This trial exists to answer a question a placebo cannot address: when metformin stops holding the line, is a weekly injection better than starting insulin? Almost every other trial in this class compares its drug against nothing. SURPASS-AP-Combo compared tirzepatide against titrated insulin glargine, the therapy it would actually be chosen instead of, in people already on one or two oral agents. The oral small-molecule route to the same vertical is in the ACHIEVE-1 article, and the head-to-head against semaglutide in diabetes is in the SURPASS-2 article.
What was preregistered, and what was demoted
917 insulin-naive adults with type 2 diabetes uncontrolled on metformin, with or without a sulfonylurea, were randomly assigned 1:1:1:1 to weekly tirzepatide at 5 mg (n = 230), 10 mg (228) or 15 mg (229), or to daily insulin glargine (230), at 66 hospitals in China, South Korea, Australia and India. The trial was open label. Entry required a glycated hemoglobin between 7.5% and 11.0% and a body-mass index of at least 23.[1][2]
The preregistered primary endpoint is narrower than the trial is usually described as being. It was the non-inferiority of the mean change in glycated hemoglobin from baseline to week 40 for the 10 mg and 15 mg doses only. The 5 mg dose was a key secondary endpoint, not a primary one, and superiority was tested only after non-inferiority was established.[2] A trial built to show a drug is not worse and which then clears superiority is a stronger result than one built to show superiority outright, and it is a different claim from the one a headline usually makes on its behalf.
The comparator is the design
An active comparator changes what every number means. Insulin glargine titrated toward a glucose target is not an inert arm: it drove fasting serum glucose down 46.2 mg/dL, most of the way to what the tirzepatide arms achieved at 58.6 to 66.2.[2] It was doing its job. That is what makes the glycated hemoglobin gap interpretable rather than trivial, and it is also why the safety and tolerability columns below cannot be read the way a placebo-controlled trial’s can.
The result, with its intervals
At week 40 the least-squares mean change in glycated hemoglobin was −2.44 percentage points at 10 mg and −2.49 at 15 mg, against −0.95 on insulin glargine — treatment differences of −1.49 (95% CI, −1.69 to −1.29) and −1.54 (95% CI, −1.74 to −1.34). The 5 mg dose returned −2.24, a difference of −1.29 (95% CI, −1.49 to −1.09), with P < 0.001 throughout.[1][2]
A glycated hemoglobin below 7.0% was reached by 75.4%, 86.0% and 84.4% of the tirzepatide groups against 23.7% on glargine — odds ratios of 14.54 (95% CI, 8.94 to 23.64), 28.76 (16.72 to 49.49) and 25.27 (14.88 to 42.95).[2] Note that 10 mg and 15 mg are indistinguishable from one another on both measures, so the top of the ladder bought nothing here that the middle rung did not already deliver — a pattern worth holding onto, because the same trial shows the two doses separating clearly on weight.
Where the two arms went opposite ways
Weight is the axis on which an insulin comparator loses by default, and the chart above shows the comparator bar crossing zero in the other direction — a control arm that did not merely fail to lose weight but gained it, which a placebo arm almost never does. Body weight changed −5.0 kg, −7.0 kg and −7.2 kg on the three tirzepatide doses and +1.5 kg on glargine — treatment differences of −6.5 kg (95% CI, −7.4 to −5.6), −8.5 kg (−9.5 to −7.6) and −8.7 kg (−9.6 to −7.7). A reduction of 5% or more was reached by 55.7%, 71.6% and 74.1% against 5.6%.[1][2]
Roughly two thirds of that measured difference is the tirzepatide arms losing weight and roughly one third is the insulin arm gaining it. Both halves are real and they are not the same fact, and someone deciding between these two options is choosing a direction rather than a magnitude. What insulin does to weight, and why, is in the insulin article.
The odds ratio nobody should quote
One secondary endpoint is a lesson in how a true number becomes a misleading one. A glycated hemoglobin below 5.7% — the non-diabetic range — was reached by 14.9%, 20.7% and 27.7% of the tirzepatide groups and by 0.00% of the insulin group. Because the comparator had zero events, the odds ratios come out at 82.54, 124.76 and 184.90, with confidence intervals of 5.13 to 1327.81, 7.79 to 1997.01 and 11.59 to 2950.73.[2]
An interval spanning three orders of magnitude is the statistic telling you it cannot estimate the thing it was asked to estimate. The finding underneath is worth having — about a quarter of the top-dose arm reached a non-diabetic range and nobody on insulin did — and the ratio attached to it is not a number to repeat. A post hoc characterization of the 598 tirzepatide-treated patients who took no rescue medication found that those reaching below 5.7% were markedly younger with markedly lower baseline glycated hemoglobin, which is most of the explanation.[4]
A quarter of this trial was not overweight
Here is what this trial adds that little else in the tirzepatide library does. Its body-mass index floor was 23, and an exploratory subgroup analysis of the 907 treated participants found 235 (25.9%) below 25, 458 (50.5%) between 25 and 30, and only 214 (23.6%) at 30 or above. Across those categories glycated hemoglobin fell 2.0 to 2.8 percentage points on tirzepatide against 0.8 to 1.0 on glargine, and weight fell 5.5% to 10.8% against a gain of 1.0% to 2.5%, with a safety profile similar across subgroups.[3]
Most of the evidence readers meet for this drug comes from trials whose mean body-mass index sits near 38. A quarter of this one sat in the normal range and still lost two percentage points of glycated hemoglobin. That is a genuinely different population from the obesity program, and it is the reason a glycemic result should not be read as a weight result: nobody with a body-mass index of 24 was enrolled here to lose weight, and the glycated hemoglobin they gained was the point of treating them.
Two arms, two different reasons for leaving
Discontinuation tells the story from both sides. Adverse events ended treatment for 7, 19 and 18 participants on the three tirzepatide doses against 4 on glargine — but withdrawal of consent ran the other way, 6, 6 and 8 against 20.[2] People left tirzepatide because of how it made them feel and left insulin because they no longer wanted to be on it. In an open-label trial both are real signals and neither is the other.
Serious adverse events were also not where a reader would guess: 15, 14 and 15 against 20 on glargine, with 0, 0 and 1 deaths against 2.[2] Hyperglycemia was reported as an adverse event in 5, 2 and 2 participants against 18 on the insulin arm. The gastrointestinal column runs the expected way — decreased appetite 74, 109 and 102 against 0, diarrhea 77, 103 and 101 against 3, nausea 46, 75 and 73 against 5, elevated lipase 32, 32 and 30 against 6.[2]
Hypoglycemia is the one people expect to separate and it did not. Rates of blood glucose below 54 mg/dL or severe hypoglycemia were 0.066, 0.089 and 0.070 events per patient-year against 0.054 on glargine, with no severe hypoglycemia reported at all,[1][2] in a trial where many participants were also taking a sulfonylurea — the agent that carries most of that risk, as the sulfonylurea article explains.
What it does not settle
It ran 40 weeks with no cardiovascular, renal or mortality endpoint. It was open label, so everything a participant reported about themselves — including the treatment-satisfaction advantage of 1.63 to 1.81 points — carries that caveat, and it is worth noting that on the questionnaire’s hypoglycemia item the arms did not differ at all (p = 0.18 to 0.38).[2] It tested no oral agent, and it did not randomize anyone against another incretin.
That last point needs stating plainly, because the numbers invite the error. The reductions here — 2.24 to 2.49 percentage points — are larger than those in the oral orforglipron trial, which ran 1.24 to 1.48. That is not a potency comparison. Mean baseline glycated hemoglobin here was 8.71% against 8.0%; these participants were already on one or two oral agents while those took nothing; the comparator here was active insulin rather than placebo; and this trial was open label while that one was double-blind. Nothing randomized the two drugs against each other, and the two populations otherwise overlap heavily — both largely Asian, both 40 weeks, both with a body-mass index floor of 23. What separates these trials is drug, route, comparator, background therapy and masking. It is not geography.
Who the result describes
763 of the 917 randomized participants (83.2%) enrolled in China.[1] Among the 907 who received a dose, 892 were recorded as Asian, 10 as White and none as Black or African American; South Korea contributed 104, India 32 and Australia 15. Mean age was 54.1 years, 400 of 907 were women, and mean baseline glycated hemoglobin was 8.71%.[2] “Asia-Pacific” in the title is doing a lot of work for a trial that is, in practice, predominantly Chinese — which its own authors say in as many words.[1][5]
Every participant received branded, FDA-approved tirzepatide, supplied without interruption and escalated on a protocol, alongside supervised background therapy and a titrated insulin comparator managed by a clinician. Sellers on the tirzepatide board dispense compounded preparations, which are not FDA-approved and are not reviewed by the FDA for safety, efficacy or quality before they are dispensed — and none of them is managing anybody’s insulin. The gap between a trial supply and a compounded vial is set out in the compounded-versus-brand article.