Table 1 Characteristics of prediction studies
From: A systematic review on passive sensing for the prediction of suicidal thoughts and behaviors
Author (year) | Predictors (device name) | Sensors/device analytics (study length) | Assessment of STB (frequency) | Main statistical analysis | Sample characteristics |
|---|---|---|---|---|---|
Barrigon et al.34 | Distance traveled, time spent at home, steps, app usage (Smartphone app: eB2) | GPS, device usage, activity (measured by Google Fit), accelerometer (6 months) | Electronic health records: suicide attempts or psychiatric emergency department visits (NA) | Unsupervised machine learning model (Heterogeneous Mixture Model) | 225 adult outpatients with a history of suicide attempt or ideation |
Bertrand et al.26 | Sleep characteristics (Wrist device: GENEActiv) | Accelerometer (14 days) | C-SSRS quick screen (2), QIDS-SR-16 item 12 (2), MADRS item 10 (2) | Spearman’s rank order correlations, multiple linear regression | 76 adult outpatients with bipolar disorder |
Coyne et al.27 | Smartphone app use, including social media (Smartphone apps: Moment or Cronicle) | Screentime (14 days) | SBQ-R (1) | Multilevel models | 281 adults from the general population |
Czyz et al.35 | Sleep, heart rate, steps (Wrist device: Fitbit) | NA (56 days) | Two EMA items: duration and intensity of ideation (4 per day) | Mixed-effects classification and regression trees | 102 young adult emergency department patients with acute suicidality |
Dogrucu et al.28 | Retrospective: GPS, browser history, call logs, text messages, contacts, social media (Smartphone app: Moodable) | GPS, microphone, calls, quantitative features of contacts, text messages, social media (14 days) | PHQ-9 item 9 (1) | Machine learning algorithms (KNN, SVM, RF) | 335 adults (general population), recruited via online platform (Mechanical Turk) |
Haines-Delmont et al.29 | Sleep, steps, smartphone usage including social media (Smartphone apps: SwiM, Apple Health kit; wrist device: Fitbit) | Accelerometer, Wi-Fi, screentime (7 days) | C-SSRS (3) | Machine learning algorithms (KNN, RF, SVM), logistic regression | 66 adult inpatients (acute mental health) |
Horwitz, Czyz et al.30 | Sleep, physical activity, heart rate (Smartphone; wrist device: Fitbit) | Accelerometer, gyroscope, light sensor (92 days) | PHQ-9 item 9 (2) | Three-step hierarchical logistic regression with suicidal subsample (n = 217) | 2881 first-year medical residents |
Horwitz, Kentopp et al.31 | Sleep, physical activity, heart rate (Smartphone; wrist device: Fitbit) | Accelerometer, gyroscope, light sensor (92 days) | PHQ-9 item 9 (2) | Machine learning algorithms (RF, ENR) | 2459 first-year medical residents |
Kleiman et al.32 | Physiological distress (electrodermal activity)b (Smartphone app: LifeData; wrist device: Empatica Embrace 2) | Accelerometer, gyroscope, thermostat, electrodermal activity sensor (Inpatient stay + 28 days after discharge) | Three EMA items (6 per day) | Multilevel models | 25 adult inpatients with acute suicidality |
Salvatore et al.36 | Circadian activity (Wrist device: Motionlogger) | Accelerometer (3 days) | EMA item: wish to die (3; one per day) | Linear regression | 83 adult outpatients with a current major depressive episode and either a bipolar or major depressive disorder |
Sheridan et al.33 | Heart rate variability (wrist device) | Light sensor (7 days) | C-SSRS (3) | Multilevel models | 51 adolescent inpatients with acute suicidality |
