One of the open questions about GLP-1 drugs is whether they are a good idea in people who are already frail — older, carrying several conditions, less able to absorb a period of appetite loss. A new analysis of the SELECT trial, published in JAMA Cardiology on 16 September 2026, sorted all 17,604 participants by a 31-item frailty index and looked again at who benefited. The short answer: the heart benefit did not detectably vary by frailty, the quality-of-life benefit was largest in the frailest, and people on semaglutide were more likely to become less frail over two years.
How the analysis was built
SELECT enrolled adults with a BMI of 27 or higher and established cardiovascular disease but no diabetes, randomising them to semaglutide 2.4 mg weekly or placebo. The participants averaged 61.6 years old and 72.3% were men. For this secondary analysis the authors built a frailty index using the Rockwood cumulative-deficit method, counting how many of 31 possible health deficits each person had at baseline, then split the trial three ways: 5,432 people (31%) not frail, 8,349 (47%) more frail, and 3,823 (22%) most frail.
Frailer participants had more cardiovascular events, as expected. The question was whether semaglutide worked differently in them. For the primary composite of cardiovascular death, non-fatal heart attack and non-fatal stroke, the hazard ratios were 0.84 (95% CI 0.65 to 1.07) in the not-frail group, 0.70 (0.59 to 0.82) in the more-frail group and 0.92 (0.76 to 1.10) in the most-frail group, with a P value for interaction of 0.09. It is worth being careful about how to read that. Two of those three intervals cross 1, and the middle one does not — but that pattern is what you expect when you cut a trial into smaller pieces and each piece has fewer events. The interaction test, which asks directly whether the effect differs across groups, did not find that it does, and neither did the version treating frailty as a continuous scale (P = 0.30). The honest reading is no detectable difference, not a benefit confined to the middle group.
The frailty findings themselves
Three results point the other way from the worry that prompted the analysis. Benefits on the heart-failure composite, on all-cause hospitalisation and on all-cause mortality were also consistent across frailty categories. Quality of life, measured on the EQ-5D-5L, improved more in frailer participants than in fitter ones (P = 0.02 for interaction) — the one place where frailty did change the answer, and in the direction of more benefit, not less. And between baseline and week 104, people on semaglutide were more likely to move into a less frail category (odds ratio 2.46, 95% CI 1.80 to 3.37) and less likely to move into a worse one (odds ratio 0.47, 0.34 to 0.65).
Tolerability ran the same way. The hazard ratios for adverse events leading to permanent discontinuation were lower in participants with higher baseline frailty (P < 0.001 for interaction), which is the opposite of the pattern clinicians tend to expect.
What is not yet known
This is a post hoc analysis, decided on after the trial finished, and the frailty index was reconstructed from data SELECT collected for other reasons rather than measured with a purpose-built instrument. A cumulative-deficit index is not the same as a clinician judging someone frail at the bedside. The trial itself constrains what this can say: everyone had established cardiovascular disease, nobody had diabetes, and 72% were men, so this is not a general statement about frail older adults, and least of all about the very old — the average participant was 61. Subgroup intervals are wide. SELECT was funded by Novo Nordisk and several co-authors are company employees. And an analysis showing that frailty scores improved is not the same as showing that people functioned better in the ways that make frailty matter.
Sources: Ostrominski JW, Plutzky J, Scirica BM, et al., for the SELECT Trial Investigators. “Efficacy and Safety of Semaglutide According to Frailty Status: A Post Hoc Analysis of the SELECT Randomized Clinical Trial.” JAMA Cardiology, 16 September 2026, doi:10.1001/jamacardio.2026.3566. Trial registration ClinicalTrials.gov NCT03574597.
