Does tirzepatide really cut all-cause mortality by 45%?

The short answer

A BMJ study reported tirzepatide cutting all-cause mortality by 45% versus sitagliptin. The randomised trial of the same drug, SURPASS-CVOT, found it merely noninferior to dulaglutide. Both are real. The gap is study design — and the giveaway is that the observational study also found a 60% drop in infection deaths.

Evidence tier: Tier 1 for SURPASS-CVOT (randomised, double-blind, 13,165 patients); Tier 2–3 for the BMJ target-trial emulation, which is well-designed observational research but cannot eliminate confounding by indication. Educational content, not medical advice.

The key points:

• The BMJ figures are accurate — MACE down 32%, mortality HR 0.55, infections HR 0.64.
• SURPASS-CVOT found noninferiority, not superiority, against dulaglutide.
• The comparators differ — sitagliptin is a much weaker drug than dulaglutide.
• The infection finding is the tell. No plausible mechanism explains a 60% drop in infection deaths.
• Who gets prescribed an expensive drug is the most likely explanation for the gap.

What did each study actually find?

Evidence tier: 1–2 — both published, both large.

The randomised trial. SURPASS-CVOT screened 16,979 people and randomised 13,165 with type 2 diabetes and established atherosclerotic cardiovascular disease — 6,586 to tirzepatide, 6,579 to dulaglutide — following them for a median of 46.9 months. On the primary composite of cardiovascular death, myocardial infarction and stroke, tirzepatide was noninferior to dulaglutide. Not superior. A cardiorenal endpoint occurred in 23.7% versus 27.4%. Gastrointestinal adverse events were more common on tirzepatide (42.5% vs 35.9%).

The observational study. A BMJ target-trial emulation used US claims data from Optum Clinformatics and Merative MarketScan to reproduce SURPASS-CVOT's eligibility criteria, identifying 52,971 patients aged 40+ with type 2 diabetes and established ASCVD who started tirzepatide (n=35,353) or sitagliptin (n=17,618) between May 2022 and May 2025.

At one year: major adverse cardiovascular events in 2.9% on tirzepatide versus 4.4% on sitagliptin — a 32% relative reduction. Myocardial infarction HR 0.67. All-cause mortality HR 0.55. Serious infections requiring hospitalisation HR 0.64, and infection-related mortality HR 0.40.

Both sets of numbers are correctly reported. The question is what they mean.

Why don't the two studies agree?

Evidence tier: 2 — study design.

Three reasons, and only the third is a problem.

The comparators are different, and this matters most. SURPASS-CVOT compared tirzepatide against dulaglutide — another GLP-1 receptor agonist with its own established cardiovascular benefit. The BMJ study compared it against sitagliptin, a DPP-4 inhibitor that is cardiovascularly neutral. Beating a neutral drug is a much lower bar than beating an active one. A bigger effect size versus sitagliptin is entirely expected and is not evidence of contradiction.

The follow-up differs. SURPASS-CVOT ran a median 46.9 months; the BMJ analysis reports one-year risk. Short-window observational estimates tend to be larger, partly because the people who stay on treatment for a year are systematically different from those who stop.

And the effect sizes are implausibly large for a glucose-lowering drug. A 45% reduction in all-cause mortality within twelve months would be an extraordinary result — larger than statins, larger than most interventions in cardiovascular medicine. That should prompt scepticism before celebration.

What gives the confounding away?

Evidence tier: 2 — the internal evidence.

The infection findings, and the study authors flag them honestly.

Serious infections requiring hospitalisation fell 36%. Infection-related mortality fell 60%. The authors note this suggests "potential nonatherosclerotic contributors to survival benefits" — a careful way of saying the mortality benefit is not coming only from the cardiovascular effects the drug is supposed to have.

But ask the direct question: by what mechanism would a GIP/GLP-1 receptor agonist stop you dying of pneumonia? Weight loss and glycaemic control plausibly reduce infection risk somewhat over years. A 60% reduction in infection deaths within one year is not something incretin pharmacology can deliver.

There is a simpler explanation, and it is the oldest problem in pharmacoepidemiology.

Who gets prescribed tirzepatide, and who gets sitagliptin? Tirzepatide is expensive, frequently requires prior authorisation, and has been supply-constrained. Sitagliptin is cheap and generic. In a US claims population, the patients who obtain tirzepatide are systematically better insured, more engaged with healthcare, more mobile, less frail and more likely to persist with treatment. The patients who end up on the cheap DPP-4 inhibitor skew older, sicker and frailer.

Frailty predicts death from infection. Frailty predicts all-cause mortality. Frailty is not well captured in claims data, so it cannot be adjusted away — and it is precisely what determines which of these two drugs a patient receives.

That is confounding by indication compounded by healthy-adherer bias, and it produces exactly this signature: implausibly large mortality benefits, spread across causes of death the drug has no mechanism to influence.

Is target trial emulation still worth anything?

Evidence tier: 2 — methodological.

Yes, and it is worth being fair about this rather than dismissive.

Target trial emulation is a genuine methodological advance. It forces observational researchers to specify eligibility, treatment assignment, follow-up and outcomes in advance, exactly as a trial protocol would — which eliminates a large class of analytical flexibility that made older observational studies unreliable. This BMJ study is a well-conducted example, using two large independent databases and explicitly emulating a real trial's criteria.

What it cannot do is randomise. Emulation controls for what is measured; it cannot control for what determined the prescription in the first place when that factor is unrecorded. When the drugs being compared differ enormously in cost and access, the populations receiving them differ in ways no covariate adjustment reaches.

Which is why the study is genuinely useful for some questions and not others. It gives a reasonable picture of real-world cardiovascular event rates in people actually taking these drugs. It cannot establish that tirzepatide prevents 45% of deaths.

How should you read claims like this in future?

Evidence tier: 3 — practical heuristics.

Four questions that resolve most of these cases quickly.

Was it randomised? If not, the effect size is an upper bound and probably an overestimate. This is the single most informative question.

What was the comparator? "Better than sitagliptin" and "better than dulaglutide" are very different claims about the same drug. Headlines almost never say which.

Is the effect biologically plausible at that size? A 45% mortality reduction from a glucose-lowering drug in one year should trigger scepticism, not enthusiasm. Real effects in cardiovascular medicine are usually modest.

Are benefits appearing where the drug has no mechanism? This is the most useful tell of all. When a cardiovascular drug also appears to prevent infection deaths, the common cause is usually the patient, not the pill.

None of this means tirzepatide does not help. SURPASS-CVOT is a large, well-run randomised trial and it found tirzepatide performing comparably to an established cardioprotective GLP-1 over nearly four years, which is a genuinely good result. It is simply a different and smaller claim than the one circulating.

Limitations

This is educational content, not medical advice.

• We did not conduct either study. This compares published findings and their designs.
• The BMJ study is not poor research. Target trial emulation is a rigorous approach and the authors flagged the infection anomaly themselves.
• Confounding by indication is an inference, not a demonstrated fact. It is the most parsimonious explanation, not the only possible one.
• Comparator differences alone explain part of the gap — sitagliptin versus dulaglutide is not a like-for-like contrast.
• SURPASS-CVOT tested noninferiority, which is a different statistical question from superiority and cannot demonstrate the absence of benefit.
• Nothing here bears on whether tirzepatide is appropriate for any individual, which is a prescribing decision.
• Marko Maal, MSc Pharmacy reviewed this article. Reviewer attribution does not constitute a doctor-patient relationship.

The bottom line

Two studies of the same drug, published months apart, both accurately reported, pointing in different directions. The randomised one says tirzepatide performs comparably to another GLP-1 over four years. The observational one says it cuts deaths by nearly half in twelve months. The second travelled much further, because it is a far more exciting number.

Part of the gap is legitimate: sitagliptin is a weaker comparator than dulaglutide, so a larger margin is expected. But the part that should stop you is the infection data. A drug that appears to prevent both heart attacks and deaths from pneumonia, at large magnitudes, in a single year, is more likely telling you something about who receives it than about what it does.

Tirzepatide is expensive and access-limited; sitagliptin is cheap and generic. The people who get the first are healthier than the people who get the second in ways claims data cannot measure. That is the most probable source of a 45% mortality difference, and it is worth recognising because the same pattern will recur with every expensive drug studied this way — including retatrutide and CagriSema when their real-world data arrives.

The honest summary of tirzepatide's cardiovascular credentials remains SURPASS-CVOT: noninferior to an established GLP-1, over 13,165 patients and nearly four years. That is a solid result. It is not a 45% reduction in death.

Related on this site

• Does retatrutide really do the job of six peptides?
• GLP-1 and cancer headlines: what the evidence actually says
• CagriSema PDUFA date and FDA approval status
• Peptide evidence grades explained
• Do GLP-1s stop working?
• Our evidence-tier framework

References

• Krüger N, et al. Tirzepatide and the risk of atherosclerotic cardiovascular events: population based cohort study. BMJ 2026. PMID 42556854 — target trial emulation, 52,971 patients; tirzepatide (n=35,353) vs sitagliptin (n=17,618); one-year MACE 2.9% vs 4.4%; MI HR 0.67; all-cause mortality HR 0.55; serious infection HR 0.64; infection-related mortality HR 0.40.
• SURPASS-CVOT: Cardiovascular outcomes with tirzepatide versus dulaglutide in type 2 diabetes. NEJM, December 2025. Article — 13,165 randomised, median 46.9 months, tirzepatide noninferior on the primary composite; cardiorenal endpoint 23.7% vs 27.4%; GI adverse events 42.5% vs 35.9%.
• Post hoc cardiorenal analysis of SURPASS-CVOT. PMID 41903177