SURPASS-4 gets quoted for a cardiovascular hazard ratio it was never built to produce. Its primary endpoint was blood sugar; its cardiovascular column was a safety attachment collected while the glucose question was being answered, and it rests on 109 adjudicated events across four arms. The comparator was not a placebo and not a fixed dose of anything — it was titrated insulin, pushed toward a target. Both of those facts change what the trial’s numbers are allowed to mean.
The comparator was a target, not a dose
The control arm received insulin glargine 100 U/mL once daily, titrated to reach a fasting blood glucose below 100 mg/dL.[1] That is a moving comparator: every participant received whatever dose their own glucose readings demanded, which makes the control arm a treatment strategy rather than a milligram figure. It also means the trial could not be blinded, and it was not — this was an open-label study, with participants and investigators aware of assignment.[1] For a hard endpoint like death that matters little; for anything reported by a patient or judged by a clinician, it matters a great deal.
Comparing an incretin against treat-to-target insulin is a fair test of glucose lowering and a poor test of almost anything else, because the two drugs do different damage. Insulin drives glucose down at the cost of hypoglycemia and weight; a GLP-1 drives it down at the cost of nausea. What that means for someone already injecting insulin is covered in the insulin article.
What was run
An open-label, parallel-group phase 3 study at 187 sites in 14 countries screened 3,045 people and randomly assigned 2,002 in a 1:1:1:3 ratio; 1,995 received at least one dose and form the analysis population — 329 on tirzepatide 5 mg, 328 on 10 mg, 338 on 15 mg and 1,000 on glargine. Eligibility required type 2 diabetes on metformin, a sulfonylurea or an SGLT2 inhibitor in any combination, glycated hemoglobin of 7.5% to 10.5%, a body-mass index of 25 or above, and established cardiovascular disease or high cardiovascular risk. All participants were treated for at least 52 weeks and continued up to 104 weeks so that cardiovascular events could accumulate and be adjudicated.[1] Median treatment duration came to 85 weeks.[3] Mean age was 63.6 years (SD 8.56) and 751 of the 2,002 randomized, 37.5%, were women.[2]
The endpoint the trial was actually built to answer
The primary objective was noninferiority of tirzepatide 10 mg or 15 mg, or both, against glargine on glycated hemoglobin change at 52 weeks, against a margin of 0.3 percentage points. Glycated hemoglobin fell 2.43 points (SD 0.05) on 10 mg and 2.58 (SD 0.05) on 15 mg against 1.44 (SD 0.03) on glargine — estimated treatment differences of −0.99 (multiplicity adjusted 97.5% CI, −1.13 to −0.86) and −1.14 (−1.28 to −1.00). The margin was met for both, and both differences sit far past it.[1] The 5 mg arm, a secondary comparison, returned −0.80 (97.5% CI, −0.93 to −0.66).[2]
Reaching below 7.0% followed the same ladder: 80.98%, 88.16% and 90.72% against 50.72%, with odds ratios of 4.78 (95% CI, 3.47 to 6.58), 9.23 (6.31 to 13.49) and 11.87 (7.88 to 17.89).[2] A trial designed to show a drug is not worse showed it substantially better, which is the cleanest result in the whole dataset and the one least often quoted.
The 109-event cardiovascular column
Adjudicated four-part major adverse cardiovascular events — cardiovascular death, myocardial infarction, stroke or hospitalization for unstable angina — occurred in 109 participants across all four arms, with a hazard ratio for tirzepatide against glargine of 0.74 (95% CI, 0.51 to 1.08).[1] The paper’s own conclusion is phrased as the absence of harm: tirzepatide was not associated with excess cardiovascular risk.
That phrasing is exact and worth respecting. A confidence interval running from a 49% reduction to an 8% increase has not excluded no-effect, the analysis was a safety assessment rather than a powered efficacy comparison, and 109 events is roughly an eighth of what a dedicated outcome trial in this class accumulates. A point estimate of 0.74 in that setting is a direction, not a finding.
The deaths do not line up with dose
Sixty deaths occurred during the study: 25 (3%) across the tirzepatide arms and 35 (4%) on glargine.[1] Pooled like that it reads as a mortality signal. The posted results split it by arm, and the split does not behave: 15 of 329 on 5 mg, 2 of 328 on 10 mg and 8 of 338 on 15 mg, against 35 of 1,000 on glargine.[2] The lowest dose recorded more than seven times the deaths of the middle dose, in arms of almost identical size, with an identical protocol.
No pharmacology produces that curve. What produces it is small numbers, and the same smallness is what sits underneath the 0.74. Reading the pooled comparison as evidence that tirzepatide reduces death requires ignoring that its own dose arms contradict each other, and a trial that cannot order its three doses cannot rank itself against a fourth arm either. Serious adverse events ran 48 of 329 (14.6%), 54 of 328 (16.5%) and 41 of 338 (12.1%) against 193 of 1,000 (19.3%) on glargine — lower on every tirzepatide arm, and equally untestable here.[2]
Where the hypoglycemia advantage gets bigger by subtracting a drug
Clinically significant hypoglycemia — glucose below 54 mg/dL, or severe — affected 6% to 9% of tirzepatide participants against 19% on glargine. Restricted to participants not taking a sulfonylurea, the same comparison is 1% to 3% against 16%.[1] The gap gets wider, not narrower, when the drug most associated with hypoglycemia is removed from both arms.
That is the opposite of the usual subgroup intuition, and it is informative rather than paradoxical: the sulfonylurea was generating most of what little hypoglycemia the tirzepatide arms had, while the insulin arm generated its own regardless. Rates per participant-year tell the same story at 0.10, 0.09 and 0.11 episodes against 0.35, and serious hypoglycemia was recorded in 1, 0 and 3 participants against 12.[2] For anyone on a sulfonylurea considering an incretin, that subgroup is the relevant one, and the hypoglycemia article sets out what it means in practice.
The kidney finding, and the ruler that could have faked it
A post hoc analysis compared kidney trajectories across the pooled tirzepatide arms and glargine. Filtration rate fell 1.4 mL/min per 1.73 m² per year (SE 0.2) against 3.6 (SE 0.2), a difference of 2.2 (95% CI, 1.6 to 2.8), widening to 3.7 (2.4 to 5.1) in participants who began below 60. Urinary albumin-to-creatinine ratio rose 36.9% (26.0 to 48.7) on glargine and did not rise on tirzepatide, at −6.8% (−14.1 to 1.1). The composite of a 40% filtration fall, end-stage kidney disease, death from kidney failure or new macroalbuminuria occurred less often on tirzepatide, hazard ratio 0.58 (95% CI, 0.43 to 0.80).[3]
There is an obvious way that result could be an artifact. Filtration rate estimated from creatinine moves when muscle mass moves, and the tirzepatide arms lost up to 11.7 kg while the insulin arm gained weight. A second post hoc analysis tested exactly that by re-estimating filtration from cystatin C, which does not depend on muscle: the creatinine-based between-group difference was 1.4 (95% CI, 0.3 to 2.4) and the cystatin C-based difference was 1.8 (95% CI, 0.8 to 2.8), with the report stating that filtration changes did not correlate with weight changes.[4] Swapping the ruler made the effect slightly larger, not smaller. What any of it means for someone with reduced kidney function belongs to the kidney disease article.
Two years, and the arm that gained weight
At 52 weeks the posted results record weight down 7.1 kg (SE 0.34), 9.5 kg (0.34) and 11.7 kg (0.33) on the three tirzepatide doses, against a 1.9 kg gain (SE 0.19) on glargine — differences of −9.0 (95% CI, −9.8 to −8.3), −11.4 (−12.1 to −10.6) and −13.5 (−14.3 to −12.8).[2] A post hoc analysis of 1,500 participants followed to 104 weeks reports the same shape holding: glycated hemoglobin down 2.3%, 2.5% and 2.6% against 1.0%, and weight down 7.6, 10.0 and 11.4 kg against a gain of 2.1 kg.[5]
The comparator gaining weight while the trial drug loses it is a thirteen-kilogram spread that is only half about tirzepatide. Half of it is what insulin does, and a reader shopping for weight loss should not read a 13.5 kg difference as a 13.5 kg loss. How the dose ladder behaves on its own is set out in the tirzepatide dose article.
Durability was examined separately. Among tirzepatide participants who had reached glycated hemoglobin of 6.5% or below at 52 weeks, 75% to 84% still held it at study end, and among those who had lost 10% or more of body weight, 79% to 82% maintained it. The analysis found predictors of sustained glucose control — greater weight loss, better beta-cell function — and reported no clinically meaningful predictor for sustained weight control.[6] Nothing measurable at 52 weeks told you who would keep the weight off.
What SURPASS-4 did not establish
It did not establish a cardiovascular benefit, and its authors do not claim one. It did not compare tirzepatide with any other incretin; that comparison is a different trial with a different comparator problem, covered in the SURPASS-2 article. It did not enroll anyone without type 2 diabetes, so the weight columns describe a diabetes population on background glucose-lowering therapy rather than an obesity cohort.
It also does not publish how many people stopped the drug. The abstract reports no study-drug discontinuation rate, the full text is paywalled with no PubMed Central record, and the registry posts reasons for leaving the study: over the first 52 weeks, 3, 2 and 0 participants across the tirzepatide arms and 3 on glargine withdrew because of an adverse event, with a further 1, 2, 0 and 7 during the extended period.[2] Those single digits describe study retention, not tolerability. Tolerability is in the symptom columns, where nausea affected 11.9%, 16.2% and 22.5% against 2.3%, diarrhea 12.5%, 19.8% and 21.9% against 4.4%, and vomiting 4.9%, 8.2% and 8.6% against 1.5%.[2]
What was in the pen
Every figure above belongs to branded, FDA-approved tirzepatide supplied free, escalated under protocol to a fixed assigned dose of 5, 10 or 15 mg, injected weekly for a median of 85 weeks on top of existing oral diabetes therapy, in people aged around 63 with high cardiovascular risk and glycated hemoglobin above 7.5%. Sellers listed on the compounded tirzepatide board dispense compounded preparations, which are not FDA-approved and are not reviewed by the FDA for safety, efficacy or quality before they reach a patient.
The specific misuse to watch for with this trial is the 0.74. It comes from 109 adjudicated events in a safety analysis inside a glucose trial, against a comparator whose own dose was adjusted patient by patient, in a population defined by diabetes and cardiac risk. It cannot be carried across to a cash-pay weight-loss purchase, and the trial’s own death counts — 15, then 2, then 8 — are the plainest demonstration of why.