JMIR Publications

ABSTRACT

Background:

Longitudinal studies using wearable sensors to track numerous attributes such as physical activity, sleep, and heart rate can benefit from reductions in missing data. Maximizing compliance through participant engagement is one method to reduce missing data and poor compliance can reduce the return on the heavy investment of time and money into large-scale studies.

Objective:

This paper aims to identify to what extent compliance can be prospectively predicted from individual attributes and initial compliance and to identify the most predictive features.

Methods:

We define compliance with Ecological Momentary Assessments (EMAs) as the response rate, and wearable compliance as the proportion of 30-minute windows with data throughout a year-long multi-modal study of information workers. study participation. We analyze 31 individual characteristics (e.g. demographics, personality) and behavioral variables (e.g early compliance, study portal use) and select 14 variables to create beta regression models to predict compliance with Ecological Momentary Assessments (EMAs) two months out and wearable compliance one year out. We survey study participation and correlate the results with compliance.

Results:

Our modeling of compliance indicates that up to 23% of the variance in compliance with EMAs and 15% of the variance in compliance with our smartwatch wearing protocol could be predicted through a survey of demographics and personality. Likelihood of higher EMA and wearable compliance was associated with being older (EMA P=.027 and wearable P<.001), not having a supervisory role (EMA P<.001 and wearable P<.001), speaking English as a first language (EMA P=.051, wearable P=.057), having had a wearable prior to joining the study (EMA P=.051 and wearable P<.001), being more conscientious (EMA P=.024 and wearable P<.001), being more introverted (EMA P<.001 and wearable P<.001). Additionally, likelihood of higher wearable compliance was also associated with being less agreeable (P<.001) and less neurotic (P=.057). Further, including early compliance in the second week of the study can help explain more variance and reduce prediction error, with up to 66% of the variance in EMA compliance and 63% in wearable compliance explained. Finally, we report that compliance in the study correlated with our participants’ self-reflection on ease of participation, usefulness of our compliance portal, timely resolution of issues, and compensation adequacy, suggesting these as avenues to improving compliance.

Conclusions:

We recommend conducting an initial two-week pilot to measure trait-like compliance and identify participants at risk of long-term non-compliance. We recommend oversampling based on participants' individual characteristics to avoid introducing bias in the sample when excluding data based on non-compliance. Finally, we recommend the use of an issue tracking portal and special care on troubleshooting to help participants maintain compliance.