Passive Smartphone Sensors for Detecting Psychopathology.

Importance: Smartphone sensors can continuously and unobtrusively collect clinically relevant behavioral data, allowing for more precise symptom monitoring in clinical and research settings. However, progress in identifying unique behavioral markers of psychopathology from smartphone sensors has been stalled by research on diagnostic categories that are heterogenous and have many nonspecific symptoms. Objective: To examine which domains of psychopathology are detectable with smartphone sensors and identify passively sensed markers for general impairment (the p-factor) and specific transdiagnostic domains. Design, Setting, and Participants: This cross-sectional study collected data from the Intensive Longitudinal Investigation of Alternative Diagnostic Dimensions study from January 1 to December 31, 2023, including a baseline survey and 15 days of smartphone monitoring. Participants were recruited from the community via a clinical research registry. A volunteer sample was selected for mental health treatment status. Main Outcomes and Measures: Transdiagnostic psychopathology dimensions of internalizing, detachment, disinhibition, antagonism, thought disorder, somatoform, and the p-factor; 27 behavior markers derived from a global positioning system, accelerometer, motion, call logs, screen on or off, and battery status. Results: A total of 557 participants were included in the study (463 [83%] female; mean [SD] age, 30.7 [8.8] years). The coefficient of multiple correlation (R) showed that the domain most strongly correlated with sensed behavior was detachment (R = 0.42; 95% CI, 0.29-0.54) followed by somatoform (R = 0.41; 95% CI, 0.30-0.53), internalizing (R = 0.37), disinhibition (R = 0.35; 95% CI, 0.19-0.51), antagonism (R = 0.33; 95% CI, 0.6-0.59), and thought disorder (R = 0.28; 95% CI, -0.19 to 0.75). Each psychopathology domain was associated with 4 to 10 smartphone sensor variables. Detachment, somatoform, and internalizing had the most behavioral markers. Of the 27 smartphone sensor variables, 14 (52%) had associations with psychopathology domains. After adjusting for shared variance between psychopathology dimensions, all domains except thought disorder retained significant, incremental associations with sensor variables, reflecting unique behavioral signatures (eg, antagonism and number of calls [standardized β = -0.11; 95% CI, -0.20 to -0.02] and disinhibition and battery charge level [standardized β = -0.24; 95% CI, -0.40 to -0.08]). The p-factor was associated with lower mobility (standardized β = -0.22; 95% CI, -0.32 to -0.12), more time at home (standardized β = 0.23; 95% CI, 0.14 to 0.32), later bed time (standardized β = 0.25; 95% CI, 0.11 to 0.38), and less phone charge (standardized β = -0.16; 95% CI, -0.30 to -0.01]). The p-factor was modeled as a latent factor estimated from common variance of the 6 psychopathology domains. All domains loaded moderately to strongly onto the p-factor as expected (standardized loadings: 0.89 for internalizing, 0.76 for somatoform, 0.70 for disinhibition, 0.62 for thought disorder, 0.51 for detachment, and 0.40 for antagonism). Conclusions and Relevance: This cross-sectional study shows how tethering transdiagnostic domains to concrete behavioral markers can maximize the potential of mobile sensing to study mechanisms driving psychopathology. Insights from these results, and future research that builds on them, can potentially be translated into symptom monitoring tools that fill the gaps in current practice and may eventually lead to more precise and effective treatment.