, 5-50 comorbidities, 10-20 comorbidities, 10-30 comorbidities, 10-40 comorbidities, 10-50 comorbidities, 20-30 comorbidities, 20-50 comorbidities, or 30-50 comorbidities. In some embodiments, the patient medical data may additionally or alternatively include socio-demographic data or health status data. In some embodiments, the comorbidity data and health status data may be obtained before or within a certain time period of diagnosis of MCI (e.g., for subjects having MCI). For example, the comorbidity data and health status data may be measured or recorded before or within a year or two years of diagnosis of MCI.

[0101] In some embodiments, the obtained training data in step 310 may concern subjects satisfying an age criterion. For example, the patient medical data may be for subjects satisfying a specific age as an inclusion criterion, such as having an age of at least 40, 50, 55, 60, 65, 70, or 80 years old in the year of the index date. In some embodiments, the obtained training data may be limited to subjects falling within one or more given age groups, such as 40-49 years old, 40-64 years old, 50-64 years old, 65-79 years old, or at least 80 years old. Additionally, or alternatively, the obtained training data in step 210 may concern subjects satisfying a cognitive impairment condition. For example, the cognitive impairment condition may be that the subject has a diagnosis of MCI, or has signs or symptoms that support a diagnosis of MCI. In some embodiments, the training data may include MCI etiology data. In some embodiments, the training data may include medical records for subjects that have a confirmed (e.g., diagnosed) clinical etiology for MCI, such as Alzheimer’s disease, Lewy body dementia, or vascular dementia.Attorney Docket No.: 08061.0066-00304

[0102] In some embodiments, platform 100 (or components thereof) can retrieve or receive at least a portion of the training data. For example, platform 100 can obtain at least a portion of the training data from a database (e.g., record(s) 101 or the like) or another system. In some embodiments, the components of platform 100 can generate at least a portion of the training data. For example, dataset creation engine 115 can identify a cohort of subjects that satisfy an MCI condition. A subject can satisfy an MCI condition when they have a diagnosis of MCI or indications of MCI. Dataset creation engine 115 may further identify subjects having an identified MCI etiology (or suitable indications of MCI etiology). In some embodiments, dataset creation engine 115 can identify subjects satisfying a sole-etiology dementia condition. A sole-etiology dementia condition may be satisfied when MCI results from a single cause. For example, a subject with MCI resulting from Alzheimer’s disease may satisfy the sole-etiology dementia condition, while a subject with MCI resulting from Alzheimer’s disease and vascular dementia may not.

[0103] Accordingly, in some embodiments, the training data can include patient medical data for subjects having an MCI diagnosis and satisfying a sole-etiology dementia condition. The subjects can be associated with labels corresponding to etiology. For example, the training data can include a training sample for a subject with MCI resulting from Alzheimer’s disease. The training sample can include patient medical data for a subject and a label corresponding to Alzheimer’s disease.

[0104] In some embodiments, dataset creation engine 115 may remove subjects having mixed dementia from the training data (e.g., subjects that do not satisfy the sole-etiology dementia condition). In some embodiments, given a superset of subjects, dataset creation engine 115 can select subjects for the training dataset according to etiology. Among subjects having the same etiology, subjects may be selected randomly. For example, dataset creation engine 115 may select ratios of subjects or a given number of subjects depending on theirAttorney Docket No.: 08061.0066-00304 etiology (e.g., obtaining patient medical data of 1200 subjects with an etiology of Alzheimer’s disease, 140 subjects with an etiology of Lewy body dementia, and 110 subjects with an etiology of vascular dementia).

[0105] In step 320 of process 300, components of platform 100 (e.g., training engine 120, or the like) can train a predictive model to predict MCI etiology, consistent with disclosed embodiments. For example, a predictive model may be trained to predict MCI etiology for a subject given patient medical data for that subject. In some embodiments, training engine 120 can create the predictive model and then store the predictive model in model storage 103. In some embodiments, training engine 120 can obtain a predictive model from model storage 103, or another database or system, and then refine the model. As described herein, the trained predictive model may include any suitable predictive model, such as a Bayesian logistic lasso regression, a stochastic gradient boosting machine, an extreme gradient boosting, or a random forest model.

[0106] In some embodiments, training engine 120 can obtain hyperparameters for training the predictive model. The particular hyperparameters obtained can depend on the type of predictive model and the disclosed embodiments are not limited to any particular set of hyperparameters. For example, a neural network model may have hyperparameters governing layer arrangement and configuration, batch size, dropout, or the like. As an additional example, a gradient boosted model may have hyperparameters governing learning rate, number of trees, bagging fraction, tree depth, or the like.

[0107] In some embodiments, a user can interact with user device 199 to provide hyperparameters to training engine 120. In some embodiments, training engine 120 can receive or retrieve hyperparameters from another component of platform 100. In some embodiments, training engine 120 can generate suitable hyperparameters. For example, training engine 120 can be configured to conduct an iterative or adaptive search of aAttorney Docket No.: 08061.0066-00304 predetermined hyperparameter space (e.g., through training predictive models, evaluating the performance of the models, and updating the selected hyperparameters based on the performance of the models).

[0108] In some embodiments, training engine 120 can train the predictive model using the hyperparameters or the training data obtained in step 310. As described herein, in some embodiments, the training dataset may include subjects satisfying an MCI condition and having an etiology for the MCI condition. A sample for a subject in the training dataset can include patient medical data for the subject and a label indicating the MCI etiology for the subject. The predictive model can be trained to predict the label given the patient medical data.

[0109] Some disclosed embodiments may involve providing the trained predictive model. For example, platform 100 may provide the trained predictive model to model storage 103. The trained predictive model may be stored together with other predictive models in model storage 103. The predictive model can be used as-is or trained further using training engine 120. Additionally, or alternatively, the trained predictive model may be provided to generate predictions. For example, the trained predictive model may be provided to predict MCI etiology using patient medical data stored in data storage 105, or received from user device 199.

[0110] In step 330 of process 300, components of platform 100 (e.g., ETL engine 110, dataset creation engine 115, prediction engine 130, or the like) can obtain patient medical data for a patient. The patient medical data may be obtained from medical record(s), insurance record(s), and / or other sources. In some embodiments, the components of platform 100 can obtain at least a portion of the patient data from a database (e.g., record(s) 101 or the like) or from another system. For example, platform 100 can be configured to accept prediction requests from other systems. Additionally, or alternatively, patient data may beAttorney Docket No.: 08061.0066-00304 obtained from user device 199. In some embodiments, the components of platform 100 can generate at least a portion of the patient medical data.

[0111] In some embodiments, the patient may have a diagnosis of MCI. In some embodiments, the patient and the subjects from which the training data was obtained can satisfy similar inclusion criteria, such as an age criterion (e.g., the patient may have an age corresponding to an age category of the training data). In some embodiments, the patient may differ from the subjects from which the training data was obtained.

[0112] In some embodiments, the patient medical data for the patient can be the same as the patient medical data included in the training data. For example, when the training patient medical data includes a combination of a certain socio-demographic data, health status data, and comorbidity data, the prediction patient medical data can include the same socio- demographic data, health status data, and comorbidity data. As may be appreciated, obtaining the prediction patient medical data can include reformatting or arranging the prediction patient medical data to match the format or arrangement of the training patient medical data. Similarly, obtaining the prediction patient medical data can include handling missing values or erroneous values in the prediction patient medical data. Furthermore, when obtaining the training patient medical data includes generating certain values (e.g., coding age groups, conditions, or the like), obtaining the prediction patient medical data can similarly include generating these values.

[0113] In step 340 of process 300, components of platform 100 (e.g., prediction engine 130, or the like) can predict MCI etiology for a patient, consistent with disclosed embodiments. Patient medical data for the patient can be input (e.g., by prediction engine 130, or the like) to a trained prediction model to predict the MCI etiology.

[0114] In some embodiments, platform 100 can provide the predicted MCI etiology data. For example, platform 100 can provide the predicted MCI etiology data to a user of platform 100Attorney Docket No.: 08061.0066-00304 (e.g., by providing the predicted MCI etiology data to user device 199 for display), store the predicted MCI etiology data in a component of platform 100, provide the predicted MCI etiology data to another system (e.g., a system that provided a prediction request), or the like. For example, a user (e.g., a clinician, the subject, a relative, family member, or caregiver of the subject, or the like) may interact with user device 199 to generate predicted MCI etiology data for a subject based on patient medical data for the subject.

[0115] FIG.4 depicts a process 400 for treating cognitive impairment, consistent with embodiments of the present disclosure. In some embodiments, process 400 is described as being performed using platform 100. However, process 400 can also be performed at least in part using another computing system. Process 400 may involve a step 410 of obtaining patient medical data for a patient. In some embodiments, the patient medical data can be obtained from medical records of the patient. Components of platform 100 (e.g., ETL engine 110, dataset creation engine 115, or the like) can obtain training data, consistent with disclosed embodiments. In some embodiments, the patient medical data may include comorbidity data. For example, the comorbidity data may include medical comorbidity data and / or behavioral comorbidity data, as described herein. In some examples, the comorbidity data may involve at least one comorbidity, such as including data for a cardiovascular comorbidity (e.g., stroke). In some examples, the comorbidity data may comprise data for at least 2, at least 5, at least 10, at least 20, at least 30, at least 40, or at least 50 comorbidities. In some embodiments, the comorbidity data may comprise data for 2-5 comorbidities, 2-10 comorbidities, 2-20 comorbidities, 2-30 comorbidities, 2-40 comorbidities, 2-50 comorbidities, 5-10 comorbidities, 5-20 comorbidities, 5-30 comorbidities, 5-40 comorbidities, 5-50 comorbidities, 10-20 comorbidities, 10-30 comorbidities, 10-40 comorbidities, 10-50 comorbidities, 20-30 comorbidities, 20-50 comorbidities, or 30-50Attorney Docket No.: 08061.0066-00304 comorbidities. Additionally, or alternatively, the patient medical data may include socio- demographic data, health status data, or medication data, as described herein.

[0116] Process 400 may involve a step 420 of predicting MCI risk. Step 420 may involve predicting an MCI risk for the patient by applying at least a portion of the obtained patient medical data (e.g., obtained in step 410) to a predictive model. In some embodiments, the predictive model may be trained to predict MCI risk using training patient medical data for subjects. The subjects may satisfy an age criterion, as described herein. In some embodiments, a portion of the subjects may be associated with an MCI label. For example, the label may indicate a diagnosis of MCI (e.g., by a clinician). In some embodiments, the label may indicate an initial diagnosis of MCI and at least one subsequent diagnosis of MCI, AD, or a related dementia. In another example, the label may indicate that the subjects exhibit certain signs or symptoms corresponding to MCI (e.g., amyloid positivity on a PET scan; abiomarker score, such as a plasma, serum, or cerebrospinal fluid p-Tau181, p-Tau217, A 1-42 score, or A 1-40 score; or the like). In some embodiments, a portion of the subjects maybe associated with a control label. For example, the control label may correspond to subjects having no diagnosis of MCI. In another example, the control label may correspond to subjects not having a diagnosis of MCI or AD, no diagnosis records in the training patient medical data for cognitive problems (e.g., delirium; disorientation; attention deficit; impairment in comprehension, concentration, speech, executive function, language, or naming; inability to find the right word, lack of judgment, lack of insight), or medication records for dementia medications.

[0117] In some embodiments, step 420 may be executed as described regarding process 200. Additionally, or alternatively, step 420 may be executed with other suitable predictive models. As described herein, the predicted cognitive impairment risk data (e.g., predicted in step 420) may involve a prediction that an individual may be at elevated risk for MCI. Thus,Attorney Docket No.: 08061.0066-00304 the predicted cognitive impairment risk data may enable treatment and / or management of MCI risk. For example, the predicted cognitive impairment risk data may include an indication that the patient seek further testing (e.g., cognitive assessment testing, biomarker testing, imaging) or monitoring (e.g., follow-up with a clinician). In some embodiments, the cognitive impairment risk data may include data for determining or predicting etiology of MCI for the patient.

[0118] In some embodiments, process 400 may proceed to step 430 based on the MCI risk predicted in step 420. For example, the predicted MCI risk data may indicate that a patient has MCI or has an elevated risk for MCI. In response, process 400 may proceed to step 430. Otherwise (e.g., should the patient be predicted to have a low risk of MCI or not to have MCI), process 400 can terminate. It will be appreciated that by predicting MCI etiologies for patients identified as likely having MCI, the accuracy of such predictions may be improved. In some embodiments, process 400 may proceed directly to step 430 in response to a diagnosis of MCI, or the like.

[0119] Process 400 may involve a step 430 of predicting cognitive impairment etiologies. Step 430 may involve predicting an MCI etiology for the patient using a predictive model. In some embodiments, step 430 may be executed as described regarding process 300. In some embodiments, step 430 may involve predicting MCI etiology for a patient by applying at least a portion of patient medical data for the patient (e.g., obtained in step 410) to a predictive model. In some examples, the patient medical data applied in step 430 and the patient medical data applied in step 420 may be inclusive of each other (e.g., the same). In some examples, a portion of the patient medical data applied in step 430 may be the same as a portion of the patient medical data applied in step 420. It is also envisioned that the patient medical data applied in step 430 may be different from patient medical data applied in step 420.Attorney Docket No.: 08061.0066-00304

[0120] In some embodiments, the predictive model in step 430 may be a model trained to predict MCI etiologies using training patient medical data for subjects. The training patient medical data for predicting MCI etiologies may be different than the training patient medical data for predicting MCI risk. For example, the training patient medical data for predicting MCI etiologies may be for subjects that satisfy an MCI condition, such as a diagnosis of MCI. In some embodiments, the subjects may be associated with MCI etiology labels. For example, the etiology labels may correspond to any dementia, such as Alzheimer’s disease, Lewy body dementia, or vascular dementia.

[0121] Process 400 may involve an optional step 440 of confirming etiology. For example, the etiology may be confirmed with platform 100, such as with a confirmation received via user device 199. In some embodiments, the etiology may be confirmed based on an assessment conducted by a clinician.

[0122] Process 400 may involve an optional step 450 of providing treatment. Step 450 may be based on the prediction of MCI etiology (e.g., step 430) or optionally on the confirmation of MCI etiology in step 440. For example, based on the predicted or confirmed MCI etiology, treatment(s) may be provided to the patient, or instructions or recommendations for such treatments can be provided by platform 100 (e.g., to a user of platform 100). In some embodiments, treatment may be provided based on other factors in addition to the predicted MCI etiology, such as factors in the patient’s medical history.

[0123] In some embodiments, providing treatment may involve administering a therapeutic agent to the patient. For example, the predicted MCI etiology for a patient may contribute to determining a therapeutic agent to administer to the patient. It will be appreciated that differing therapeutic agents may be administered to patients based on the etiology, as some therapeutic agents may be more effective, or may be clinically indicated, for some etiologies but not for other clinical etiologies. For example, in step 450, based on the determination thatAttorney Docket No.: 08061.0066-00304 the predicted etiology is Alzheimer’s disease, a first therapeutic agent may be administered. In an example, based on the determination that the predicted etiology is Lewy body dementia, a second therapeutic agent may be administered. In another example, based on the determination that the predicted etiology is vascular dementia, a third therapeutic agent may be administered. A therapeutic agent may refer to any substance, treatment, or compound capable of treating, curing, mitigating, or preventing diseases or conditions. For example, a therapeutic agent may be a biologically active compound capable of treating dementia-related conditions. In some embodiments, when the predicted etiology is Alzheimer’s disease, the therapeutic agent may be an anti-amyloid beta protofibril antibody. For example, the antibody may be lecanemab.

[0124] It will be appreciated that the disclosed embodiments provide improvements to predicting MCI risk, predicting MCI etiologies, and providing treatment for MCI. As described herein, the disclosed embodiments may be provided as a screening tool (e.g., for clinicians or the like) to improve their ability to identify individuals at elevated risk of MCI, thereby providing an entry point for triage. The disclosed embodiments may assist in early identification of MCI and may alert clinicians to the possibility of undetected risks in patients. The disclosed embodiments may also provide improvements in monitoring and managing MCI by predicting data that may indicate a patient should seek follow up (e.g., with a specialist) and / or seek further testing.

[0125] Examples

[0126] Multiple investigations were performed into the training and use of predictive models consistent with disclosed embodiments. These investigations concerned prediction of MCI risk and prediction of MCI etiologies.

[0127] Example 1Attorney Docket No.: 08061.0066-00304

[0128] FIGs.5A-5E depict the results of an investigation into development of a model for the prediction of cognitive impairment etiologies using electronic health record (EHR) data. In this investigation, predictive models were developed that used comorbidity data, socio- demographic data, and health status data to predict etiologies of Alzheimer’s disease (AD), Lewy body dementia (LBD), and vascular dementia (VD) in patients with MCI. The models were developed using the National Alzheimer’s Coordinating Center (NACC) Uniform Data Set (UDS). The dataset included subjects with a confirmed MCI diagnosis and clinical etiology (e.g., including AD, LBD, and VD). The inclusion criteria specified that subjects have an MCI diagnosis, be at least 40 years old, have recorded health status markers and comorbidities at the time of MCI diagnosis, and have AD, LBD, or VD as the sole MCI etiology. Subjects having mixed dementia were excluded. In the investigation, the sample size was 1211 subjects having AD, 139 subjects having LBD, and 110 subjects having VD.

[0129] Patient medical data considered in this investigation included comorbidity data (e.g., hypertension, stroke / transient ischemic attack, congestive heart failure, heart attack / cardiac arrest, atrial fibrillation, diabetes, hypercholesterolemia, thyroid disease, depression, psychiatric disorder other than depression, seizures, traumatic brain injury, insomnia / hyposomnia, sleep apnea, and other sleep disorders), socio-demographic data (e.g., age, gender, race, education, marital status), and health status data (e.g., BMI, smoking history, vision impairment, health impairment).

[0130] Predictive models used in this investigation included a Bayesian logistic lasso algorithm with a “mixture double exponential” spike-and-slab prior to predict etiology data (e.g., classify patients according to etiology). Additional algorithms were also implemented, such as gradient boosting. In this investigation, training and test sets were created via a random 70-30 split, and model performance was assessed using the area under the curve of the receiver operating characteristic curve (ROC-AUC).Attorney Docket No.: 08061.0066-00304

[0131] FIG.5A depicts a summary of select socio-demographic data for subjects, such as age at MCI diagnosis, gender, race, education, and marital status, organized by MCI etiology. The socio-demographic features displayed different distributions across different MCI etiologies.

[0132] FIG.5B depicts a summary of select comorbidity data for subjects, such as atrial fibrillation, stroke / TIA, diabetes, hypercholesterolemia, sleep disorder, insomnia, depression, and traumatic brain injury, organized by MCI etiology. Several comorbidities displayed associations with etiologies, such as stroke / TIA with VD and diabetes with VD.

[0133] FIG.5C depicts the differential predictive effect between AD versus LBD for select predictors. The strongest predictors for AD included hypertension and age, and the strongest predictors for LBD included gender (male), psychiatric disorder, anti-psychiatric agent, total medication, and “other sleep disorder.” For differential predictions between AD and LBD, the area under the receiver operating characteristic curve (AUC-ROC) was 0.742 on the test set.

[0134] FIG.5D depicts the differential predictive effect between AD versus VD for select predictors. The strongest predictors for VD included stroke / TIA and diabetes. For differential predictions between AD and VD, the area under the receiver operating characteristic curve (AUC-ROC) was 0.705 on the test set.

[0135] FIG.5E depicts the differential predictive effect between LBD versus VD for select predictors. The strongest predictors for VD included hypertension, stroke, and diabetes. The strongest predictors for LBD included gender (female) and total medication. For differential predictions between LBD and VD, the area under the receiver operating characteristic curve (AUC-ROC) was 0.858 on the test set.

[0136] Overall, predictive models using socio-demographic, health status, and comorbidity data were shown to accurately differentiate AD, LBD, and VD MCI etiologies. In particular,Attorney Docket No.: 08061.0066-00304 patients with VD had a higher prevalence of cardiovascular and metabolic comorbidities including stroke / TIA. Such predictive models can be used in the management of MCI, helping clinicians select appropriate treatments for MCI patients.

[0137] Example 2

[0138] FIGs.6A-6I depict results of an investigation into prediction of MCI risk data. In the investigation, a predictive model was developed using a cohort of individuals with an MCI diagnosis (but without an AD diagnosis) and a cohort without an MCI or AD diagnosis. The datasets used included the Merative MarketScan Commercial and Medicare Databases. These databases included enrollment records and inpatient, outpatient, ancillary, and drug claims. The index date for each individual in the MCI cohort was the date of the first MCI diagnosis. Individuals in the non-MCI cohort were matched three to one to an individual in the MCI cohort using age, sex, geographic region, and the year in which the individual had data entered into the MarketScan databases via the propensity score method. The index dates for non-MCI individuals were the index dates of their matched MCI individuals.

[0139] The pre-index period was defined as 2 years before the index date, and the follow-up period was defined as the time from the index date to the last visit recorded (a minimum of one year of follow-up required). All individuals in the study were required to be at least 50 years old in the year of the index date. An MCI diagnosis was defined by an International Classification of Disease (ICD) code, either ICD-9 code 331.83 or ICD-10 code G31.84. To be included in the non-MCI cohort, individuals could not have an MCI or AD diagnosis throughout their inclusion in the MarketScan databases. Individuals were excluded if they (1) received a diagnosis of Parkinson’s disease at any time or (2) received donepezil, memantine, memantine / donepezil, galantamine, or rivastigmine during the pre-index period. A total of 5185 individuals with an MCI diagnosis and 15,555 matched non-MCI individuals met the eligibility criteria.Attorney Docket No.: 08061.0066-00304

[0140] Individuals in the MCI and non-MCI cohorts were categorized by age group (50 to 64 years, 65 to 79 years, 80+ years). To predict the risk for MCI, several predictive models were trained using various machine learning approaches (Bayesian logistic lasso regression (BLLR) with a mixture of double-exponential prior, stochastic gradient boosting machine, extreme gradient boosting (XGBoost), and regularized random forests). The AUC-ROC was used to evaluate each model. The patient data used for prediction included age, sex, and 25 comorbidities, including cardiovascular diseases (hypertension, stroke / transient ischemic attack, ischemic heart disease, congestive heart failure, myocardial infarction, atherosclerosis, atrial fibrillation), metabolic disorders (diabetes, hyperlipidemia, obesity, metabolic syndrome, weight loss), psychiatric disorders (depression, insomnia, bipolar, schizophrenia, psychosis, alcohol abuse, drug abuse), and other diseases (chronic kidney disease, chronic pulmonary disease, hearing loss, obstructive sleep apnea, disturbances of sensation of smell and taste, hypothyroidism). Indices of the comorbidities and their treatments within 2 years before the index date were derived using ICD-9 and ICD-10 codes from inpatient and outpatient visits. Training and test sets were created via a random 70-30% split, stratified by MCI / non-MCI status and by age.

[0141] FIGs.6A-6B (of which FIG.6B is a continuation of FIG.6A) depict select characteristics for the pre-index period. As depicted, the MCI cohort had a statistically significantly higher frequency of patients with one or more of the 25 comorbidities (95.6% vs.81.4%) with odds ratio (OR) (MCI vs. non-MCI, 95% confidence interval (CI)) of 4.9 (4.3, 5.7). The 15 comorbidities with the largest frequency in the MCI cohort were as follows: hyperlipidemia (66.0%), hypertension (65.2%), depression (31.3%), obstructive sleep apnea (27.8%), diabetes (25.2%), hypothyroidism (25.0%), chronic pulmonary disease (24.2%), ischemic heart disease (21.9%), stroke / TIA (21.0%), hearing loss (21.0%), obesity (17.9%), insomnia (14.6%), atherosclerosis (10.0%), atrial fibrillation (9.6%), and congestiveAttorney Docket No.: 08061.0066-00304 heart failure (8.0%). The 15 comorbidities with the largest OR (MCI vs. non-MCI) were as follows: psychosis with an OR (95% CI) of 8.2 (5.3, 12.8), bipolar 6.7 (4.8, 9.2), schizophrenia 4.9 (2.4, 9.7), disturbances of sensation of smell and taste 4.4 (2.6, 7.4), depression 3.8 (3.5, 4.1), stroke / TIA 3.3 (3.0, 3.6), drug abuse 3.1 (2.4, 3.9), obstructive sleep apnea 2.9 (2.7, 3.1), weight loss 2.8 (2.4, 3.2), insomnia 2.7 (2.4, 2.9), alcohol abuse 2.6 (2.1, 3.3), hearing loss 2.1 (1.9, 2.3), metabolic syndrome 2.1 (1.8, 2.5), ischemic heart disease 1.7 (1.6, 1.8), and hyperlipidemia 1.6 (1.5, 1.7). Seven comorbidities appeared on both lists of the highest frequency and highest OR. These seven were presented in the order of OR as follows: depression, stroke / TIA, obstructive sleep apnea, insomnia, hearing loss, ischemic heart disease, and hyperlipidemia.

[0142] FIGs.6C-6E depict forest plots of the prevalence odds ratio of AD comorbidities at baseline by age group.

[0143] FIG.6F depicts the performance of Bayesian logistic lasso regression for all individuals and by age group. The BLLR had AUC-ROC = 0.72 for the analysis of all individuals. Performance of the model for MCI risk prediction varied across the three age groups with AUC-ROC values of 0.75, 0.70, and 0.66 for the age groups 50 to 64 years, 65 to 79 years, and 80+ years, respectively. The 50 to 64 years group had a sensitivity of 61.9% and a specificity of 77.4%. The 65 to 79 years group had a sensitivity of 59.3% and a specificity of 71.7%. The 80+ years group had a sensitivity of 55.3% and a specificity of 66.8%.

[0144] FIGs.6G-6I depict the comorbidities that the Bayesian logistic lasso regression (BLLR) model identified as significant predictors of MCI for each age group. FIG.6G depicts odds ratios for age group 50-64 years old, FIG.6H depicts odds ratios for age group 65-79 years old, and FIG.6I depicts age group 80+ years. For the age group 50 to 64 years, 12 comorbidities were significant predictors of MCI diagnosis in the multivariate model withAttorney Docket No.: 08061.0066-00304 p-values < 0.05: depression, stroke / TIA, obstructive sleep apnea, hearing loss, bipolar, psychosis, hypothyroidism, insomnia, weight loss, chronic pulmonary disease, alcohol abuse, and drug abuse. For the age group 65 to 79 years, five comorbidities were significant predictors of MCI diagnosis with p-values < 0.05: stroke / TIA, depression, obstructive sleep apnea, hearing loss, and weight loss. For the age group 80+ years, four comorbidities were significant predictors of MCI diagnosis with p-values < 0.05: depression, hearing loss, stroke / TIA, and weight loss.

[0145] The results of this investigation of the MarketScan Commercial and Medicare Databases suggest that AD comorbidities are also comorbidities for MCI. The selected AD comorbidities were also significant risk factors for MCI in this population. Individuals in the MCI cohort had a higher frequency of comorbidities compared with the non-MCI cohort. The differences between cohorts for depression and stroke / TIA were the largest. Depression, stroke / TIA, obstructive sleep apnea, insomnia, hearing loss, ischemic heart disease, and hyperlipidemia appeared on both the list of the highest frequency comorbidities and the list of comorbidities with the highest ORs, suggesting these seven comorbidities may be predictive of MCI. The BLLR results in this study demonstrated better model predictivity in the younger age group. Depression, stroke / TIA, hearing loss, and weight loss were significant predictors across all age groups. Identifying the comorbidities with the strongest connections and quantifying the relationships can help inform the development of a screening tool to identify high-risk individuals.

[0146] Example 3

[0147] FIGs.7A-7D depict results of an investigation into developing a risk prediction model for MCI using electronic health record (EHR) data. This model could provide a rapid, adaptable, and low-cost tool for clinicians (e.g., primary care physicians) to detect potential MCI in patients. The dataset used in the investigation included the Optum EHR database.Attorney Docket No.: 08061.0066-00304 Potential MCI risk factors included socio-demographic variables (e.g., age, gender, race, ethnicity, US census region), health status variables (e.g., body mass index, smoking status, blood pressure, cholesterol, hemoglobin A1c; values closest to the index date were captured), medical comorbidities (e.g., cancer, central nervous system, cognition, cardiovascular, digestive, endocrine / metabolic, infection, musculoskeletal and movement, respiratory, urinary, sensory, and sleep diseases or conditions), and behavioral comorbidities (e.g., diseases or symptoms relating to behavioral health (e.g., depression, hallucination) or emotional health (e.g., anxiety, combativeness)). All individuals in the study sample were at least 40 years old and were observed with EHR records for 24 months before their index date. Individuals had no diagnosis records for Parkinson’s disease or parkinsonism in the study period. Individuals in the MCI cohort had a diagnosis record with an International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10) diagnosis code for MCI (G31.84) during the identification period. The date of the first MCI diagnosis was the index date. Individuals in the non-MCI cohort had no diagnosis or NLP-derived records for MCI, AD, diagnosis records for cognitive problems, or medication records for dementia medications in the study period. Individuals from the MCI cohort were randomly selected to achieve a 30:1 ratio of non-MCI to MCI.

[0148] FIG.7A depicts selected socio-demographic characteristics (gender, race, ethnicity, and US census region) at baseline. The study sample comprised 652,829 individuals.

[0149] FIG.7B depicts selected health status characteristics (BMI, diastolic blood pressure, systolic blood pressure, high-density lipoprotein, low-density lipoprotein, total cholesterol, hemoglobin A1c, smoking status) at baseline. As illustrated, the MCI cohort had lower mean body mass index, diastolic blood pressure, low-density lipoprotein, and total cholesterol compared with the non-MCI cohort. The MCI cohort had higher mean systolic blood pressureAttorney Docket No.: 08061.0066-00304 and hemoglobin A1c versus the non-MCI cohort, and was more likely to include previous smokers.

[0150] FIG.7C depicts selected comorbidities (stroke, weight loss, kidney disease, atherosclerosis, hearing loss, hypertension, depression, hyperlipidemia, COPD, diabetes, insomnia, sleep apnea) at baseline. The MCI cohort had significantly higher unadjusted odds of baseline stroke, weight loss, kidney disease, atherosclerosis, hearing loss, hypertension, depression, hyperlipidemia, chronic obstructive pulmonary disease (COPD), diabetes, insomnia, and sleep apnea compared with the non-MCI cohort.

[0151] FIG.7D depicts selected emotional symptoms (living alone and loneliness, apathy, agitation, anxiety) at baseline. The MCI cohort also had significantly higher unadjusted odds of being alone, apathy, agitation, and anxiousness relative to the non-MCI cohort.

[0152] Overall, this investigation found higher unadjusted odds of selected comorbidities and emotional symptoms in the MCI cohort as compared to the non-MCI cohort. The results may enable an adaptable MCI risk prediction tool that can facilitate MCI detection by primary care providers.

[0153] Example 4

[0154] FIGs.8A-8D depict results of an additional investigation into a risk prediction model for MCI using electronic health record (EHR) data. Such a model could provide a rapid, adaptable, and low-cost tool for clinicians (e.g., primary care physicians) to detect potential MCI in patients.

[0155] The dataset used in the investigation included the Optum EHR database described in Example 3. An MCI cohort and a non-MCI cohort were generated using the Optum EHR database. Records were analyzed for explicit diagnoses (e.g., diagnosis codes) and, using natural language processing, for signs of disease symptoms abstracted from clinical note records (“NLP-derived records”). Both cohorts included patients identified from Jan 2016 –Attorney Docket No.: 08061.0066-00304 Mar 2021 (the “ID period”). The included patients were at least 40 years old, had EHR records for 24 months before an index date, and had no diagnosis records for Parkinson’s disease or parkinsonism from Jan 2007 – Mar 2022 (the “study period”).

[0156] The MCI cohort included patient satisfying inclusion criteria. The inclusion criteria included a requirement that the medical record of a patient include an initial MCI diagnosis in the ID period, and at least one subsequent MCI or Alzheimer’s disease or related dementia (ADRD) diagnosis. A medical record of a patient included an MCI diagnosis when the medical record included an International Classification of Diseases, Tenth Revision, Clinical Modification diagnosis code for MCI (G31.84). The date of the initial MCI diagnosis was taken to be the index date. The inclusion criteria further included a requirement that the medical records for the patient lack a diagnosis or NLP-derived record for MCI, ADRD, cognitive problems, or dementia medication from 2007 until the index date.

[0157] The non-MCI cohort included patients satisfying inclusion criteria. The inclusion criteria included a requirement that the medical records for a patient lack a diagnosis or NLP- derived record for MCI, ADRD, cognitive problems, or dementia medication during the study period. The inclusion criteria included a further requirement that the medical records for a patient lack any NLP-derived records for delirium, disorientation, attention deficit, impairment in comprehension, concentration, speech, executive function, language, or naming, inability to find the right word, lack of judgment and / or insight in the study period. Exclusion of patients with these NLP-derived records associated with cognitive impairment may have mitigated the risk of individuals with undiagnosed MCI being included in the non- MCI cohort. The index dates for the patients in the non-MCI cohort were set from a randomly selected diagnosis record during the ID period, and patients in the non-MCI cohort were required to have had two or more years of follow-up after the index date. Patients satisfyingAttorney Docket No.: 08061.0066-00304 the MCI cohort inclusion criteria were randomly selected to achieve a 30:1 ratio between the sizes of the non-MCI and MCI cohorts.

[0158] A dataset was created using the MCI and non-MCI patient cohorts. The dataset included patient data for the 24-month period preceding the index date for each patient and an indication of whether the patient was an MCI or non-MCI patient. The patient data included socio-demographic data (e.g., age, sex, race, ethnicity, US Census region), health status measures (e.g., smoking status, blood pressure, cholesterol, hemoglobin A1C, and other laboratory test results), medical comorbidities from diagnosis and NLP-derived records (e.g., cancer, central nervous system, cognition, cardiovascular, digestive, endocrine / metabolic, infection, musculoskeletal and movement, respiratory, urinary, sensory, or sleep comorbidities), and behavioral comorbidities (e.g., depression, irritability) including emotional symptoms (e.g., anxiety, combativeness) from diagnosis and NLP-derived records.

[0159] The dataset was divided into a 70% random sample and a remaining 30% test set. Patient data for the 70% random sample was used as a training dataset. Prediction models were trained to predict patient MCI risk using the training dataset. Medical comorbidities and behavioral comorbidities were initially selected for use in predicting MCI risk by comparing odds ratios (ORs) and differences in means with 95% confidence intervals (CIs). Patients were stratified into age groups: 40-49, 50-64, 65-79, and at least 80 years old. MCI vs non- MCI ORs were compared for each medical comorbidity and behavioral comorbidity item by age group using logistic regression, with cohort as response variable and the comorbidity, age group, and the interaction of the comorbidity by age group. The p-values for testing the equality of ORs among age groups were obtained by assessing comorbidity by age group interaction. Medical comorbidities and behavioral comorbidities were selected as predictors when the MCI cohort had a statistically significantly higher frequency of the comorbidities compared with the non-MCI cohort.Attorney Docket No.: 08061.0066-00304

[0160] Age-specific prediction models were generated for the 50-64, 65-79, and at least 80 year old age groups. Predictions for patients in the 40-49 year old age group were obtained using the model trained for the 50-64 year old age group. The prediction models were generated by training a machine-learning algorithm (a Bayesian logistic lasso regression model with a spike-and-slab mixture double-exponential prior) to discriminate between MCI and non-MCI individuals. Model performance was evaluated via 10-fold cross-validation using a 70% random sample of subjects and then validated in a remaining 30% test set. For each age group, three predictive models were generated, based on origin of the comorbidity data. The first model included only medical and behavioral comorbidities identified using diagnosis records (e.g., diagnosis codes). For example, in the first model, a patient was coded as having a medical comorbidity or behavioral comorbidity when the patient’s medical records included a diagnosis (e.g., an International Classification of Diseases (ICD) diagnosis code) for the medical comorbidity or behavioral comorbidity. The second model included only medical and behavioral comorbidities identified using NLP-derived records. For example, in the second model, a patient was coded as having a medical comorbidity or behavioral comorbidity when relevant signs or symptoms of the comorbidity were abstracted from clinical notes in medical records for the patient using natural language processing. The third model included medical and behavioral comorbidities included in diagnosis records and / or present in NLP-derived records. For example, in the third model, a patient was coded as having a medical or behavioral comorbidity when the patient was diagnosed with the medical or behavioral comorbidity, or relevant signs of disease symptoms were abstracted from note records for the patient using natural language processing. Prediction model performance was measured with sensitivity, specificity, and area under the curve (AUC).

[0161] FIG.8A depicts demographic characteristics and risk factors at baseline. In the study, the sample comprised 291,551 individuals. The table in FIG.8A depicts differences inAttorney Docket No.: 08061.0066-00304 demographic characteristics and selected medical and behavioral comorbidities that were applied in the predictive model, by cohort. The selected medical and behavioral comorbidities included delusion, traumatic brain injury, psychosis, stroke or transient ischemic attack, cerebral palsy and paralytic syndromes, gait abnormalities, schizophrenia, concussion, polyneuropathies, vision loss, constipation, aplastic / other anemias, urinary incontinence, demyelinating disease, kidney disease, bipolar disorder, chronic obstructive pulmonary disease, depression, insomnia, anxiety, soft tissue disorder, sleep apnea, and obesity, as identified using diagnostic records (e.g., diagnosis codes). The selected medical comorbidities and behavioral comorbidities further included apathy, irritability, and agitation, as primarily identified using NLP-derived records.

[0162] FIG.8B depicts a comparison across age groups of odds ratios for selected medical and behavioral comorbidities. In particular, FIG.8B provides a table showing odds ratios and 95% confidence intervals for the selected medical and behavioral comorbidities. The selected medical comorbidities and behavioral comorbidities were stroke / TIA, gait abnormalities, constipation, urinary incontinence, kidney disease, depression, anxiety, sleep apnea, and irritability. All selected medical comorbidities and behavioral comorbidities displayed in FIG.8B were significantly different at p<0.05 between MCI and non-MCI cohorts for the 50-64, 65-79, and at least 80 year old age groups. In general, odds ratios decreased as age increased; that is, differences in occurrence of the selected comorbidities were larger between the MCI and non-MCI cohorts in the younger age groups.

[0163] As described herein, predictive models were developed that utilized medical and behavioral comorbidities i) included in diagnosis records (e.g., identified using diagnosis codes, such as ICD codes), ii) present in NLP-derived records (e.g., identified using signs of the comorbidities extracted from clinical notes using natural language processing) , or iii)Attorney Docket No.: 08061.0066-00304 included in diagnosis records and / or present in NLP-derived records. Each of these approaches performed similarly. In the test set the AUC for all 3 approaches was 0.89.

[0164] FIG.8C depicts the receiver operator curves (ROCs) and AUC for predictive models that used diagnosis and NLP-derived records. These predictive models used the input patient data depicted in FIG.8A (e.g., 23 medical and behavioral comorbidities identified using diagnosis records and 3 behavioral comorbidities, agitation, apathy, and irritability, derived from diagnosis and NLP-derived records – though principally NLP-derived records). The ROC for the 40-49 year old age group was obtained by fitting the model estimated for the 50- 64 year old age group to individuals in the 40-49 year old age group. The overall AUC for the predictive models, considered overall, was 0.87. The AUC was 0.81 for the 40-49 year old age group, 0.80 for the 50-64 year old age group, 0.76 for the 65-79 year old age group, and 0.71 for the at least 80 year old age group. As depicted in FIG.8D, the distribution of MCI and non-MCI cohorts in the study varied by age group. The age-specific AUCs were generally smaller because the ratio of non-MCI to MCI individuals was larger in the younger age groups. Consistent with the decline in odds ratios with age depicted in FIG.8B, AUC decreased as age increased.

[0165] Example 5

[0166] FIGs.9A-9H depict results of another investigation into developing a risk prediction model for MCI using EHR data. The dataset used in the investigation included the Optum EHR database described above, and an MCI cohort and non-MCI cohort were generated using the Optum EHR database as described in Example 4. As discussed in Example 4, the dataset was randomly divided into a 70% training and cross-validation set and a 30% test set.

[0167] Predictive models for predicting patient MCI risk based on socio-demographic data and selected medical and behavioral comorbidities were developed. The socio-demographicAttorney Docket No.: 08061.0066-00304 data included age and race. The selected medical and behavioral comorbidities included stroke / transient ischemic attack; kidney disease; depression; bipolar; psychosis; hallucination; schizophrenia; substance abuse; anxiety; aphasia; atherosclerosis; gait abnormalities; concussion; delusion; cerebral palsy; incontinence; traumatic brain injury; rapid eye movement (REM) disorder; inability to follow a command; irritability; confusion, inability to communicate, falling, feeling or living alone, poor mental state, and impulsivity. The following comorbidities were not selected for inclusion in the predictive models: cancer- related conditions, polyneuropathies and other disorders of peripheral nervous system, demyelinating diseases of the central nervous system, and chronic obstructive pulmonary disease (COPD).

[0168] FIG.9A depicts estimated odds ratios for age, race, and the selected medical and behavioral comorbidities. An odds ratio greater than 1 indicates an increase in MCI risk, and an odds ratio less than1 indicates protection, or a reduction in MCI risk. A predictive model constructed using this patient data exhibited an AUC of 0.86.

[0169] FIG.9B depicts ROC curves and their areas under curve (AUC) for the 40-49, 50-64, 65-79, and at least 80 year old age groups. An AUC of greater than or equal to 0.70 is generally considered a good performance. The AUC for the 40-49 year old age group was obtained by fitting the model for the 50-64 year old age group. The model for the 40-49 year old age group had an AUC of 0.78, the model for the 50-64 year old age group had an AUC of 0.78, the model for the 65-79 year old age group had an AUC of 0.77, and the model for the at least 80 year old age group had an AUC of 0.72.

[0170] FIGs.9C, 9D, and 9E depict estimated odds ratios for age, race, and the selected medical and behavioral comorbidities for the 50-64 year old, 65-79 year old, and at least 80 year old age groups, respectively. An odds ratio greater than 1 indicates an increase in MCIAttorney Docket No.: 08061.0066-00304 risk, and an odds ratio less than 1 indicates protection, or a reduction in MCI risk. The impact of medical and behavioral comorbidities varied between age groups.

[0171] FIG.9F depicts ROC curves and their areas under curve (AUC) for the age-specific predictive models generated using age, race, and the selected medical and behavioral comorbidities. As shown, AUC decreased with increasing age (e.g., 40-49 years, AUC=0.78; 50-64 years, AUC=0.77; 65-79 years, AUC=0.77; 80+ years, AUC=0.73). The decline in AUC may be due to decreasing odds ratios (MCI vs non-MCI) as age group increases, or an increasing proportion of undiagnosed MCI in the non-MCI cohort as age group increases.

[0172] FIG.9G depicts a comparison of the three approaches taken for obtaining medical comorbidity data and / or behavioral comorbidity data. In the first model, the patient data included medical comorbidities and behavioral comorbidities obtained from diagnosis records (e.g., diagnosis codes) only (labeled “Dx”). In the second model, the patient data included medical and behavioral comorbidities identified using NLP-derived records (labeled “SDS”). In the third model, the patient data included medical and behavioral comorbidities identified using diagnosis records and / or NLP-derived records (labled “Dx+SDS”). Sensitivity, specificity, and AUC performance measures are shown for each approach for both the test set and the cross-validation set. The “Dx + SDS” model had a slightly better performance than the “Dx” or “SDS” approaches.

[0173] FIG.9H depicts sensitivity, specificity, and AUC performance measures for different age groups for both the test set and the cross-validation set. As observed, the performance measures generally decreased with increasing age.

[0174] The foregoing description has been presented for purposes of illustration. It is not exhaustive and is not limited to precise forms or embodiments disclosed. Modifications and adaptations of the embodiments will be apparent from consideration of the specification and practice of the disclosed embodiments. For example, the described implementations includeAttorney Docket No.: 08061.0066-00304 hardware, but systems and methods consistent with the present disclosure can be implemented with hardware and software. In addition, while certain components have been described as being coupled to one another, such components may be integrated with one another or distributed in any suitable fashion.

[0175] Embodiments herein include systems, methods, and tangible non-transitory computer- readable media. The methods may be executed, at least in part for example, by at least one processor that receives instructions from a tangible non-transitory computer-readable storage medium. Similarly, systems consistent with the present disclosure may include at least one processor and memory, and the memory may be a tangible non-transitory computer-readable storage medium. As used herein, a tangible non-transitory computer-readable storage medium refers to any type of physical memory on which information or data readable by at least one processor may be stored. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, non-volatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, registers, caches, and any other known physical storage medium. Singular terms, such as “memory” and “computer-readable storage medium,” may additionally refer to multiple structures, such a plurality of memories or computer-readable storage media. As referred to herein, a “memory” may comprise any type of computer-readable storage medium unless otherwise specified. A computer-readable storage medium may store instructions for execution by at least one processor, including instructions for causing the processor to perform steps or stages consistent with embodiments herein. Additionally, one or more computer-readable storage media may be utilized in implementing a computer-implemented method. The term “non-transitory computer-readable storage medium” should be understood to include tangible items and exclude carrier waves and transient signals.

[0176] Moreover, while illustrative embodiments have been described herein, the scope includes any and all embodiments having equivalent elements, modifications, omissions,Attorney Docket No.: 08061.0066-00304 combinations (e.g., of aspects across various embodiments), adaptations or alterations based on the present disclosure. The elements in the claims are to be interpreted broadly based on the language employed in the claims and not limited to examples described in the present specification or during the prosecution of the application, which examples are to be construed as nonexclusive. Further, the steps of the disclosed methods can be modified in any manner, including reordering steps or inserting or deleting steps.

[0177] The features and advantages of the disclosure are apparent from the detailed specification, and thus, it is intended that the appended claims cover all systems and methods falling within the true spirit and scope of the disclosure. As used herein, the indefinite articles “a” and “an” mean “one or more.” Similarly, the use of a plural term does not necessarily denote a plurality unless it is unambiguous in the given context. Further, since numerous modifications and variations will readily occur from studying the present disclosure, it is not desired to limit the disclosure to the exact construction and operation illustrated and described, and accordingly, all suitable modifications and equivalents may be resorted to, falling within the scope of the disclosure. Therefore, it is intended that the disclosed embodiments and examples be considered as examples only, with a true scope of the present disclosure being indicated by the following claims and their equivalents.

[0178] The embodiments may further be described using the following clauses:

[0179] 1. A system, comprising: at least one processor; and at least one computer-readable, non-transitory medium containing instructions that, when executed by the at least one processor, cause the system to perform operations comprising: obtaining patient medical data for a patient, the patient medical data including comorbidity data; predicting a mild cognitive impairment risk for the patient by applying a first portion of the patient medical data to a first predictive model trained to predict mild cognitive impairment risk using first medical record training data for first subjects; and based on the predicted risk of mild cognitive impairment,Attorney Docket No.: 08061.0066-00304 predicting a mild cognitive impairment etiology for the patient by applying a second portion of the patient medical data to a second predictive model trained to predict mild cognitive impairment etiologies using second medical record training data for second subjects that satisfy a mild cognitive impairment condition, the second subjects associated with mild cognitive impairment etiology labels.

[0180] 2. The system of clause 1, wherein at least one of the first predictive model or the second predictive model comprises at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest.

[0181] 3. The system of any one of clauses 1-2, wherein a first portion of the first subjects are associated with a mild cognitive impairment label, and a second portion of the first subjects are associated with a control label.

[0182] 4. The system of any one of clauses 1-3, wherein the first and second subjects satisfy an age criterion.

[0183] 5. The system of any one of clauses 1-4, wherein the patient medical data further comprises at least one of socio-demographic data or health status data.

[0184] 6. The system of any one of clauses 1-5, wherein the first medical record training data comprises at least one of socio-demographic data or health status data.

[0185] 7. The system of any one of clauses 1-6, wherein the second medical record training data comprises at least one of socio-demographic data or health status data.

[0186] 8. The system of any one of clauses 1-7, wherein the comorbidity data comprises at least one of cardiovascular condition data, metabolic condition data, neurological condition data, psychiatric condition data, respiratory condition data, sleep condition data, or substance abuse data.

[0187] 9. The system of clause 8, wherein the comorbidity data comprises data for at least two comorbidities.Attorney Docket No.: 08061.0066-00304

[0188] 10. The system of any one of clauses 1-9, wherein the first portion of the patient medical data is the same as the second portion of the patient medical data.

[0189] 11. The system of any one of clauses 1-9, wherein the first portion of the patient medical data is different from the second portion of the patient medical data.

[0190] 12. A method of treatment comprising: obtaining patient medical data for a patient, the patient medical data including comorbidity data; predicting a mild cognitive impairment risk for the patient by applying a first portion of the patient medical data to a first predictive model trained to predict mild cognitive impairment risk using first medical record training data for first subjects; based on the predicted risk of mild cognitive impairment, predicting a mild cognitive impairment etiology for the patient by applying a second portion of the patient medical data to a second predictive model trained to predict mild cognitive impairment etiologies using second medical record training data for second subjects that satisfy a mild cognitive impairment condition, the second subjects associated with mild cognitive impairment etiology labels; and based at least in part on the prediction of the mild cognitive impairment etiology, applying a therapeutic agent to the patient.

[0191] 13. The method of clause 12, wherein at least one of the first predictive model or the second predictive model comprises at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest.

[0192] 14. The method of any one of clauses 12-13, wherein a first portion of the first subjects are associated with a mild cognitive impairment label, and a second portion of the first subjects are associated with a control label.

[0193] 15. The method of any one of clauses 12-14, wherein the first and second subjects satisfy an age criterion.

[0194] 16. The method of any one of clauses 12-15, wherein the patient medical data further comprises at least one of socio-demographic data or health status data.Attorney Docket No.: 08061.0066-00304

[0195] 17. The method of any one of clauses 12-16, wherein the first medical record training data comprises at least one of socio-demographic data or health status data.

[0196] 18. The method of any one of clauses 12-17, wherein the second medical record training data comprises at least one of socio-demographic data or health status data.

[0197] 19. The method of any one of clauses 12-18, further comprising: based on a determination that the predicted etiology is Alzheimer’s disease, applying a first therapeutic agent; or based on a determination that the predicted etiology is Lewy body dementia, applying a second therapeutic agent; or based on a determination that the predicted etiology is Vascular dementia, applying a third therapeutic agent.

[0198] 20. The method of any one of clauses 12-19, wherein the therapeutic agent is an anti- amyloid beta protofibril antibody.

[0199] 21. The method of clause 20, wherein the antibody is lecanemab.

[0200] 22. The method of any one of clauses 12-21, wherein the comorbidity data comprises at least one of cardiovascular condition data, metabolic condition data, neurological condition data, psychiatric condition data, respiratory condition data, sleep condition data, or substance abuse data.

[0201] 23. The method of clause 22, wherein the comorbidity data comprises data for at least two comorbidities.

[0202] 24. A system, comprising: at least one processor; and at least one computer-readable, non-transitory medium containing instructions that, when executed by the at least one processor, cause the system to perform operations comprising: obtaining a training dataset including patient medical data, the training dataset comprising a first portion including subjects satisfying a mild cognitive impairment condition and a second portion including control subjects; training a predictive model, using the training dataset, to predict a risk ofAttorney Docket No.: 08061.0066-00304 mild cognitive impairment using the training dataset; and providing the trained predictive model.

[0203] 25. The system of clause 24, wherein the patient medical data comprises at least one of comorbidity data, socio-demographic data, or health status data.

[0204] 26. The system of clause 25, wherein the comorbidity data comprises data for at least 2 comorbidities.

[0205] 27. The system of clause 26, wherein the comorbidity data comprises data for 2-30 comorbidities.

[0206] 28. The system of any one of clauses 24-27, wherein the comorbidity data comprises medical comorbidity data or behavioral comorbidity data.

[0207] 29. The system of clause 28, wherein the medical comorbidity data comprises data on one or more of stroke, transient ischemic attack, kidney disease, aphasia, atherosclerosis, gait abnormalities, concussion, cerebral palsy, incontinence, traumatic brain injury, or rapid eye movement (REM) disorder.

[0208] 30. The system of any one of clauses 28-29, wherein the medical comorbidity data does not comprise data on one or more of cancer, polyneuropathies, demyelinating disease, or COPD.

[0209] 31. The system of any one of clauses 28-30, wherein the behavioral comorbidity data comprises data on one or more of depression, bipolar disorder, psychosis, hallucination, schizophrenia, substance abuse, anxiety, delusion, inability to follow a command, irritability, confusion, inability to communicate, falls, feeling or living alone, poor mental state, or impulsivity.

[0210] 32. The system of any one of clauses 25-31, wherein the socio-demographic data comprises data on age or race.Attorney Docket No.: 08061.0066-00304

[0211] 33. The system of any one of clauses 24-32, wherein the predictive model comprises at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest.

[0212] 34. The system of any one of clauses 24-33, wherein subjects satisfying the mild cognitive impairment condition and the control subjects satisfy an age criterion.

[0213] 35. The system of any one of clauses 24-34, wherein the patient medical data for each subject satisfying the mild cognitive impairment condition comprises an initial diagnosis of MCI and at least one subsequent diagnosis of MCI or AD.

[0214] 36. The system of any one of clauses 24-35, wherein the patient medical data for each control subject does not comprise a record of MCI, AD, a cognitive problem, or a dementia medication.

[0215] 37. A system, comprising: at least one processor; and at least one computer-readable, non-transitory medium containing instructions that, when executed by the at least one processor, cause the system to perform operations comprising: obtaining a training dataset including patient medical data, the training dataset comprising subjects satisfying a mild cognitive impairment condition and having different etiologies for the mild cognitive impairment condition; training a predictive model, using the training dataset, to predict an etiology for the subjects; and providing the trained predictive model.

[0216] 38. The system of clause 37, wherein the patient medical data comprises at least one of comorbidity data, socio-demographic data, or health status data, wherein the subjects satisfy a sole-etiology dementia condition.

[0217] 39. The system of any one of clauses 37-38, wherein the comorbidity data comprises at least one of cardiovascular condition data, metabolic condition data, neurological condition data, psychiatric condition data, respiratory condition data, sleep condition data, or substance abuse data.Attorney Docket No.: 08061.0066-00304

[0218] 40. The system of any one of clauses 37-39, wherein the comorbidity data and health status data are obtained before or within a year of diagnosis of the mild cognitive impairment condition.

[0219] 41. The system of any one of clauses 37-40, wherein the predictive model comprises at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest.

[0220] 42. The system of any one of clauses 37-41, wherein the predicted etiology comprises at least one of Alzheimer’s disease, Lewy body dementia, or Vascular dementia.

[0221] 43. The system of any one of clauses 37-42, wherein the predicted etiology is the sole etiology.

[0222] 44. The system of any one of clauses 37-43, further comprising training the predictive model to predict a treatment based on the predicted etiology.

[0223] 45. A system, comprising: at least one processor; and at least one computer-readable, non-transitory medium containing instructions that, when executed by the at least one processor, cause the system to perform operations comprising: obtaining training data for first subjects, the training data comprising patient medical data for each subject, wherein the patient medical data comprises comorbidity data and socio-demographic data, wherein each subject satisfies an age criterion, wherein at least a portion of the first subjects satisfy a mild cognitive impairment condition; training a predictive model, using the training data, to predict cognitive impairment risk data for a first subject; obtaining patient medical data for a second subject; predicting cognitive impairment risk data for the second subject by applying the patient medical data for the second subject to the trained predictive model; and providing the predicted cognitive impairment risk data.

[0224] 46. The system of clause 45, wherein the comorbidity data comprises at least one of medical comorbidity data or behavioral comorbidity data.Attorney Docket No.: 08061.0066-00304

[0225] 47. The system of clause 46, wherein the medical comorbidity data comprises data on one or more of stroke, transient ischemic attack, kidney disease, aphasia, atherosclerosis, gait abnormalities, concussion, cerebral palsy, incontinence, traumatic brain injury, or rapid eye movement (REM) disorder.

[0226] 48. The system of any one of clauses 46-47, wherein the medical comorbidity data does not comprise data on one or more of cancer, polyneuropathies, demyelinating disease, or COPD.

[0227] 49. The system of any one of clauses 46-48, wherein the behavioral comorbidity data comprises data on one or more of depression, bipolar disorder, psychosis, hallucination, schizophrenia, substance abuse, anxiety, delusion, inability to follow a command, irritability, confusion, inability to communicate, falls, feeling or living alone, poor mental state, or impulsivity.

[0228] 50. The system of any one of clauses 45-49, wherein the patient medical data includes at least one diagnosis or record abstracted using natural language processing from clinical notes.

[0229] 51. The system of any one of clauses 45-50, wherein the predictive model comprises at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest.

[0230] 52. The system of any one of clauses 45-51, wherein the patient medical data for each of the first subjects satisfying the mild cognitive impairment condition comprises an initial diagnosis of MCI and at least one subsequent diagnosis of MCI or AD.

[0231] 53. A system, comprising: at least one processor; and at least one computer-readable, non-transitory medium containing instructions that, when executed by the at least one processor, cause the system to perform operations comprising: obtaining training data for first subjects satisfying a mild cognitive impairment condition, the training data comprisingAttorney Docket No.: 08061.0066-00304 patient medical data for each subject, wherein the patient medical data comprises comorbidity data and socio-demographic data, wherein each subject satisfies an age criterion; training a predictive model, using the training data, to predict MCI etiology data for a first subject; obtaining patient medical data for a second subject; predicting MCI etiology data for the second subject by applying the patient medical data for the second subject to the trained predictive model; and providing the predicted MCI etiology data.

[0232] 54. The system of clause 53, wherein the comorbidity data comprises at least one of medical comorbidity data or behavioral comorbidity data.

[0233] 55. The system of any one of clauses 53-54, wherein the socio-demographic data comprises one or more of gender, age at MCI diagnosis, race, education, or marital status.

[0234] 56. The system of any one of clauses 53-55, wherein the patient medical data includes at least one diagnosis or record abstracted using natural language processing from clinical notes.

[0235] 57. The system of clause 56, wherein the at least one diagnosis or record includes at least one of stroke, ITA, diabetes, hypercholesterolemia, hypertension, atrial fibrillation, insomnia, depression, traumatic brain injury, or a sleep disorder.

[0236] 58. The system of any one of clauses 53-57, wherein the predictive model comprises at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest.

[0237] 44. Methods corresponding to the systems of any one of clauses 1-11 or 24 to 58.

[0238] 45. Systems each including at least one processor; and at least one computer-readable, non-transitory medium containing instructions that, when executed by the at least one processor, cause the systems to perform the methods of any one of clauses 12-23.Attorney Docket No.: 08061.0066-00304

[0239] 46. Non-transitory computer readable media instructions that, when executed by at least one processor of a system, cause the system to perform the methods of any one of clauses 12 – 23 or the operations of any one of clauses 1-11 or 28 to 43.

[0240] As used herein, unless specifically stated otherwise, the term “or” encompasses all possible combinations, except where infeasible. For example, if it is stated that a component may include A or B, then, unless specifically stated otherwise or infeasible, the component may include A, or B, or A and B. As a second example, if it is stated that a component may include A, B, or C, then, unless specifically stated otherwise or infeasible, the component may include A, or B, or C, or A and B, or A and C, or B and C, or A and B and C.

[0241] Other embodiments will be apparent from consideration of the specification and practice of the embodiments disclosed herein. It is intended that the specification and examples be considered as example only, with a true scope and spirit of the disclosed embodiments being indicated by the following claims.Attorney Docket No.: 08061.0066-00304 SEQUENCE TABLES Table 1. Amino acid sequences of monoclonal antibody (mAb) CDRsTable 2. Amino acid sequences of mAb variable regionsAttorney Docket No.: 08061.0066-00304 Table 3. Amino acid sequences of mAb heavy and light chainsAttorney Docket No.: 08061.0066-00304 Table 4. Amino acid sequences of mAb constant regions

Claims

Attorney Docket No.: 08061.0066-00304 WHAT IS CLAIMED IS:

1. A system, comprising: at least one processor; and at least one computer-readable, non-transitory medium containing instructions that, when executed by the at least one processor, cause the system to perform operations comprising: obtaining patient medical data for a patient, the patient medical data including comorbidity data; predicting a mild cognitive impairment risk for the patient by applying a first portion of the patient medical data to a first predictive model trained to predict mild cognitive impairment risk using first medical record training data for first subjects; and based on the predicted risk of mild cognitive impairment, predicting a mild cognitive impairment etiology for the patient by applying a second portion of the patient medical data to a second predictive model trained to predict mild cognitive impairment etiologies using second medical record training data for second subjects that satisfy a mild cognitive impairment condition, the second subjects associated with mild cognitive impairment etiology labels.

2. The system of claim 1, wherein at least one of the first predictive model or the second predictive model comprises at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest.

3. The system of any one of claims 1-2, wherein a first portion of the first subjects are associated with a mild cognitive impairment label, and a second portion of the first subjects are associated with a control label.

4. The system of any one of claims 1-3, wherein the first and second subjects satisfy an age criterion.Attorney Docket No.: 08061.0066-00304 5. The system of any one of claims 1-4, wherein the patient medical data further comprises at least one of socio-demographic data or health status data.

6. The system of any one of claims 1-5, wherein the first medical record training data comprises at least one of socio-demographic data or health status data.

7. The system of any one of claims 1-6, wherein the second medical record training data comprises at least one of socio-demographic data or health status data.

8. The system of any one of claims 1-7, wherein the comorbidity data comprises at least one of cardiovascular condition data, metabolic condition data, neurological condition data, psychiatric condition data, respiratory condition data, sleep condition data, or substance abuse data.

9. The system of claim 8, wherein the comorbidity data comprises data for at least two comorbidities.

10. The system of any one of claims 1-9, wherein the first portion of the patient medical data is the same as the second portion of the patient medical data.

11. The system of any one of claims 1-9, wherein the first portion of the patient medical data is different from the second portion of the patient medical data.

12. A method of treatment comprising: obtaining patient medical data for a patient, the patient medical data including comorbidity data; predicting a mild cognitive impairment risk for the patient by applying a first portion of the patient medical data to a first predictive model trained to predict mild cognitive impairment risk using first medical record training data for first subjects; based on the predicted risk of mild cognitive impairment, predicting a mild cognitive impairment etiology for the patient by applying a second portion of the patient medical data to a second predictive model trained to predict mild cognitive impairment etiologies using second medical record training data forAttorney Docket No.: 08061.0066-00304 second subjects that satisfy a mild cognitive impairment condition, the second subjects associated with mild cognitive impairment etiology labels; and based at least in part on the prediction of the mild cognitive impairment etiology, applying a therapeutic agent to the patient.

13. The method of claim 12, wherein at least one of the first predictive model or the second predictive model comprises at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest.

14. The method of any one of claims 12-13, wherein a first portion of the first subjects are associated with a mild cognitive impairment label, and a second portion of the first subjects are associated with a control label.

15. The method of any one of claims 12-14, wherein the first and second subjects satisfy an age criterion.

16. The method of any one of claims 12-15, wherein the patient medical data further comprises at least one of socio-demographic data or health status data.

17. The method of any one of claims 12-16, wherein the first medical record training data comprises at least one of socio-demographic data or health status data.

18. The method of any one of claims 12-17, wherein the second medical record training data comprises at least one of socio-demographic data or health status data.

19. The method of any one of claims 12-18, further comprising: based on a determination that the predicted etiology is Alzheimer’s disease, applying a first therapeutic agent; or based on a determination that the predicted etiology is Lewy body dementia, applying a second therapeutic agent; or based on a determination that the predicted etiology is Vascular dementia, applying a third therapeutic agent.

20. The method of any one of claims 12-19, wherein the therapeutic agent is an anti- amyloid beta protofibril antibody.Attorney Docket No.: 08061.0066-00304 21. The method of claim 20, wherein the antibody is lecanemab.

22. The method of any one of claims 12-21, wherein the comorbidity data comprises at least one of cardiovascular condition data, metabolic condition data, neurological condition data, psychiatric condition data, respiratory condition data, sleep condition data, or substance abuse data.

23. The method of claim 22, wherein the comorbidity data comprises data for at least two comorbidities.

24. A system, comprising: at least one processor; and at least one computer-readable, non-transitory medium containing instructions that, when executed by the at least one processor, cause the system to perform operations comprising: obtaining a training dataset including patient medical data, the training dataset comprising a first portion including subjects satisfying a mild cognitive impairment condition and a second portion including control subjects; training a predictive model, using the training dataset, to predict a risk of mild cognitive impairment using the training dataset; and providing the trained predictive model.

25. The system of claim 24, wherein the patient medical data comprises at least one of comorbidity data, socio-demographic data, or health status data.

26. The system of claim 25, wherein the comorbidity data comprises data for at least 2 comorbidities.

27. The system of claim 26, wherein the comorbidity data comprises data for 2-30 comorbidities.

28. The system of any one of claims 24-27, wherein the comorbidity data comprises medical comorbidity data or behavioral comorbidity data.Attorney Docket No.: 08061.0066-00304 29. The system of claim 28, wherein the medical comorbidity data comprises data on one or more of stroke, transient ischemic attack, kidney disease, aphasia, atherosclerosis, gait abnormalities, concussion, cerebral palsy, incontinence, traumatic brain injury, or rapid eye movement (REM) disorder.

30. The system of any one of claims 28-29, wherein the medical comorbidity data does not comprise data on one or more of cancer, polyneuropathies, demyelinating disease, or COPD.

31. The system of any one of claims 28-30, wherein the behavioral comorbidity data comprises data on one or more of depression, bipolar disorder, psychosis, hallucination, schizophrenia, substance abuse, anxiety, delusion, inability to follow a command, irritability, confusion, inability to communicate, falls, feeling or living alone, poor mental state, or impulsivity.

32. The system of any one of claims 25-31, wherein the socio-demographic data comprises data on age or race.

33. The system of any one of claims 24-32, wherein the predictive model comprises at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest.

34. The system of any one of claims 24-33, wherein subjects satisfying the mild cognitive impairment condition and the control subjects satisfy an age criterion.

35. The system of any one of claims 24-34, wherein the patient medical data for each subject satisfying the mild cognitive impairment condition comprises an initial diagnosis of MCI and at least one subsequent diagnosis of MCI or AD.

36. The system of any one of claims 24-35, wherein the patient medical data for each control subject does not comprise a record of MCI, AD, a cognitive problem, or a dementia medication.

37. A system, comprising: at least one processor; andAttorney Docket No.: 08061.0066-00304 at least one computer-readable, non-transitory medium containing instructions that, when executed by the at least one processor, cause the system to perform operations comprising: obtaining a training dataset including patient medical data, the training dataset comprising subjects satisfying a mild cognitive impairment condition and having different etiologies for the mild cognitive impairment condition; training a predictive model, using the training dataset, to predict an etiology for the subjects; and providing the trained predictive model.

38. The system of claim 37, wherein the patient medical data comprises at least one of comorbidity data, socio-demographic data, or health status data, wherein the subjects satisfy a sole-etiology dementia condition.

39. The system of any one of claims 37-38, wherein the comorbidity data comprises at least one of cardiovascular condition data, metabolic condition data, neurological condition data, psychiatric condition data, respiratory condition data, sleep condition data, or substance abuse data.

40. The system of any one of claims 37-39, wherein the comorbidity data and health status data are obtained before or within a year of diagnosis of the mild cognitive impairment condition.

41. The system of any one of claims 37-40, wherein the predictive model comprises at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest.

42. The system of any one of claims 37-41, wherein the predicted etiology comprises at least one of Alzheimer’s disease, Lewy body dementia, or Vascular dementia.

43. The system of any one of claims 37-42, wherein the predicted etiology is the sole etiology.

44. The system of any one of claims 37-43, further comprising training the predictive model to predict a treatment based on the predicted etiology.Attorney Docket No.: 08061.0066-00304 45. A system, comprising: at least one processor; and at least one computer-readable, non-transitory medium containing instructions that, when executed by the at least one processor, cause the system to perform operations comprising: obtaining training data for first subjects, the training data comprising patient medical data for each subject, wherein the patient medical data comprises comorbidity data and socio-demographic data, wherein each subject satisfies an age criterion, wherein at least a portion of the first subjects satisfy a mild cognitive impairment condition; training a predictive model, using the training data, to predict cognitive impairment risk data for a first subject; obtaining patient medical data for a second subject; predicting cognitive impairment risk data for the second subject by applying the patient medical data for the second subject to the trained predictive model; and providing the predicted cognitive impairment risk data.

46. The system of claim 45, wherein the comorbidity data comprises at least one of medical comorbidity data or behavioral comorbidity data.

47. The system of claim 46, wherein the medical comorbidity data comprises data on one or more of stroke, transient ischemic attack, kidney disease, aphasia, atherosclerosis, gait abnormalities, concussion, cerebral palsy, incontinence, traumatic brain injury, or rapid eye movement (REM) disorder.

48. The system of any one of claims 46-47, wherein the medical comorbidity data does not comprise data on one or more of cancer, polyneuropathies, demyelinating disease, or COPD.Attorney Docket No.: 08061.0066-00304 49. The system of any one of claims 46-48, wherein the behavioral comorbidity data comprises data on one or more of depression, bipolar disorder, psychosis, hallucination, schizophrenia, substance abuse, anxiety, delusion, inability to follow a command, irritability, confusion, inability to communicate, falls, feeling or living alone, poor mental state, or impulsivity.

50. The system of any one of claims 45-49, wherein the patient medical data includes at least one diagnosis or record abstracted using natural language processing from clinical notes.

51. The system of any one of claims 45-50, wherein the predictive model comprises at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest.

52. The system of any one of claims 45-51, wherein the patient medical data for each of the first subjects satisfying the mild cognitive impairment condition comprises an initial diagnosis of MCI and at least one subsequent diagnosis of MCI or AD.

53. A system, comprising: at least one processor; and at least one computer-readable, non-transitory medium containing instructions that, when executed by the at least one processor, cause the system to perform operations comprising: obtaining training data for first subjects satisfying a mild cognitive impairment condition, the training data comprising patient medical data for each subject, wherein the patient medical data comprises comorbidity data and socio- demographic data, wherein each subject satisfies an age criterion; training a predictive model, using the training data, to predict MCI etiology data for a first subject; obtaining patient medical data for a second subject; predicting MCI etiology data for the second subject by applying the patient medical data for the second subject to the trained predictive model; andAttorney Docket No.: 08061.0066-00304 providing the predicted MCI etiology data.

54. The system of claim 53, wherein the comorbidity data comprises at least one of medical comorbidity data or behavioral comorbidity data.

55. The system of any one of claims 53-54, wherein the socio-demographic data comprises one or more of gender, age at MCI diagnosis, race, education, or marital status.

56. The system of any one of claims 53-55, wherein the patient medical data includes at least one diagnosis or record abstracted using natural language processing from clinical notes.

57. The system of claim 56, wherein the at least one diagnosis or record includes at least one of stroke, ITA, diabetes, hypercholesterolemia, hypertension, atrial fibrillation, insomnia, depression, traumatic brain injury, or a sleep disorder.

58. The system of any one of claims 53-57, wherein the predictive model comprises at least one of a Bayesian logistic lasso regression, a stochastic gradient boosting machine, extreme gradient boosting, or random forest.

Images