To the Editor:
In the September 2019 edition of CHEST, Javaheri et al1 proposed changes to the design of randomized controlled trials (RCTs) exploring the relation between CPAP and prevention of cardiovascular disease (CVD). The authors attributed the negative findings of the Sleep Apnea Cardiovascular Endpoints (SAVE) study2 and other trials in this area to limitations of patient selection, choice of end points, inadequate sleep monitoring, low CPAP adherence, and underpowered samples. These negative RCTs generally recruited patients from CVD clinics and excluded individuals with excessive sleepiness and severe hypoxemia. Javaheri et al proposed a redesigned RCT with a larger sample (N = 16,000-24,000) recruited from sleep centers and a more complex design in which all patients receive stimulant therapy to overcome the ethical dilemma of randomizing sleepy patients to no treatment.
There are compelling reasons to include excessively sleepy patients in future trials. A recent study of clinical subtypes of OSA observed that the increased risk of CVD events was almost exclusively in an excessively sleepy subtype.3 However, randomizing a large sample of excessively sleepy patients to no effective CPAP therapy is ethically questionable because CPAP reduces sleepiness and risk of accidents. There is insufficient evidence that stimulant therapy adequately ameliorates this risk.
We propose an alternate design that includes excessively sleepy patients from sleep centers and addresses this question more efficiently, cost-effectively, and ethically: an observational cohort study using a rigorous two-stage propensity score (PS) design capable of providing valid causal estimates of treatment effects.4 The PS is formally defined as the conditional probability of being treated (eg, CPAP adherent) vs untreated (eg, a nonuser) as a function of baseline covariates. PS methodologies have been used extensively in CVD research.5 The first design stage is similar to that of an RCT, including specification of hypotheses and power analyses, with additional focus on identifying a rich set of baseline covariates associated with CVD outcomes that may differ between those with and without adequate CPAP adherence. Once CPAP adherence is evaluable in all subjects, a second design stage implementing the PS methodology is completed, blinded to CVD outcomes to avoid the bias associated with traditional covariate adjustment. Subclassification, matching, and weighting are PS design approaches capable of achieving at least as much covariate balance as an RCT, effectively reducing bias and improving causal estimates.
About 30% to 40% of patients with severe OSA do not accept CPAP, creating an ideal opportunity for a natural comparative effectiveness study on CVD outcome prevention in CPAP users vs nonusers. This approach is appealing for several reasons: (1) patients attending sleep centers routinely undergo polysomnography and expert review to diagnose OSA accurately; (2) there are no ethical issues around including excessively sleepy patients; (3) comparisons are made between patients who adhere to CPAP and those who fail to accept or adhere to CPAP; (4) the study findings are clinically applicable because the sample is representative of typical clinical patients; (5) because most procedures are routine, the study is less costly than an RCT, facilitating larger sample sizes; and (6) the study could be conducted over a considerably longer duration than an RCT.
Thus, longer term observational studies of incident CVD among clinical cohorts from sleep centers are an effective, alternative approach to examining the role of CPAP in preventing CVD events. Rigorous PS design approaches that go beyond traditional covariate adjustment can create balance in baseline CVD covariates between patients adherent and nonadherent to CPAP, ensuring that outcome differences are attributable to CPAP. The immediate challenge is to define a sufficiently rich set of covariates on which to create balance. Although important unrecognized and unmeasured covariates could exist, current approaches allow estimation of the unobserved effect sizes required to nullify conclusions. Ultimately, application of the proposed PS design allows valid causal estimates of the “real-world” benefit of CPAP on CVD outcomes.
Acknowledgments
∗Sleep Apnea Global Interdisciplinary Consortium collaborators: The authors represent the Sleep Apnea Global Interdisciplinary Consortium (SAGIC) (https://www.med.upenn.edu/sleepctr/sagic.html) which is a collaboration of 10 sites, conducting research projects worldwide on a variety of topics related to obstructive sleep apnea. The 10 sites and members at each site include: University of Pennsylvania (Allan I. Pack, Richard Schwab, Diane C. Lim, Greg Maislin, Brendan T. Keenan, Olivia J. Veatch, Diego Mazzotti, Mary Boland, Francis Pack, Jinyoung Kim, now at the University of Nevada, Las Vegas), Ohio State University (Ulysses J. Magalang, Jesse Mindel, M. Melanie Lyons, Steven Holfinger, Samantha Rojas), University of Iceland (Thorarinn Gislason, BryndSs BenediktsdLttir), Charite Universitatsmedizin Berlin (Thomas Penzel, Bernd Sanner, Ingo Fietze, Maria Franczyk, Naima Laharnar, Hua Qin), Peking University (Fang Han, Adele Liyue Xu, Jing Jing Guo), Shanghai University (Qing Yun Li, Yingni Lin), Chang Gung Memorial Hospital (Ning-hung Chen, Li-Pang Chuang, Yu-Sheng Lin, Shih-Wei Lin, Hung-Yu Huang), Korea University (Chol Shin, Seung Ku Lee), University of Sydney (Peter A. Cistulli, Philip deChazal, Kate Sutherland), University of Western Australia (Bhajan Singh, Nigel McArdle, Peter Eastwood).
Footnotes
FINANCIAL/NONFINANCIAL DISCLOSURES: None declared.
Contributor Information
Bhajan Singh, Email: Bhajan.Singh@health.wa.gov.au.
Sleep Apnea Global Interdisciplinary Consortium:
Allan I. Pack, Richard Schwab, Diane C. Lim, Greg Maislin, Brendan T. Keenan, Olivia J. Veatch, Diego Mazzotti, Mary Boland, Francis Pack, Jinyoung Kim, Ulysses J. Magalang, Jesse Mindel, M. Melanie Lyons, Steven Holfinger, Samantha Rojas, Thorarinn Gislason, BryndSs BenediktsdLttir, Thomas Penzel, Bernd Sanner, Ingo Fietze, Maria Franczyk, Naima Laharnar, Hua Qin, Fang Han, Adele Liyue Xu, Jing Jing Guo, Qing Yun Li, Yingni Lin, Ning-hung Chen, Li-Pang Chuang, Yu-Sheng Lin, Shih-Wei Lin, Hung-Yu Huang, Chol Shin, Seung Ku Lee, Peter A. Cistulli, Philip deChazal, Kate Sutherland, Bhajan Singh, Nigel McArdle, and Peter Eastwood
References
- 1.Javaheri S., Martinez-Garcia M.A., Campos-Rodriguez F. CPAP treatment and cardiovascular prevention: we need to change the design and implementation of our trials. Chest. 2019;156(3):431–437. doi: 10.1016/j.chest.2019.04.092. [DOI] [PubMed] [Google Scholar]
- 2.McEvoy R.D., Antic N.A., Heeley E., et al. CPAP for prevention of cardiovascular events in obstructive sleep apnea. N Engl J Med. 2016;375(10):919–931. doi: 10.1056/NEJMoa1606599. [DOI] [PubMed] [Google Scholar]
- 3.Mazzotti D.R., Keenan B.T., Lim D.C., Gottlieb D.J., Kim J., Pack A.I. Symptom subtypes of obstructive sleep apnea predict incidence of cardiovascular outcomes. Am J Respir Crit Care Med. 2019;200(4):493–506. doi: 10.1164/rccm.201808-1509OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Rubin D.B. Using propensity scores to help design observational studies: application to the tobacco litigation. Health Services Outcomes Res Method. 2001;2:169–188. [Google Scholar]
- 5.Deb S., Austin P.C., Tu J.V., et al. A review of propensity-score methods and their use in cardiovascular research. Can J Cardiol. 2016;32(2):259–265. doi: 10.1016/j.cjca.2015.05.015. [DOI] [PubMed] [Google Scholar]