Statistical and Methodological Reviews for Cardiologists
Reviewer: Robert Long
Position: Principal Consultant
Company: The Data Guru Ltd
Date: 20 April 2025
Document Version: 1.9
The CAPRICORN trial evaluated the efficacy of carvedilol, a third-generation β-blocker, in patients who had experienced a myocardial infarction (MI) and exhibited significant left ventricular (LV) dysfunction (EF ≤40%). Previous β-blocker trials often excluded high-risk populations, leading to uncertainty about their applicability to contemporary clinical practice (Dargie & CAPRICORN Steering Committee, 2000).
CAPRICORN was a randomised, double-blind, placebo-controlled trial that enrolled 1,959 patients between 1997 and 1999. The initial primary endpoint was all-cause mortality, with the study powered at 80% to detect significant differences at an α-level of 0.05 (Dargie & CAPRICORN Steering Committee, 2000). A pre-planned interim analysis was scheduled after 125 deaths to assess early efficacy or futility — a common practice in clinical trials to determine whether a study should continue as planned, be modified, or be stopped early for ethical or scientific reasons.
Midway through the trial, the interim analysis revealed a lower-than-expected mortality rate, raising concerns about the study’s statistical power. Consequently, the primary endpoint was revised to include co-primary endpoints: all-cause mortality (α=0.005) and all-cause mortality or cardiovascular hospital admission (α=0.045). This adjustment effectively split the original α-level between the two endpoints. Owen (2001) criticised this change, suggesting it introduced interpretive bias and compromised the trial’s validity, especially since the mortality result was emphasised despite not meeting the stricter significance threshold.
The original authors responded to Owen (2001) in the same editorial, emphasising the amendment's ethical and practical motivations. They argued that the change was made before unblinding, approved by the ethics committee, and aimed to maintain clinical relevance given the unexpectedly low mortality rate. They also stated that the α-split was transparently incorporated into the revised protocol.
While these arguments have merit, concerns remain. The lack of formal α-spending approaches, such as O’Brien-Fleming boundaries, raises questions about statistical rigour. Furthermore, emphasising the mortality result — despite it not meeting the revised threshold — suggests a disconnect between formal statistical claims and narrative presentation. The rebuttal, though sincere, does not fully alleviate concerns about post-hoc adaptation and selective emphasis.
In clinical trials, the pre-specified α-level should represent the maximum tolerable Type I error — the probability of incorrectly concluding treatment efficacy. Altering the α-level after interim data analysis is problematic, as it can inflate the risk of false-positive conclusions and undermine the integrity of the hypothesis test.
Adaptive changes in trials facing operational challenges, like lower-than-expected event rates, can be acceptable if proper safeguards are employed. These include group sequential designs or α-spending functions, which mathematically preserve the overall Type I error rate. CAPRICORN did not utilise these methods. Although the revised thresholds were agreed upon before unblinding, the revision appears influenced by interim trends, introducing ambiguity about the independence of the statistical plan from emerging results.
Such changes are uncommon in well-powered cardiovascular outcome trials and even rarer without formal statistical correction. CAPRICORN exemplifies how statistical flexibility can blur the line between clinical relevance and methodological rigour.
The CAPRICORN authors cited ethical reasons for the amendment, arguing that continuing to collect mortality data alone, when unlikely to yield definitive answers, would be inappropriate. They claimed the change preserved patient value and avoided unnecessary risk or prolongation.
However, this ethical defence warrants scrutiny. If equipoise had been lost, continuing placebo administration might have been unethical; if equipoise remained, the trial’s planned endpoints should have persisted. This reveals an inconsistency. Moreover, it is unclear whether participants were informed of the amendment or if subsequent consent materials reflected the updated objectives. Transparency to participants is an ethical imperative.
Beyond individual patient protection, trials have an ethical duty to the scientific community: to produce trustworthy, reproducible knowledge. By changing endpoints mid-trial and later highlighting a nominally significant result, CAPRICORN potentially undermined that trust — despite intentions. Ethical conduct encompasses beneficence, respect for persons, and scientific integrity.
For the composite endpoint, results showed no significant benefit (HR = 0.92; 95% CI: 0.80–1.07; p = 0.296). Mortality alone reached nominal significance (HR = 0.77; 95% CI: 0.60–0.98; p = 0.03), failing to meet the revised pre-specified threshold (α = 0.005). Secondary results included reductions in cardiovascular mortality (HR = 0.75; p = 0.024) and non-fatal MI (HR = 0.59; p = 0.014) — both of which were statistically unadjusted and exploratory, warranting cautious interpretation.
Despite methodological criticisms, strong biological substantiation emerged from two substudies. The Echo substudy demonstrated clear beneficial effects on LV remodelling, significantly reducing LV end-diastolic and end-systolic volumes (Pfeffer et al., 2004). Furthermore, McMurray et al.’s (2005) arrhythmia substudy reported marked reductions in malignant ventricular arrhythmias (HR = 0.24; 95% CI: 0.11–0.49; p < 0.0001), strengthening the clinical justification for carvedilol use post-MI.
CAPRICORN’s outcomes align with earlier meta-analyses of β-blockers post-MI (eg., Freemantle et al., 1999), which demonstrated mortality benefit in broader populations. Importantly, CAPRICORN extended this evidence to higher-risk individuals with LV dysfunction. The American Heart Association’s 2001 guidelines referenced CAPRICORN as supporting evidence for carvedilol’s inclusion in post-MI regimens. Historical β-blocker trials such as BHAT and the Norwegian Timolol Study showed mortality reductions in less complex populations, and CAPRICORN importantly demonstrated additive benefit when combined with contemporary therapies such as ACE inhibitors.
The statistical and ethical limitations associated with CAPRICORN’s mid-trial amendments are non-trivial. Clinicians must interpret the reported mortality benefit with appropriate caution. That said, the overall therapeutic narrative for carvedilol is supported by consistent mechanistic data and wider trial evidence. CAPRICORN thus contributes meaningfully to practice — albeit with caveats regarding methodological integrity.
Carvedilol remains a rational choice for post-MI patients with LV dysfunction, underpinned by mechanistic plausibility and external evidence. However, CAPRICORN is a textbook example of why strict adherence to pre-specified statistical analysis plans is critical. Clinical researchers must balance ethics, practicality, and methodological discipline to safeguard credibility.
The pre-specified threshold for statistical significance. Commonly set at 0.05, it represents the probability of making a Type I error — ie., rejecting a true null hypothesis.
The error of falsely declaring a treatment effective when it is not. Also called a “false positive.”
A formal method used in interim analyses to ensure that the cumulative probability of a Type I error does not exceed the pre-specified threshold. Examples include O’Brien-Fleming and Pocock boundaries.
A type of clinical trial design that allows for multiple interim analyses with pre-specified rules for early stopping, while maintaining control of the Type I error rate.
A measure of relative risk over time, used in survival analysis. A hazard ratio less than 1 indicates a lower risk in the treatment group compared to the control.
CAPRICORN Investigators. (2001). Effect of carvedilol on outcome after myocardial infarction in patients with left-ventricular dysfunction: The CAPRICORN randomised trial. The Lancet, 357(9266), 1385–1390. https://doi.org/10.1016/S0140-6736(00)04560-8
Dargie, H. J., & CAPRICORN Steering Committee. (2000). Design and methodology of the CAPRICORN trial: A randomised double-blind placebo-controlled study of the impact of carvedilol on morbidity and mortality in patients with left ventricular dysfunction after myocardial infarction. European Journal of Heart Failure, 2(3), 325–332. https://doi.org/10.1016/S1388-9842(00)00098-2
Freemantle, N., Cleland, J., Young, P., Mason, J., & Harrison, J. (1999). β-blockade after myocardial infarction: Systematic review and meta-regression analysis. BMJ, 318(7200), 1730–1737. https://doi.org/10.1136/bmj.318.7200.1730
McMurray, J. J. V., Køber, L., Robertson, M., Dargie, H. J., Colucci, W., López-Sendón, J., Remme, W. J., Sharpe, D. N., & Ford, I. (2005). Antiarrhythmic effect of carvedilol after acute myocardial infarction: Results of the Carvedilol Post-Infarct Survival Control in Left Ventricular Dysfunction (CAPRICORN) trial. Journal of the American College of Cardiology, 45(4), 525–530. https://doi.org/10.1016/j.jacc.2004.09.076
Owen, A. (2001). Benefit of β-blockers after myocardial infarction [Correspondence]. The Lancet, 358(9291), 1457–1458. https://doi.org/10.1016/S0140-6736(01)06501-1
Pfeffer, M. A., et al. (2004). Prevention of left ventricular remodeling by carvedilol in patients with acute myocardial infarction. Circulation, 109(2), 201–206. https://doi.org/10.1161/01.CIR.0000108928.25690.94