Evaluating cellular communication sensing for lapse risk prediction during early recovery from alcohol use disorder: A longitudinal observational study

22/08/2026

AI Summary

We evaluated a machine learning model predicting next-day alcohol lapse among individuals in early recovery from AUD using contextualized cellular communication data and baseline alcohol use, demographic, and psychiatric and personality characteristics.

These findings demonstrate that cellular communication data capture some risk-relevant signal for alcohol lapse but do not provide incremental predictive value beyond baseline measures.

Nevertheless, several communication features were retained in the full model with moderately sized coefficients, suggesting that aspects of social communication may still be clinically relevant for understanding lapse risk.

Basic summary

PLoS One. 2026 Aug 21;21(8):e0355396. doi: 10.1371/journal.pone.0355396. eCollection 2026.

ABSTRACT

Alcohol Use Disorder (AUD) is a chronic, relapsing condition, and identifying periods of elevated lapse risk remains a major challenge in supporting recovery. An automated recovery monitoring and support system using personal sensing and machine learning may help detect when individuals are at heightened risk. Cellular communication sensing may be a promising approach for passively capturing risk-relevant information about social interactions, particularly when these data are contextualized with participant-specific meaning. We evaluated a machine learning model predicting next-day alcohol lapse among individuals in early recovery from AUD using contextualized cellular communication data and baseline alcohol use, demographic, and psychiatric and personality characteristics. A total of 144 participants (49% male; mean age = 40; 87% non-Hispanic White) with a goal of abstinence provided cellular communication data and alcohol use reports via a 4x daily EMA for up to three months. Models were trained and evaluated using repeated k-fold cross-validation. The best-performing full model used an elastic net algorithm and retained 10 features (median posterior auro C = 0.67, 95% Bayesian credible interval (CI; [0.64, 0.71]). A comparison model including only baseline features demonstrated comparable performance (median auROC = 0.69, 95% CI [0.65, 0.72]). Cellular communication features on their own performed poorly, but still above chance performance (median auROC = 0.59, 95% CI [0.55, 0.62]). These findings demonstrate that cellular communication data capture some risk-relevant signal for alcohol lapse but do not provide incremental predictive value beyond baseline measures. Nevertheless, several communication features were retained in the full model with moderately sized coefficients, suggesting that aspects of social communication may still be clinically relevant for understanding lapse risk. However, limitations inherent to cellular communication sensing may outweigh its added utility in lapse prediction models.

PMID:42627809 | DOI:10.1371/journal.pone.0355396

Document this CPD