Is obstructive sleep apnea a risk factor for hypertension?--differences between the Wisconsin Sleep Cohort and the Sleep Heart Health Study.
P. Peppard
DOI: https://doi.org/10.5664/JCSM.27592
2009-10-15
Abstract:Results of CPAP intervention trials examining changes in blood pressure following obstructive sleep apnea (OSA) treatment give an overall impression that CPAP remediation of OSA can be expected to result, on average, in a small but non-trivial reduction in systemic blood pressure (BP)—on the order of a few mm Hg.1,2 However, while CPAP trials can address clinically relevant questions regarding the effectiveness of OSA treatment, they are not ideal for addressing the entire pathophysiologic role of OSA in the elevation of BP. Reasons for this include: OSA-related elevations in BP may not be entirely reversible; trials may need to run longer than weeks-to-months to achieve the potential impact of CPAP; and less than complete CPAP compliance is the norm. That leaves observational studies the task of attempting to measure the “full” role of OSA in elevated BP. To date, two large observational longitudinal studies based on non-clinic populations have attempted to measure the relation of baseline OSA severity and the risk of subsequent development of hypertension in cohorts initially free of hypertension over a 2- to 5-year follow-up period. These are reports published this year from the Sleep Heart Health Study (SHHS)3 and from our analyses of the Wisconsin Sleep Cohort Study (WSCS)4,5 in 2000 (WSCS results discussed here will refer primarily to those presented in citation 5). Both the WSCS and SHHS reported elevated risks of developing hypertension among subjects with the highest severity categories of baseline OSA compared to subjects without OSA. However, the SHHS associations were smallish and not statistically significant (SHHS subjects with an apnea-hypopnea index ≥ 30 events/hr had an adjusted 1.5-fold increased odds of developing hypertension compared to subjects without OSA at baseline) whereas the WSCS association was moderate and statistically significant (subjects with an apnea-hypopnea index ≥ 15 events/hr had a 3.2-fold increased odds of developing hypertension relative to subjects without OSA at baseline). I have been asked to provide my perspective on this apparent discrepancy.
In my view, there is a straightforward, if superficial, two-fold explanation for these “discrepant” results. First, considering chance effects, both cohort studies are consistent with small-to-modest OSA-BP association implied by CPAP intervention trials. Second, the studies were built on different infrastructures which underlie varying susceptibilities to potential selection, measurement, and confounding biases. Notably, the study populations were quite distinct: the SHHS sample (N = 2470) was older, geographically and somewhat racially diverse, and sampled from a number of “parent” cardiovascular cohort studies; the WSCS sample (N = 641) was selected from a relatively homogenous employed population. Furthermore, substantially different OSA-assessment techniques were employed: SHHS used unattended in-home polysomnography; WSCS used attended in-laboratory polysomnography. Thus, it would be surprising if both studies had arrived at precisely convergent conclusions. Nevertheless, some observers may feel further explanation is required for the weaker nonsignificant positive associations from SHHS compared to the WSCS. It is difficult to predict how all of the differences in the WSCS and SHHS study designs would impact their findings. However, I will describe three factors that would be expected to yield systematically elevated associations in the WSCS compared to the SHHS.
First, the elevated odds of developing hypertension for persons with increasing categories of OSA severity were calculated relative to different definitions of baseline “no OSA” groups in SHHS and the WSCS. The baseline comparison category for SHHS was an apnea hypopnea index (AHI) < 5 events/hr, and for the WSCS it was AHI = 0. I have calculated that the magnitude of the estimated associations from the WSCS would have been diminished by approximately 9% (i.e., moved closer to the SHHS estimates) on the scale of the odds ratio had the WSCS used the same baseline “no OSA” definition as the SHHS.
Second, due to differences in OSA measurement methods and scoring, the SHHS and WSCS are likely to have measured the AHI using, essentially, different scales. The WSCS had approximately half the prevalence of AHI ≥ 15 events/hr at baseline (7.5% vs. 13.4% in the SHHS) and less than half the median AHI. These large differences seem unlikely to be explained simply by varying population demographics: although the WSCS population was, on average, younger (47 years vs. 60 years in the SHHS), the WSCS was heavier (baseline BMI = 29 kg/m2 vs. 28 in the SHHS) and disproportionately male (56% men vs. 45% in SHHS). More likely, a “unit of AHI” as assessed by in-home polysomnography in the SHHS does not directly correspond to that assessed by in-laboratory polysomnography in the WSCS. We demonstrated this discrepancy in a subset of 34 WSCS participants who had in-home polysomnography using the same equipment (Compumedics, P-series) as used in the SHHS. When comparing in-lab to in-home polysomnography (performed within a 1-year period) in these WSCS participants, the average in-home AHIs were more than 2-fold higher than the in-lab measurements (mean in-lab AHI = 3.4 events/hr, mean in-home AHI = 8.9 events/hr). The reasons for these differences are likely numerous, and the point here is not that either SHHS or WSCS evaluated AHI the “wrong” way (the correlation between the two methods was r = 0.91, indicating that these measures provide essentially the same information)—only that the studies are likely to be using different OSA severity metrics even if they were both labeled “AHI.” Importantly, the WSCS metric would likely yield larger odds ratios (i.e., a unit increment in WSCS' AHI traverses a larger portion of the OSA pathophysiologic spectrum than a SHHS unit of AHI). In particular, the highest severity categories for the WSCS (AHI ≥ 15 events/hr) is more comparable to the SHHS severity category of AHI ≥ 30 than to the SHHS severity category of 15 ≤ AHI < 30 events/hr.
Third, in order to perform an analysis of incident hypertension, baseline hypertensive subjects in the SHHS were excluded from follow-up. This was 51% of the otherwise eligible SHHS sample. The analogous percentage for the WSCS was 27%. Thus, if: 1) OSA is a cause of elevated BP, and, 2) there is variability among persons in the susceptibility to developing hypertension due to OSA, then the SHHS would have excluded a greater proportion of subjects at baseline who were relatively more susceptible to developing hypertension due to OSA compared to the WSCS. This might be viewed as a form of “survivor bias.”6,7 The authors of the SHHS study allude to this possibility noting that “sleep-disordered breathing may have a relatively rapid effect to raise blood pressure but relatively little impact on future hypertension risk after excluding persons who are normotensive despite sleep-disordered breathing at baseline.” (citation 3, p. 1163). While some putative mechanisms for an OSA-hypertension association (e.g., sympathetic activation) may produce such a “relatively rapid effect,” a survivor bias can also result from longer term effects of OSA on BP—a variation among individuals in susceptibility to OSA effects on BP is primarily what is required. I would accept such variation in susceptibility as a given as I am unaware of any investigation, whether observational or highly controlled experiment (human or otherwise), that has failed to show appreciable interindividual variation in effects of pressors (or antihypertensive therapies). Any cohort study that had subjects excluded due to prevalent hypertension at baseline might be subject to this association-attenuating “bias,” and the 51% exclusion of baseline hypertensives for SHHS vs. 27% for WSCS makes stark the potential relative impact on the two studies.
With available data it is difficult to estimate the combined impact of these three factors, but I suspect that together they substantially contribute to the superficially discrepant results for the most severe OSA categories, and chance effects would readily explain any residual discrepancy. It is also notable that, in addition to the comparison of the 2009 SHHS study and the 2000 WSCS study, there is another analysis that bears comparison with the 2009 SHHS study: a large cross-sectional analysis of the association of OSA and prevalent hypertension published in 2000 also from the SHHS.8 That earlier SHHS article has been widely cited as evidence of an OSA-BP association, reporting a statistically significant 1.37-fold increased odds of hypertension in persons with an AHI ≥ 30 events/hr relative to no OSA. It is difficult to see the results of the 2009 SHHS report (with an adjusted odds ratio of 1.50 for AHI ≥ 30 vs. no OSA—a 35% increase in estimated effect size) as anything but an affirmation of the earlier cross-sectional SHHS finding, at least for more severe OSA.
Unfortunately, despite the strong implication of a modest causal association between OSA and BP from well-designed observational and experimental studies, there may not yet be clear evidence to guide the clinician in the utility of treating mild to moderate OSA if blood pressure reduction is the primary goal of such treatment. However, given the high (and likely growing) prevalence of OSA, the population-level impact of a modest SDB-BP association on cardiovascular outcomes is likely substantial, and elevated BP may be only one of several mechanisms by which OSA increases cardiovascular risk.9