BACKGROUND
Ecological Momentary Assessment (EMA) can capture highly dynamic processes and intense variability patterns suitable to the study of suicidal ideation and behaviors. Artificial Intelligence (AI), and in particular Machine Learning (ML) strategies, have increasingly been applied to EMA data in suicide research.
OBJECTIVE
The review aims to (1) synthesize empirical research applying AI strategies to EMA data in the study of suicidal ideation and behaviors, (2) identify methodologies used, data collection procedures employed, suicide outcomes studied, AI applied, results reported, and (3) develop a standardized reporting framework for researchers applying AI to EMA data in future.
METHODS
PsycINFO, PubMed, SCOPUS and EMBASE were searched for articles published until June 2024. Studies that applied AI to EMA data in the investigation of suicide outcomes (suicidal ideation, suicide attempt, suicide death), collected across devices (Smartphone, Personal Digital Assistant, PC, tablet) and settings (clinical, community), were included. The Preferred Reporting Items for Systematic Reviews and Meta Analyses (PRISMA) guidelines were used to identify relevant studies while minimizing bias. Specific EMA data reported included EMA sampling method, monitoring period, prompt latency, compliance, attrition, and treatment of missing data. Quality appraisal was performed using an adapted checklist for reporting EMA studies (CREMAS).
RESULTS
1,201 records were identified across databases. After full text review, 12 articles, comprising 4398 participants were included. Studies were conducted in psychiatric hospitals (n = 5), emergency departments (n = 2), outpatient clinics (n = 2), medical residency programs (n = 1), and university mental health clinics (n = 1), with some conducted across settings. Design features reported (sampling strategy, prompting frequency, response latency, device used, compliance, and treatment of missing data) varied across studies. In the application of AI to EMA data to predict suicidal ideation, studies reported mean AUCs (0.74 to 0.86), sensitivity (0.64 to 0.81), specificity (0.73 to 0.86), and positive predictive values (0.72 to 0.77).
CONCLUSIONS
The application of AI to EMA data within suicide research is a small but burgeoning area with high heterogeneity apparent in data collection and reporting standards. Findings indicate some promise in the application of ML to self-report EMA data in the prediction of near-term suicidal ideation. The development by the research team of a reporting framework aims to standardize reporting on the application of AI to EMA data in mental health research going forward.
CLINICALTRIAL
PROSPERO: CRD42023440218
Open Science Framework: https://doi.org/10.17605/OSF.IO/NZWUJ