OBJECTIVES Patients’ preferences regarding attributes of oral anti-coagulant (OAC) therapy were shown to be heterogeneous. The objective of this study was to understand preference heterogeneity in patients with atrial fibrillation (AF) using latent class analysis (LCA).

METHODS : The health preference survey consisted of 12 discrete choice questions. The attributes of convenience were: intake frequency; need for routine monitoring of coagulation; diet and drug interactions; medication intake and type. Background information regarding gender, age, current OAC therapy and medication adherence was elicited. 508 AF Patients from five European countries were surveyed in August 2017. LCA was performed for 1-5 preference classes and 1-3 certainty classes.

RESULTS : We selected the 2-sClass-4-Class model based on model fit and interpretability. For certainty, 45% of patients had strong preferences while 55% had weaker preferences. In the “no INR monitoring only” class (58% of patients) omitting the need for monitoring was the only relevant attribute, and patients were more likely to be current DOAC users and least adherent. The “once daily, no INR monitoring” class (19%) attached equal importance to both aspects of treatment, and patients were more often current VKA users and moderately adherent. Besides a “no INR monitoring, interactions likely pattern” class (16%) there was a small group of patients (7%) who strongly preferred to be monitored; patients were more likely to be current VKA users and highly adherent. Current OAC, adherence and country were significant predictors of class membership (p<0,05), while age, gender and burden of medication were not.

CONCLUSIONS : Different preferences in patients can be partly explained by background characteristics of patients, however, it is unknown whether patients align preferences with therapy or receive therapy or receive therapy they prefer. Latent class analysis of preference data can result in increased insight in predictors of patient preferences compared to traditional regression models.