Reply to Patel and Althouse: Robust Methods Are Needed to Evaluate the Pharmacologic Treatment of Obstructive Sleep Apnea

AHI is known to vary from night to night (4), restricting an analysis to subjects with the highest AHI on one night of placebo treatment results in a subgroup whose AHI would likely be lower if the subjects were simply treated with placebo for a second night. If the analysis is restricted to a subgroup of patients whose AHI on placebo is an overestimate of their true mean AHI, the effect of atomoxetine–oxybutynin in lowering the AHI compared with placebo will also be overestimated. Regression to the mean can explain why those with the highest AHI on placebo showed not only the greatest difference (atomoxetine–oxybutynin2 placebo) in AHI but also the greatest differences in variables that are correlated with the AHI, such as arousal index, sleep efficiency, and sleep quality. An alternative approach that would provide an unbiased estimate of the true effect of atomoxetine–oxybutynin would be to stratify results on the AHI determined before enrollment rather than on the AHI observed on placebo. The inclusion criteria for this study reported on clinicaltrials.gov include an AHI of .15 events/h, so presumably the authors have an AHI assessment before randomization. Surprisingly, this AHI is not reported in the article and is not used for stratification purposes. This would enable a more valid assessment of whether the response to pharmacologic therapy is greater in patients with more severe OSA. Patients and clinicians eagerly await a pharmacologic treatment for OSA that will be better tolerated than currently available therapies. Despite the hunger for a magic cure, it is important to preserve methodological rigor to ensure that treatments are actually as effective as we say they are. n