How Many Nights Are Really Needed to Diagnose Obstructive Sleep Apnea?

To the Editor:

We read with great interest the report by Lechat and colleagues (1) on characterizing the prevalence, variability, and diagnostic misclassification of obstructive sleep apnea (OSA) using multinight testing. The authors are to be commended on leveraging observations from the largest community-based sample with home recordings to address an issue of immense clinical relevance. The amassed data are impressive given the number of people included and the volume of nocturnal data used to describe the variability and misclassification of OSA. The authors were indeed crafty in using crowdsourced data from scalable technology and have thus paved the way for future studies that can leverage the ongoing explosive growth in sensors. Without doubt, the report by Lechat and colleagues (1) adds to the accepted notion that one night of monitoring, which is common in clinical decision making, is insufficient to case identify and classify OSA severity. Because the data on OSA diagnosis were derived at home, the issues of variability and misclassification, a phenomenon that is well known with in-lab studies, has been further addressed in the home setting (2).

Despite the many valuable insights, however, their report also raises several issues. First, the terminology used to describe the prevalence, variability, and misclassification uses “OSA” without further qualification. In their methods, the authors state OSA was defined as an apnea–hypopnea index (AHI) of ⩾15 events/h. However, the qualifier, “…at least moderate severity…”, does not consistently permeate the report, particularly with regard to the global estimate of OSA prevalence. It is important to recognize that the estimate of 22.6% is for moderate to severe OSA and not just OSA. This is not a trivial issue, because the prevalence of OSA of any severity will be much higher than 22.6%. In fact, analyses presented in Figure 2 show that data on prevalent mild, moderate, and severe OSA from the contactless sensor are available. Given that the previous validation study (3) and the supplementary data comparing the contactless sensor and polysomnography showed no difference in AHI between the two tests, reporting the prevalence of different OSA categories using AHI thresholds would be of value.

Second, the authors have opted to use the mean AHI of all available nights to calculate the reference AHI against which the reliability of a subset of the nights is compared. It could be easily argued that the median AHI may be a better estimator of central tendency than the mean AHI, particularly if a person has extreme AHI values that may result from factors such as being in the supine position only or consumption of alcohol on any particular night. A possible alternative to the median could be the mode of the AHI distribution from each person. Although we are not proponents, an argument could also be made that the “diagnosis of OSA” should be based on the highest AHI value. Did the authors examine whether the prevalence and misclassification of OSA would be different if the median or mode were used for the reference AHI instead of the mean?

Third, the data on operating characteristics of multinight testing suggest that the increase in positive predictive and the drop in negative predictive values when comparing 7 with 14 nights is relatively small. Thus, what is the minimum number of nights of monitoring necessary to reliably estimate AHI in clinical practice within a ±5% margin of error? Having such information would help change the paradigm of clinical testing in which 1 night is always used despite the capability for multinight testing. It is time that multinight testing became mainstream practice, because the body of empirical evidence on AHI variability is unquestionable (4). One night is just not enough!