Presentation (CTI)

OBJECTIVES: Medication non-adherence in chronic disease management remains a major contributor to preventable complications, hospitalizations, and healthcare costs. Traditional methods often fail to detect early behavioral and clinical predictors of non-adherence. This review summarizes current evidence on artificial intelligence (AI) and machine learning (ML) models to predict medication adherence and describes their reported clinical and economic impact.
METHODS: A narrative literature review was conducted to identify published studies applying AI or ML techniques to predict medication adherence. Extracted information included model type, data sources, input variables, adherence definitions, and predictive performance metrics. Studies reporting clinical and economic outcomes associated with adherence-related interventions were also reviewed. Included data encompassed clinical history, behavioural and psychosocial factors, pharmacy dispensing patterns, digital engagement measures, and technology- derived metrics.
RESULTS: AI- based adherence prediction models demonstrated wide variability in performance depending on data depth and feature quality. Models relying only on administrative or demographic data showed lower accuracy, whereas models integrating clinical, behavioral, psychosocial, and digital inputs consistently achieved stronger performance, including reported accuracy above 85% or error rates near 12.9%. Frequently identified predictors included regimen complexity, prior adherence history, disease duration, comorbidity burden, and indicators of social or technological engagement. Studies describing downstream effects of AI-supported adherence prediction included improved chronic disease control, reduced likelihood of hospitalization or complications, and meaningful reductions in healthcare utilization. Several studies described associated cost savings and lower resource use resulting from targeted adherence interventions informed by AI-generated risk insights.
CONCLUSIONS: AI- enabled adherence prediction offers a promising strategy for strengthening chronic disease management by facilitating earlier identification of patients at risk for non-adherence. When incorporated into clinical or pharmacy-led interventions, these models may support improved outcomes and generate economic benefits through reduced utilization and better disease control. Further research is needed to standardize model inputs, validate performance, and evaluate long-term economic effects.