Pooling two trials, stopping beta-blockers years after a heart attack met noninferiority

A large number of people are still taking a beta-blocker years after a heart attack, without a current reason for it. The drug went on at discharge, nobody took it off, and the evidence for continuing it indefinitely in patients whose heart function recovered has never been strong [s1].

A pooled analysis of individual patient data from two randomised trials, published in The Lancet on 30 August alongside ESC Congress 2026, is the largest attempt yet to answer whether it can safely come off [s1]. It is accompanied by a commentary from Borja Ibanez under the title "Stopping beta-blockers long after myocardial infarction" [s2].

What was pooled

The analysis combined individual patient data from the ABYSS and SMART-DECISION randomised trials [s1]. Both enrolled stable patients more than six months after a myocardial infarction, with a left ventricular ejection fraction of at least 40% and no heart failure — that is, patients with no remaining indication for the drug [s1].

Among 6,238 patients (3,698 from ABYSS and 2,540 from SMART-DECISION), 3,092 were assigned to discontinuation and 3,146 to continuation [s1]. The median time from the index myocardial infarction to randomisation was 3.6 years (IQR 1.6 to 7.5), and median follow-up was 3.0 years (IQR 2.3 to 3.8) [s1].

The median of 3.6 years is worth pausing on. This is not a trial about whether to start a beta-blocker after a heart attack, or whether to stop one at three months. It is about patients who have been stable on the drug for years.

Noninferiority was assessed for a composite primary endpoint of all-cause death, myocardial infarction, stroke or hospitalisation for cardiovascular reasons, and for a key secondary endpoint of all-cause death, myocardial infarction or hospitalisation for heart failure [s1]. The analysis used a one-stage mixed-effects Cox model with trial as a random effect, with noninferiority margins of 1.25 for the primary endpoint and 1.40 for the key secondary [s1]. The review protocol was registered with PROSPERO as CRD420261299839 [s1].

What happened

The primary endpoint occurred in 532 of 3,092 patients (17.2%) in the discontinuation group and 500 of 3,146 (15.9%) in the continuation group, a hazard ratio of 1.09 (95% CI 0.97 to 1.24) with a noninferiority p value of 0.016 [s1].

The key secondary endpoint occurred in 201 of 3,092 patients (6.5%) in the discontinuation group and 201 of 3,146 (6.4%) in the continuation group, a hazard ratio of 1.01 (95% CI 0.83 to 1.23) with a noninferiority p value of 0.0006 [s1]. Findings were consistent irrespective of background left ventricular ejection fraction [s1].

Two endpoints, two different kinds of reassurance

The key secondary result is the cleaner one. A hazard ratio of 1.01 for death, myocardial infarction or heart failure hospitalisation is as close to identical as trial data get, and the confidence interval is symmetric around no effect [s1].

The primary result is more awkward, and the authors say so themselves. Their stated interpretation is that discontinuation met the noninferiority criteria for both endpoints, but that interpretation of the primary endpoint is limited by its hospitalisation-heavy and trial-dependent composition [s1].

That is an unusually candid caveat to place in an abstract's interpretation. What it means in practice: the primary composite is dominated by hospitalisation for cardiovascular reasons rather than by hard events, hospitalisation thresholds differ between health systems and between the two trials, and a noninferiority margin of 1.25 applied to such a composite is a weaker test than the same margin applied to death and infarction alone.

The hazard ratio of 1.09 with an upper bound of 1.24 also sits only just inside the 1.25 margin [s1]. Noninferiority was declared, but not with room to spare.

What the analysis does not cover

It covers stable patients more than six months out, with ejection fraction of at least 40% and no heart failure, and no other indication for a beta-blocker [s1]. Patients with reduced ejection fraction, ongoing angina, arrhythmia or hypertension requiring the drug were not the subject of this question.

It pools two trials, not twenty. A one-stage model with trial as a random effect handles between-trial variation statistically, but two trials give little information about how much such variation there is.

And a median follow-up of 3.0 years is short relative to the decades over which these prescriptions persist [s1].

The analysis reports no funding [s1].

What to watch

Whether guideline committees now write a positive recommendation for deprescribing rather than merely withdrawing the recommendation to continue. Those are different acts, and the distinction determines whether anything changes in practice.

The accompanying Lancet commentary is the place where that argument will be joined [s2]. The pooled data give the strongest case so far that beta-blockers can be stopped in this specific, narrow group. The authors' own warning about the primary endpoint's composition is the strongest case for reading the result carefully rather than as a general licence [s1].

This article describes trial results. It is not advice about any medicine, and nothing here should be used to start, stop or change treatment.

Sources

• Discontinuation of beta-blockers in stable patients with previous myocardial infarction, preserved left ventricular ejection fraction, and no heart failure: a pooled analysis of individual patient data, The Lancet, 30 August 2026
• Stopping beta-blockers long after myocardial infarction, The Lancet, 30 August 2026