Cardiovascular disease accounts for most of the excess mortality in severe mental illness, so any drug that reduces it in this group matters. This target trial emulation used the TriNetX federated electronic health record network, comparing adults starting a GLP-1 receptor agonist with adults starting an SGLT2 inhibitor — a new-user, active-comparator design with propensity score matching, which is the right way to do this kind of study.
The primary four-year analysis matched 764,115 pairs, of which 195,184 pairs had serious mental illness. Mortality was 4.91% with GLP-1 receptor agonists against 6.45% with SGLT2 inhibitors, hazard ratio 0.76 (95% CI 0.74 to 0.78), absolute difference 1.54 percentage points. Among those with serious mental illness and type 2 diabetes, semaglutide was associated with lower three-point major adverse cardiovascular events (hazard ratio 0.77), and lower myocardial infarction, stroke, heart failure and bypass grafting. Ten-year exploratory analyses showed lower mortality across major depressive disorder, bipolar disorder and schizophrenia subgroups.
Now the caution, because it is substantial. In a separately matched one-year cohort, mortality was 1.46% against 2.84% — a relative risk of 0.52, meaning the drug apparently halved mortality within twelve months. No cardiometabolic drug does that. A halving of all-cause mortality in one year is far outside what the randomised GLP-1 outcome trials have shown, and the most likely explanation is residual confounding: prescribers do not start a GLP-1 agonist in someone who looks frail or terminally unwell, and propensity matching on recorded variables cannot capture that judgement.
So read the direction, not the magnitude. It is biologically plausible that GLP-1 agonists help this population disproportionately — weight gain from antipsychotics, high metabolic burden, poor access to preventive care. The comparator was an active drug rather than nothing, which is a real strength. But the effect size here is not credible as a causal estimate, and the authors are right that randomised trials are needed.
The practical position for now: where a patient with serious mental illness has an established indication — type 2 diabetes, obesity with comorbidity — this supports choosing a GLP-1 agonist and supports not withholding it because of the psychiatric diagnosis. It is not a reason to prescribe one to prevent cardiovascular death in the absence of an indication.
- Do not withhold GLP-1 agonists from patients with serious mental illness who have a metabolic indication.
- Do not prescribe them for cardiovascular prevention alone on this evidence.
- Treat the one-year halving of mortality as implausible and a marker of residual confounding.
- The active-comparator design is a genuine strength; the observational design is still the limiting factor.
- Coordinate with whoever manages the patient's diabetes rather than assuming someone else has considered it.
The statistics, in plain English
Target trial emulation is a method for making observational data behave more like a trial: defining eligibility, a time zero and an active comparator, then matching on measured confounders. It is a genuine improvement on naive database analysis, and it cannot fix confounding by indication — the fact that the choice between two drugs is made by a clinician who knows things about the patient that the record does not contain. The tell here is the one-year result. A relative risk of 0.52 for all-cause death within twelve months exceeds anything demonstrated in randomised trials of these agents, and when an observational estimate is much larger than the randomised evidence, the difference is usually the confounding rather than the drug.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for psychiatry, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free