JMIR Publications

ABSTRACT

Background:

Smartphone-based digital phenotyping uses built-in sensors and usage patterns to passively capture behavioral and environmental data relevant to a person’s health. This practice has been applied extensively in the context of mental health and chronic disease management.

Objective:

This review aims to synthesize digital phenotyping articles that are solely smartphone- based. That is, they use data exclusively from onboard sensors on smartphones to characterize specific health conditions.

Methods:

We conducted a scoping review of English-language, peer-reviewed articles published between 2012 and 2024 in Google Scholar, IEEE Xplore, ACM Digital Library, and PubMed using terms such as “mobile sensing” and “digital phenotyping.” Eligible articles used onboard smartphone sensors to assess health, allowed validation with other devices, and went beyond self-report (e.g., ecological momentary assessments). We excluded articles that focused solely on algorithm development or data capture without linking results to a health context.

Results:

The search identified 111 articles, of which 65 articles met the inclusion criteria. Most articles described observational and used passive sensing. Mental health conditions were the most frequently examined health conditions. These included: depression (n = 17), stress or anxiety (n = 14), bipolar disorder (n = 11) and schizophrenia (n = 8). Other, relatively less common conditions included: substance use disorders (n = 7), Parkinson’s disease (n = 4), and sleep apnea (n = 2). The smartphone sensor data used in these articles were: screen state or device usage logs (n = 35), GPS (n = 34), accelerometer (n = 26), microphone data (n = 18), call logs (n = 21), app usage statistics (n = 15), WiFi connectivity (n = 5), Bluetooth (n = 4), battery state (n = 3), and keystroke dynamics (n = 2). Ground-truth measurements relied on validated clinical scales (e.g., PHQ-9, GAD- 7, YMRS, PSQI) (n = 41), followed by ecological momentary assessments (n = 18), clinician-confirmed diagnoses (n = 9), and physiological measures, such as polysomnography (n = 3). In terms of study focus, articles could be categorized into monitoring symptoms (n = 43), diagnostic applications (n = 15), and intervention strategies (n = 7). Overall in these articles presented several methodological gaps. These included: (1) inconsistent reporting of sensor streams, (2) limited data quality descriptions, and (3) lack of standard in outcome validation. These gaps underscore the need for standardized reporting and greater data availability to improve reproducibility and comparability in the smartphone-based digital phenotyping space.

Conclusions:

This review shows that smartphone-based digital phenotyping is a versatile tool for understanding health-related behaviors across diverse conditions. Most research targets mental health, but work also extends to other disorders. Future work should expand applications, standardize reporting, improve rigor, and develop shared datasets to advance the field. Clinical Trial: N/A