This guy stops breathing when he sleeps, do you think it would help if we blow air in his mouth real fast? The SAVE trial and CPAP for sleep apnea.

Seriousy, think about it. Obstructive sleep apnea (OSA) is basically when you spend a third of your life holding your breath every few minutes until your body gets so low on oxygen that you wake up and start breathing again. And what’s our solution? Put your face in this wind tunnel so you can keep breathing. Some things will always be mad science to me. Still, continuous positive airway pressure (CPAP) is well-demonstrated to decrease the apnea-hypopnea index (AHI), the number of times over an hour that patients with OSA stop or nearly stop breathing for at least 10 seconds. Given that sleep apnea has associations with everything from coronary artery disease to a fib to stroke, it just seems like a good idea to stop the apneic episodes however we can.

But patients, in general, hate it. Somewhere between 46-83% of people can’t even manage to use it for four hours a night. And while we’ve tried getting blue in the face telling patients that prospective cohort studies show that CPAP adherence is associated with significantly lower rates of major adverse cardiac events (MACE), there has been a paucity of the holy grail of data: blinded randomized controlled trials. Dr. R Doug McEvoy and his colleagues designed the SAVE trial to be the definitive answer to the question of whether CPAP really decreases adverse cardiac events. If you’re not curious enough to keep reading, hey man, I can’t help you.

If You’re Only Going to Read One Paragraph

In a randomized trial of CPAP vs usual care in patients with moderate-to-severe OSA, CPAP did not produce a significant difference in rates of MACE (HR for CPAP 1.10; 95% confidence interval [CI], 0.91 to 1.32; P=0.34), but did reduce symptoms of OSA as measured by the Epworth Sleepiness Scale (mean change −2.5; 95% CI, −2.8 to −2.2; P<0.001). The major limitation of the trial is the relatively low median adherence to CPAP, though the 3.3 hours per night observed is similar to the average reported in clinical practice. Patients who used their CPAP at least 4 hours had a non-significant trend towards benefit (HR 0.80, CI 0.60 to 1.07, P=0.13). The population was 63% Asian, which probably does not represent your patient panel, and lumping together patients with moderate and severe sleep apnea may obscure a possible benefit among patients with severe OSA (who have higher event rates in observational studies). At the end of the day, everyone can agree that patients with moderate apnea won’t live longer if they use their CPAP less than four hours per night, but there are still unanswered questions about patients who manage better adherence or those with a worse AHI at baseline.

A Bit of Context

I covered most of this above, but we’ll recap quickly. OSA, defined as an AHI of at least 5 (with 15-30 qualifying as moderate and ≥ 30 as severe), has been shown to increase sympathetic activation and produce associated increases in blood pressure as well as inflammatory markers. It is reasonable to assume that this will produce badness when happening for eight hours at a time every night, but we don’t have to assume — the biggest and most recent metanalysis, by Yiqun Fu and colleagues, showed significant independent associations of OSA with both all-cause and cardiovascular mortality:

Compared to the control group, the pooled HR of all-cause mortality was 1.19 (95 % CI, 0.86–1.65) for mild OSA, 1.28 (0.96–1.69) for moderate OSA, and 2.13 (1.68–2.68) for severe OSA.

Note the significantly higher hazard ratio for severe disease as compared to mild or moderate disease — we’re going to talk about that a little later. Other studies have demonstrated associations with endpoints upstream from this one, including CAD, a fib, heart failure, and stroke. This metanalysis and several cohort studies have demonstrated association between receiving CPAP treatment and lower hazard ratios. In this particular study, CPAP treated patients had a hazard ratio for all cause mortality of HR 0.66 (CI 0.59–0.73) compared to OSA patients not on CPAP.

Two prior RCTs have investigated CPAP’s impact on a composite outcome of adverse cardiovascular events. Both a 725 patient study in primary prevention as well as a study of 224 patients who had just undergone coronary revascularization were negative for the primary outcome. However, planned subgroup analyses in both of these trials did demonstrate a benefit in patients who used their CPAP for ≥ 4 hours per night compared to either control patients or those unable to use CPAP for this amount of time.

So, to recap, clear association, no clear proof that CPAP per se helps to ameliorate event rates, but the trials were small and subgroup analysis demonstrated that their might be a threshold effect with regard to CPAP use that could produce a meaningful benefit. The SAVE trial attempted to combat each of these problems individually.

Patients, Intervention, Comparator, and Outcomes

The SAVE trial recruited 3246 patients from 7 countries. They initially shot for 5000 patients to ensure they were powered appropriately given the two prior negative smaller trials. Patients were eligible for inclusion if they were 45 to 75 years old, carried an existing diagnosis of cardiovascular or cerebrovascular disease (definitions of both of which are reasonable), and moderate-to-severe sleep apnea. As opposed to the standard use of the AHI, investigators used the oxygen desaturation index, which measures decrease in oxygen saturation (SpO2) from baseline of at least 4%. This index has been previously validated to correlate almost 1:1 with the AHI. The authors used a cutoff of 12 in an attempt to capture “moderate to severe” sleep apnea patients — with a cutoff of 12 you could imagine some patients on the higher end of the “mild” range sneaking in, but the mean desaturation index was 28, so it seems they probably got the population they were looking for.

Patients were excluded for desaturation to less than 80% SpO2 or Epworth Sleepiness Scale score greater than 15 — effectively people on the sicker end of the apnea spectrum. Cheyne-Stokes breathing was reason for exclusion, as this pattern is unlikely to improve completely with CPAP and so could confound results. Interestingly, patients with NYHA stage III or IV heart failure were also excluded, nominally because of their high pre-test probability of having central apnea.

Given that CPAP only works if you use it, the authors used a 1 week unblinded run-in period in which all patients received subtherapeutic pressure CPAP to assess adherence. Of the initial 3246, 2717 were able to attain the minimal level of CPAP use required for inclusion (3 hours per night). Patients were then randomized 1:1. Patients on CPAP received auto-titration of pressures at home for one week (i.e. the machine increases pressure as needed to prevent apnea) and then were fixed to the 90th percentile of pressure delivered by the machine in automatic mode.

Then, something interesting. Well, if you think power calculations are interesting, which obviously you do I mean come on you’re only human. The authors had initially calculated a need for 5000 patients, but had difficulty recruiting enough patients. Four years into the trial, they unblinded the data and recalculated their power — at that point, they found that an AHI 10 points higher increased the event rate 25-32% in the patients they had accrued to date. On this basis, they calculated they would need only 2500 patients to detect an event rate reduction from CPAP of 25% with 90% power over 4.7 years of follow up. On the one hand, this feels super weird. We don’t often see investigators revising their sample size calculations four years into enrollment (and deciding they already have enough patients). But on the other, this 25% reduction is consistent with the degree of benefit found in prior metanalyses, and it’s not unreasonable to make an informed decision based on your observed event rate rather than the a prior expected rate. Ultimately I think that if the findings had been a non-significant trend towards benefit with CPAP, this would be an issue, but, uhhh, that’s not what happened. Spoilers, I guess.

Going to 7 countries for patients is very impressive and all (and probably necessary to get 5000 patients), but the authors point out that CPAP was “not well established in clinical practice” in many of these countries prior to study commencement. Given our prior issues with multinational trials, I’m glad the authors feel that their “substantial time and effort” prior to the study and good oversight during the trial prevented any misdiagnosis or other monkey business. They did cut all 21 patients at one center out of the study due to concerns for study conduct, which I mean, hey, maybe means they did a good job. There is one more interesting effect of the multinational aspect, which is that ~63% of patients were Asian, with 25% of patients being white and the remainder in the ever-exciting “other” category, with 30 total black patients. Not super representative of the population I care for.

The primary outcome was a composite of cardiovascular death, myocardial infarction, stroke, hospitalization for heart failure, acute coronary syndrome, or transient ischemic attack (TIA). Secondary outcomes included each of the above components plus any revascularization procedures, new-onset atrial fibrillation, new-onset diabetes mellitus, and all-cause mortality. Sleep apnea symptomatic outcomes also served as secondary outcomes.


So it uhh, it didn’t work.


For the primary outcome, hazard ratio for CPAP was 1.10; 95% confidence interval [CI], 0.91 to 1.32; P=0.34. So no difference. There was a decrease in symptoms of OSA as measured by the Epworth Sleepiness Scale (mean change −2.5; 95% CI, −2.8 to −2.2; P<0.001). However, among the remaining secondary endpoints, the only one to reach significance was TIA, which occurred at a rate of 2.29 in the CPAP population compared to placebo; 95% CI, 1.05 to 4.99; P=0.04. It should go without saying that this is probably more noise than signal. Planned subgroup comparisons (including an interesting analysis of patients in China versus at all other centers) revealed no significant differences.

So a negative study. But remember the problems we talked about with the prior studies — yes, the increased sample size should have solved that problem, but what about issues in CPAP adherence? In short, mean adherence was pretty close to what we see in clinical practice, at 3.3 hours per night. That was enough to decrease the overall AHI as planned by the authors’ power calculations — if you decrease the AHI close to zero for that many hours per night, the mean AHI overall decreases to a degree that could reasonably be expected to change the event rate. However, is it clear that 3.3 hours is actually enough to move the needle? Well, there was a planned propensity-matched analysis of those patients who wore CPAP for at least 4 hours per night. 561 of these patients were matched with 561 usual care patients, and here we see a trend towards reduction in event rates with CPAP compared to usual care; the hazard ratio was 0.80; 95% CI, 0.60 to 1.07; P=0.13.

The study is not, of course, powered to pick up benefits in such a small subgroup. I’m not about to re-do their power calculation, but it seems to me that even if the risk reduction were significantly greater in these adherent patients (and it’s not — the 20% observed reduction in rate is less than the 25% they calculated around when re-establishing their sample size), 1100 patients just aren’t going to get you to the point where you can confidently say there is no effect. So once again, we are left to wonder if there truly is a threshold here, and if you could somehow study only patients who used CPAP consistently for 4 hours or more a night, would we still be looking at a negative study?

Last, remember the difference in event rates from the moderate vs severe group we discovered above? Relative risk 1.28 (0.96–1.69) for moderate OSA and 2.13 (1.68–2.68) for severe OSA vs patients without apnea, more than a 50% increase. By recruiting patients with moderate as well as severe disease, the authors were able to accrue many more patients, but we can wonder if patients with a higher AHI might still have a benefit from CPAP even at normal clinical adherence rates. The authors did not break out the severe patients separately in this study or do any subgroup analysis based on baseline AHI.

So how does this change my practice? Well if we assume that adherence to CPAP is normally distributed among the population, a given patient has more or less a coin flip’s chance of using their CPAP enough to encounter a potential, technically unproven benefit. But, I cannot say with confidence that there is no benefit for any patient. So, certainly I will continue to consider the diagnosis of sleep apnea in patients with excessive daytime sleepiness — CPAP can relieve their symptoms and make them feel better. And for patients with witnessed apnea by a partner (a fairly specific sign for OSA), it’s probably worth the sleep study to get a sense of the AHI and the severity of their disease. But what about patients who are simply at risk for sleep apnea and have a high risk for atherosclerotic disease? Well I think that conversation starts with willingness to use CPAP. These patients have to be quite motivated, and willing to modify their lives significantly in order to receive any potential benefit. If a patient isn’t enthusiastic about CPAP or even a sleep study, that probably means they are unlikely to derive clinical benefit even if they do undergo screening.

So for me, in the litany of 100 things to screen for and treat, sleep apnea stays pretty near the bottom of the list. I’m effectively following the USPSTF guidelines that say “Should you screen for sleep apnea? Man, good question.” What are you doing in your practice?

Coming up on IM HEAT

CANVAS came out, so now we’re stuck talking about that. Also EMPA-REG. Really, the SGLT2 inhibitors and a new age of choosing diabetic medications that don’t rhyme with “Januvia.” I… I couldn’t actually think of a word that rhymes with Januvia. Friday! Check it out!


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s