Machine Learning and Medication Adherence: Scoping Review.

Area of Science:

• * Biomedical Informatics
• * Artificial Intelligence in Healthcare

Background:

• * This is the first scoping review to broadly examine machine learning (ML) applications in medication adherence.
• * Existing literature highlights the growing interest in leveraging ML for improving patient adherence to prescribed treatments.

Purpose of the Study:

• * To systematically categorize, summarize, and analyze existing literature on the use of machine learning for medication adherence.
• * To identify key predictors, methods, and outcomes in ML-driven medication adherence research.

Main Methods:

• * A comprehensive search of major scientific databases (PubMed, Scopus, ACM, IEEE, Web of Science) was conducted.
• * Inclusion criteria were applied, leading to the analysis of 43 relevant studies following PRISMA-ScR guidelines.
• * Studies were systematically charted and categorized based on their approach to medication adherence actions.

Main Results:

• * Twenty strong predictors of medication adherence were identified across studies, with self-reported questionnaires and pharmacy claims being common data sources.
• * Machine learning models like logistic regression, neural networks, random forest, and support vector machines were frequently used, achieving prediction accuracies as high as 77.6%.
• * Monitoring systems demonstrated high accuracy (e.g., >93% for inhaler use), with AI-powered reminders significantly improving adherence rates compared to traditional methods.

Conclusions:

• * Machine learning shows significant potential for accurately predicting medication adherence, enabling targeted interventions to prevent nonadherence.
• * Monitoring systems, particularly for inhaler use and Parkinson's disease, achieve high accuracy, offering valuable insights into medication management.
• * Conversational AI reminders effectively enhance adherence, though context-aware systems may raise user intrusiveness concerns.