Postoperative Outcomes in Obstructive Sleep Apnea: Matched Cohort Study

The cohort study by Mutter et al.¹  is the first observational study with large polysomnography data allowing a comparison of clinically important outcomes of patients with undiagnosed obstructive sleep apnea (OSA) and diagnosed OSA (DOSA) versus controls (chosen from the general population) after a wide variety of surgical procedures. For the first time, large polysomnography data may allow answer the questions whether the severity of OSA is related to poor postoperative outcomes and whether the diagnosis of OSA before surgery affects postoperative outcomes.

The conclusions of this study seem to go much beyond what the data suggest and should be viewed in the light of following limitations:

1. First, it is not realistic to conclude that the diagnosis and presumed treatment of OSA in the DOSA group had any bearing on fewer cardiac complications compared with postoperative pulmonary complications in the first place. No details of treatment of OSA in the DOSA group are provided in the article, which, barring measures of adherence and compliance and the use of positive airway pressure therapy during the preoperative period, may be possible to obtain from their database. In the absence of that information, it is only presumptive that diagnosis and possible treatment of OSA were associated with the reduction of postoperative cardiovascular complications and not those of postoperative respiratory complications.

2. The investigators use data from a large polysomnography database (1990–2006) with more than 3,000 patients with polysomnographically confirmed OSA, and they may have been able to find non-OSA matches for a lot of these patients, some from within the polysomnography database (1990–2006) itself and others who tested negative in the general population. However, it is understandable that besides the advantage of having numerous general population controls who were never tested, it was also easier to use the general population as controls. As the authors themselves report as many as 90% of those afflicted by OSA are not yet diagnosed, this methodology introduces a bias of many such population controls having OSA particularly of the mild-to-moderate variety. This could be one of the reasons that the investigators found only severe OSA associated with significant postoperative respiratory and cardiac complications, which has not been shown among most studies till date.^2–4 

3. Given that the authors chose to use general population controls as they were numerous and easy to find, they did not seem to do a good job with matching and adjustment for comorbidities. Also given the total number of postoperative complications, it is quite possible that the regression models are overfitted.

4. Last, in studies reporting postoperative cardiac outcomes among patients with OSA, significant heterogeneity exists in the types of reported cardiac events and it then not surprisingly enough contributes to the difficulty of uncovering the relationship, if any, between OSA and postoperative cardiac events, which in general has been harder to prove even in this study in the DOSA group. By way of example, when used as an International Classification of Diseases-9 diagnosis, it is difficult to believe that acute respiratory distress syndrome has the same connotation as what is meant in the clinical sense where it is based on hemodynamic measurements. Similarly, in the case of this study, it is not clear what the International Classification of Diseases diagnosis of cardiac arrest and shock actually pertains to when reporting postoperative cardiac outcomes. Looking specifically at the results of this study, it is harder to believe that the outcomes of cardiac arrest and shock differ significantly between the OSA groups and controls, whereas those of acute coronary syndrome and atrial fibrillation/flutter, which on an average are more common (and often times, the basis for the more serious event of cardiac arrest and shock), not differ between the two groups. As much the authors do not have any way of explaining this difference, they should probably recognize this important limitation stemming from the use of outcomes based on administrative data in their article. If they believe that such an outcome does have biologic plausibility, then they should at least try to explain the possible mechanisms.

The author declares no competing interests.