GLP-1 drugs, death rates and serious mental illness: a very large, very soft signal

People with schizophrenia, bipolar disorder and major depressive disorder die earlier than people without them, and the excess is driven principally by cardiovascular disease [s1]. That fact has been stable for decades and has proved almost immune to intervention. A study published in JAMA Psychiatry on 26 August reports something that would matter a great deal if it holds: adults with serious mental illness who started a GLP-1 receptor agonist had lower mortality over the following four years than matched adults who started an SGLT2 inhibitor instead [s1].

The size of the analysis is unusual. Working inside TriNetX, a multinational federated electronic-health-record network, the authors used a new-user, active-comparator design with propensity-score matching, then ran psychiatric and non-psychiatric cohorts separately [s1]. The primary four-year analysis matched 1,528,230 adults — 764,115 pairs, of which 195,184 pairs had serious mental illness and 568,931 did not [s1].

What the numbers say

Among the participants with serious mental illness, 4.91% of GLP-1 initiators died within four years (9,585 of 195,184) against 6.45% of SGLT2-inhibitor initiators (12,584 of 195,184) — a hazard ratio of 0.76 (95% CI, 0.74–0.78) and an absolute risk difference of −1.54 percentage points (95% CI, −1.68 to −1.39) [s1]. In a separately matched one-year cohort the gap was proportionally larger still: 1.46% versus 2.84%, a relative risk of 0.52 (95% CI, 0.49–0.54) [s1].

Cardiovascular endpoints moved in the same direction. Among participants with serious mental illness and type 2 diabetes, starting semaglutide rather than an SGLT2 inhibitor was associated with a lower risk of three-point major adverse cardiovascular events (HR, 0.77; 95% CI, 0.76–0.79), and of five-point MACE, myocardial infarction, stroke, heart failure and coronary artery bypass grafting [s1]. Ten-year exploratory analyses in participants with type 2 diabetes found lower mortality in the major depressive disorder (RR, 0.55; 95% CI, 0.53–0.56), bipolar disorder (RR, 0.57; 95% CI, 0.51–0.63) and schizophrenia (RR, 0.67; 95% CI, 0.59–0.75) groups [s1]. The authors report that the primary mortality association survived their prespecified sensitivity analyses, which included stratifying by diabetes, excluding baseline heart failure and chronic kidney disease, and censoring at crossover [s1].

Why the effect size is the problem, not the reassurance

A halving of one-year mortality is not what cardiovascular pharmacology usually delivers, and that is the reason to slow down rather than to celebrate. The comparison here is not against nothing; it is against SGLT2 inhibitors, a drug class with its own established mortality benefit. For a GLP-1 drug to cut deaths by roughly half within twelve months relative to that comparator implies an effect larger than anything randomised trials have demonstrated for either class.

A second paper published the same week, in Diabetes, Obesity and Metabolism, makes the point from the inside. It used the same kind of design — a target trial emulation in the TriNetX US Collaborative Network, propensity-score matched, active comparators — in a different population: adults with type 2 diabetes and a body-mass index below 27 kg/m² [s2]. It found GLP-1 use associated with lower all-cause mortality than DPP-4 inhibitors (HR, 0.61; 95% CI, 0.52–0.71), lower MACE (HR, 0.83; 95% CI, 0.71–0.98) and lower coded heart failure (HR, 0.71; 95% CI, 0.60–0.85) [s2]. Against SGLT2 inhibitors, MACE and coded heart failure were lower but all-cause mortality was not (HR, 0.89; 95% CI, 0.76–1.05) [s2].

Those authors draw the conclusion explicitly: the mortality association "should be interpreted cautiously as likely reflecting residual confounding rather than a benefit of this magnitude" [s2]. The mechanism is familiar to anyone who has watched observational drug studies fail to replicate. People who are prescribed and can stay on a GLP-1 drug are, on average, healthier, better engaged with care and more able to tolerate a titrated injectable than people who are not. Propensity-score matching balances what is coded in the record. It cannot balance what is not.

What would settle it

The JAMA Psychiatry authors say prospective randomised trials are warranted [s1], which is the right conclusion and also an admission of what this study is not. Two features would make the finding more believable: a mortality benefit that shrinks toward the trial-established range once better confounding control is applied, and consistency across networks that are not TriNetX.

There is a narrower claim inside the paper that is more robust and arguably more useful. People with serious mental illness were not excluded from benefit. Across bipolar disorder, major depressive disorder and schizophrenia subgroups the direction of effect was the same [s1], and the absolute reductions were larger in the serious-mental-illness cohorts than in those without [s1] — which is what you would expect from a group starting at higher baseline cardiovascular risk. Psychiatric populations have been routinely under-represented in cardiometabolic trials. Evidence that they are being prescribed these drugs at scale, and that outcomes are not visibly worse, is worth having even when the headline number is not trustworthy.

Nothing here establishes that a GLP-1 drug prevents deaths in serious mental illness. It establishes that the question is now large enough, and consequential enough, to justify a trial designed to answer it.

Sources

• [s1] "Glucagon-Like Peptide 1 Receptor Agonists, Mortality, and Cardiovascular Outcomes in Serious Mental Illness," JAMA Psychiatry, 26 August 2026. https://doi.org/10.1001/jamapsychiatry.2026.2574
• [s2] "Cardiorenal Mortality and Safety Outcomes of GLP-1 Receptor Agonists in Type 2 Diabetes With BMI Below 27 kg/m2: A Target Trial Emulation," Diabetes, Obesity and Metabolism, 25 August 2026. https://doi.org/10.1111/dom.71266