Inhaled treatment adherence patterns and related barriers across three monthly follow-up waves among patients with cough variant asthma: a clinical natural language processing study

Objective To characterize adherence and adherence-related text features across three monthly telephone follow-ups among adults with cough variant asthma (CVA) newly initiating maintenance inhaled therapy and to evaluate their contemporaneous associations. Methods This single-center, three-wave observational cohort classified adherence during the preceding month as high or low based on reported treatment implementation; records with insufficient information were unevaluable. A manually validated, rule-enhanced clinical natural language processing approach extracted seven non-mutually exclusive text features, including four prespecified patient-perceived barriers. Logistic generalized estimating equations (GEE) estimated population-averaged contemporaneous associations with high adherence. Complete-case GEE and wave-specific Firth models were conducted as sensitivity analyses. Results Among 767 patients, adherence was evaluable for 732, 697, and 602 at Waves 1, 2, and 3, respectively; high adherence was documented in 94.4%, 73.9%, and 47.0% of evaluable records. Intentional discontinuation or dose reduction increased from 4.8% to 26.5%. The primary GEE included 2,031 observations. Lower odds of high adherence were associated with access or financial barriers (OR 0.28, 95% CI 0.16–0.49; P<0.001), illness- or treatment-related beliefs (OR 0.19, 95% CI 0.06–0.63; P = 0.007), and side-effect concerns or poor tolerability (OR 0.43, 95% CI 0.19–0.96; P = 0.040). Similar associations for access or financial barriers and illness- or treatment-related beliefs were observed in the complete-case and Wave 3 Firth analyses. Conclusions High adherence was documented less frequently and patient-initiated treatment reduction more frequently at later waves. Access or financial barriers showed the most consistent contemporaneous association with lower odds of high adherence, whereas associations with illness- or treatment-related beliefs and side-effect concerns were less consistent across analyses. These record-based observational findings require prospective confirmation.