Suicidal thoughts and behavior predictions and treatment

Abstract

Certain aspects of the disclosure provide systems and methods for diagnosis and treatment of suicidal thought and behavior (STB) through reward-aversion judgment and contextual variables. Methods include generating a set of STB parameters associated with a subject, the set of STB parameters based on reward-aversion judgment variables and contextual variables and processing the set of STB parameters with a machine learning model to generate an STB prediction. The subject may then be treated based on the STB prediction.

Description

2024-068 / 10738-1206 SUICIDAL THOUGHTS AND BEHAVIOR PREDICTIONS AND TREATMENT CROSS-REFERENCE TORELATEDAPPLICATION

[0001] This application claims priority to U.S. Provisional Application Serial No.63 / 551,326, filed February 08, 2024, the entire contents of which are incorporated herein by reference. STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT

[0002] This invention was made with government support under N000142112216 and N00014-23-1-2396 awarded by the Office of Naval Research. The government has certain rights in the invention. TECHNICALFIELD

[0003] Aspects of the present disclosure relate to predictions of suicidal thoughts and behavior, and treatment thereof, based on quantification of reward-aversion judgment. BACKGROUND

[0004] Suicide is a leading cause of death in the United States and worldwide. It is a serious public health problem with lasting effects on individuals, families, and communities. In the United States, suicide rates increased over 30% between 2000 and 2020. Further, during the COVID-19 pandemic, rates of suicide, self-harm, anxiety, depression, and stress increased. There were approximately 45,000 deaths from suicide in the U.S. in 2020. Suicide attempts are even more frequent, with an estimated 1.3 million suicide attempts for U.S. adults each year. The estimated cost of such attempts, including for injury treatment costs and loss of productivity, is over $93.5 billion.

[0005] Current methods for suicide prevention and reduction include prediction of incidence of suicidal thought and behavior (STB) for an individual. STB may include suicidal ideation, suicidal attempt(s), and completed suicide. Clinical measures may be predictive, or otherwise correlated with STB, such as post-traumatic stress disorder, anger, severe depression, psychotic symptoms, and hopelessness.2024-068 / 10738-1206

[0006] Additionally, social and behavioral measures may also play a role in prediction of incidence of STB. For example, measures of social integration and social media behavior may be associated with incidence of STB. SUMMARY

[0007] Certain aspects include a method of diagnosis and treatment for suicidal thought and behavior (STB), comprising: generating a set of STB parameters associated with a subject, wherein the set of STB parameters comprises a set of reward-aversion judgment variables, and at least one contextual variable; processing the set of STB parameters with a machine learning model to generate a STB prediction; and treating the subject with an STB treatment based on the STB prediction.

[0008] Other aspects include a method of diagnosis of STB, comprising generating a set of STB parameters associated with a subject, wherein the set of STB parameters comprises a set of reward-aversion judgment variables, and at least one contextual variable; and processing the set of STB parameters with a machine learning model to generate a STB prediction.

[0009] In some aspects, the method further comprises presenting to the subject, via a user interface, a set of stimuli; receiving, from the subject via the user interface, a set of ratings associated with the set of stimuli, wherein each respective rating in the set of ratings corresponds with a respective stimuli in the set of stimuli; and determining, based on the set of ratings, the set of reward-aversion judgment variables.

[0010] In some aspects, the method further comprises presenting, to a subject via the user interface, at least one contextual question; and receiving, from the subject via the user interface, the at least one contextual variable.

[0011] In some aspects, the set of stimuli comprises one or more subsets of stimuli, and each subset of the one or more subset of stimuli comprises one or more stimuli associated with a category.

[0012] In some aspects, determining the set of reward-aversion judgment variables based on the set of ratings comprises: determining one or more subsets of ratings of the set of ratings, wherein each one of the one or more subsets of ratings corresponds to the respective stimuli in the2024-068 / 10738-1206 one or more subsets of the set of stimuli; determining an average of each respective subset of the one or more subsets of ratings; determining a variance of each respective subset of the one or more subsets of ratings; and determining an uncertainty or entropy of each respective subset of the one or more subsets of ratings.

[0013] In some aspects, determining the set of reward-aversion judgment variables based on the set of ratings comprises: determining one or more reward-aversion judgment variables in the set of reward-aversion judgment variables based on a relative preference graph.

[0014] In some aspects, the set of reward-aversion judgment variables comprises one or more of: a loss aversion variable, a risk aversion variable, a loss resilience variable, an ante variable, an insurance variable, a peak positive risk variable, a peak negative risk variable, a total reward variable, a total aversion risk variable, or a tradeoff range variable.

[0015] In some aspects, the machine learning model comprises a linear regression, a random forest, a Gaussian Process, a Gaussian Mixture Model, a Support Vector Machine, or an artificial neural network.

[0016] In some aspects, the machine learning model comprises a balanced random forest.

[0017] In some aspects, the at least one contextual variable comprises one or more of: a loneliness variable, a prior suicidal attempts variable, a health variable, or an anxiety variable.

[0018] In some aspects, the set of STB parameters further comprises biographical data associated with the subject.

[0019] In some aspects, the biographical data associated with the subject comprises one or more of: age data, income data, marital status data, employment data, ethnicity data, education level data, sex data, gender data or handedness data.

[0020] In some aspects, the STB prediction comprises one or more measures of STB, the one or more measures of STB comprising a passive ideation measure, an active ideation measure, a suicide planning measure, or a planning for safety measure.

[0021] In some aspects, the STB treatment comprises one or more of a selective serotonin2024-068 / 10738-1206 reuptake inhibitor, a serotonin-norepinephrine reuptake inhibitor, an atypical antidepressant, a tricyclic antidepressant, a monoamine oxidase inhibitor, mood stabilizers, antipsychotics, anti- anxiety medications, or stimulant medication.

[0022] Other aspects provide processing systems configured to perform the aforementioned methods as well as those described herein; non-transitory, computer-readable media comprising instructions that, when executed by a processors of a processing system, cause the processing system to perform the aforementioned methods as well as those described herein; a computer program product embodied on a computer readable storage medium comprising code for performing the aforementioned methods as well as those further described herein; and a processing system comprising means for performing the aforementioned methods as well as those further described herein.

[0023] The following description and the related drawings set forth in detail certain illustrative features of one or more aspects. BRIEFDESCRIPTION OF THEDRAWINGS

[0024] The appended figures depict certain aspects and are therefore not to be considered limiting of the scope of this disclosure.

[0025] FIG. 1 depicts an exemplary system for generating predictions of STB, according to one or more aspects of the disclosure.

[0026] FIGS.2A-2B depict exemplary processing devices configured to generate predictions of STB, according to one or more aspects of the disclosure.

[0027] FIG. 3 depicts an exemplary workflow for generating predictions of STB, according to one or more aspects of the disclosure.

[0028] FIG. 4 depicts an exemplary workflow for quantification of reward-aversion judgment, according to one or more aspects of the disclosure.

[0029] FIG. 5 depicts an exemplary workflow for training a machine learning model to generate predictions of STB, according to one or more aspects of the disclosure.2024-068 / 10738-1206

[0030] FIG. 6 depicts an exemplary method for generating predictions of STB, according to one or more aspects of the disclosure.

[0031] FIG.7 depicts an exemplary method for training a machine learning model to generate predictions of STB, according to one or more aspects of the disclosure.

[0032] FIG.8A depicts an exemplary rating task for quantification of judgment, according to one or more aspects of the disclosure.

[0033] FIG. 8B depicts a first exemplary set of variables for quantification of judgment, according to one or more aspects of the disclosure.

[0034] FIG. 8C depicts a second exemplary set of variables for quantification of judgment, according to one or more aspects of the disclosure.

[0035] FIG. 8D depicts a third exemplary set of variables for quantification of judgment, according to one or more aspects of the disclosure.

[0036] FIG.9 depicts an exemplary set of reward-aversion judgment variables and interaction with four measures of STB, according to one or more aspects of the disclosure.

[0037] FIG. 10 depicts an exemplary workflow for training and inferencing with a machine learning model configured to generate predictions of STB, according to one or more aspects of the disclosure.

[0038] FIG. 11 depicts another exemplary workflow for training and inferencing with a machine learning model configured to generate predictions of STB, according to one or more aspects of the disclosure.

[0039] FIGS. 12A-12B depict exemplary sets of variables for quantification of judgment, according to one or more aspects of the disclosure.

[0040] To facilitate understanding, identical reference numerals have been used, where possible, to designate identical elements that are common to the drawings. It is contemplated that elements and features of one aspect may be beneficially incorporated in other aspects without further recitation.2024-068 / 10738-1206 DETAILEDDESCRIPTION

[0041] Aspects of the present disclosure provide apparatuses, methods, processing systems, and computer-readable mediums for predicting incidence of suicidal thoughts and behavior (STB), and treatment thereof, based on quantification of reward-aversion judgment.

[0042] Early and accurate prediction of STB enables earlier and more effective treatment and intervention. Conventional evaluation and risk diagnosis rely on accurate and honest self-reporting and self-evaluation, and previous STB. This can lead to incorrect and / or inaccurate evaluation methods, often delaying treatment for suffering individuals, and decreasing interventions to prevent future self-harm.

[0043] Aspects of the present disclosure relate to systems and methods for prediction and treatment of STB, through quantification of reward-aversion judgment with contextual information, including contextual mental health information. For example, aspects described herein provide for systems and methods for generating predictions of STB through machine learning based on reward-aversion judgment variables and contextual variables. In some aspects, reward-aversion judgment variables may be generated based on a rating task, whereby a subject indicates approach or avoidance (e.g., positive or negative) responses to various stimuli. Contextual variables may include variables related to overall health, prior suicidal attempts, loneliness, or anxiety. The reward-aversion judgment variables and one or more contextual variables may be processed with a machine learning model trained to predict STB, and / or treatment of the subject.

[0044] Beneficially, the rating task is inconspicuous compared to conventional surveys and may elicit more honest answers, e.g., by not asking questions directly related to STB. For example, the rating task limits biases involved in STB. The rating task involves presenting sets of stimuli, such as pictures, to users and receiving user ratings. The rating task, including the stimuli, does not have a perceivable relation to STB. Thus, the machine learning model does not incorporate such bias in generating the STB predictions.

[0045] Furthermore, the rating task may be readily deployed to personal computing devices of users, for example, smart devices or computers, enabling convenient completion and unobtrusive patient assessment and monitoring.2024-068 / 10738-1206

[0046] The responses from the rating task may be used to determine a set of reward-aversion judgment variables. The set of reward-aversion judgment variables may be a quantification of aspects of judgments of the subject. Each judgment variable in the set of judgment variables may describe a quantitative component of a user’s approach, avoidance, or judgment behavior. For example, in individuals with STB may show increased aversion to risk and loss, with lower focus on negative consequences of decisions, discounting of delayed rewards, and higher bias to escape aversive situations. Abnormalities in reward / aversion judgment have been linked to dopamine dysfunction in major depressive disorder, addiction, anxiety, chronic stress, and STB.

[0047] Additionally, one or more contextual variables may be utilized in conjunction with the set of judgment variables to predict STB, and / or treatment of the subject. These contextual variables may be related to STB, but do not directly reference STB, for example, by not specifically referencing STB such as self-harm. In some cases, additional biographic information may be utilized to increase predictive power of the machine learning model.

[0048] Certain aspects described herein including generating and training the machine learning model to predict STB and treatment, based on the reward-aversion judgment variables derived from the subject’s ratings and the contextual variables. Beneficially, the input data (e.g., independent variables) may be beneficially few, but very predictive. For example, a relatively small number of judgment variables may be utilized, in some cases, from 20 to 40 variables, to generate an STB prediction and / or treatment prediction.

[0049] In certain aspects described herein, the machine learning model may comprise a classification model, for example, a balanced random forest model or Gaussian mixture model. Beneficially, a balanced random forest model, and other classification models may be readily trained, generated, and maintained. Thus, the computation resources to train, generate, and maintain the machine learning model may be fewer compared to more computationally and resource intensive methods. For example, compared to large language models (LLMs) or natural language processing, a classification model, such as a balanced random forest or linear regression, comprises fewer parameters for training and maintaining the model. Further, a smaller volume of training data may be utilized by the classification model to generate a performative model. Thus, beneficially, aspects described herein may be computationally and resource efficient, while providing powerful performance.2024-068 / 10738-1206

[0050] In some aspects, prediction of STB may include prediction of one or more measures of STB. In some aspects, measures of STB may correspond to severity of ideation. A first measure of STB may be a passive ideation measure. A passive ideation measure may correspond to either wishing to be dead or wishing to go to sleep and not wake up.

[0051] A second measure of STB may be an active ideation measure. An active ideation measure may correspond to wanting to hurt themselves or taking their own life, but does not explicitly measure intent.

[0052] A third measure of STB may be a suicide planning measure. A suicide planning measure may correspond to planning for suicide, including thoughts about how to accomplish hurting themselves or taking their own life, thoughts about doing something to make themselves not alive anymore, and active planning steps, such as started to do anything or preparing to do anything to end their life.

[0053] A fourth measure of STB may be a planning for safety measure. A planning for safety measure may correspond to a safety plan to prevent or reduce hurting themselves when these feelings arise. A safety plan is a short-term intervention to help individuals survive suicidal crises by developing a set of steps to reduce the likelihood of engaging in suicidal behavior. In some cases, a safety plan may include identification of warning signs and triggers that indicate suicidal ideation is likely to occur, internal coping strategies for when the signs and triggers occur, social contacts and locations to provide distraction, supportive contacts who can provide assistance, emergency resources, and steps to ensure safety of the environment to minimize the subject’s ability to act on suicidal thoughts or urges.

[0054] Although described here as four measures of STB, additional or fewer measures of STB may be predicted.

[0055] In some aspects, a treatment for STB may generated based on the STB prediction, such as a medication treatment, a therapy treatment, and the like. Medications may include one or more of selective serotonin reuptake inhibitors (SSRIs), serotonin-norepinephrine reuptake inhibitors (SNRIs), monoamine oxidase inhibitors (MAOIs), atypical antidepressants, tricyclic antidepressants, mood stabilizers, anti-anxiety medications, or stimulant medications. In some aspects, a treatment method may comprise a therapy, for example, cognitive-behavior therapy, interpersonal therapy, electroconvulsive therapy, transcranial magnetic stimulation, and the like.2024-068 / 10738-1206

[0056] In some aspects, a treatment for STB may be based on one or more measures of STB. For example, in some aspects, where a first measure of STB may be high, a first treatment may be utilized, such as a therapy treatment. Similarly, wherein a third measure of STB may be high, a third treatment may be utilized, such as a medication treatment. In some aspects, one or more treatments may be utilized based on the measures of STB.

[0057] Reference will now be made in detail to aspects of the present disclosure, examples of which are illustrated in the accompanying drawings. It is to be understood that other aspects may be utilized, and structural and functional changes may be made without departing from the scope of the present disclosure. Moreover, features of the aspects may be combined, switched, or altered without departing from the scope of the present disclosure, e.g., features of each disclosed embodiment may be combined, switched, or replaced with features of the other disclosed aspects. As such, the following description is presented by way of illustration and does not limit the various alternatives and modifications that may be made to the illustrated aspects and still be within the spirit and scope of the present disclosure.

[0058] As used herein, the words "example" and "exemplary" mean an instance, or illustration. The words "example" or "exemplary" do not indicate a key or preferred aspect or embodiment. The word "or" is intended to be inclusive rather an exclusive, unless context suggests otherwise. As an example, the phrase "A employs B or C," includes any inclusive permutation (e.g., A employs B; A employs C; or A employs both B and C). As another matter, the articles "a" and "an" are generally intended to mean "one or more" unless context suggest otherwise. Example System for Prediction and Treatment of STB

[0059] FIG.1 depicts an example system 100 for generating prediction and treatment of STB. In some aspects, system 100 is configured to predict one or more of the measures of STB, and / or a treatment for STB. For example, in some aspects, system 100 is configured to predict a first measure of STB, a second measure of STB, a third measure of STB, or a fourth measure of STB.

[0060] System 100 may include a user 102 interacting with a computing device 108 to complete a task. Computing device 108 comprises a user interface 110. The user interface 110 may run on a variety of computing devices, including personal computers, tablet computers, smart2024-068 / 10738-1206 devices, and others. The user interface 110 may be a graphical user interface for a website, application, software program, and the like. The user interface 110 is configured to be displayed on a monitor, television, touchscreen, and the like, of computing device 108.

[0061] In some aspects, system 100 is implemented in a doctor’s office, a clinic, via a web survey, via a social media link or message, via a public health message, and the like. For example, system 100, and / or method 600 of FIG. 6 may be implemented as part of a physical, well-check, annual check-up, a mental health screening, and the like. In some aspects, system 100 is implemented at an emergency department.

[0062] In some aspects, the user 102 comprises a child, an adolescent, or an adult.

[0063] In FIG. 1, a user 102 interacts with task 106 presented on the user interface 110 of a computing device 108. The user interface 110 may be rendered by computing device 108, such as a monitor of a touch screen of computing device 108. Computing device 108 may control the user interface 110 to present the task 106. In some aspects, the task 106 may comprise a rating task, a contextual task, and / or a biographical task. In some aspects, as part of a rating task, the computing device 108 is configured to present one or more stimuli to the user 102, for example, one or more pictures. Other example stimuli may include one or more sounds or one or more videos, for example, through an audio component coupled to the computing device 108, such as a speaker. The one or more stimuli may comprise one or more subsets, wherein all the pictures in a subset are associated with a category (e.g., based on content, type, effect, etc.). In some aspects, the contextual task comprises one or more contextual questions, for example, related to STB of the user 102. In some aspects, the biographical task comprises one or more biographical questions, for example, related to biographical data of the user 102.

[0064] Further, the computing device 108 is configured to receive one or more responses 104 from the user 102. In some aspects, a response 104 may include a rating, as part of a rating task. For example, and as described in further detail with respect to workflow 500 of FIG. 5, the user 102 may select or input a rating as a response 104, for example through a rating user interface element displayed on user interface 110. In some aspects, the user 102 may select one or more rating options displayed on a user interface element of user interface 110. In some aspects, the user 102 may enter, for example through a touchscreen of user interface, through one or more control elements of user interface, or through one or more control elements of computing device2024-068 / 10738-1206 108 (e.g., keys, mouse, buttons, etc.), text comprising a response. Each of the one or more ratings may correspond to a respective stimulus in the one or more stimuli presented via computing device 108 as part of task 106. For example, one stimulus of the one or more stimuli may be presented to the user 102 and one rating may be received from user 102. The one rating corresponds to the one picture presented, for example, rating scale 804 in FIG. 8A, depicts an example discrete rating scale associated with an example picture 802. A subset of ratings may be determined based on each rating corresponding to a picture in a given subset. For example, a subset of pictures may be associated with the category of nature, e.g., each picture in the subset depicts nature. Each subset may be identified by association to a category, for example, a nature category. Each rating entered by the user 102 for a nature picture in the nature category subset of pictures may be determined to be part of a nature category subset of ratings. Thus, each subset of ratings may correspond to a subset of pictures, wherein each rating in the subset of ratings corresponds to one of the pictures in the subset of pictures.

[0065] A rating may be based on the user’s preferences, emotions, and / or attention, for example, a user may rate the stimulus based on an initial response. In some aspects, a rating may comprise a continuous scale. For example, a continuous scale may be a numerical rating, such as from 0 to 5, from -3 to +3, etc. As another example, a continuous scale may be based on a sliding scale, such as a rating between “dislike very much” to “like very much” endpoints. In some aspects, a rating may comprise a discrete scale, or one or more categories. For example, a user may select a rating from a set of rating options, such as one or more emoticons (e.g., smiley face, sad face, angry face, etc.), one or more terms (e.g., “dislike very much”, “dislike”, “neutral”, “like”, “like very much”, etc.), one or more emotions (e.g., “happy”, “sad”, “angry”, etc.).

[0066] In some aspects, a response 104 may comprise a response to a contextual question, such as a contextual variable, as part of a contextual task. In some aspects, a contextual question may comprise a health related question, an anxiety related question, a loneliness related question, a prior self-harm attempt related question, and / or the like. For example, in some aspects, a health related question might comprise a health questionnaire, such as the eight-item Patient Health Questionnaire (PHQ-8) scale (e.g., the Patient Health Questionnaire-9 without the question on suicide. In some aspects, an anxiety related question may comprise an anxiety question such as a State-Trait Anxiety Inventory (STAI). In some aspects, a loneliness related question may comprise2024-068 / 10738-1206 a question regarding self-reporting loneliness. In some aspects, a prior self-harm attempt related question may comprise a question regarding self-reported number of attempts at self-harm in the last 12 months.

[0067] In some aspects, a response may comprise a response to a biographic question, such as biographical data, in a biographical task. Biographical data may include data related to age, income, marital status, employment, ethnicity, education level, sex, gender, handedness, and the like.

[0068] The computing device 108 is further configured to interface with an STB prediction component 112, for example to send the responses 104 received from user 102. STB prediction component 112 may be run on computing device 108, or an external component (e.g., accessed through an application programming interface (API), such as in a microservices-type deployment). STB prediction component 112 is configured to generate predictions of STB, STB related behaviors (e.g., diminished self-care), and broader health management associated with user 102. As described in further detail, for example, with respect to workflow 300 of FIG. 3 and method 600 of FIG. 6, STB prediction component 112 is configured to generate STB prediction 120 associated with user 102. In some aspects, the STB prediction 120 may comprises an STB prediction, a prediction of a measure of STB prediction, or a treatment method for a STB. STB prediction component 112 comprises a ratings component 114, a contextual component 116, and a machine learning component 118.

[0069] Ratings component 114 is configured to determine a set of reward-aversion judgment variables based on the one or more responses 104 from the user 102, for example, based on one or more ratings associated with one or more stimuli of the task 106. The set of reward-aversion judgment variables may be a quantification of aspects of reward-aversion judgments made by the user 102 in rating the stimuli. Each reward-aversion judgment variable in the set of reward- aversion judgment variables may describe a quantitative component of a user’s approach, avoidance, or judgment behavior. The set of reward-aversion judgment variables may include, for example, loss aversion, risk aversion, loss resilience, ante, insurance, peak positive risk, peak negative risk, reward tipping point, aversion tipping point, total reward risk, total aversion risk, reward aversion tradeoff, tradeoff range, reward aversion consistency, and consistency range. For example, Table 1 depicts example reward-aversion judgment variables and associated abbreviations.2024-068 / 10738-1206 Table 12024-068 / 10738-1206

[0070] Ratings component 114 is configured to determine a set of approach / avoidance variables based on the one or more ratings from the user 102 in response to the set of stimuli. The set of approach / avoidance variables may include, for example, a mean, variance, and uncertainty of the one or more ratings from the user 102. In some aspects, a set of approach / avoidance variables may be determined for each subset of ratings, e.g., each subset of ratings corresponding to a subset of pictures (e.g., based on category). Ratings component 114 is further configured to graph the set of approach / avoidance variables, and as described in further detail below, for example, with respect to workflow 400 in FIG. 4, and method 600 in FIG. 6, and method 700 in FIG. 7 to determine one or more reward-aversion judgment variables based on the graph(s).

[0071] Contextual component 116 is configured to determine, based on the response 104 to contextual questions, one or more contextual variables. A contextual variable may comprise an answer to a health question, an answer to an anxiety question, a loneliness rating, a number of2024-068 / 10738-1206 prior self-harm attempts, and / or the like. For example, a health contextual variable may comprise a score of a health questionnaire, such as a PHQ-8 score. As another example, an anxiety contextual variable may comprise a score of an anxiety questionnaire, such as a STAI score. In yet another example, a loneliness contextual variable may comprise a self-reported loneliness rating. In yet another example, a self-harm attempt contextual variable may comprise a number of self-harm attempts in a prior period of time, such as a prior 1 month, a prior, 3 months, a prior 6 months, a prior 12 months, etc.

[0072] Machine learning component 118 is configured to process the set of reward-aversion judgment variables and the contextual variable(s) with a machine learning model to generate the STB prediction 120. The machine learning model may be trained to predict an occurrence of STB for the user 102. In some aspects, the machine learning model may be trained to predict a measure of STB as prediction 120 associated with the user 102.

[0073] For example, the machine learning model may be trained to predict a first measure of STB. In some aspects, the prediction of the first measure of STB may comprise a rating of occurrence of passive ideation of the user 102. For example, a rating of occurrence of passive ideation may be based on a Likert scale of 1-5, where 1 is no suicidal ideation (e.g., no occurrence of suicidal ideation) and 2 through 5 increasing degrees of suicidal ideation (e.g., increasing occurrence of suicidal ideation). The scale may comprise 1 being “never”; 2 being “rarely”, 3 being “sometimes”; 4 being “often”, and 5 being “always.”

[0074] In some aspects, the machine learning model may be trained to predict a second measure of STB. For example, the prediction of the second measure of STB may comprise a rating of occurrence of active ideation of the user 102. A rating of occurrence of active ideation may be based on a Likert scale of 1-5, where 1 is no active ideation (e.g., no occurrence of active suicidal ideation) and 2 through 5 increasing degrees of active ideation (e.g., increasing occurrence of active suicidal ideation).

[0075] In some aspects, the machine learning model may be trained to predict a third measure of STB. For example, the prediction of the third measure of STB may comprise a rating of occurrence of suicide planning of the user 102. A rating of occurrence of suicide ideation may be2024-068 / 10738-1206 based on a Likert scale of 1-5, where 1 is no suicide planning (e.g., no occurrence of suicide planning), and 2 through 5 increasing degrees of suicide planning (e.g., increasing occurrence or intensity of planning).

[0076] In some aspects, the machine learning model may be trained to predict a fourth measure of STB. For example, the prediction of the fourth measure of STB may comprise a rating of occurrence of safety planning of the user 102. A rating of occurrence of safety planning may be based on a Likert scale of 1-5, where 1 is no safety plan (e.g., no occurrence of safety plan), and 2 through 5 increasing degrees of safety planning (e.g., increasing occurrence or intensity of safety plan).

[0077] In some aspects, machine learning component 118 may comprise a set of machine learning models, wherein each machine learning model may be trained to predict a measure of STB. For example, a first machine learning model trained to predict a first measure of STB, a second machine learning model trained to predict a second measure of STB, a third machine learning model trained to predict a third measure of STB, and a fourth machine learning model trained to predict a fourth measure of STB.

[0078] In some aspects, machine learning component 118 comprises a classification model, for example, a binary classification model. Examples of classification models include a linear regression, a random forest, a Gaussian Process, a Gaussian Mixture Model classifier (GMM), a Support Vector Machine (SVM), or an artificial neural network (ANN). In some aspects, the machine learning component 118 comprises a balanced random forest (BRF) classifier machine learning model.

[0079] A machine learning model is a mathematical representation or algorithm which is used to recognize patterns, make predictions, or perform specific tasks, called “inferencing”. Generally, a model uses input data and its training to generate an output. The machine learning component 118 may be trained to generate a prediction of STB, for example, as described with respect to FIG. 5.

[0080] In some aspects, machine learning component 118 is further configured to process biographical data in generating the STB prediction 120. Biographical data may include biographical data associated with user 102. In aspects, user 102 may input biographical data via a user interface of computing device 108 in response to a biographical task. In some aspects,2024-068 / 10738-1206 biographical data may be obtained from a database (not pictured), for example, a health record, such as an electronic medical record (EMR) system. Biographical data may include age data, income data, marital status data, employment data, ethnicity data, education level data, sex data, gender data or handedness data associated with a user.

[0081] In some aspects, the STB prediction 120 is presented to user 102, for example, on the user interface 110 of computing device 108. In some aspects, the STB prediction 120 facilitates one or more downstream processes. For example, a prediction component 122 may be further configured to send STB prediction 120 to one or more downstream components, such as to a treating physician, to a health record or EMR, and the like.

[0082] In some aspects, a treatment component 124 is configured to utilize the STB prediction 120 to generate a treatment for the user 102. For example, treatment component 124 may be configured to generate a treatment recommendation based on the STB prediction 120. In some examples, treatment component 124 may be configured to provide one or more messages, for example, a message comprising educational materials, lifestyle changes and / or techniques, stress management techniques, support groups, mindfulness and relaxation techniques, therapy recommendations, and the like. In some aspects, the one or more messages may be provided to the user 102. In some examples, treatment component 124 is configured to generate a treatment recommendation based on a measurement of STB.

[0083] In some aspects, prediction component 122 and / or treatment component 124 may be configured to transmit to the user 102, a communication based on the STB prediction 120, for example, communication comprising a diagnosis or a treatment for a STB associated with the user 102, a predicted behavior of user 102 associated with STB, and the like. Example Processing Device for STB Predictions

[0084] FIGS. 2A-2B depict example processing devices 200 and 250 configured to perform the methods described herein, for example, to generate a STB prediction. In some aspects, processing device 200 is configured to perform aspects of workflow 300 described with respect to FIG. 3, or aspects of workflow 400 as described with respect to FIG. 4.2024-068 / 10738-1206

[0085] Processing device 200 includes one or more processors 202. Generally processor(s) 202 may be configured to execute computer-executable instructions (e.g., software code) to perform various functions, as described herein.

[0086] Processing device 200 further includes a network interface(s) 204, which generally provides data access to any sort of data network, including personal area networks (PANs), local area networks (LANs), wide area networks (WANs), the Internet, and the like. In some aspects, processing device 200 may be implemented in a closed network, such as a local clinical network, etc.

[0087] Processing device 200 further includes input(s) and output(s) 206, which generally provide means for providing data to and from processing device 200, such as via a connecting to computing device peripherals, including user interface peripherals.

[0088] Processing device 200 further includes a memory 210 (e.g., a non-transitory, processor readable storage medium) configured to store various types of components and data, in this example, including presenting component 220, stimuli data 240, receiving component 222, and ratings data 242, task data 246, contextual variable data 248, and biographical data 244.

[0089] Presenting component 220 is configured to present one or more tasks, such as tasks stored in task data 246, for example, a rating task, a contextual information task, or a biographical information task. In some aspects, presenting component 220 may be configured to present one or more stimuli as part of a ratings task, such as stimuli stored as stimuli data 240. For example, presenting component 220 may present a picture on a user interface, present a video on a user interface, and present a sound via a speaker, which may be an example of an output 206. Presenting component 220 is configured to present a set of stimuli, in some aspects, the set of stimuli comprises one or more subsets of stimuli, and each subset of the one or more subset of stimuli comprises one or more stimuli associated with a category. In some aspects, presenting component 220 may be configured to present a contextual information task, for example, one or more contextual questions to a user. In some aspects, presenting component 220 may be configured to present a biographical information task, for example, one or more biographical information questions. For example, presenting component 220 may be configured to present a set of stimuli to a user, such as described with respect to block 404 of FIG. 4, below.2024-068 / 10738-1206

[0090] Receiving component 222 is configured to receive responses from a user, for example, via a user interface of processing device 200. In some aspects, a response may comprise one or more ratings, for example, receive a rating and store a rating in rating data 242. For example, receiving component 222 may be configured to receive a set of ratings from a user, such as described with respect to block 406 of FIG. 4, below. In some aspects, receiving component 222 is further configured to receive contextual variable data associated with a user, for example, as part of a contextual information task, and store it as contextual variable data 248. In some aspects, receiving component 222 is configured to receive biographical data associated with a user, for example, as part of a biographical information task, and store it as biographical data 244. Biographical data may include, for example, age data, income data, marital status data, employment data, ethnicity data, education level data, or sex data associated with a user.

[0091] Processing device 200 may be implemented in various ways. For example, processing device 200 may be implemented within on-site, remote, or cloud-based processing equipment.

[0092] In some aspects, processing device 250 is configured to perform aspects of workflow 300 described with respect to FIG.3, aspects of workflow 500 as described with respect to FIG. 5, or aspects of workflow 500 as described with respect to FIG. 5, or aspects of method 600 as described with respect to FIG. 6, or aspects of method 700 as described with respect to FIG. 7.

[0093] Processing device 250 includes one or more processors 252. Generally processor(s) 252 may be configured to execute computer-executable instructions (e.g., software code) to perform various functions, as described herein.

[0094] Processing device 250 further includes a network interface(s) 254, which generally provides data access to any sort of data network, including personal area networks (PANs), local area networks (LANs), wide area networks (WANs), the Internet, and the like. In some aspects, processing device 250 may be implemented in a closed network, such as a local clinical network, etc.

[0095] Processing device 250 further includes input(s) and output(s) 256, which generally provide means for providing data to and from processing device 250, such as via a connecting to computing device peripherals, including user interface peripherals.2024-068 / 10738-1206

[0096] Processing device 250 further includes a memory 260 (e.g., a non-transitory, processor readable storage medium) configured to store various types of components and data, in this example, including determining component 224, generating component 226, training component 228, training data 232, machine learning model 230, prediction data 234, and treatment data 236.

[0097] Determining component 224 is configured to determine a set of reward-aversion judgment variables, for example, based on one or more ratings. For example, determining component 224 is configured to determining one or more judgment variables based on the one or more ratings, such as described with respect to block 408 of FIG. 4, below. In some aspects, determining component 224 is configured to determine one or more subsets of ratings of the set of ratings, wherein each one of the one or more subsets of ratings corresponds to the respective stimuli in the one or more subsets of the set of stimuli. In some aspects, determining component 224 is configured to determine an average of each respective subset of the one or more subsets of ratings. In some aspects, determining component 224 is configured to determine a variance of each respective subset of the one or more subsets of ratings. In some aspects, determining component 224 is configured to determine an uncertainty of each respective subset of the one or more subsets of ratings. In some aspects, determining component 224 is further configured to generate a relative preference graph based on the average of each respective subset of the one or more subsets of ratings, the variance of each respective subset of the one or more subsets of ratings, and the uncertainty of each respective subset of the one or more subsets of ratings. In some aspects, determining component 224 is further configured to determine one or more judgment variables in the set of judgment variables based on the relative preference graph and store the one or more judgment variables in ratings data 242.

[0098] Generating component 226 is configured to generate a set of STB parameters associated with a user based on the set of judgment variables, the one or more contextual variables, and / or the biographical data. In some aspects, generating component 226 is further configured to process biographical data 244 in generating the STB prediction. Biographical data 244 may include biographical data associated with user. In some aspects, biographical data 244 may be obtained from a database, for example, a health record such as an electronic medical record (EMR) system. Biographical data 244 may include age data, income data, marital status data, employment data, ethnicity data, education level data, or sex data associated with a user.2024-068 / 10738-1206

[0099] Training component 228 is configured to train machine learning model 230 based on training data stored in training data 232, such as described with respect to workflow 500 of FIG. 5. Training a machine learning model may include supervised learning, unsupervised learning, and reinforcement learning.

[0100] Supervised learning involves training the model with a training dataset comprising labeled sets of pairs. The pair includes input data and output data. The model is trained to generate a prediction based on the input. During training, the parameters of the model are adjusted to minimize the different between the prediction and the output.

[0101] Unsupervised learning involves training the model with a training dataset comprising unlabeled data. The model is tasked to find patterns, relationships, groups, or clusters between the input data without guidance (e.g., from labels).

[0102] Reinforcement learning involves training the model based on feedback for actions taken by the model. The model includes an agent configured to take actions in an environment. Each action taken by the agent receives feedback, either a reward (e.g., positive), or penalty (e.g., negative). The agent learns by taking actions to maximize rewards and minimize penalties.

[0103] In some aspects, the type of training utilized by the training component 228 may depend on the machine learning model 230 utilized. For example, machine learning model 230 may comprise a classification model, for example, a linear regression, a random forest, a Gaussian Process, a SVM, or an ANN. In some aspects, the machine learning model comprises a GMM.

[0104] Processing device 250 may be implemented in various ways. For example, processing device 250 may be implemented within on-site, remote, or cloud-based processing equipment.

[0105] Processing devices 200 and 250 are just examples, and other configurations are possible. For example, in an alternative aspect, aspects described with request to processing device 200 or processing device 250 may be omitted, added, or substituted for alternative aspects. Example Workflow for STB Prediction

[0106] FIG.3 depicts an example workflow 300 for generating and treating a STB prediction, for example, STB prediction 120 in FIG.1. In some aspects, workflow 300 may be performed by STB prediction component 112 in FIG. 1.2024-068 / 10738-1206

[0107] At step 304, one or more tasks may be completed by a subject 302, for example, one or more task 106 in FIG. 1. In some aspects, a task may comprise a rating task. In some aspects, a task may comprise a contextual information task. In some aspects, a task may comprise a biographical information task.

[0108] In some aspects, in response to presentation of the one or more tasks, one or more responses may be received from the subject 302. For example, the one or more response may include one or more ratings as part of a rating task, one or more contextual variables in response to a contextual information task, or one or more biographical data in response to a biographical information task.

[0109] A set of STB parameters may be generated based on completion of the one or more tasks at step 304. In some aspects, the STB parameters may include a set of reward-aversion judgment variables, and at least on contextual variable. The set of reward-aversion judgment variables may be generated based on a rating task. The at least one contextual variables may be generated based on a contextual information task. In some aspects, the set of STB parameters further includes biographical data, generated based on a biographical information task.

[0110] In some aspects, the set of reward-aversion judgment variables comprise a quantification of a subject’s judgment. In some aspects, the set of reward-aversion judgment variables comprises a quantification of how a subject makes judgments prior to decision-making, for example, including gender effects, ethnicity, marital status, and / or employment. For example, in some aspects, a subject’s judgment is determined based on relative preference theory (RPT) framework. The RPT framework may quantify and capture judgments about the valence of judgment, for example, positive vs. negative, approach vs. avoidance, etc., as well as magnitude of judgment (e.g., intensity of rating) to describe a user’s preferences. For example, in some aspects, a subject’s judgment may be quantified through a picture rating task of the RPT, such as described with respect to workflow 400 in FIG. 4.

[0111] In some aspects, the at least one contextual variable may comprise an answer to a health question, an answer to an anxiety question, a loneliness rating, a number of prior self-harm attempts, and / or the like, such as part of a contextual information task. For example, a health contextual variable may comprise a score of a health questionnaire, such as a PHQ-8 score. As another example, an anxiety contextual variable may comprise a score of an anxiety questionnaire,2024-068 / 10738-1206 such as a STAI score. In yet another example, a loneliness contextual variable may comprise a self-reported loneliness rating. In yet another example, a self-harm attempt contextual variable may comprise a number of self-harm attempts in a prior period of time, such as a prior 1 month, a prior, 3 months, a prior 6 months, a prior 12 months, etc.

[0112] In some aspects, the health questionnaire may be a Patient Health Questionnaire-8 (PHQ-8); an example is depicted as Table 2. PHQ-8 is a screening tool for assessing severity of depression. The PHQ-8 consists of 8 questions focusing on the symptoms of depression, as outlined in the Diagnostic and Statistical Manual of Mental Disorders (DSM). Each question corresponds to one of the 8 diagnostic criteria for major depressive disorder, according to the DSM. A user responds to each question based on how often they have experienced the symptom over the past two weeks, based on a scale. For example, a 0 rating indicates the user has not experienced the symptom over the past two weeks, while a 3 rating indicates the user has experienced the symptom nearly every day for the past two weeks. Table 2

[0113] The scores for each question are summed to produce a total score ranging from 0 to 27, with a higher score indicating more severe depression. For example, scores 0-4 may indicate minimal depression corresponding to a level 1. Scores 5-8 may indicate mild depression2024-068 / 10738-1206 corresponding to a level 2. Scores 10-14 may indicate moderate depression corresponding to a level 3. Scores 15-19 may indicate moderately severe depression corresponding to a level 4. Scores 20-27 may indicate a severe depression corresponding to a level 5.

[0114] In some aspects, the anxiety questionnaire may comprise a STAI questionnaire, and example is depicted as Table 3. A STAI score is a measure of trait and state anxiety and is often used in clinical settings to diagnose anxiety and to distinguish it from depressive syndromes. Subjects rate items on a 4-point scale from “almost never” to “almost always.” Higher scores may indicate greater anxiety. Table 32024-068 / 10738-1206

[0115] At step 306, the STB parameters are processed with a machine learning model, for example, machine learning component 118 in FIG. 1, to generate a STB prediction 326. The machine learning model may comprise, in some aspects, a classification model, for example, a linear regression, a random forest, a Gaussian Process, a SVM, a ANN, and the like.

[0116] In one aspect, the machine learning model comprises a GMM. A GMM is a probabilistic model used to represent a distribution of data as a combination of multiple Gaussian distributions, each corresponding to a different cluster or subgroup within the dataset, also called a component. A GMM may be useful for modeling complex data distributions where the data points may belong to different underlying distributions.

[0117] In one aspect, the machine learning model comprises a balanced random forest (BRF). A BRF is a version of a random forest, which is a tree-based machine learning model and a BRF is especially useful for class imbalanced problems. A BRF uses an ensemble of decision trees to generate an aggregate prediction to provide robust and generalized predictions. Overall, a BRF may have improved accuracy and recall for the minority class. The machine learning model may be trained as described with respect to workflow 500 in FIG. 5.

[0118] In some aspects, additional data may be provided to the machine learning model to generate the STB prediction. The additional data may include biographical data associated with user. In some aspects, biographical data may be obtained from a database, for example, a health record such as an electronic medical record (EMR) system. Biographical data may include age data, income data, marital status data, employment data, ethnicity data, education level data, sex data, gender data or handedness data associated with a user. In some aspects, biographical data may be generated based on a biographical information task.

[0119] In some aspects, the STB prediction 326 may include prediction of one or more2024-068 / 10738-1206 measures of STB. In some aspects, measures of STB may correspond to severity of ideation. A first measure of STB may be a passive ideation measure. A passive ideation measure may correspond to either wishing to be dead or wishing to go to sleep and not wake up.

[0120] A second measure of STB may be an active ideation measure. An active ideation measure may correspond to wanting to hurt themselves or taking their own life.

[0121] A third measure of STB may be a suicide planning measure. A suicide planning measure may correspond to planning for suicide, including thoughts about how to accomplish hurting themselves or taking their own life, thoughts about doing something to make themselves not alive anymore, and active planning steps, such as started to do anything or prepared to do anything to end their life.

[0122] A fourth measure of STB may be a planning for safety measure. A planning for safety measure may correspond to a safety plan to prevent or reduce hurting themselves when these feelings arise. A safety plan is a short-term intervention to help individuals survive suicidal crises by developing a set of steps to reduce the likelihood of engaging in suicidal behavior. In some cases, a safety plan may include identification of warning signs and triggers that indicate suicidal ideation is likely to occur, internal coping strategies for when the signs and triggers occur, social contacts and locations to provide distraction, supportive contacts who can provide assistance, emergency resources, and steps to ensure safety of the environment to minimize the subject’s ability to act on suicidal thoughts or urges.

[0123] In some aspects, at step 306, the set of STB parameters 324 may be processed by a set of machine learning models, wherein each machine learning model may be trained to predict a measure of STB. For example, a first machine learning model trained to predict a first measure of STB, a second machine learning model trained to predict a second measure of STB, a third machine learning model trained to predict a third measure of STB, and a fourth machine learning model trained to predict a fourth measure of STB.

[0124] At block 308, the STB prediction 326 may be used to generate a treatment for subject 302 based on the STB prediction. The STB prediction 326 may indicate a type of treatment for STB of the subject 302. For example, in certain aspects, a treatment method comprises a medication treatment. In some aspects, medication may be determined based on the measure of STB generated at step 306. For example, medications may include one or more of selective2024-068 / 10738-1206 serotonin reuptake inhibitors (SSRIs), such as citalopram, escitalopram, fluoxetine, paroxetine, sertraline, or vilazodone. In another example, medications may include one or more of serotonin- norepinephrine reuptake inhibitors (SNRIs) such as duloxetine, venlafaxine, desvelafaxine, levomilnacipran. In another example, medications may include monoamine oxidase inhibitors (MAOIs) such as tranylcypromine, phenelzine, isocarboxazid, or selegiline. In further examples, medications may include atypical antidepressants such as bupropion, mirtazapine, nefazodone, trazodone, or vortioxetine; tricyclic antidepressants such as imipramine, doxepin, trimipramine, desipramine, or protripyline; mood stabilizers, such as lithium, carbamazepine, lamotrigine, or valproate, antipsychotics such as quetiapine, aripiprazole; anti-anxiety medications such as alonazepam, alprazolam, lorazepam, or diazepam, or stimulant medications such as methylphenidate, amphetamines, or lisdexamfetamine. In some aspects, a treatment method may comprise a therapy, for example, cognitive-behavior therapy, interpersonal therapy, electroconvulsive therapy, transcranial magnetic stimulation, and the like.

[0125] In some aspects, the STB prediction 326 may be used to determine one or more behaviors associated with STB of the subject 302, such as occurrence of one or more measures of STB, such as passive ideation, active ideation, suicidal planning, and / or safety planning.

[0126] In some aspects, the STB prediction 326 may be used to generate a message for the subject 302 regarding suicide. The STB prediction 326 may indicate a message, a type of message, content of a message, or a medium for messaging for the subject 302 regarding suicide. For example, messaging may include for example, a message comprising educational materials, lifestyle changes and / or techniques, stress management techniques, support groups, mindfulness and relaxation techniques, therapy recommendations, and the like. In some aspects, the messaging may be determined based on the measure of STB of the subject 302, a type of treatment, or a predicted behaviors of the subject 302. In some aspects, the message, including type, content, and / or medium for the message, may depend on a forum, medium, or setting task completion at step 304, for example, a doctor’s office, a clinic, a web survey, a social media link or message, a public health message, and the like.

[0127] Workflow 300 provides many benefits, including, for example, for generating a model for prediction of STB, including diagnosis, treatment, behavior, and / or messaging. STB prediction has many beneficial uses, as described herein, for example, improved diagnosis (including prior to self-harm) and ready treatment. Further, prediction of behavior may improve adherence to2024-068 / 10738-1206 treatment or predict adherence (e.g., medication or therapy adherence), as well as propensity for lifestyle changes. Additionally, messaging regarding all of the above may be generated to provide education, resources, and recommendations to subjects in need thereof.

[0128] Note that FIG. 3 is just one example of a workflow, and other workflows including fewer, additional, or alternative steps are possible consistent with this disclosure. Example Workflow for Quantifying a Subject’s Judgment

[0129] FIG.4 depicts an example workflow 400 for quantifying a subject’s judgment through a rating task, such as to generate a set of reward-aversion judgment variables, part of the set of STB parameters generated at step 302 of FIG. 3.

[0130] At block 404, a set of stimuli 420 is presented to user 402, for example, via a user interface 110 of computing device 108 in FIG.1. In some aspects, a digital or computer processor, or analog processor may be used. The set of stimuli 420 may comprise one or more stimuli, for example, one or more pictures, one or more sounds, one or more videos, or combinations thereof. For example, picture 802 in FIG. 8A depicts an example picture from a picture rating task. In some aspects, the set of stimuli 420 may comprise one or more subsets of stimuli. The subsets may be organized by a category, for example, all stimuli in a subset are associated with the category. A category may be based on content of the stimuli, effect on the recipient (e.g., user 402), or type of stimuli (e.g., pictures, sounds, video).

[0131] In an example, a set of stimuli 420 may comprise a set of pictures. In this example, the set of pictures may comprise six subsets, wherein each subset is associated with a category. Example categories include, sports, disasters, cute animals, aggressive animals, nature, and / or food. Each subset may comprise, for example, about 8 pictures, about 6 pictures, less than 8 pictures, less than 6 pictures, greater than 8 pictures, greater than 6 pictures, etc. In some aspects, the set of pictures may be presented in an ordered sequence or in an unordered sequence.

[0132] Each stimulus in the set of stimuli 420 may be presented individually, for example, one picture or video on a user interface, or one sound played by the computing device. One or more stimuli in the set of stimuli 420 may be presented together, for example, multiple pictures on a user interface, a picture presented on the user interface and a sound played by the computing device simultaneous, and the like.2024-068 / 10738-1206

[0133] At block 406, a set of ratings 422 is receive from user 402, for example, through the user interface. In some aspects, a rating may comprise a continuous scale. For example, a continuous scale may be a numerical rating, such as from 0 to 5, from -3 to +3, etc. As another example, a continuous scale may be based on a sliding scale, such as a rating between “dislike very much” to “like very much” endpoints. In some aspects, a rating may comprise a discrete scale (e.g., an integer scale), or one or more categories. For example, a user may select a rating from a set of rating options, such as one or more emoticons (e.g., smiley face, sad face, angry face, etc.), one or more terms (e.g., “dislike very much”, “dislike”, “neutral”, “like”, “like very much”, etc.), one or more emotions (e.g., “happy”, “sad”, “angry”, etc.). For example, rating scale 804 in FIG. 8A depicts an example discrete rating scale.

[0134] A rating may be based on the user’s preferences, emotions, and / or attention, for example, a user may rate a stimulus based on an initial response.

[0135] Each rating in the set of ratings 422 may correspond to one rating in the set of stimuli 420. In certain aspects, user 402 may be presented with one stimuli in the set of stimuli 420 (e.g., at block 404), and user 402 inputs one rating associated with the presented stimuli. Once the rating is received (e.g., at block 406), an additional stimuli in the set of stimuli 420 is presented to user 402 and user 402 inputs a rating associated with the presented additional stimuli. The presentation of stimuli and input of ratings may continue until each stimuli in the set of stimuli 420 has been presented and a rating associated with each stimuli has been received.

[0136] In some aspects, blocks 404-406 may together be referred to as a “rating task”.

[0137] At block 408, a set of judgment variables 424 is determined based on the set of ratings 422. In some aspects, the set of judgment variables may be determined based on a Relative Preference Theory (RPT) framework. The RPT framework may quantify and capture judgments about the valence of judgment, for example, positive vs. negative, approach vs. avoidance, etc., as well as magnitude of judgment (e.g., intensity of rating) to describe a user’s preferences. As described herein, , such judgment variables may be used to predict risk, including risk behavior of a user, for example, a STB prediction 326 in FIG. 3.

[0138] The set of judgment variables 424 may include, for example, loss aversion, risk aversion, loss resilience, ante, insurance, peak positive risk, peak negative risk, reward tipping point, aversion tipping point, total reward risk, total aversion risk, reward aversion tradeoff,2024-068 / 10738-1206 tradeoff range, reward aversion consistency, and consistency range. For example, Table 1 depicts an example judgment variables and associated abbreviations. The set of judgment variables 424 may be determined based on approach / avoidance variables in the pattern of ratings in the set of ratings 422.

[0139] These approach / avoidance variables included the mean magnitude (K), variance (e.g., standard deviation) (σ), and the uncertainty of the pattern of ratings (e.g., Shannon entropy (H)) related to a user’s preference behavior. K reflects the average (mean) of positive ratings a subject made (^^) or negative ratings (^^) within each stimulus category. Variance in positive ratings (^^) and variance in negative ratings (^^) within each stimulus category may be determined. The Shannon entropy of positive ratings (^^) or negative ratings (^^) for stimuli within each category may also be determined. The Shannon entropy may characterize uncertainty across the set of ratings, for example, by quantifying the pattern of judgments in the set of ratings. It may also be considered a memory variable.

[0140] In some aspects, one or more subsets of ratings may be identified from the sets of ratings based on the approach / avoidance variables. For example, a subset of ratings may be identified as a category of ratings for a given user where K=0, for example where the user rated all stimuli in the category as neural. The Shannon entropy H, cannot be computed where K=0 ^ because, the H computation results in evaluating log10( ^), which is undefined which is undefined. In these cases, the Shannon entropy was set to H=0 for categories in which the user rated “0” for all the stimuli.

[0141] At block 410, one or more graphs of the approach / avoidance variables may be generated. In some aspects, one or more of the quantified variables may be plotted against one or more of the other. For example, (K, H), (K,^), andgraphs may be generated. (^^, ^^)and ( ^^ , ^^ ) are plotted separately from ( ^^ ,and ( ^^ , ^^ ) thereby calibratingapproach / avoidance, K, to the pattern of prior judgments, H, and their variance, ^. A cure may be fit to each graph. In some examples, features may be generated for example, ratios between the variables, intersections with graph axes, parameters fit, or other features derivable from the graph(s). For example, FIGS. 8B-D depict various example graphs.

[0142] One example graph, graph 806 in FIG. 8B, the (K, H) curve may comprise (^^, ^^) on the positive x-axis and (^^, ^^) on the negative x-axis. One or more of the set of judgment2024-068 / 10738-1206 variables 424 may be determined based on the (K, H) curve, including Risk Aversion, Loss Resilience, Loss Aversion, Ante, and Insurance. Loss aversion, risk aversion, loss resilience, ante, and insurance are derived from the logarithmic or power-law fit of mean keypresses (K) versus the entropy of keypresses (H); this is referred to as the value function, for example depicted as graph 808 in FIG. 8B.

[0143] Loss Aversion (LA) may be the absolute value of the ratio of the linear regression slopeof ^log^^ , log ^^^ to the linear regression slope of ^log^^ , log ^^^. LA may measure the degreeto which an individual person overweighs negative stimuli to positive stimuli, for example due to a cognitive bias.

[0144] Risk Aversion (RA) may be determined as the ratio of the second derivative of the^^^,^^^curve to its first derivative, which also produces a curve. RA may measure the degree to which an individual prefers an uncertain high value outcome to something certain, but lower value outcome.

[0145] Loss Resilience (LR) may be the absolute value of the ratio of the second derivative of the (^^, ^^) curve to its first derivative, which also produces a curve. LR may be the degree to which an individual prefers to lose a small, defined amount in comparison to losing a greater amount with more uncertainty associated with this loss.

[0146] Ante may be the value of ^^ when setting ^^ = 0. This intuitively measures the anteone needs to engage in a game of chance and models the amount of a bid an individual is willing to make to enter a game of chance (e.g., poker).

[0147] Insurance may be the value of ^^when setting ^^= 0. Insurance may measure how much security an individual is willing to acquire to avoid negative outcomes. Insurance mirrors the ante, but in the framework of potential losses.

[0148] Another example graph, graph 810 in FIG. 8C, the (K, ^) curve comprises (^^, ^^) on the positive x-axis and (^^, ^^) on the negative x-axis. Preference magnitude (e.g., K) may be compared with respect to variance in rewards and sanctions. For example, the (K, ^) curve may model the following question: “Would an individual prefer a dollar with probability one, or value drawn from a normal distribution with mean of two and variance of two?” One or more of the set of judgment variables may be determined based on the (K, ^) curve including Peak Positive Risk,2024-068 / 10738-1206 Peak Negative Risk, Reward Tipping Points, Aversion Tipping Point, Total Reward Risk, and Total Aversion Risk. Peak positive risk, peak negative risk, reward tipping point, aversion tipping point, total reward risk, and total aversion risk are derived from the quadratic fit of K versus the standard deviation of keypresses (^); this is referred to as the limit function, for example depicted as graph 812 in FIG. 8C.

[0149] Peak Positive Risk (Peak PR) may be the value of ^^ for the derivative ^^^ = 0. PeakPR may represent the maximum variance for approach behavior. ^^models where increases in positive value transition from a relationship with increases in risk, to a relationship with decreases in risk. The positive apex models when variance changes from weighing against a decision to facilitating a decision.

[0150] Peak Negative Risk (Peak NR) may be the value of ^^where the derivative ^^^= 0. Peak NR may represent the maximum variance for avoidance behavior. Like with the Peak PR, this transition point may relate to avoidance decisions.

[0151] Reward Tipping Point (Reward TP) may be the value of ^^ when the derivative ^^^= 0. Reward TP represents the rating intensity with maximum variance for approach behavior, potentially when an individual decides to approach a goal-object.

[0152] Aversion Tipping Point (Aversion TP) may be the value of ^^where the derivative ^^^= 0. Aversion TP represents the rating intensity with maximum variance for approach behavior, potentially when an individual decides to avoid a goal-object.

[0153] Total Reward Risk (Total RR) may be the area under the curve (AUC) of the first quadrant of the^^^,^^^curve. Total RR represents the relationship between ^^and ^^and may be a quantity that measures the amount of value an individual associates to positive stimuli.

[0154] Total Aversion Risk (Total AR) may be the area under the curve on the negative quadrant of the graph ofTotal AR represents the relationship between ^^and ^^and may be a quantity that measures the amount of overall value a person associates to a negative stimulus.2024-068 / 10738-1206

[0155] As another example graph, graph 814 in FIG. 8D, the (^^, ^^) curve may compare patterns in approach and avoidance judgments. One or more of the set of judgment variables may be determined based on the curve (^^, ^^) including Reward Aversion Tradeoff, Tradeoff Range, Reward Aversion Consistency, and Consistency Range. Risk aversion tradeoff, tradeoff range, risk aversion consistency, and consistency range are derived from the radial fit of the pattern of avoidance judgments (H-) versus the pattern of approach judgments (H+); this is referred to as the tradeoff function for example depicted as graph 816 in FIG. 8D.

[0156] Reward Aversion Tradeoff (RA Tradeoff) may be the mean of the polar angles of the points in the (^^, ^^) plane. RA Tradeoff may be the mean ratio of entropies or patterns in approach to avoidance behavior.

[0157] Tradeoff Range may be the standard deviation of the polar angles of the points in the (^^, ^^) plane. Tradeoff Range may measure the standard deviation in the patterns of approach to avoidance behavior. This variance represents the spread for positive preferences and negative preferences across a set of potential goal-objects and is one measure of the breadth of an individual’s (or group’s) preferences.

[0158] Reward Aversion Consistency (RA Consistency) may measures the mean of the distances of the data points in the (^^, ^^) curve to the origin. RA Consistency defines how individuals can have strong preferences (e.g., biases) for the same thing, reflecting conflict, or having low preferences for something, reflecting indifference. The consistency or compatibility of approach and avoidance may be quantified, for example, how humans can both like and dislike something, or be indifferent to both its positive and negative features.

[0159] Consistency Range may measures the standard deviation of the distances of the data points in the (^^, ^^) plane to the origin. Consistency Range measures how the points in the HH plane vary with regard to the radial distance from the origin. The variance in this radial distance will reflect how much an individual goes between having conflicting preferences and having indifferent ones.

[0160] Then, the set of judgment variables 424 may be outputted. In some examples, the set of judgment variables may be used by a machine learning model to generate a STB prediction,2024-068 / 10738-1206 such as described with respect to FIG.3. In some examples, the set of judgment variables may be used to train a machine learning model to generate a STB prediction, such as described with respect to FIG. 5.

[0161] Mutual information scoring of RPT features with respect to (a) passive suicidal ideation, (b) active suicidal ideation, (c) planning for suicide and (d) having a plan for safety are depicted in FIG. 9. The mutual information between two variables informally expresses the amount of information gained about one variable by observation of another. In this context, this relates to the amount of information gained about STB variables by knowledge of the judgment variables. The length of the bars in the figures represents the mutual information (x-axis) of the RPT variables (y- axis). Longer bars are indicate a higher mutual information between RPT variables and STB. This alludes to a larger predictive value. A mutual information score of zero implies that the two variables are independent, and therefore that prediction of one variable based on another is unlikely.

[0162] In some aspects, one or more statistical mechanisms may be used to determine preferences, based on the set of judgment variables, for example, mediation or hold-out analyses or moderation analysis.

[0163] Beneficially, the rating task is inconspicuous compared to conventional surveys and may elicit more honest answers. For example, the rating task limits biases involved in STB and / or mental health. The rating task involves presenting sets of stimuli, such as pictures, to users and receiving user ratings. The rating task, including the stimuli, does not have a perceivable relation to STB. Thus, follow on predictions, such as a STB prediction, does not incorporate such bias. Furthermore, the rating task may be readily deployed to personal computing devices of users, for example, smart devices or computers, enabling convenient completion and unobtrusive patient assessment and monitoring.

[0164] Note that FIG. 4 is just one example of a workflow, and other workflows including fewer, additional, or alternative steps are possible consistent with this disclosure. Example Workflow for Training a Machine Learning Model

[0165] FIG. 5 depicts an example workflow 500 for training a classification model for STB prediction, for example, a classification model of machine learning component 118 in FIG. 1.2024-068 / 10738-1206

[0166] Initially, a rating task implementing a RPT framework may be presented to a user to generate a set of judgment variables, for example, as described with respect to workflow 400 in FIG. 4. In particular, at block 502, as described herein, stimuli may be presented to a user, such as described with respect to block 404 of workflow 400, for example via a computing device. The set of stimuli may comprise one or more stimuli, for example, one or more pictures, one or more sounds, one or more videos, or combinations thereof. In some aspects, the set of stimuli may comprise one or more subsets of stimuli. The subsets may be organized by a category, for example, all stimuli in a subset are associated with the category. A category may be based on content of the stimuli, an intended effect on the recipient, or type of stimuli (e.g., pictures, sounds, and video).

[0167] Then, at block 504, a set of ratings may be received from a user in response to the set of stimuli, for example, as described with respect to block 406 of workflow 400. Each rating in the set of ratings may correspond to a stimulus in the set of stimuli, for example, one rating for one picture.

[0168] Then, at block 506, a set of judgment variables may be determined based on the set of ratings for each stimuli in the set of stimuli. The set of judgment variables may include, for example, loss aversion, risk aversion, loss resilience, ante, insurance, peak positive risk, peak negative risk, reward tipping point, aversion tipping point, total reward risk, total aversion risk, reward aversion tradeoff, tradeoff range, reward aversion consistency, and consistency range. The set of judgment variables may be determined based on a set of approach / avoidance variables determined based on the sets of ratings.

[0169] As described, these approach / avoidance variables included the mean magnitude (K), variance (e.g., standard deviation) (σ), and the uncertainty of the pattern of ratings (e.g., Shannon entropy (H)) related to a user’s preference behavior, for example, as described with respect to block 408 in FIG. 4.

[0170] In some aspects, one or more subsets of ratings may be identified from the sets of ratings based on the approach / avoidance variables. For example, a subset of ratings may be identified as a category of ratings for a given user where K=0, for example where the user rated all stimuli in the category as neural. The Shannon entropy H, cannot be computed where K=02024-068 / 10738-1206 ^ because, the H computation results in evaluating log10( ^), which is undefined which is undefined. In these cases, the Shannon entropy was set to H=0 for categories in which the user rated “0” for all the stimuli.

[0171] At block 508, a mental health assessment may be received for each user. The mental health assessment may correspond to the STB measures, for example, based on the Massachusetts General Hospital Subjective Question screener (MGH SQ). Four questions corresponding to the four STB measures may be used to question users based on the last month.

[0172] Workflow 500 may repeat blocks 504-508 for many users.

[0173] At block 510, training data may be generated based on sets of judgment variables determined at block 506 and the mental health assessment received from users at block 508.

[0174] In some aspects, training data may include additional data associated with training users, such as biographical data. Biographical data may include, for example, age data, income data, marital status data, employment data, ethnicity data, education level data, or sex data associated with a user.

[0175] At block 512, the training data may be used to train a machine learning model to generate a diagnosis.

[0176] The machine learning model may be trained in a supervised manner. Supervised learning may be used to generate a model of relationships between one or more input features (e.g., a feature vector) and a target output, in this example, a risk prediction. Training data is labeled including input data and the desired output. Models may be trained with supervised learning to perform tasks including classification and regression. A classification task is a task to predict discrete values. A regression task is to predict continuous values. Example models trained through supervised learning include a linear regression, a random forest, a Gaussian Process, a GMM, a SVM, or an ANN.

[0177] In some aspects, one or more levels of PHQ-8 may be merged together, for example, to generate a zero class comprising mildly depressed users and a one class comprising severely depressed. For example, in some aspects, a zero class comprises users with a level 1 score and a one class comprises users with a level 2, 3, 4, or 5 score. In some aspects, a zero class comprises2024-068 / 10738-1206 users with a level 1 or 2 score, and a one class comprises users with a level 3, 4, or 5 score. In some aspects, a zero class comprises users with a level 1, 2, or 3 score, and a one class comprises users with a level 4 or 5 score.

[0178] In one aspect, the machine learning model comprises a GMM. A GMM is a probabilistic model used to represent a distribution of data as a combination of multiple Gaussian distributions, each corresponding to a different cluster or subgroup within the dataset, also called a component. A GMM may be useful for modeling complex data distributions where the data points may belong to different underlying distributions.

[0179] A Gaussian distribution is an approximately normal distribution characterized by its mean and covariance. The shape and orientation of the Gaussian distribution may vary depending on the data itself. Each Gaussian distribution has an associated weight indicating the proportion of total data that is represented by that distribution. Together, all the weights of each Gaussian distribution sums to 1. The overall model is expressed as a weighted sum of the individual Gaussian distributions, thus, enabling capture of various subgroups of within the dataset, including various shapes and clusters.

[0180] The probability density function of a Gaussian Mixture Model may be expressed by Equation 1:Where ^ is the number of Gaussian components; ^^ is the weight of the k-th Gaussiancomponent; and ^(^ ∣ ^^ ,^^) is the Gaussian density function with mean ^^ and covariance ^^.

[0181] The number of clusters (e.g., distribution or component) may be determined using a Bayesian Information Criterion (BIC) selection. The gradient of the BIC may be plotted and the lowest number of clusters before the gradient stabilizes is the optimal number of clusters. For example, the number of clusters may be set from between 2 to 20 clusters for each Gaussian Mixture Model.

[0182] In some aspects, a separate GMM may be trained for each measure of STB (e.g., STB1, STB2, STB3, and STB4).2024-068 / 10738-1206

[0183] FIG. 11 depicts an example schema 1100 for using a GMM classifier, comprising a STB positive model GMM+ 1120B and a STB negative model GMM—1120A, together, the GMM classifier 1120. Gaussian mixture models correspond to a mixture of Gaussian distributions, each being drawn from with a respective probability. Two Gaussian mixture models, STB positive model GMM+ 1120B and a STB negative model GMM—1120A, were fit to the training data 1104A, where the training data was randomly selected to contain 77% of the overall data 1102. A first GMM 1120B was fit to the feature set corresponding to the STB-positive class 1106B and a second GMM 1120A was fit to the feature set corresponding to the STB-negative class 1106A. The number of Gaussian mixtures in each of the two GMMs were chosen by evaluating the number of mixtures that minimized the Bayesian Information Criteria (BIC) score, with the constraint that the number of mixtures be between one and ten. After completing this process, the performance was evaluated on the testing data 1104B. Given a sample from the test set, the probability of that sample of belonging to each GMM, STB positive model GMM+ 1120B and a STB negative model GMM—1120A , a negative class 1116A or positive class 1116B, was evaluated using the probability density function of each GMM. The sample was subsequently classified as belonging to the class of the GMM which produced the greater probability, either a negative classification 1118A or a positive classification 1118B. The accuracy, sensitivity, specificity and balanced accuracy were calculated for the testing set. The entire procedure was then repeated ten times, which each iteration producing a randomized 77 / 23 train / test split of the data. The average of the accuracy, sensitivity, specificity and balanced accuracy over the ten iterations was subsequently reported.

[0184] Once the GMM classifier 1120 is trained, input data may be used to inference with GMM classifier 1120.

[0185] In one aspect, the machine learning model comprises a balanced random forest (BRF). A BRF is an ensemble of decision trees, similar to a random forest, but uses balanced bootstrapping sampling methods to handle imbalanced classes. A balanced bootstrapped sample is drawn from the training data set such that the sample includes an equal number of instances from each class to generate a balanced ratio. A decision tree is generated for each sample with a focus on maintaining the class balance. Each decision tree splits nodes based on criteria, for example, Gini impurity, and is grown to a specified depth or until stopping criteria are met. Further, a BRF may assign different weights to classes during the tree building process to assign a high penalty to misclassifications of the minority class.2024-068 / 10738-1206

[0186] Then, during inferencing, an input is processed by the ensemble of trees and predictions are aggregated to make a final classification. For example, majority voting may be used where the most common class prediction among the trees is selected as the final classification.

[0187] FIG. 10 depicts an example schema 1000 for using a BRF classifier 1020. After performing recursive feature elimination (RFE) with 3-fold cross validation (RFECV), the best feature subsets were identified by Stratified 10-fold-cross validation to maximize accuracy and recall, independently, while also producing a model that was generalizable and explainable. In example schema 1000, ratings 1002 where collected, divided into positive approach 1004A and negative approach 1004B variables, and used to generate graphs 1006. The RPT variables may be extracted from the graphs at block 1008. The RPT variables provides the input data 1010, which may be used as training data 1012 to generate BRF model 1020. Training data 1012 may be dividing into multiple bootstrapped samples 1014, in the depicted schema 3 samples, although 2- 200 samples may be used. For each sample 1014, a decision tree 1016 is constructed using feature bagging. At each node, a random subset of features is selected to determine the best split. Once each tree is gown to maximum depth, the trees may be aggregated to form BRF model 1020.

[0188] Once the BRF model 1020 is trained, input data 1010 may be used to inference with BRF model 1020.

[0189] In some aspects, a model is used for each measure of STB, e.g., STB1, STB2, STB3, and STB4. For example, each measure may be classified with a different weighting of the input parameters, based on distinct aspects of reward-aversion judgment variables have differing importance for each measure.

[0190] Then, beneficially, the trained model may be implemented, such as described with respect to workflow 300 in FIG. 3.

[0191] Workflow 500 provides many benefits, including, for example, for generating a model for prediction of STB, including diagnosis, treatment, behavior, and / or messaging. STB prediction has many beneficial uses, as described herein, for example, improved diagnosis (including prior to extended suffering of symptoms) and ready treatment. Further, prediction of behavior may2024-068 / 10738-1206 improve adherence to treatment or predict adherence (e.g., medication or therapy adherence), as well as propensity for lifestyle changes. Additionally, messaging regarding all of the above may be generated to provide education, resources, and recommendations to subjects in need thereof.

[0192] Note that FIG. 5 is just one example of a workflow, and other workflows including fewer, additional, or alternative steps are possible consistent with this disclosure. Example Method for Generating STB Predictions

[0193] FIG. 6 depicts an example method 600 for generating STB predictions, for example, with system 100 in FIG. 1.

[0194] Method 600 begins at block 602 with generating a set of STB parameters associated with a subject, wherein the set of STB parameters comprises a set of reward-aversion judgment variables, and at least one contextual variable.

[0195] In some aspects, the method 600 comprises presenting to the subject, via a user interface, a set of stimuli; receiving, from the subject via the user interface, a set of ratings associated with the set of stimuli, wherein each respective rating in the set of ratings corresponds with a respective stimuli in the set of stimuli; and determining, based on the set of ratings, the set of reward-aversion judgment variables.

[0196] In some aspects, the set of stimuli comprises one or more subsets of stimuli, and each subset of the one or more subset of stimuli comprises one or more stimuli associated with a category.

[0197] In some aspects, determining the set of reward-aversion judgment variables based on the set of ratings comprises: determining one or more subsets of ratings of the set of ratings, wherein each one of the one or more subsets of ratings corresponds to the respective stimuli in the one or more subsets of the set of stimuli; determining an average of each respective subset of the one or more subsets of ratings; determining a variance of each respective subset of the one or more subsets of ratings; and determining an uncertainty or entropy of each respective subset of the one or more subsets of ratings.

[0198] In some aspects, determining one or more reward-aversion judgment variables in the set of reward-aversion judgment variables based on the relative preference graph.2024-068 / 10738-1206

[0199] In some aspects, the set of reward-aversion judgment variables comprises one or more of a loss aversion variable, a risk aversion variable, a loss resilience variable, an ante variable, an insurance variable, a peak positive risk variable, a peak negative risk variable, a total reward variable, a total aversion risk variable, or a tradeoff range variable.

[0200] In some aspects, the method further comprises presenting, to a subject via the user interface, at least one contextual question; and receiving, from the subject via the user interface, the at least one contextual variable.

[0201] In some aspects, the set of STB parameters further comprises biographical data associated with the subject.

[0202] In some aspects, the biographical data associated with the subject comprises one or more of age data, income data, marital status data, employment data, ethnicity data, education level data, sex data, gender data or handedness data.

[0203] Method 600 continues at block 604 with processing the set of STB parameters with a machine-learning model to generate a STB prediction.

[0204] In some aspects, wherein the machine-learning model comprises a linear regression, a random forest, a Gaussian Process, a Gaussian Mixture Model, a Support Vector Machine, or an artificial neural network.

[0205] In some aspects, the machine-learning model comprises a balanced random forest.

[0206] In some aspects, the at least one contextual variable comprises one or more of a loneliness variable, a prior suicidal attempts variable, a health variable, or an anxiety variable.

[0207] In some aspects, the STB prediction comprises one or more measure of STB, the one or more measures of STB comprising a passive ideation measure, an active ideation measure, a suicide planning measure, or a planning for safety measure.

[0208] Method 600 continues at block 606 with treating the subject with an STB treatment based on the STB prediction.

[0209] In some aspects, wherein the STB treatment comprises one or more of an selective2024-068 / 10738-1206 serotonin reuptake inhibitor, a serotonin-norepinephrine reuptake inhibitor, an atypical antidepressant, a tricyclic antidepressant, a monoamine oxidase inhibitor, mood stabilizers, antipsychotics, anti-anxiety medications, or stimulant medication.

[0210] Note that FIG. 6 is just one example of a method, and other methods including fewer, additional, or alternative steps are possible consistent with this disclosure. Example Method for Training a Model to Generate STB Predictions

[0211] FIG.7 depicts an example method 700 for generating and training a machine-learning model to generate STB predictions, for example, with system 100 in FIG. 1.

[0212] Initially, method 700 begins at step 702 with generating training data based on STB parameters.

[0213] In some aspects, the STB parameters may include one or more sets of ratings, comprising: presenting, to one or more subjects via a subject interface, a set of stimuli; receiving, from the one or more subjects via the subject interface, one or more sets of ratings associated with the set of stimuli; and determining a set of judgment variables based on the sets of ratings, such as described with respect to workflow 500 in FIG. 5.

[0214] In some aspects, the set of stimuli comprises one or more subsets of stimuli, and each subset of the one or more subset of stimuli comprises one or more stimuli associated with a category. In some aspects, a stimulus in the set of stimuli comprises one or more of an image, a sound, or a video.

[0215] In some aspects, determining the set of judgment variables based on the sets of ratings comprises: determining one or more subsets of ratings of each set of the sets of ratings, wherein each one of the one or more subsets of ratings corresponds to the respective stimuli in the one or more subsets of the set of stimuli; determining an average of each respective subset of the one or more subsets of ratings; determining a variance of each respective subset of the one or more subsets of ratings; and determining an uncertainty or entropy of each respective subset of the one or more subsets of ratings.

[0216] In some aspects, method 700 further comprises generating a relative preference graph2024-068 / 10738-1206 based on the average of each respective subset of the one or more subsets of ratings, the variance of each respective subset of the one or more subsets of ratings, and the uncertainty or entropy of each respective subset of the one or more subsets of ratings.

[0217] In some aspects, the set of judgment variables comprises one or more of a loss aversion variable, a risk aversion variable, a loss resilience variable, an ante variable, an insurance variable, a peak positive risk variable, a peak negative risk variable, a total reward variable, a total aversion risk variable, or a tradeoff range variable.

[0218] In some aspects, method 700 further comprises presenting, to the one or more subjects via the subject interface, a mental health assessment comprising a set of questions, receiving, from the one or more subjects via the subject interface, a set of responses associated with the set of questions; and adding the set of responses to the training data.

[0219] In some aspects, method 700 further comprises generating a score for each subject of the one or more subjects based on the set of responses.

[0220] In some aspects, the training data further comprises biographical data associated with the one or more subjects. In some aspects, the biographical data associated with the one or more subjects comprises at least one or more of age data, income data, marital status data, employment data, ethnicity data, education level data, or sex data.

[0221] Method 700 proceeds to step 704 with training the STB prediction model with the training data to generate a STB prediction, such as described with respect to FIG. 5.

[0222] Method 700 provides many benefits, including, for example, for generating a model for prediction of STB, including diagnosis, treatment, behavior, and / or messaging. STB prediction has many beneficial uses, as described herein, for example, improved diagnosis (including prior to extended suffering of symptoms) and ready treatment. Further, prediction of behavior may improve adherence to treatment or predict adherence (e.g., medication or therapy adherence), as well as propensity for lifestyle changes. Additionally, messaging regarding all of the above may be generated to provide education, resources, and recommendations to subjects in need thereof.

[0223] Note that FIG. 7 is just one example of a method, and other methods including fewer, additional, or alternative steps are possible consistent with this disclosure.2024-068 / 10738-1206 EXAMPLES

[0224] Example 1: Use of Rating Task for Machine Learning Prediction of the Incidence of Suicidal Thought and Behavior

[0225] Adults (ages 18-70) across the United States were surveyed in December 2021. High quality data from 3476 participants included: (1) PHQ8 (absent the question on suicidality), (2) State Trait Anxiety Inventory – State (STAI), (3) perceived loneliness (self-report), (4) prior attempts at self-harm in past 1-12 months, (5) five demographic variables known to affect human neuroscience studies (i.e., age, ethnicity, education level, sex, and handedness) (6) 15 judgment variables computationally derived from a simple picture rating task (Table 1), and (7) four questions about passive ideation (STB1), active ideation (STB2), suicide planning (STB3), planning for safety (STB4) on a five-point Likert-scale (collectively referred to as STB variables). The predictive power of variables in (1) – (6) was tested using balanced random forest (BRF) and Gaussian mixture models (GMM) to discriminate between low and high thresholds of the four STB variables. To provide a baseline against these analyses, the inventors also performed the following four standard machine learning analyses: Random Forest (RF), Logistic Regression (LR), Neural Network (NN) and Support Vector Machine (SVM). Given potential personal reluctances or cultural norms against reporting past self-harm, variables from (1) – (3), (5) and (6) were initially tested, followed by a minimal predictor set of (4) – (6). The full set of (1) – (6) was further tested. The relative importance of features used in prediction was evaluated using mutual information (MI) scoring (where higher MI of a feature and predictor suggest predictive power) and Gini score plots. Mechanistic relationships between the top predictors were assessed using statistical mediation and moderation where the four STB variables were dependent variables. 1. Prediction of STB Variables

[0226] Given ML results for judgment variables, PHQ8 score, STAI score, and loneliness, compared with results for judgment variables, prior attempts, and loneliness, and with results with inclusion of all predictors (judgment variables, PHQ8 score, STAI score, loneliness and prior attempts) are described. (a) Passive Suicidal Ideation (STB1).2024-068 / 10738-1206

[0227] BRF prediction of STB1 (rated on a Likert scale of 1-5, where 1 = no suicidal ideation and 2-5 = increasing degrees of suicidal ideation; i.e., threshold=1) using judgment variables yielded higher accuracies with PHQ8, STAI, and loneliness variables included (59.0%-78.8%). Sensitivities and specificities improved from 51.4% to 83.3% and from 61.0% to 77.7%, respectively. Adding demographics improved these metrics less than two percent. Fusion of the PHQ8 score with judgment variables had a consistent boost of about 18% for accuracy and 32% for sensitivity. Results with inclusion of judgment, PHQ8, STAI, and loneliness features were similar (61.9%-78.5%; Table 4) when the threshold for passive suicidal ideation was set to 2.

[0228] When judgment variables were fused with reports of prior suicide attempts and loneliness, predictive accuracy of STB1 at threshold=1 was 78.4% and 78.1%, respectively. Sensitivity improved from 52.0% to 74.2% when loneliness was fused with judgment variables and prior attempts, whereas specificity showed a decrease from 85.4% to 79.1%. Prediction of STB1 at threshold=2 when judgment variables were fused with reports of prior suicide attempts and loneliness was 83.4% and 80.4%, respectively, with similar sensitivities and specificities to threshold=1.

[0229] Analysis with all predictors achieved maximum AUC scores of 0.905 for STB1 threshold=1, and 0.907 when threshold=2, achieving sensitivities of 84.8% and 85.1% respectively. (b) Active Suicidal Ideation (STB2).

[0230] BRF prediction of STB2 (threshold=1) using judgment variables yielded higher accuracies as PHQ8, STAI, and loneliness variables were successively included (63.8%-78.7%). Sensitivities and specificities improved from 56.0% to 86.1% and 65.1% to 77.5%, respectively. Further adding demographics improved these metrics by less than one percent. Fusion of the PHQ8 score with judgment variables boosted measures by 12% for accuracy and 28% for sensitivity. For threshold=2, judgment variables yielded higher accuracies as PHQ8, STAI, and loneliness features were successively included (62.7%-78.0%; Table 4), with similar outcomes for sensitivity and specificity.

[0231] Prediction accuracy of STB2 (threshold=1) when judgment variables were fused with reports of prior suicide attempts and loneliness was 86.4% and 84.7%, respectively (Table 4). Sensitivities and specificities were in the high 60-70s and 80-90s, respectively. Prediction of2024-068 / 10738-1206 active suicidal ideation (threshold=2) when judgment variables were fused with reports of prior suicide attempts and loneliness was 88.0% and 86.4%, respectively. Sensitivities and specificities were in the 70s and high 80s, respectively.

[0232] Analysis with all predictor variables achieved maximum AUC scores of 0.935 at threshold=1 and 0.931 for threshold=2, achieving sensitivities in each case of 87.2% and 86.6% respectively. (c) Suicide Planning (STB3).

[0233] BRF prediction of STB3 (threshold=1) using judgment variables yielded higher accuracies as PHQ8, STAI, and loneliness variables were successively fused with them (64.4.8%- 79.4%). Sensitivities and specificities improved from 57.5% to 84.9% and 65.3% to 78.7%, respectively. Further adding demographics improved these metrics by less than two percent. Fusion of the PHQ8 score with judgment variables consistently boosted prediction by approximately 10% for accuracy and 20% for sensitivity. For threshold=2, judgment variables yielded higher accuracies as PHQ8, STAI, and loneliness features were successively fused with them (66.8%-79.7%; Table 4).

[0234] Prediction accuracy of STB3 (threshold=1) when judgment variables were fused with reports of prior suicide attempts and loneliness variables was 92.2% and 90.8%, respectively. Sensitivities and specificities were in the high 70s and low 90s. Prediction of STB3 (threshold=2) when judgment variables were fused with reports of prior suicide attempts and loneliness was 91.4% and 90.5%, respectively. Sensitivities and specificities were in the low 80s and low 90s, respectively.

[0235] Analysis with all predictor variables achieved maximum AUC scores of 0.953 with threshold=1 and 0.948 for threshold=2, achieving sensitivities of 86.4% and 88.7% respectively. (d) Planning for Safety (STB4).

[0236] BRF prediction of STB4 (threshold=1) using judgment variables yielded higher accuracies as PHQ8, STAI, and loneliness features were successively fused with them (59.4%- 73.8%). Further adding demographics improved these metrics by less than two percent. Sensitivities and specificities improved from 55.0% to 74.8% and 60.3% to 73.6%, respectively. Fusion of the PHQ8 score with judgment variables consistently boosted the prediction by2024-068 / 10738-1206 approximately 14% for accuracy and 15% for sensitivity. For threshold=2 judgment variables also yielded higher accuracies as PHQ8, STAI, and loneliness features were fused with them (71.7%- 79.9%; Table 4), with similar outcomes for other metrics.

[0237] Prediction accuracy of STB4 (threshold=1) using judgment variables fused with prior suicide attempts and loneliness variables was 81.0% and 79.2%, respectively (Table 4). Sensitivities and specificities were in the high 50s and mid 80s, respectively. Prediction of planning for safety (threshold=2) when judgment features were fused with prior suicide attempts and loneliness was 77.7% and 61.9%, respectively (Table 4). Sensitivities and specificities were in the 50-60 range and 60-80 range, respectively.

[0238] Analysis with all predictor variables achieved maximum AUC scores of 0.837 at threshold=1 and 0.831 at threshold=2, while maintaining sensitivities of 73.1% and 74.1%, respectively. Table 42024-068 / 10738-12062024-068 / 10738-12062. Variable Contributions to STB Prediction

[0239] Distinct sets of judgment variables contributed to prediction of the four STB measures, as measured through normalized MI scoring (FIG. 9). LA had zero MI for each STB measure. For passive suicidal ideation, no judgment variable predominated by MI value; three had zero- value MI’s. This profile was different for active suicidal ideation where variables for aversion TP and tradeoff range had the highest MI. For suicide planning, MI with the tradeoff range was far more than other judgment variables. For planning for safety, variables for aversion TP and reward TP were predominant. Despite the distinct patterns of MI for the 15 judgment variables between the four STB measures, regressions between these variables had consistent valences (excepting Reward TP and Total AR). On the basis of the valence between each judgment variable and STB measure, passive suicidal ideation and suicide planning shared the same patterns, and both differed from active suicidal ideation and planning for safety.2024-068 / 10738-1206

[0240] Gini score plots revealed that some survey variables were consistently highest in importance, but the full set of judgment variables were consistently grouped together (FIG. 12A- 12B). In all analyses, the grouped judgment variables produced summed Gini scores of 0.404 to 0.638 — the highest summed Gini scores in 14 of the 24 analyses (Table 5). The 15 judgment variables were consistently more important than education, race, gender, and handedness variables.

[0241] The rank ordering of Gini scores for the survey variables was distinct for each STB measure, as it was for the 15 judgment variables. Despite this, STAI measures tended to have one of the top two Gini scores, and Age was consistently one of the bottom of the five survey variables. Table 52024-068 / 10738-1206

[0242] FIGS. 12A and 12B depict Gini importance values of the entire set of features used to predict planning for safety (STB 4) at t=2 using a BRF, e.g., according to schema 1000 of FIG. 10. Specifically, FIG. 12A depicts the Gini importance values, while FIG. 12B depicts a visual representation of the Gini importance values. The numerical values displayed are the mean decrease in Gini coefficient that occurs by removing each variable. The larger the mean decrease in Gini coefficient, the greater importance the feature has in the classifier. Note that two boxes are shown; the highest box highlights the top variables in terms of feature importance; the middle boxes represent the Gini values of the judgment variables.

[0243] Feature importance was assessed in two frameworks, one via a mutual information (MI) assessment and the other via a Gini importance score assessment. For this study, MI was computed between each of the 15 judgment variables and the four prediction outcomes of interest [(passive suicidal ideation (STB1), active suicidal ideation (STB2), planning for suicide (STB3), and planning for safety (STB4)]. The MI was used as a proxy for the dependence between the two random variables, with a larger mutual information between judgment variables and STB metrics2024-068 / 10738-1206 suggesting higher importance of the feature in predicting STB. MI scores for the 15 judgment variables are expected to add up to 1 for each of the STB measures tested. Note that the STB1-4 nomenclature is used only for figures and tables. Analyses were conducting using the package sklearn to calculate the mutual information scores in Python.

[0244] Gini importance scores reflected the feature importance (i.e., rank) as determined using the function model.feature_importances_ in Python, and results were reported and were plotted using matplotlib.

[0245] The sensitivity of a binary classifier measures its ability to accurately label a symptom- positive class. The sensitivity plays an important role in STB prediction due to the relevance of correctly identifying STB positive individuals for potential intervention. However, imbalanced datasets typically result in poor sensitivities when the positive class is significantly less in size compared to the negative class. To enable prediction with high sensitivity, two classifiers with the reported ability to handle data imbalance were chosen, to be compared to a set of standard classifiers. Balanced Random Forest (BRF) and Gaussian Mixture Model (GMM) classifiers were thus chosen to predict the four STB measures, and compared to four standard classifiers (i.e., Random Forest (RF), Logistic Regression (LR), Neural Network (NN) and Support Vector Machine (SVM)). Machine learning analyses used parameters detailed below. Across analyses, code was implemented in Python 3.9 using the packages imblearn 0.0, sklearn 1.2.2, pandas 2.0.2, and pandas 2.0.2. Feature importance (Gini scores and mutual information scores) were obtained with sklearn 1.2.2. Figures relating to variable importance were plotted with seaborn 0.12.2.

[0246] Classifiers used the following features: (1) PHQ8 (absent the question on suicidality), (2) State Trait Anxiety Inventory – State (STAI), (3) perceived loneliness (self-report), (4) prior attempts at self-harm in past 1-12 months, (5) five demographic variables (i.e., age, ethnicity, education level, sex, and handedness) and (6) 15 judgment variables computationally extracted from a simple picture rating task. The predictive power of variables in (1) – (6) was tested using BRF, GMM and the four standard classifiers to discriminate between low and high measures of the four STB variables. Namely, each STB measure was partitioned as binary data in two different ways for their 1-5 Likert ratings: 1 vs. 2-5 (threshold=1), and 1,2 vs. 3-5 (threshold =2). In each case, the binary data was analyzed with both BRF, GMM and four standard classifiers. Given potential cultural norms against reporting past self-harm variables from (1) – (3), (5) and (6) were initially tested, followed by a minimal predictor set of (4) – (6). As a third framework, the full set2024-068 / 10738-1206 of (1) – (6) was also tested. In the analysis of variables from (1) – (3), (5) and (6), judgment variables from (6) were first tested, and then the other variables added in incrementally. The same was done with the other two analysis frameworks (i.e., using just a minimal predictor set for (4) – (6), and using all variables).

[0247] BRF Analysis: BRF (FIG. 10) was implemented in Python with the package imblearn with ten-fold cross validation. Random forest classifiers contain an ensemble of decision trees, from which majority voting is performed to output a class label, and are typically trained by optimizing a Gini or an information score. In the BRF approach, an ensemble of 200 trees was constructed, where each boostrapped sample was randomly under-sampled to create a balanced dataset of both classes (i.e., 50% of STB positive and 50% of STB negative data); balancing was used for training only, and not for testing within cross-validation. No hyperparameter tuning was performed. Subsequent analysis was performed internally using the sklearn Random Forest package through imblearn. Soft labels were used for majority voting, so that the majority vote was weighted on the probability of the sample belonging to the STB positive class. The BRF was trained using the Gini criterion, and no maximum tree depth was used, so that nodes expanded until leaves were pure or contained at most one sample. The mean accuracy, sensitivity, specificity, and AUC were reported.

[0248] Analyses for GMM RF, LR, NN, and SVM classification was conducted with sklearn in Python (FIG. 11). 3. Mediation / Moderation Analysis

[0249] Mediation (Me) assesses the causal pathway between variables and moderation (Mo) assesses the interaction between such variables to predict a third variable. Given the number of associations tested prior to Me / Mo, and the potential for skewed distributions and outliers in human samples, inventors integrated Cook’s distance outlier analysis with Me to protect against false positives and increase the analytic power.

[0250] Mediation models suggest that instead of a direct causal relationship between the independent variable (X) and the dependent variable (Y), there is an intermediary variable (M) so that X influences M, which in turn influences Y. Mediation analyses, similar to moderation analyses, were conducted for all combinations of STB variables as dependent variables, RPT features as moderators, and Prior Attempts, Age, Loneliness, PHQ8-Score and STAI score as independent variables. STB2024-068 / 10738-1206 variables involved no thresholding. Beta coefficients and their standard error(s) terms from the following linear regression equations were used from a regression in order to calculate Sobel p-values and mediation effect percentages: ^^^^ 1∶^= ^^+ ^(^)+^^ ^^^^ 2 :^=^^+ ^(^)+^^ ^^^^ 3 : ^= ^^+ ^′(^)+ ^(^)+^^ Step 4: Sobel’s test was then used to test if ^′ was significantly lower than ^: ^^^^^ ^−^^^^^= ^−^′√^^^^^+ ^^^^^=^^√^^^^^+ ^^^^^.

[0251] Using a standard 2-tail z-score table, the Sobel p-value was determined from the Sobel z-score and the mediation effect percentage (Teff) was calculated using the following equation: ^^^=!""∗[!−^′^].

[0252] For mediation to be considered significant, inventors required that all three regressions between X predicting Y (i.e., pc), X predicting M (i.e., pa), and M predicting Y (i.e., pb') show nominal significance with p < 0.05. Significant mediation further required the following: ^^^^^^ < 0.05 and ^^^ > 50%, following prior publications. Secondary mediation analysis was run by switching variables assigned to X and M to see if the mediation effects were directed. For secondary analysis if ^^^^^^ > 0.05 and #^^^ < 50%, this added to the evidence of M laying in the causal pathway between X and Y.

[0253] Moderation models suggest that a moderator variable (Mo) controls the magnitude of the relationship between the independent variable (X) and the dependent variable (Y). Moderation analyses were conducted for all combinations of STB variables as dependent variables, RPT features as moderators, and Prior Attempts, Age, Loneliness, PHQ8-Score and STAI score as independent variables. No thresholds were considered for the STB variables. The original data for STB variables, involving five categories of severity were used. The moderation analyses involved fitting a logistic regression to the data, described by the equation below: ^^$(^%%^)= &^+&^^+&^^^+&^(^∗^^)+^.

[0254] The standard approach was used, where the moderation was deemed significant if the p- value of the interaction term ( ^^3) and the p-value of the overall model (^^^^^^^^) were both less than2024-068 / 10738-1206 or equal to 0.05 through Likelihood Ratio tests. The Likelihood Ratio test for the full model was implemented with the following null and alternative hypotheses: '^: &^=&^=&^=" '^: &^≠"; (^^ ^)^^^^) ^*^ &^ ;+,^^^ -=!,., / .

[0255] The Likelihood Ratio test for the ^3 coefficient was implemented with the following null and alternative hypothesis respectively: '^: &^=" '0: & / ≠" where the restricted model is shown below: ^^$(^%%^)= &^+&^^+&^^^+^.

[0256] The majority of judgment variables were involved in Me / Mo relationships (1=.05), excluding LA, Total RR, Reward TP for Me, and excluding Total RR, Total AR for Mo. Survey variables statistically mediated the relationship between 11 judgment variables and passive suicidal ideation whereas they statistically moderated the relationship between eight judgment variables and passive suicidal ideation. For the other three STB measures, there was minimal Me involving perceived loneliness, PHQ8, and STAI survey variables. Instead, there were salient Mo effects for these three survey measures with 12 of the 15 judgment variables.

[0257] Of the five survey variables, prior suicide attempts demonstrated Me with four judgment variables to predict STB2-4 and moderated three judgment variables to predict STB1- 2. Age showed Me with six judgment variables, and Mo with one judgment variable.

[0258] This study sought a short, objective, and automatable framework for predicting four STB measures using 15 variables for biases in reward / aversion judgment and a very limited set of demographic and mental health survey indices. Given reward / aversion judgment is known to be affected by demographic and mental health indices, inventors fused demographics with anxiety, depression and judgment variables for ML. This produced five primary results. (1) All four STB metrics were predicted with small sets of predictors within a range of 78%-92% accuracy and 0.796-.953 AUC; this compares favorably with the literature. (2) Judgment variables and limited survey indices were most effective at predicting planning for suicide, producing accuracies in the upper range of what other studies have reported for suicide risk and suicide attempts (e.g., AUC = 0.857 and 0.99, respectively, in the literature and 0.953 herein) without using complex big data2024-068 / 10738-1206 approaches (e.g., 100+ variables or inclusion of neuroimaging) or retrospective data. (3) Prediction of active suicidal ideation and planning for suicide was improved by the addition of one self-report measure of prior attempts of self-harm, similar to the addition of depression and other mental health indices. (4) Mediation / moderation analyses showed that depression, anxiety, loneliness, and age variables had significant moderation effects on judgment variables indicating that the interaction of mental health and contextual indices with judgment variables statistically predicted STB. (5) BRF prediction far outperformed GMM prediction and standard ML prediction (i.e., RF, LR, NN and SVM prediction), particularly for the sensitivity index.

[0259] Collection of the limited set of variables used for prediction was more feasible, and less time consuming, when compared to previous studies using larger datasets with hundreds to thousands of variables for prediction; this task can be easily implemented on any digital device. When the other mental health indices and demographics are included, data acquisition takes approximately 5 minutes to complete but has comparable prediction results to big data approaches. Inclusion of the prior self-harm variable greatly improved the prediction of active suicidal ideation and planning for suicide, but did not improve prediction of passive suicidal ideation or planning for safety. This suggests that prior history might impact intention for harm, but not its inverse – the planning for self-preservation. Of the other mental health indices, the neurovegetative symptoms of depression in the PHQ8 most improved the predictive accuracy of the four STB variables, although by themselves each mental health index and age had higher Gini scores compared to the judgment variables. Loneliness is commonly considered a risk factor for STB, and it was found to moderate the largest number of judgment variables in the prediction of STB variables. The addition of demographic variables was also not consistently beneficial for the prediction of STB, and competition between variables, leading to poorer prediction, cannot be ruled out.

[0260] The sensitivity metric is important for evaluating the prediction efficacy of STB variables, particularly if intervention might be considered. Per a review of studies since 2017 that reported a sensitivity metric for predicting suicidal ideation sensitivities ranged from 41% to 87%, with AUC scores ranging from .61 to .94, in line with our findings. Higher AUC scores and sensitivities were typically provided for peer-reviewed work with large feature sets consisting of neuroimaging data, although none of these publications segregated passive and active ideation, as done commonly in clinical interviews. Furthermore, a number of studies relied on anonymized electronic health records, and did not explicitly report the number of features used, which can vary2024-068 / 10738-1206 per participant. A similar range of sensitivities has been reported for prediction of suicide attempt(s) (30.8% to 100%) with AUC scores ranging from .59 to .99 in cohorts of between 75 to 2,959,689 participants. The prediction of suicidal attempt(s) appear to be more common in the literature than other STB outcomes, but has the caveat of being predominantly retrospective reporting. For studies in the past 5-6 years providing sensitivity results, prediction of suicide risk provided similar sensitivities to suicidal ideation and suicidal attempt(s) (between 59% and 85.3%), and AUC scores of up to .857. Lastly, it must be noted the prediction of completed suicide reports showed much lower sensitivities (between 28% to 69%) and lower AUC scores (between .66 - .8). Relative to this literature, prior studies have not (1) segregated passive from active ideation in prediction of suicidal ideation, (2) explicitly predicted planning for harm, or (3) explicitly predicted planning for safety, all while achieving results in the upper range of what is reported for prediction of suicidal ideation, suicidal attempts and completed suicides.

[0261] The type of ML used for imbalanced data has become a topic of significant research. In this study BRF prediction outperformed GMM prediction and four standard approaches to ML, consistent with the literature. With BRF prediction, each STB variable further had a unique profile of judgment variables that contributed to their prediction, in that the MI metric of variable contribution was unique (FIG. 12A). The only exception was loss aversion (LA), which had no mutual information with any STB variable, and was not involved in any mediation / moderation relationship. The values for LA observed in this study were quite low, consistent with other work using picture ratings where there is no consequence for making a rating unlike an operant keypress that changes view time. Each of the four STB variables was best classified by a unique weighting of the 15 judgment variables, arguing that distinct aspects of reward / aversion judgment are important for each of the four STB metrics.

[0262] The current work found that 15 judgment variables and limited mental health and demographic information, predicted four STB measures with sensitivities and specificities around 80% using a BRF approach that produced the highest sensitivities of the approaches used. There appear to be few studies that integrate quantitative judgment features, as from a short behavioral task, to predict distinct STB measures. This work supports publications suggesting that social and behavioral measures play a key role in prediction of STB, sometimes surpassing clinical variables in predictive accuracies. Contextual risk factors are also not typically studied in machine learning applications of STB, yet predictive variables of suicidality can be contextual and the reported mediation / moderation results strongly support these reports. The current results contrast tendencies in the literature to either2024-068 / 10738-1206 (1) use large feature sets for prediction (e.g., 100s to 1000s of variables) or (2) collect expensive clinical or biological measures for prediction. The data needed for prediction in this study can readily be acquired by smart phones and other digital devices, which are currently available for 92% of the US population, and 85% of the world population. The analysis does not require a supercomputer and thus can be scaled to populations for which big data and expensive clinical or biological measures are not available, meeting frameworks proposed by others for development of a scalable detection platform for prediction of suicidality. By combining multiple variables around STB, including assessments of planning for safety, this framework suggests a digital approach for early assessment and triage, which is particularly needed now. Example Clauses

[0263] Implementation examples are described in the following numbered clauses:

[0264] Clause 1: A method of diagnosis and treatment for suicidal thought and behavior (STB), comprising; generating a set of STB parameters associated with a subject, wherein the set of STB parameters comprises a set of reward-aversion judgment variables, and at least one contextual variable; processing the set of STB parameters with a machine learning model to generate a STB prediction; and treating the subject with an STB treatment based on the STB prediction.

[0265] Clause 2: The method of clause 1, further comprising: presenting to the subject, via a user interface, a set of stimuli; receiving, from the subject via the user interface, a set of ratings associated with the set of stimuli, wherein each respective rating in the set of ratings corresponds with a respective stimuli in the set of stimuli; and determining, based on the set of ratings, the set of reward-aversion judgment variables.

[0266] Clause 3: The method of any one of clauses 1-2, further comprising: presenting, to a subject via the user interface, at least one contextual question; and receiving, from the subject via the user interface, the at least one contextual variable.

[0267] Clause 4: The method of clause 3, wherein the set of stimuli comprises one or more subsets of stimuli, and each subset of the one or more subset of stimuli comprises one or more stimuli associated with a category.2024-068 / 10738-1206

[0268] Clause 5: The method of clause 4, wherein determining the set of reward-aversion judgment variables based on the set of ratings comprises: determining one or more subsets of ratings of the set of ratings, wherein each one of the one or more subsets of ratings corresponds to the respective stimuli in the one or more subsets of the set of stimuli; determining an average of each respective subset of the one or more subsets of ratings; determining a variance of each respective subset of the one or more subsets of ratings; and determining an uncertainty or entropy of each respective subset of the one or more subsets of ratings.

[0269] Clause 6: The method of clause 5, further comprising determining one or more reward- aversion judgment variables in the set of reward-aversion judgment variables based on the relative preference graph.

[0270] Clause 7: The method of clause 6, wherein the set of reward-aversion judgment variables comprises one or more of: a loss aversion variable, a risk aversion variable, a loss resilience variable, an ante variable, an insurance variable, a peak positive risk variable, a peak negative risk variable, a total reward variable, a total aversion risk variable, or a tradeoff range variable.

[0271] Clause 8: The method of any one of clauses 1 - 7, wherein the machine learning model comprises a linear regression, a random forest, a Gaussian Process, a Gaussian Mixture Model, a Support Vector Machine, or an artificial neural network.

[0272] Clause 9: The method of any one of clauses 1 - 7, wherein the machine learning model comprises a balanced random forest.

[0273] Clause 10: The method of any one of clauses 1 – 9, wherein the at least one contextual variable comprises one or more of: a loneliness variable, a prior suicidal attempts variable, a health variable, or an anxiety variable.

[0274] Clause 11: The method of any one of clauses 1 - 9, wherein the set of STB parameters further comprises biographical data associated with the subject.

[0275] Clause 12: The method of clause 10, wherein the biographical data associated with the subject comprises one or more of: age data, income data, marital status data, employment data, ethnicity data, education level data, sex data, gender data or handedness data.2024-068 / 10738-1206

[0276] Clause 13: The method of any one of clauses 1-12, wherein the STB prediction comprises one or more measure of STB, the one or more measures of STB comprising a passive ideation measure, an active ideation measure, a suicide planning measure, or a planning for safety measure.

[0277] Clause 14: The method of any one of clauses 1-13, wherein the STB treatment comprises one or more of an selective serotonin reuptake inhibitor, a serotonin-norepinephrine reuptake inhibitor, an atypical antidepressant, a tricyclic antidepressant, a monoamine oxidase inhibitor, mood stabilizers, antipsychotics, anti-anxiety medications, or stimulant medication.

[0278] Clause 15: A processing system, comprising: a memory comprising computer- executable instructions; and a processor configured to execute the computer-executable instructions and cause the processing system to perform a method in accordance with any one of Clauses 1-14.

[0279] Clause 16: A processing system, comprising means for performing a method in accordance with any one of Clauses 1-14.

[0280] Clause 17: A non-transitory computer-readable medium storing program code for causing a processing system to perform the steps of any one of Clauses 1-14.

[0281] Clause 18: A computer program product embodied on a computer-readable storage medium comprising code for performing a method in accordance with any one of Clauses 1-14. Additional Considerations

[0282] The preceding description is provided to enable any person skilled in the art to practice the various aspects described herein. The examples discussed herein are not limiting of the scope, applicability, or aspects set forth in the claims. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects. For example, changes may be made in the function and arrangement of elements discussed without departing from the scope of the disclosure. Various examples may omit, substitute, or add various procedures or components as appropriate. For instance, the methods described may be performed in an order different from that described, and various steps may be added, omitted, or combined. Also, features described with respect to some examples may be combined in some other examples. For example, an apparatus may be implemented or a method2024-068 / 10738-1206 may be practiced using any number of the aspects set forth herein. In addition, the scope of the disclosure is intended to cover such an apparatus or method that is practiced using other structure, functionality, or structure and functionality in addition to, or other than, the various aspects of the disclosure set forth herein. It should be understood that any aspect of the disclosure disclosed herein may be embodied by one or more elements of a claim.

[0283] As used herein, the word “exemplary” means “serving as an example, instance, or illustration.” Any aspect described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other aspects.

[0284] As used herein, a phrase referring to “at least one of” a list of items refers to any combination of those items, including single members. As an example, “at least one of: a, b, or c” is intended to cover a, b, c, a-b, a-c, b-c, and a-b-c, as well as any combination with multiples of the same element (e.g., a-a, a-a-a, a-a-b, a-a-c, a-b-b, a-c-c, b-b, b-b-b, b-b-c, c-c, and c-c-c or any other ordering of a, b, and c). Reference to an element in the singular is not intended to mean only one unless specifically so stated, but rather “one or more.” For example, reference to an element (e.g., “a processor,” “a memory,” etc.), unless otherwise specifically stated, should be understood to refer to one or more elements (e.g., “one or more processors,” “one or more memories,” etc.). The terms “set” and “group” are intended to include one or more elements, and may be used interchangeably with “one or more.” Where reference is made to one or more elements performing functions (e.g., steps of a method), one element may perform all functions, or more than one element may collectively perform the functions. When more than one element collectively performs the functions, each function need not be performed by each of those elements (e.g., different functions may be performed by different elements) and / or each function need not be performed in whole by only one element (e.g., different elements may perform different sub- functions of a function). Similarly, where reference is made to one or more elements configured to cause another element (e.g., an apparatus) to perform functions, one element may be configured to cause the other element to perform all functions, or more than one element may collectively be configured to cause the other element to perform the functions. Unless specifically stated otherwise, the term “some” refers to one or more.

[0285] As used herein, the term “determining” encompasses a wide variety of actions. For example, “determining” may include calculating, computing, processing, deriving, investigating, looking up (e.g., looking up in a table, a database or another data structure), ascertaining and the2024-068 / 10738-1206 like. Also, “determining” may include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory) and the like. Also, “determining” may include resolving, selecting, choosing, establishing and the like.

[0286] The methods disclosed herein comprise one or more steps or actions for achieving the methods. The method steps and / or actions may be interchanged with one another without departing from the scope of the claims. In other words, unless a specific order of steps or actions is specified, the order and / or use of specific steps and / or actions may be modified without departing from the scope of the claims. Further, the various operations of methods described above may be performed by any suitable means capable of performing the corresponding functions. The means may include various hardware and / or software component(s) and / or module(s), including, but not limited to a circuit, an application specific integrated circuit (ASIC), or processor. Generally, where there are operations illustrated in figures, those operations may have corresponding counterpart means-plus-function components with similar numbering.

[0287] The following claims are not intended to be limited to the aspects shown herein, but are to be accorded the full scope consistent with the language of the claims. Within a claim, reference to an element in the singular is not intended to mean “one and only one” unless specifically so stated, but rather “one or more.” Unless specifically stated otherwise, the term “some” refers to one or more. No claim element is to be construed under the provisions of 35 U.S.C. §112(f) unless the element is expressly recited using the phrase “means for” or, in the case of a method claim, the element is recited using the phrase “step for.” All structural and functional equivalents to the elements of the various aspects described throughout this disclosure that are known or later come to be known to those of ordinary skill in the art are expressly incorporated herein by reference and are intended to be encompassed by the claims. Moreover, nothing disclosed herein is intended to be dedicated to the public regardless of whether such disclosure is explicitly recited in the claims.

Claims

2024-068 / 10738-1206 CLAIMS1. A method of diagnosis and treatment for suicidal thought and behavior (STB), comprising; generating a set of STB parameters associated with a subject, wherein the set of STB parameters comprises a set of reward-aversion judgment variables, and at least one contextual variable; processing the set of STB parameters with a machine learning model to generate a STB prediction; and treating the subject with an STB treatment based on the STB prediction.

2. The method of claim 1, further comprising: presenting to the subject, via a user interface, a set of stimuli; receiving, from the subject via the user interface, a set of ratings associated with the set of stimuli, wherein each respective rating in the set of ratings corresponds with a respective stimuli in the set of stimuli; and determining, based on the set of ratings, the set of reward-aversion judgment variables.

3. The method of any one of claims 1-2, further comprising: presenting, to a subject via the user interface, at least one contextual question; and receiving, from the subject via the user interface, the at least one contextual variable.

4. The method of claim 3, wherein the set of stimuli comprises one or more subsets of stimuli, and each subset of the one or more subset of stimuli comprises one or more stimuli associated with a category.

5. The method of claim 4, wherein determining the set of reward-aversion judgment variables based on the set of ratings comprises: determining one or more subsets of ratings of the set of ratings, wherein each one of the one or more subsets of ratings corresponds to the respective stimuli in the one or more subsets of the set of stimuli; determining an average of each respective subset of the one or more subsets of ratings;2024-068 / 10738-1206 determining a variance of each respective subset of the one or more subsets of ratings; and determining an uncertainty or entropy of each respective subset of the one or more subsets of ratings.

6. The method of claim 5, further comprising determining one or more reward- aversion judgment variables in the set of reward-aversion judgment variables based on a relative preference graph.

7. The method of claim 6, wherein the set of reward-aversion judgment variables comprises one or more of: a loss aversion variable, a risk aversion variable, a loss resilience variable, an ante variable, an insurance variable, a peak positive risk variable, a peak negative risk variable, a total reward variable, a total aversion risk variable, or a tradeoff range variable.

8. The method of any of the preceding claims, wherein the machine learning model comprises a linear regression, a random forest, a Gaussian Process, a Gaussian Mixture Model, a Support Vector Machine, or an artificial neural network.

9. The method of claim 1, wherein the machine learning model comprises a balanced random forest.

10. The method of any of the preceding claims, wherein the at least one contextual variable comprises one or more of: a loneliness variable, a prior suicidal attempts variable, a health variable, or an anxiety variable.

11. The method of any of the preceding claims, wherein the set of STB parameters further comprises biographical data associated with the subject.

12. The method of claim 11, wherein the biographical data associated with the subject comprises one or more of: age data, income data, marital status data, employment data, ethnicity data, education level data, sex data, gender data or handedness data.

13. The method of any of the preceding claims, wherein the STB prediction comprises one or more measure of STB, the one or more measures of STB comprising a passive ideation measure, an active ideation measure, a suicide planning measure, or a planning for safety measure.2024-068 / 10738-1206 14. The method of any of the preceding claims, wherein the STB treatment comprises one or more of an selective serotonin reuptake inhibitor, a serotonin-norepinephrine reuptake inhibitor, an atypical antidepressant, a tricyclic antidepressant, a monoamine oxidase inhibitor, mood stabilizers, antipsychotics, anti-anxiety medications, or stimulant medication.

15. A processing system, comprising: a memory comprising computer-executable instructions; and a processor configured to execute the computer-executable instructions and cause the processing system to: generate a set of suicidal thought and behavior (STB)parameters associated with a subject, wherein the set of STB parameters comprises a set of reward-aversion judgment variables, and at least one contextual variable; process the set of STB parameters with a machine learning model to generate a STB prediction; and treat the subject with an STB treatment based on the STB prediction.

16. The processing system of claim 15, wherein the processor is further configured to: present to the subject, via a user interface, a set of stimuli; receive, from the subject via the user interface, a set of ratings associated with the set of stimuli, wherein each respective rating in the set of ratings corresponds with a respective stimuli in the set of stimuli; and determine, based on the set of ratings, the set of reward-aversion judgment variables.

17. The processing system of any one of claims 15-16, wherein the processor is further configured to: present, to a subject via the user interface, at least one contextual question; and receive, from the subject via the user interface, the at least one contextual variable.

18. The processing system of claim 17, wherein the set of stimuli comprises one or more subsets of stimuli, and each subset of the one or more subset of stimuli comprises one or more stimuli associated with a category.

19. The processing system of claim 18, wherein to determine the set of reward- aversion judgment variables based on the set of ratings wherein the processor is further configured to:2024-068 / 10738-1206 determine one or more subsets of ratings of the set of ratings, wherein each one of the one or more subsets of ratings corresponds to the respective stimuli in the one or more subsets of the set of stimuli; determine an average of each respective subset of the one or more subsets of ratings; determine a variance of each respective subset of the one or more subsets of ratings; and determine an uncertainty or entropy of each respective subset of the one or more subsets of ratings.

20. The processing system of claim 19, wherein to determine the determining one or more reward-aversion judgment variables in the set of reward-aversion judgment variables based on a relative preference graph.

21. The processing system of claim 20, wherein the set of reward-aversion judgment variables comprises one or more of: a loss aversion variable, a risk aversion variable, a loss resilience variable, an ante variable, an insurance variable, a peak positive risk variable, a peak negative risk variable, a total reward variable, a total aversion risk variable, or a tradeoff range variable.

22. The processing system of claim 15, wherein the machine learning model comprises a linear regression, a random forest, a Gaussian Process, a Gaussian Mixture Model, a Support Vector Machine, or an artificial neural network.

23. The processing system of claim 15, wherein the machine learning model comprises a balanced random forest.

24. The processing system of any of the preceding claims, wherein the at least one contextual variable comprises one or more of: a loneliness variable, a prior suicidal attempts variable, a health variable, or an anxiety variable.

25. The processing system of any of the preceding claims, wherein the set of STB parameters further comprises biographical data associated with the subject.2024-068 / 10738-1206 26. The processing system of claim 25, wherein the biographical data associated with the subject comprises one or more of: age data, income data, marital status data, employment data, ethnicity data, education level data, sex data, gender data or handedness data.

27. The processing system of any of the preceding claims, wherein the STB prediction comprises one or more measure of STB, the one or more measures of STB comprising a passive ideation measure, an active ideation measure, a suicide planning measure, or a planning for safety measure.

28. The processing system of any of the preceding claims, wherein the STB treatment comprises one or more of an selective serotonin reuptake inhibitor, a serotonin-norepinephrine reuptake inhibitor, an atypical antidepressant, a tricyclic antidepressant, a monoamine oxidase inhibitor, mood stabilizers, antipsychotics, anti-anxiety medications, or stimulant medication.

29. A non-transitory computer-readable medium storing program code for causing a processing system to perform the steps of: generating a set of STB parameters associated with a subject, wherein the set of STB parameters comprises a set of reward-aversion judgment variables, and at least one contextual variable; processing the set of STB parameters with a machine learning model to generate a STB prediction; and treating the subject with an STB treatment based on the STB prediction.

30. The non-transitory computer-readable medium of claim 29, wherein the steps further comprise: presenting to the subject, via a user interface, a set of stimuli; receiving, from the subject via the user interface, a set of ratings associated with the set of stimuli, wherein each respective rating in the set of ratings corresponds with a respective stimuli in the set of stimuli; and determining, based on the set of ratings, the set of reward-aversion judgment variables.

31. The non-transitory computer-readable medium of any one of claims 29-30, wherein the steps further comprise: presenting, to a subject via the user interface, at least one contextual question; and2024-068 / 10738-1206 receiving, from the subject via the user interface, the at least one contextual variable.

32. The non-transitory computer-readable medium of claim 31, wherein the set of stimuli comprises one or more subsets of stimuli, and each subset of the one or more subset of stimuli comprises one or more stimuli associated with a category.

33. The non-transitory computer-readable medium of claim 32, wherein determining the set of reward-aversion judgment variables based on the set of ratings comprises: determining one or more subsets of ratings of the set of ratings, wherein each one of the one or more subsets of ratings corresponds to the respective stimuli in the one or more subsets of the set of stimuli; determining an average of each respective subset of the one or more subsets of ratings; determining a variance of each respective subset of the one or more subsets of ratings; and determining an uncertainty or entropy of each respective subset of the one or more subsets of ratings.

34. The non-transitory computer-readable medium of claim 33, wherein the steps further comprise determining one or more reward-aversion judgment variables in the set of reward-aversion judgment variables based on a relative preference graph.

35. The non-transitory computer-readable medium of claim 34, wherein the set of reward-aversion judgment variables comprises one or more of: a loss aversion variable, a risk aversion variable, a loss resilience variable, an ante variable, an insurance variable, a peak positive risk variable, a peak negative risk variable, a total reward variable, a total aversion risk variable, or a tradeoff range variable.

36. The non-transitory computer-readable medium of any one of claims 29-35, the machine learning model comprises a linear regression, a random forest, a Gaussian Process, a Gaussian Mixture Model, a Support Vector Machine, or an artificial neural network.

37. The non-transitory computer-readable medium of any one of claims 29-35, wherein the machine learning model comprises a balanced random forest.2024-068 / 10738-1206 38. The non-transitory computer-readable medium of any one of claims 29-37, wherein the at least one contextual variable comprises one or more of: a loneliness variable, a prior suicidal attempts variable, a health variable, or an anxiety variable.

39. The non-transitory computer-readable medium of any one of claims 29-38, wherein the set of STB parameters further comprises biographical data associated with the subject.

40. The non-transitory computer-readable medium of claim 39, wherein the biographical data associated with the subject comprises one or more of: age data, income data, marital status data, employment data, ethnicity data, education level data, sex data, gender data or handedness data.

41. The non-transitory computer-readable medium of any one of claims 29-40, wherein the STB prediction comprises one or more measure of STB, the one or more measures of STB comprising a passive ideation measure, an active ideation measure, a suicide planning measure, or a planning for safety measure.

42. The non-transitory computer-readable medium of any one of claims 29-41, wherein the STB treatment comprises one or more of an selective serotonin reuptake inhibitor, a serotonin-norepinephrine reuptake inhibitor, an atypical antidepressant, a tricyclic antidepressant, a monoamine oxidase inhibitor, mood stabilizers, antipsychotics, anti-anxiety medications, or stimulant medication.

Images