Presentation (CTI)

OBJECTIVES: Group-based trajectory modeling (GBTM) is a common method for identifying medication adherence patterns over time. In GBTM, the proportion of days covered (PDC) can be modeled as either a continuous measure or a binary threshold (e.g., PDC ≥0.80); however, the value of these two modeling approaches remains unclear. Therefore, we aim to compare GBTM approaches involving continuous versus binary PDC in trajectory-based class assignments.
METHODS: Utilizing Medicare claims data (2014-2019), we identified a national cohort of 3,317 beneficiaries with chronic hepatitis B (CHB) who initiated first-line oral nucleos(t)ide analogue therapy. Monthly adherence was calculated using PDC over a 12-month follow-up. Parallel GBTMs were fit as a censored normal (CNORM) model using continuous PDC (0-1) and a logistic (LOGIT) model using a binary indicator (PDC ≥0.80), in SAS 9.4. Two to five latent classes were evaluated, and the final model was selected based on Bayesian Information Criterion, interpretability, average posterior probability (AvePP), odds of correct classification (OCC), and entropy. The concordance between CNORM and LOGIT class assignments was evaluated by kappa analyses.
RESULTS: Both models found five clinically interpretable adherence patterns, ranging from early nonadherence to persistently high adherence. In both models, the resulting classes demonstrated good classification performance, with high entropy (CNORM=0.862; LOGIT=0.839) and strong AvePP (CNORM=0.98-0.87; LOGIT=0.85-0.96). Modal agreement was 72.9% (n=2,418). Weighted kappa ordered from lowest to highest adherence yielded near-perfect agreement (κw=0.82, 95%CI: 0.81-0.83). OCCs tend to be higher for continuous GBTM. Disagreements were observed between intermediate adherence groups, with perfect agreement for the extreme classes in the two models.
CONCLUSIONS: The modeling of continuous and binary adherence trajectories produced primarily consistent adherence trajectories, exhibiting high overall agreement. However, the binary GBTM approach tended to shift/merge intermediate adherence patterns, indicating that continuous PDC trajectories may better preserve the gradation of adherence patterns.