The Challenges of Estimating Causal Effects of Continuous Positive Airway Pressure Therapy from Observational Data

To the Editor:

We read with great interest the recent work by Gervès-Pinquié and colleagues seeking to estimate the causal effect of continuous positive airway pressure (CPAP) on long-term outcomes among patients with obstructive sleep apnea (1). Obtaining unbiased causal estimates from observational data is challenging, and Gervès-Pinquié and colleagues are to be congratulated for their efforts. Nevertheless, we had several concerns about interpreting the findings from this work.

First, the authors identified a stronger impact of CPAP on lowering the relative risk of major adverse cardiovascular events (MACEs) in those without preexisting cardiovascular disease compared with those with prior history of cardiovascular disease. In considering the clinical implications of this finding, it is important to recognize that the potential benefit of CPAP therapy is based on the absolute risk reduction rather than the relative risk reduction. Given that those with a prior history of cardiovascular disease are at a much higher baseline risk, a smaller relative risk reduction can still translate into a larger absolute risk reduction. It would be helpful if the authors could estimate the absolute risk reduction (with confidence intervals) for MACEs to be obtained from CPAP therapy in the primary and secondary prevention subgroups.

Second, while we congratulate the authors for reporting E-values, we disagree with their interpretation. E-values estimate the necessary effect size that unmeasured confounders would need to explain a given finding’s point estimate and upper confidence interval (2). For the comparison of CPAP use of 7 hours or more relative to less than 4 hours in Table 2 of their manuscript, Gervès-Pinquié and colleagues estimate an impact on incident MACEs with a relative risk of 0.78 with an upper confidence interval of 0.93, translating to E-values of 1.88 and 1.36, respectively (1). Based on these estimates, the authors conclude there is a “low risk” that their findings stem from residual confounding. However, the healthy adherer effect is known to impact mortality with effect sizes in this range. In a meta-analysis of eight randomized trials, nonadherence to placebo medications was associated with a 1.79-fold increase in mortality (3). While the authors accounted for adherence to common medications in their analyses, concern for residual confounding remains. Adherence to medications does not fully predict CPAP adherence (4), and residual confounding by the healthy adherer effect is difficult to address. For instance, even after accounting for demographics, socioeconomic status, comorbidities, and healthy behaviors such as physical activity and diet, randomized trials in both women and men have estimated the risk of mortality is 50% greater in those who are nonadherent to placebo medications compared to those who are adherent (5, 6). Further research is needed to better understand how CPAP adherence is related to the general healthy adherer effect and to develop robust methods to account for these effects. Until then, efforts need to continue to develop feasible strategies for the conduct of long-term randomized trials of CPAP in patients with obstructive sleep apnea.