Project Grant R43MH125461

IMPROVING THE PERFORMANCE OF AN AI-ENABLED CLINICAL DECISION SUPPORT TOOL FOR DETECTING SUICIDAL RISK AND OTHER MENTAL HEALTH CONCERNS - PROJECT SUMMARY-CLARIGENT HEALTH IS REFINING AN AI-ENABLED CLINICAL DECISION SUPPORT TOOL THAT USES LINGUISTIC AND ACOUSTIC PATTERNS IN RECORDED PATIENT INTERVIEWS TO IDENTIFY PATIENTS AT RISK OF HARM DUE TO SUICIDAL IDEATION OR OTHER MENTAL HEALTH CONCERNS. IN AN ONGOING STUDY, CLARIGENT IS COLLECTING > 6,000 PATIENT INTERVIEWS FROM ADOLESCENTS AND ADULTS IN SCHOOLS, PRIMARY CARE OFFICES, AND OTHER SETTINGS TO IMPROVE RISK PREDICTION AND REFINE THE PLATFORM FOR DATA COLLECTION AND REPORTING. PRIOR STUDIES INDICATE THE TOOL WILL HAVE BROAD CLINICAL UTILITY, BUT THE IMPACTS OF PATIENT CHARACTERISTICS AND INTERVIEW SETTING ON THE TOOL'S PERFORMANCE ARE UNKNOWN. CHARACTERIZING THESE EFFECTS WOULD SUPPORT TAILORED MODELS, INFORM PROSPECTIVE TRIALS, AND AID IN SELECTING THE CORRECT REGULATORY PATHWAYS AND SETTINGS FOR DEPLOYMENT. FURTHER DEVELOPMENT IS EXPECTED TO PROVIDE A RAPID AND RELIABLE OBJECTIVE ASSESSMENT OF SUICIDAL IDEATION AND OTHER ACTIONABLE MENTAL HEALTH CONCERNS TO REDUCE LOSS OF LIFE AND IMPROVE MENTAL HEALTH AMONG ADOLESCENTS AND ADULTS. EACH YEAR, ~13% OF ADOLESCENTS AND 7% OF ADULTS IN THE US EXPERIENCE A MAJOR DEPRESSIVE EPISODE. MANY CONTEMPLATE OR ATTEMPT SUICIDE, RESULTING IN > 45,000 DEATHS. EARLY RISK IDENTIFICATION AND TREATMENT COULD PREVENT MANY OF THESE, BUT AT- RISK INDIVIDUALS RARELY RECEIVE SCREENING UNTIL SYMPTOMS ARE SEVERE. UNIVERSAL SCREENING AND DIAGNOSTIC TOOLS DEPLOYED IN SCHOOLS AND PHYSICIANS' OFFICES COULD ADDRESS THIS PROBLEM, BUT THE LACK OF AN OBJECTIVE RISK ASSESSMENT TOOL FOR USE IN THESE SETTINGS IS A MAJOR BARRIER. IN THE ABSENCE OF BLOOD-BASED OR GENETIC BIOMARKERS, VERBAL AND NONVERBAL LANGUAGE CUES CAN BE USED TO IDENTIFY RISK. THE COMPLEXITY AND BROAD VARIABILITY OF THESE THOUGHT MARKERS RESISTS EASY CLASSIFICATION BY A CLINICIAN, BUT AI APPROACHES CAN IDENTIFY CONSISTENT, MEANINGFUL PATTERNS. TWO TRIALS HAVE DEMONSTRATED THE ABILITY OF AI-ENABLED ALGORITHMS TO USE LINGUISTIC AND ACOUSTIC FEATURES OF A BRIEF INTERVIEW TO CORRECTLY CATEGORIZE SUICIDE RISK. CLARIGENT, WHICH WAS FORMED TO FURTHER DEVELOP AND COMMERCIALIZE THIS TECHNOLOGY FOR USE IN DIVERSE SETTINGS, IS COLLECTING > 6,000 PATIENT INTERVIEWS TO IMPROVE THE PERFORMANCE OF THIS TOOL FOR PREDICTING SUICIDALITY, DEPRESSION, AND ANXIETY. THIS PHASE I SBIR WILL ASSESS THE EFFECTS OF AGE, SEX, RACE/ETHNICITY, SOCIOECONOMIC STATUS, GEOGRAPHIC AREA, OR INTERVIEW ENVIRONMENT ON MODEL PERFORMANCE; VALIDATE MODIFIED MODELS; AND INFORM KEY DECISIONS REGARDING COMMERCIALIZATION PATH AND FUTURE TRIAL DESIGNS. AIM 1. CHARACTERIZE THE EFFECTS OF PATIENT AND SETTING CHARACTERISTICS ON MODEL PERFORMANCE. FACTORS THAT IMPROVE MODEL PERFORMANCE BY = 0.05 AUC WILL BE CONSIDERED FOR SEPARATE MODEL DEVELOPMENT. AIM 2. VALIDATE UPDATED ALGORITHMS WITH A HOLDOUT DATA SET. SUCCESSFUL MODELS WILL CORRECTLY CLASSIFY RISK CATEGORIES WITH AN AUC = 0.80. MILESTONES FOR PROGRESSION TO PHASE II-THE GOAL IS TO IDENTIFY PATIENT OR SETTING CHARACTERISTICS THAT IMPROVE MODEL PERFORMANCE = 0.05 AUC. MODELS THAT CORRECTLY CLASSIFY RISK CATEGORIES WITH AN AUC = 0.80 WILL BE ADVANCED FOR EVALUATION IN PROSPECTIVE TRIALS. CLARIGENT WILL CONSULT WITH FDA AND ADVISORS TO DETERMINE THE CORRECT REGULATORY PATH DEPENDING ON MODEL PERFORMANCE AND INTENDED USE.