Abstract
BACKGROUND: Suboptimal maintenance medication (MM) adherence remains a clinical problem among Medicare beneficiaries with chronic obstructive pulmonary disease (COPD). OBJECTIVE: To inform risk-based personalized decision-making, this study sought to develop and validate prediction models of nonadherence to COPD MMs for Medicare beneficiaries. METHODS: This was a retrospective cohort study of beneficiaries aged 65 years and older with COPD and inhaled MMs. Nonadherence (proportion of days covered < 0.8) was measured in 12 months following the first MM fill after COPD diagnosis. Logistic and least absolute shrinkage selector operator regressions were implemented, and area under the receiver operating characteristic curve (AUROC) evaluated model accuracy, as well as positive predictive values and negative predictive values. Our models evaluated different sets of predictors for two cohorts: those with an MM prescription before COPD diagnosis (prevalent users) and those without (new users). RESULTS: Among 16,157 prevalent and 40,279 new users of MMs, 11,271 (69.8%) and 34,009 (84.4%), respectively, were nonadherent. The best-performing logistic models achieved AUROCs of 0.8714 and 0.881, positive predictive values of 0.881 and 0.881, and negative predictive values of 0.559 and 0.578, respectively, for prevalent and new users. The least absolute shrinkage selector operator models had similar accuracy. Models with baseline-only predictors had average performance (AUROC < 0.72). The most important predictors were initial MM adherence, short-acting bronchodilator use, and asthma. CONCLUSIONS: To our knowledge, this study is the first to develop predictive models of nonadherence to COPD MMs. Generated models achieved good discrimination and underlined the importance of early adherence. Well-performed models can be useful for care decision-making and interventions to improve COPD medication adherence after the first critical few months following a treatment episode. DISCLOSURES: All authors declared no conflicts of interest.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have