GLP-1 Response Differences in Three Trials

A 2026 exploratory reanalysis found little evidence that routinely measured baseline characteristics produce large differences in response to GLP-1 receptor agonists among participants from three cardiovascular outcome trials. Baseline hemoglobin A1c ranked highest as a possible effect modifier, and kidney function ranked next in one relative-risk analysis. Yet the amount of response variation jointly explained by the tested covariates was small^[1]. The practical result is restrained: average treatment effects remained a reasonable summary for most trial participants, rather than a personalized rule for choosing treatment.

The study combined two semaglutide programs with a liraglutide program. The scope is narrow. It speaks to patterns across selected GLP-1 trial data, not to every medicine in the class and not to people who would have been excluded from those trials. Readers looking for the broader medicine profile can use our semaglutide evidence overview, while this article stays with the reanalysis question.

What the researchers reanalyzed

The authors used participant-level data from SUSTAIN 6, PIONEER 6 and LEADER. SUSTAIN 6 randomized 3,297 people to once-weekly subcutaneous semaglutide or placebo. PIONEER 6 randomized 3,183 to once-daily oral semaglutide or placebo. LEADER randomized 9,340 to subcutaneous liraglutide or placebo. After exclusions, the new pooled analysis contained 13,670 participants^[1]. Pooling increased the data available for exploring response patterns, but it also joined different drugs, formulations and follow-up periods.

All three parent studies enrolled people with type 2 diabetes and elevated cardiovascular risk. Their original purpose was cardiovascular-outcome assessment, not the development of a bedside personalization tool. SUSTAIN 6 followed participants for 104 weeks and tested semaglutide's cardiovascular safety. PIONEER 6 was event-driven, with a median 15.9 months in the trial. LEADER followed participants for a median 3.8 years and evaluated liraglutide^[2][3][4]. Those design differences matter when interpreting a pooled secondary analysis.

How treatment differences were explored

The reanalysis defined endpoints using hemoglobin A1c, body mass index and blood pressure measurements. It applied causal forests, a machine-learning approach that can search for variation in estimated treatment effects across combinations of baseline characteristics. The authors used global tests for heterogeneity, ranked candidate modifiers by variable importance and estimated the magnitude of the variation^[1]. This was exploratory modeling of already completed trials. Participants were not newly randomized according to A1c, kidney function or a model-generated treatment recommendation.

A high variable-importance rank does not automatically mean a large or clinically decisive interaction. A variable can help a model split participants while the total difference it explains remains modest. That distinction is central here: baseline A1c repeatedly ranked first among candidate modifiers, but the magnitude analysis indicated that the tested covariates together explained little treatment-effect variation.

What the findings do and do not show

Kidney function ranked second in the relative-risk forest model, not necessarily in every model or on every endpoint. The abstract does not provide subgroup-specific treatment estimates, thresholds, confidence intervals or a validated decision rule^[1]. It would therefore be inappropriate to turn the ranking into a claim that one level of kidney function predicts a better response, or that treatment should be selected from an individual laboratory value.

The parent trials also evaluated active treatment against placebo rather than semaglutide against liraglutide. SUSTAIN 6 reported cardiovascular events in 6.6% of participants assigned semaglutide and 8.9% assigned placebo. PIONEER 6 was designed for noninferiority and reported major cardiovascular events in 3.8% and 4.8%, respectively. LEADER reported its primary outcome in 13.0% with liraglutide and 14.9% with placebo^[2][3][4]. Those separate results cannot rank the two drugs, and the 2026 reanalysis does not create a head-to-head comparison.

Why small heterogeneity still matters

A finding of small covariate-explained heterogeneity is informative because it pushes back against an easy assumption: visible differences between patients do not necessarily translate into large, predictable differences in treatment effect. In these data, routinely available baseline measures did not divide participants into sharply different response groups. That supports using the trial average as a useful summary at the population level while preserving normal clinical judgment about contraindications, tolerability and treatment goals.

It does not mean every person responds identically. Individual outcomes can differ because of adherence, exposure, concurrent care, chance, unmeasured biology and outcomes not represented in the models. The analysis instead asks a narrower question: how much systematic variation could the included baseline covariates explain within these trials? Its answer was that large covariate-driven differences could be ruled out in this setting.

What we do not yet know

The PubMed record does not expose full model specifications, exclusion counts by trial, missing-data handling, numerical heterogeneity estimates or performance in an external validation cohort^[1]. Without those details, this readout cannot judge model calibration or identify a clinically useful cutoff. Future work would need prespecified validation in a separate population before any prediction method could support treatment selection.

The enrolled populations were also selected for cardiovascular-risk trials. Results may not transfer unchanged to younger people, lower-risk populations, people with different kidney disease, or patients using current combinations of glucose-lowering therapies. The analysis covered A1c, BMI and blood-pressure endpoints, not every outcome that may matter to a patient. It should not be used to start, stop or choose a GLP-1 medicine without clinical assessment.

Medical disclaimer: This article is for educational and journalistic purposes only and does not constitute medical advice. Peptides discussed may be classified as prescription medicines or research chemicals depending on your jurisdiction. Always consult a qualified healthcare professional before using any peptide product. PeptideMethods.com does not sell, distribute, or facilitate the sale of any peptide product.