Session (CTI)

ISSUE: Value assessments of health technologies are increasingly incorporating medication adherence data. Yet these analyses often rely on static metrics, such as the proportion of days covered (PDC) or the medication possession ratio (MPR), typically applying an arbitrary 80% cutoff to classify patients as “adherent” or “nonadherent”. While simple and widely used, these metrics mask substantial heterogeneity in real-world medication taking behavior, including intermittent use, early discontinuation, or overuse. Relying on static measures risks can misestimate the true clinical and economic impact of nonadherence and undervalue adherence enhancing interventions. Newer approaches, such as group-based trajectory modeling (GBTM) and latent class mixed models, combined with artificial intelligence (AI) and machine-learning (ML) models offer a way to move from static thresholds to dynamic, pattern-based classifications of nonadherence. The purpose of this workshop is to examine how advanced trajectory- and AI/ML-based adherence measures can be appropriately defined, validated, and incorporated into value assessment, and to identify the conditions under which they add real decision value beyond traditional metrics and can be practically embedded into economic evaluations and broader value frameworks. OVERVIEW: This session explores why measuring adherence is a critical structural input in in the value assessment of health technologies. Dweeti Nayak will introduce the importance of adherence measurement in these contexts (5 minutes). Bijan Borah will compare the traditional static measures with emerging trajectory-based approaches and illustrate how they capture dynamic adherence patterns (15 minutes). Chao Cai will introduce latent mixed models and compare their performance with GBTM and unsupervised learning methods for classifying adherence behaviors (15 minutes). Tamas Agh will discuss the integration of AI/ML methods into adherence trajectory modeling and present a real-world case demonstration (15 minutes). The final 10 minutes will be dedicated to audience discussion and Q&A, facilitated through ISPOR’s polling tools to encourage active engagement.