JMIR Publications

ABSTRACT

Background:

Suicide is a leading cause of preventable mortality worldwide, with over 700,000 deaths annually. Although suicidal ideation fluctuates rapidly, conventional risk assessments rely on retrospective self-report collected infrequently, and the detection of short-term suicide risk remains limited. Passive digital sensing using smartphones and wearable devices enables continuous monitoring of behavioral and physiological signals associated with suicide-related outcomes. However, current evidence remains fragmented, without a clear framework for translation into clinically interpretable risk indicators.

Objective:

This scoping review synthesized and mapped passive digital markers associated with suicide-related outcomes via the layered hierarchical sense-making framework (LHSF), which structures information from raw sensor data to high-level behavioral markers. We aimed to illustrate a clinically interpretable mapping of digital markers for suicide-specific digital phenotyping.

Methods:

Following Arksey and O’Malley and the Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews guidelines, this scoping review was conducted using the Population-Concept-Context framework (Population: not restricted; Concept: passively collected digital data from smartphones or wearable devices; Context: suicide-related outcomes). PubMed, CINAHL, PsycINFO, and IEEE Xplore were searched for studies published between 2015 and 2025. Studies were included if they (i) collected passive digital data from smartphones or wearable devices and (ii) measured suicide-related outcomes. Narrative mapping was conducted using LHSF to distinguish between low-level features (i.e., measurable properties extracted from sensors) and high-level behavioral markers (i.e., clinically meaningful constructs interpreted from low-level features).

Results:

Of 626 studies identified, 14 met inclusion criteria. Six studies employed predictive modeling, nine conducted correlational analyses, and one applied both approaches. Among the six predictive studies (AUC = .56–.89), lower heart rate variability predicted imminent suicide risk in one study (AUC = .89). Of the nine correlational studies, eight reported at least one significant association between passive sensor data and suicide-related outcomes. When mapped to the LHSF, low-level features spanned seven domains (location type and variability, movement intensity, sleep parameters, autonomic physiological indices, respiratory rate, smartphone use metrics, and linguistic features), and high-level behavioral markers included autonomic dysregulation, sleep disturbance, social withdrawal, physical inactivity, phone use patterns, suicide-related expression, and psychological and behavioral patterns. Physiological indicators of autonomic regulation were most consistently associated with suicide-related outcomes. Location type and variability, smartphone use metrics, and linguistic features were significantly associated with suicide-related outcomes in several studies. In contrast, sleep parameters and movement intensity showed limited associations with suicide-related outcomes.

Conclusions:

Physiological indicators showed the most consistent associations with suicide-related outcomes, followed by smartphone-derived behavioral and linguistic features, whereas sleep and activity-related indicators showed limited or inconsistent evidence. To improve understanding of clinically interpretable indicators, future research should prioritize multimodal data integration, algorithmic refinement, and external validation to strengthen clinical utility in digital suicide phenotyping based on the LHSF.