Sociodemographic characteristics of missing data in digital phenotyping

Area of Science:

• Digital Health
• Behavioral Science
• Data Science

Background:

• Smartphones offer a powerful tool for collecting longitudinal behavioral data through digital phenotyping.
• Understanding data collection and missing data patterns is crucial for effective smartphone-based research.
• Existing research highlights the promise of digital phenotyping in psychiatry and neuroscience.

Purpose of the Study:

• To investigate data collection and non-collection (missing data) patterns in smartphone-based digital phenotyping.
• To examine factors influencing accelerometer and GPS sensor data non-collection across diverse populations.

Main Methods:

• A meta-study analyzing accelerometer and GPS data from 211 participants across six studies (29,500 person-days).
• Utilized Bayesian hierarchical negative binomial regression with study- and user-level random intercepts.
• Conducted sensitivity analyses with alternative model specifications and stratified models.

Main Results:

• iOS users exhibited lower GPS data non-collection rates compared to Android users.
• GPS data non-collection did not vary significantly by race/ethnicity, education, age, or gender.
• Black participants showed higher accelerometer data non-collection rates; no significant differences by sex, education, or age.
• Weekly non-collection rates for both sensors increased by 0.5% to 0.9%.

Conclusions:

• Smartphone-based digital phenotyping is feasible for diverse populations and extended study durations.
• Findings provide critical insights for designing, planning, and analyzing digital phenotyping studies.
• Understanding sensor data non-collection is essential for maximizing the utility of smartphone data in research.