JMIR Publications

ABSTRACT

Background:

Body-worn sensors paired with ecological momentary assessment (EMA) could detect substance use, craving and related states in daily life. Judging such detection depends on how data were combined, how detectors were validated and what studies report.

Objective:

This review aimed to map studies that pair a body-worn sensor with EMA in the daily lives of people who use substances: what they sensed and detected, whether their detection results can be compared, and how completely they reported details needed to interpret them.

Methods:

Following Joanna Briggs Institute methods, we searched PubMed/MEDLINE, the ACM Digital Library, IEEE Xplore, Google Scholar and the references of other papers from 25 April to 14 October 2026. We included peer-reviewed reports in English (2015 to February 2026) that combined a body-worn sensor recording in daily life with digital EMA in people recruited for their substance use or its treatment. One reviewer screened titles and abstracts and confirmed every full-text eligibility decision and all charted data; a second reviewer checked a random sample of included studies. Data were summarized descriptively.

Results:

Of 776 records, 77 studies (51 cohorts) were included. Just over half were published in 2023 or later (n=42, 55%), and almost all were conducted in the United States (n=71, 92%). Alcohol was the most studied substance (n=44, 57%), followed by nicotine (n=16, 21%) and opioids (n=14, 18%). Movement was the most analyzed signal (n=38, 49%). The most common aim was testing a temporal association (n=35, 45%). Just over half evaluated sensor-based detection or classification against a reference (n=42, 55%), almost all with ground-truth labels from EMA (n=39, 93%). Of these 42, only a few clearly met three evaluation criteria in the same analysis (explicit or verified non-events, testing on participants not used to build or tune the detector, and false alarms estimated on unselected free-living data; n=3, 7%), all detecting substance use, and only a few made forecasts (n=4, 10%). Studies scored performance on different units of analysis. Of the 70 studies that sent prompts, more than half did not report an EMA compliance rate with its denominator (n=40, 57%). Over a third of studies reported per-participant wear time (n=30, 39%); of the 29 studies with an optical sensor, only a few reported skin tone (n=2, 7%).

Conclusions:

Studies pairing body-worn sensors with EMA have become more common, but few of their detection results were fully evaluated or could be compared, and many studies omitted diagnostic and wearable details. Studies should report how substance problems were established, the reference data source, how non-events were defined, how detectors were validated, false alarms on unselected free-living data, and wear time and device details. Clinical Trial: OSF Registries osf.io/hrzyb; https://osf.io/hrzyb