OBJECTIVES:

Obstructive sleep apnea (OSA) is prevalent in patients with cardiovascular diseases, such as atrial fibrillation (AFib) and heart failure (HF). Positive airway pressure (PAP) therapy is used to treat OSA, yet patients need to be adherent to realize benefits. Propensity score methods are used to address confounding in retrospective analyses, by balancing observed covariates. We sought to compare outcomes of healthcare resource utilization (HCRU), by PAP adherence, in OSA patients with AFib, systolic HF, and diastolic HF using two propensity score methods, propensity score matching (PSM) and inverse probability weighting (IPW).

METHODS:

US administrative claims data linked to objective PAP usage data was analyzed. Propensity scores were calculated from a logistic regression model predicting non-adherence to PAP therapy from baseline covariates. For PSM, propensity scores were used in caliper matching, with additional exact matching on several covariates. For IPW, weights were calculated from propensity scores and applied to create a weighted pseudo-population that mirrored the distribution of the overall cohort. Mean HCRU visits (emergency room visits and hospitalizations) were compared between adherent and non-adherent patients 1y before and after starting PAP therapy for both PSM and IPW approaches in each cardiovascular cohort.

RESULTS:

The propensity model fit well for all 3 cohorts (McFadden's pseudo R²: AFib 0.04, systolic HF 0.07, diastolic HF 0.08). Both propensity score methods created well balanced groups at baseline. Both methods led to the same conclusions in terms of statistical comparisons between adherent and non-adherent groups for 1y HCRU visits. PSM reduced the sample size by 47%, 34%, and 36% in AFib, systolic HF, and diastolic HF cohorts, respectively.

CONCLUSIONS:

Both propensity score methods generated similar results for associations between PAP adherence and 1y HCRU visits in all cohorts. The reduction in sample size with PSM may impact generalizability of findings.