How can a trial say you are both more and less likely to die after an intervention?

Short answer: Composite endpoints

Medium answer: Many clinical trials report composite outcomes such as: “stroke, acute coronary syndrome (acute myocardial infarction and hospitalization for unstable angina), acute decompensated heart failure, coronary revascularization, atrial fibrillation, or death from cardiovascular causes.” [1] In this particular trial evaluating blood pressure control regimens in 9,624 Chinese adults, the primary composite outcome favored more aggressive control (hazard ratio [HR] 0.74, 95% CI 0.60-0.92, p = 0.007). However, all-cause mortality was actually increased in the intense treatment group, although not significantly so (HR 1.11, 95% CI 0.78-1.56). While not a particularly striking example (I just clicked on a random recent paper in NEJM), one can imagine this being very confusing to patients. To paraphrase a hypothetical scenario from a review on composite outcomes in the BMJ:

“Mr. Chen, here is a drug that will reduce your combined risk of getting a heart attack that will not kill you, stroke, specific arrythmias, or of dying from heart disease.”

“Doctor, I am not sure I quite understood you, but please give me this drug.”

“But I should also mention that the drug will increase your risk of dying.”

“Didn’t you just tell me that the drug would decrease my risk of dying? I am confused.”

Adapted from Cordoba et al. [2]

Full Answer: The American Heart Association meeting took place from 11/16/21-11/18/21 and there has been a rush of medical podcasts based on the content from this meeting. Full disclosure: as an aspiring ophthalmologist, my working knowledge of body medicine is to “keep the heart full of blood, keep the lung full of air.” Nonetheless, while skimming some of the papers from the AHA meeting, I was once again reminded of the preponderance of composite outcomes in clinical trials.

In particular, I was thinking about a paper I presented at rounds a few weeks ago, EMPEROR-Preserved, (Empagliflozin in Heart Failure with a Preserved Ejection Fraction), published in NEJM in August 2021. Empagliflozin (Jardiance) an SGLT-2 inhibitor, has already been shown in the EMPEROR-reduced trial to have benefit in all HFrEF patients regardless of type 2 diabetes (T2DM) status. This double-blind trial compared empagliflozin 10mg vs. placebo (in addition to usual therapy) in approximately 6000 patients with HFpEF (EF >= 40, NYHA II-IV) assessing for a primary outcome of cardiovascular death or hospitalization for HF. About 50% of patients in both groups had T2DM at baseline. Primary outcome in 13.8% in empagliflozin group vs 17.1% in placebo group (HR 0.79, NNT = 31). This benefit was seen in patients with and without diabetes (Table 2). The benefit was driven by decreased hospitalizations (8.6% vs 11.8%, HR 0.71), but no benefit was seen for cardiovascular death (7.3% vs 8.2%, HR 0.91 [CI 95% 0.76-10.9]). Adverse effects of UTI and genital infections are more common in the empagliflozin group. For this group of patients in which there have been so few effective treatments, SGLT-2 inhibitors are certainly a game-changer. However, I couldn’t help but think that the trial would have been even more convincing if it had been powered to detect benefit in either cardiovascular or all-cause mortality alone.

The question of composite outcomes has plagued the field for a long time. For example, the recent ASCNED trial showed that aspirin was not shown to significantly decrease the risk of developing dementia in diabetics. This work reminded me of some of the original work on aspirin for primary prevention of CV events in diabetics (Ogawa et al. JAMA 2008). Cordoba et al. singled out this study for particularly egregious abuse of composite outcomes:

Because components can be combined in so many ways, it is easy to find significant results. In one of the trials we included (Ogawa et al.), the composite consisted of eight cardiovascular end points, but there were also secondary composites that consisted of “combinations of primary end points as well as death from any cause.” These combinations were not specified, but nine end points can be combined, as two or more components, in 502 possible ways (29−1(empty sample)−9(samples with only one component)). The result for the composite was not statistically significant, but the abstract noted that the hazard ratio was 0.10 for a combined end point of fatal coronary events and fatal cerebrovascular events (P=0.0037)—that is, a cherry picked result. One would expect 25 of 502 possible combinations to be significant purely by chance.

Unsurprisingly, more modern evidence has suggested aspirin may be of much more limited benefit than described in this study for primary prevention (note: this is not true for secondary prevention – don’t stop taking aspirin without asking your doctor!) The US Preventive Services Task Force no longer routinely recommends aspirin in patients younger than 50 or older than 60. Similarly, AHA guidelines say to not be given routinely to adults over 70 or at increased risk of bleeding (diabetes or not).

As another recent example highlighted by Dr. Mandrola in This Week in Cardiology, the AVATAR trial asked whether surgical aortic valve replacement is indicated for asymptomatic patients with severe aortic stenosis by ultrasound criteria. 157 patients were randomized in this trial to early surgery vs conservative treatment with a median follow-up of 35 months. The primary outcome was a composite of all-cause mortality, acute MI, stroke, and hospitalization for heart failure. The primary composite outcome favored early surgery (HR 0.46, 95% CI 0.23-0.90, p = 0.02). However, cardiovascular death did not differ between the two groups (HR 1.02, CI 0.40-2.58), while all-cause mortality strongly favored early surgery (HR 0.56, CI 0.24-1.27, 16 vs. 9 patients.). Turns out, 3 patients in the conservative management group died of COVID-19 – if two of these patients had been in the early surgery arm, the trial would not have been significant.

Some final thoughts on composite outcomes:

  1. Studies need to be better powered. Clinical trials are experiments on human beings! If we are going to do that, we must strive to get a clear answer on a meaningful clinical question.
  2. Studies should assess composite endpoints with fewer components (corollary of running better powered studies). Dr. Gregg Stone (EXCEL trial lead discussed below) even acknowledges the importance of doing so in a panel last year.
  3. If composite endpoints are necessary, studies should seek to ensure components of composite outcomes are both objective, clinically relevant, and of equal importance. This means avoiding lumping death together with outcomes like silent MIs, TIA, or doubling of baseline creatinine. It also means carefully considering if all-cause mortality, cardiovascular mortality, or cardiovascular mortality in patients proven to be COVID-19 positive is the more appropriate outcome depending on the clinical context.
  4. We need to do a better job of communicating outcomes of trials to patients. This goes beyond simply explaining what composite outcomes mean – we should also consider making explicit the value judgments baked into things like:
    • unequal impact of components of composite outcomes on patients’ quality of lives
    • non-inferiority margins (What percent difference do we consider non-inferior?)
    • number needed to treat (What our threshold for an NNT small enough that we would recommend an intervention?)

For those interested in further reading – I’d recommend persuing the controversy around the EXCEL trial (Stone et al. NEJM 2016) and the recent meta-analysis of the evidence regarding PCI vs CABG (Sabatine et al. Lancet, 2021). This debate between surgeons and interventionalists highlights the challenge of interpreting trials in which the results of the primary composite outcome and all-cause mortality conflict. I may eventually write a separate post about this whole saga – but I feel like it is already well-contextualized/discussed in great detail in the linked articles.

If anybody is out there reading, I’d love to hear your thoughts on this. As an early-career trainee, I obviously haven’t been involved in actually designing these kinds of trials – and I’m always striving to learn how to more critically evaluate evidence.


  1. Zhang W, Zhang S, Deng Y, et al. Trial of Intensive Blood-Pressure Control in Older Patients with Hypertension. N Engl J Med. 2021;385(14):1268-1279. doi:10.1056/NEJMoa2111437
  2. Cordoba G, Schwartz L, Woloshin S, Bae H, Gøtzsche PC. Definition, reporting, and interpretation of composite outcomes in clinical trials: systematic review. BMJ. 2010;341:c3920. Published 2010 Aug 18. doi:10.1136/bmj.c3920
  4. Anker SD, Butler J, Filippatos G, et al. Empagliflozin in Heart Failure with a Preserved Ejection Fraction. N Engl J Med. 2021;385(16):1451-1461. doi:10.1056/NEJMoa2107038
  5. Packer M, Anker SD, Butler J, et al. Cardiovascular and Renal Outcomes with Empagliflozin in Heart Failure. N Engl J Med. 2020;383(15):1413-1424. doi:10.1056/NEJMoa2022190
  7. Ogawa H, Nakayama M, Morimoto T, et al. Low-dose aspirin for primary prevention of atherosclerotic events in patients with type 2 diabetes: a randomized controlled trial [published correction appears in JAMA. 2009 May 13;301(18):1882] [published correction appears in JAMA. 2012 Nov 14;308(18):1861]. JAMA. 2008;300(18):2134-2141. doi:10.1001/jama.2008.623
  10. Banovic et al. Aortic Valve ReplAcemenT versus Conservative Treatment in Asymptomatic SeveRe Aortic Stenosis: The AVATAR Trial. Preprint in Circulation.
  12. Stone GW, Sabik JF, Serruys PW, et al. Everolimus-Eluting Stents or Bypass Surgery for Left Main Coronary Artery Disease [published correction appears in N Engl J Med. 2019 Oct 31;381(18):1789]. N Engl J Med. 2016;375(23):2223-2235. doi:10.1056/NEJMoa1610227
  14. Sabatine MS, Bergmark BA, Murphy SA, et al. Percutaneous coronary intervention with drug-eluting stents versus coronary artery bypass grafting in left main coronary artery disease: an individual patient data meta-analysis [published online ahead of print, 2021 Nov 12]. Lancet. 2021;S0140-6736(21)02334-5. doi:10.1016/S0140-6736(21)02334-5

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

%d bloggers like this: