Machine learning models for temporally precise lapse prediction in alcohol use disorder.

We developed three machine learning models that predict hour-by-hour probabilities of a future lapse back to alcohol use with increasing temporal precision (i.e., lapses in the next week, next day, and next hour). Model features were based on raw scores and longitudinal change in theoretically implicated risk factors collected through ecological momentary assessment. Participants (N = 151, 51% male, Mage = 41, 87% White, 97% non-Hispanic) in early recovery (1-8 weeks of abstinence) from alcohol use disorder provided 4 × daily ecological momentary assessment for up to 3 months. We used grouped, nested cross-validation to select the best models and evaluate the performance of those best models. Models yielded median areas under the receiver operating curves of 0.89, 0.90, and 0.93 in the 30 held-out test sets for week-, day-, and hour-level models, respectively. Some feature categories consistently emerged as being globally important to lapse prediction across our week-, day-, and hour-level models (i.e., past use, future self-efficacy). However, most of the more punctate, time-varying constructs (e.g., craving, past stressful events, arousal) appear to have a greater impact within the next-hour prediction model. This research represents an important step toward the development of a smart (machine learning guided) sensing system that can both identify periods of peak lapse risk and recommend specific supports to address factors contributing to this risk. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Acceptability of Personal Sensing Among People With Alcohol Use Disorder: Observational Study.

BACKGROUND: Personal sensing may improve digital therapeutics for mental health care by facilitating early screening, symptom monitoring, risk prediction, and personalized adaptive interventions. However, further development and the use of personal sensing requires a better understanding of its acceptability to people targeted for these applications. OBJECTIVE: We aimed to assess the acceptability of active and passive personal sensing methods in a sample of people with moderate to severe alcohol use disorder using both behavioral and self-report measures. This sample was recruited as part of a larger grant-funded project to develop a machine learning algorithm to predict lapses. METHODS: Participants (N=154; n=77, 50% female; mean age 41, SD 11.9 years; n=134, 87% White and n=150, 97% non-Hispanic) in early recovery (1-8 weeks of abstinence) were recruited to participate in a 3-month longitudinal study. Participants were modestly compensated for engaging with active (eg, ecological momentary assessment [EMA], audio check-in, and sleep quality) and passive (eg, geolocation, cellular communication logs, and SMS text message content) sensing methods that were selected to tap into constructs from the Relapse Prevention model by Marlatt. We assessed 3 behavioral indicators of acceptability: participants' choices about their participation in the study at various stages in the procedure, their choice to opt in to provide data for each sensing method, and their adherence to a subset of the active methods (EMA and audio check-in). We also assessed 3 self-report measures of acceptability (interference, dislike, and willingness to use for 1 year) for each method. RESULTS: Of the 192 eligible individuals screened, 191 consented to personal sensing. Most of these individuals (169/191, 88.5%) also returned 1 week later to formally enroll, and 154 participated through the first month follow-up visit. All participants in our analysis sample opted in to provide data for EMA, sleep quality, geolocation, and cellular communication logs. Out of 154 participants, 1 (0.6%) did not provide SMS text message content and 3 (1.9%) did not provide any audio check-ins. The average adherence rate for the 4 times daily EMA was .80. The adherence rate for the daily audio check-in was .54. Aggregate participant ratings indicated that all personal sensing methods were significantly more acceptable (all P<.001) compared with neutral across subjective measures of interference, dislike, and willingness to use for 1 year. Participants did not significantly differ in their dislike of active methods compared with passive methods (P=.23). However, participants reported a higher willingness to use passive (vs active) methods for 1 year (P=.04). CONCLUSIONS: These results suggest that active and passive sensing methods are acceptable for people with alcohol use disorder over a longer period than has previously been assessed. Important individual differences were observed across people and methods, indicating opportunities for future improvement.

Prospective Prediction of Lapses in Opioid Use Disorder: Protocol for a Personal Sensing Study.

BACKGROUND: Successful long-term recovery from opioid use disorder (OUD) requires continuous lapse risk monitoring and appropriate use and adaptation of recovery-supportive behaviors as lapse risk changes. Available treatments often fail to support long-term recovery by failing to account for the dynamic nature of long-term recovery. OBJECTIVE: The aim of this protocol paper is to describe research that aims to develop a highly contextualized lapse risk prediction model that forecasts the ongoing probability of lapse. METHODS: The participants will include 480 US adults in their first year of recovery from OUD. Participants will report lapses and provide data relevant to lapse risk for a year with a digital therapeutic smartphone app through both self-report and passive personal sensing methods (eg, cellular communications and geolocation). The lapse risk prediction model will be developed using contemporary rigorous machine learning methods that optimize prediction in new data. RESULTS: The National Institute of Drug Abuse funded this project (R01DA047315) on July 18, 2019 with a funding period from August 1, 2019 to June 30, 2024. The University of Wisconsin-Madison Health Sciences Institutional Review Board approved this project on July 9, 2019. Pilot enrollment began on April 16, 2021. Full enrollment began in September 2021. CONCLUSIONS: The model that will be developed in this project could support long-term recovery from OUD-for example, by enabling just-in-time interventions within digital therapeutics. INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID): DERR1-10.2196/29563.