Reply to Khoury and Hakim

From the Authors:

We thank Khoury and Hakim for highlighting the clinical relevance and major impact of our study on the basis of the strengths of our real-world sample of patients with obstructive sleep apnea and chronic obstructive pulmonary disease (COPD), the overlap syndrome. The dataset comprises objectively measured positive airway pressure (PAP) therapy and administrative claims data (1). We agree with the authors that our study does have some limitations; however, we would like to comment on several issues they raised.

We expected to see differences in covariates between our comparison cohorts, and therefore, we used multiple statistical analysis methods to control for potential confounding. Although there was still some imbalance in the propensity score matching analysis, the sensitivity analysis we ran using inverse probability of treatment weighting (IPTW) achieved balance on the covariates mentioned by Khoury and Hakim. The balance between the cohorts was assessed using standardized mean differences (SMDs) of baseline covariates; the absolute value of SMDs of less than 0.1 generally indicates negligible difference (2). The IPTW analysis demonstrated the following results: psychotic disorders, 10.8% in the adherent and 11.6% in the nonadherent cohort (SMD = −0.03); morbid obesity, 40.7% in the adherent and 39.7% in the nonadherent cohort (SMD = 0.02); obesity not listed, 31.4% in the adherent and 32% in the nonadherent cohort (SMD = −0.01); and tobacco use, 47.4% in the adherent and 47.6% in the nonadherent cohort (SMD = 0.0).

As previously stated in our study, results from the IPTW sensitivity analysis were consistent with the propensity score analysis results, showing increased hospitalizations, emergency room visits, and costs for overlap patients not adherent to PAP. Therefore, we believe the covariates mentioned by Khoury and Hakim are unlikely to explain major differences observed in healthcare resource usage between adherent and nonadherent patients.

Khoury and Hakim incorrectly stated that we “[excluded] previous COPD exacerbations in patients with COPD as inclusion criteria”. Although we did not specifically create a covariate for prior COPD exacerbations, the claims-based algorithm adapted from Mapel and colleagues used for the number of “2-year severe acute exacerbation model” includes the number of respiratory-related hospitalizations and COPD-related emergency room visits (3). On the basis of this criterion, most prior COPD exacerbations are captured in prior hospitalizations and emergency room visits. Thus, the inclusion of this specific variable would have caused multicollinearity issues and violated statistical modeling principles.

As discussed extensively in the limitations, we acknowledge that we lack laboratory data, and therefore, we were unable to characterize patients as hypercapnic or nonhypercapnic on the basis of concentrations of Pa_CO2 from an arterial blood gas draw. We speculate that our observed PAP benefits would be particularly pronounced in patients who are hypercapnic (4, 5). Of note, in our study, 85% of patients used automatic PAP or continuous PAP mode, and 15% used bilevel modes, predominantly spontaneous mode (i.e., without a backup rate). None of the patients in our analytic cohort used other modes of noninvasive ventilation. Future research should assess the impact of hypercapnia in overlap syndrome and the potential benefits of bilevel PAP over automatic/continuous PAP in these patients.

Although we are supportive of further randomized clinical trials, challenges regarding baseline differences in covariates and healthy user effect can still be an issue in this context. Moreover, multicentered randomized clinical trials are logistically challenging and are unlikely to be conducted in as large of sample sizes as our present analyses. Nonetheless, we hope our new data encourage further rigorous research in overlap syndrome. Until more definitive data are available, we believe our data may be clinically directive because they are consistent with the existing mechanistic and clinical literature (6).