Digital medicine systems that combine an ingestible sensor, oral medication, and a wearable receiver can provide objective evidence of medication ingestion. However, interpreting days without a detected ingestion requires knowing whether the wearable receiver (patch) was actually in contact with the participant's body. This study integrated ingestion records from the Abilify MyCite system with patch-contact data to quantify how often apparent nonadherence reflected patch nonwear rather than true missed doses.
The analysis included 997 person-days contributed by 20 outpatients prescribed aripiprazole through the Abilify MyCite platform. Ingestion detection events were combined with patch-contact information to classify days as observed (patch-worn) or unobserved (patch-off or insufficient wear). In addition, 17 participants contributed smartphone sensing data that permitted exploratory analyses linking passive behavioral measures to ingestion detection on days when the patch was worn.
IRB approval for the protocol (Beth Israel Deaconess Medical Center Institutional Review Board, protocol 2023P-000628) and written informed consent from participants were reported. Participant-level data are restricted from public release due to sensitive high-resolution location and digital phenotyping information but may be made available to qualified researchers subject to approvals and data-use agreements.
The authors emphasize a three-state classification for the adherence outcome rather than a binary detected/undetected approach. The three states are:
Across the 997 person-days, there were 322 days without a detected ingestion. Of these, 53.7% were attributable to insufficient patch wear. Counting every no-detection day as nonadherence would yield an apparent nonadherence rate of 32.3%. However, conditioning the numerator on days with adequate wear reduces the observed nonadherence to 18.1%.
Because adherence on patch-off days cannot be observed, the study reports a bounded range for the true nonadherence rate: the lower bound is 14.9% and the upper bound is 32.3%, reflecting uncertainty introduced by unobserved patch-off periods.
Restricting to the 744 days when the patch was worn and ingestion could be observed, the investigators explored associations between passive smartphone-derived behavioral measures and the likelihood of a detected ingestion. Two exploratory findings were reported:
The authors explicitly describe these behavioral associations as exploratory and as characterizing the context of medication-taking behavior rather than implying causal relationships.
Key implications from the analysis are that device wear is integral to correctly interpreting adherence outcomes in digital medicine systems that rely on a wearable receiver. When a wearable is not in contact with the body, a no-detection day cannot reliably be interpreted as a missed dose. Without accounting for patch nonwear, studies and clinical monitoring could substantially overestimate nonadherence.
The proposed three-state outcome aims to distinguish observed confirmed adherence events, confirmed nonadherence when the device is demonstrably worn, and indeterminate periods when the device is not sufficiently worn. This approach preserves information about observation status and yields a bounded estimate of adherence that reflects uncertainty from unobserved days.
Passive behavioral measures from smartphones appear most interpretable after establishing observation status (patch on), since their association with ingestion detection depends on days when ingestion could actually be observed.
Details beyond the reported abstract are not available in this summary. From the provided information, limitations include the small sample size (20 participants, 997 person-days) and the fact that behavioral associations were exploratory and not presented as causal. The restricted availability of participant-level data—owing to privacy concerns and high-resolution location data—limits independent validation without approved data-use agreements. The study population and context (outpatients prescribed aripiprazole using a specific digital medicine product) may limit generalizability to other drugs, devices, or populations.
In this cohort using the Abilify MyCite system, more than half of apparent nonadherence days without a detected ingestion were explained by patch nonwear. Conditioning adherence estimates on adequate device wear substantially lowered measured nonadherence from 32.3% to 18.1%, while the true rate remains bounded between 14.9% and 32.3% because ingestion during patch-off days is unobserved. The authors recommend distinguishing detected ingestion, confirmed nonadherence during adequate wear, and indeterminate patch-off days when reporting adherence outcomes from digital medicine systems. Passive behavioral sensing may supplement interpretation but is most useful after observation status is established.