SURPASS-3 is the tirzepatide trial that scanned people. Its main result is a clean win over basal insulin on glucose and weight, which is what gets quoted. Its three substudies are where the trial earns its place, because they measured what was happening inside the body while those two columns moved — liver fat, abdominal fat, thigh muscle and twenty-four-hour glucose — in a way no other trial in the program did. A separate trial compared tirzepatide with a different basal insulin in a higher-risk population, and its architecture is covered in the SURPASS-4 article. This page is about what the scanner found.
What was run, and what the control arm was doing
An open-label, parallel-group phase 3 study at 122 sites in 13 countries assessed 1,947 people and randomized 1,444 in a 1:1:1:1 ratio; 1,437 formed the modified intention-to-treat population — 358 on tirzepatide 5 mg, 360 on 10 mg, 359 on 15 mg and 360 on insulin degludec. Participants were insulin-naive adults with type 2 diabetes on metformin alone or with an SGLT2 inhibitor, with a glycated hemoglobin of 7.0% to 10.5%, a body-mass index of at least 25, and stable weight. Tirzepatide started at 2.5 mg and rose by 2.5 mg every four weeks to the assigned dose.[1]
The comparator was not a fixed milligram figure. Insulin degludec started at 10 units a day and was titrated weekly toward a fasting self-monitored blood glucose below 5.0 mmol/L (90 mg/dL), following a treat-to-target algorithm, for 52 weeks.[1] Every participant in that arm received whatever dose their own readings demanded, which makes this a comparison against a well-run insulin strategy rather than against a dose someone chose in advance. Baseline glycated hemoglobin was 8.17% (SD 0.91) and baseline weight 94.3 kg (SD 20.1).[1] What starting insulin involves for someone already on a GLP-1 is set out in the insulin article.
The glucose result, and the part of it the insulin arm earned
The primary endpoint was noninferiority of tirzepatide 10 or 15 mg against insulin degludec on glycated hemoglobin change at 52 weeks, with a margin of 0.3 percentage points. Glycated hemoglobin fell 1.93 points (SE 0.050), 2.20 (0.051) and 2.37 (0.050) across the three tirzepatide doses, against 1.34 (0.049) on insulin. The margin was met and every dose then cleared superiority: differences of −0.59 (95% CI, −0.73 to −0.45), −0.86 (−1.00 to −0.72) and −1.04 (−1.17 to −0.90), all at p < 0.001.[2]
Reaching below 7.0% followed the ladder: 82.44%, 89.71% and 92.63% against 61.25% on insulin, with odds ratios of 3.45 (95% CI, 2.38 to 5.01), 7.02 (4.55 to 10.84) and 10.79 (6.65 to 17.48).[2] Six in ten on titrated basal insulin reached target, which is a working comparator rather than a straw one; the trial drug simply reached it more often.
Weight moves in opposite directions
Body weight fell 7.5 kg (SE 0.37), 10.7 (0.37) and 12.9 (0.37) on the three tirzepatide doses and rose 2.3 kg (0.37) on insulin. The estimated treatment differences were −9.8 kg (95% CI, −10.8 to −8.8), −13.0 (−14.0 to −11.9) and −15.2 (−16.2 to −14.2).[2]
A fifteen-kilogram spread between arms is not fifteen kilograms of weight loss. Twelve-point-nine of it is what tirzepatide did and 2.3 is what insulin did, and only the first half describes anyone not starting insulin. That distinction matters more here than in a placebo trial, because the comparator has a known and opposite effect on the endpoint being advertised. The dose ladder on its own is set out in the tirzepatide dose article.
Liver fat, in 296 people who were scanned
A substudy at 45 centers in eight countries enrolled 296 of the trial’s participants who additionally had a fatty liver index of 60 or above, and imaged them at baseline and week 52 — 71, 79, 72 and 74 across the four arms. From a mean baseline liver fat content of 15.71% (SD 8.93), the pooled 10 and 15 mg groups fell 8.09 percentage points (SE 0.57) against 3.38 (0.83) on insulin, an estimated treatment difference of −4.71 (95% CI, −6.72 to −2.70; p < 0.0001). Visceral and abdominal subcutaneous fat volumes also fell more on tirzepatide.[3]
The substudy reports what the liver fat reduction tracked: baseline liver fat most strongly, at a correlation of −0.71, then reductions in subcutaneous fat at 0.33, body weight at 0.34 and visceral fat at 0.29, all at p ≤ 0.0006 within the tirzepatide groups.[3] Weight change explains part of it and nothing like all of it. The wider evidence on fatty liver in this class is in the fatty liver article.
The muscle finding, and the benchmark that defuses it
A later post hoc analysis of the same scans examined thigh muscle in the 246 participants with a valid week-52 image — 63, 60, 67 and 56 across the arms, at a mean age of 56.0 years (SD 9.9), a median diabetes duration of 6.7 years (IQR 3.7 to 10.7) and a mean body-mass index of 33.4. Pooled across doses, tirzepatide reduced muscle fat infiltration by 0.36 percentage points (95% CI, −0.48 to −0.25), muscle volume by 0.64 L (−0.74 to −0.54) and the muscle volume Z score by 0.22 (−0.29 to −0.15), all at p < 0.0001. Insulin degludec was associated with a modest significant increase in both body weight and muscle volume, and no significant change in the other two measures.[4]
Read alone, that is the worst-sounding column in the trial: the drug arm lost muscle and the insulin arm gained it. The analysis then supplies the comparison that decides what it means. Using longitudinal imaging from 2,942 UK Biobank participants to model how much muscle volume ordinarily moves with a given change in body weight, the observed tirzepatide losses were statistically indistinguishable from the prediction — a mean difference against the population-based estimate of −0.04 L (95% CI, −0.11 to 0.03; p = 0.22).[4]
Two results in that same comparison run in different directions and both belong on the page. Muscle fat infiltration fell more than predicted, by 0.42 percentage points (95% CI, −0.54 to −0.31; p < 0.0001), which is the favorable direction: less fat inside the muscle than the weight change alone would produce. But at 15 mg the muscle volume Z score fell further than predicted, by 0.18 (95% CI, −0.29 to −0.07; p = 0.0016).[4] The highest dose is the one place the muscle loss exceeds what the weight loss accounts for. What is known about lean mass on this class generally is in the muscle loss article.
Twenty-four-hour glucose, and what patients could feel
A third substudy fitted 243 participants with continuous glucose monitors for roughly seven days at baseline, week 24 and week 52 — 64, 51, 73 and 55 across the arms. The primary comparison was time in a tight target range of 71 to 140 mg/dL at 52 weeks: pooled 10 and 15 mg spent 25% more time there than insulin (95% CI, 16 to 33; p < 0.0001). Per dose the differences were 12% (1 to 22; p = 0.031), 24% (13 to 35) and 25% (14 to 35). At 24 weeks the 5 mg arm had not yet separated, while 10 and 15 mg had, at 19% (8 to 30) and 21% (11 to 31).[5]
Hypoglycemia ran the way basal insulin makes it run. Episodes below 54 mg/dL or severe occurred at rates of 0.0137, 0.0108 and 0.0275 per patient-year across the tirzepatide doses against 0.1020 on insulin, which the abstract reports as 5, 4 and 8 participants (1%, 1% and 2%) against 26 (7%).[2][1] The general picture on low blood sugar with this class is in the hypoglycemia article.
Participants could not reliably feel that difference. On the Diabetes Treatment Satisfaction Questionnaire change version, perceived frequency of hypoglycemia differed from insulin by −0.41 at 5 mg (p = 0.014), −0.18 at 10 mg (95% CI, −0.51 to 0.15; p = 0.280) and −0.26 at 15 mg (−0.59 to 0.07; p = 0.129). Perceived hyperglycemia separated only at the top dose, at −0.47 (p = 0.003), despite glycated hemoglobin differences of nearly a full point at every dose. Overall treatment satisfaction, by contrast, separated cleanly everywhere: 3.01, 2.90 and 2.99 points (p < 0.001 for each).[2]
A near-tenfold reduction in measured low-glucose episodes that patients did not detect on a questionnaire, alongside a satisfaction score that moved regardless, is a reminder that the perceived and measured versions of a benefit are separate things. The satisfaction gain here was not being driven by felt hypoglycemia.
Who left, and the column that runs against expectation
No study-drug discontinuation rate is published. The abstract states only that discontinuation due to an adverse event was more common in the tirzepatide groups than on insulin, without a figure, and the full text is paywalled with no PubMed Central record.[1] What the registry posts is why people left the study: 26, 40, 19 and 34 of the four arms did not complete. Adverse events accounted for 5, 7, 3 and 1 of those.[2]
The largest single reason runs the other way. Withdrawal by the participant — leaving because they chose to, for no recorded clinical reason — was 8, 17 and 7 across the tirzepatide arms and 22 on insulin, the highest count in the trial.[2] Serious adverse events were 29 of 358, 20 of 360, 26 of 359 and 22 of 360, with one, two, one and one death, none considered treatment-related.[2][1] The symptom columns behave as expected: nausea 12% to 24%, diarrhea 15% to 17%, decreased appetite 6% to 12% and vomiting 6% to 10% on tirzepatide, against 2%, 4%, 1% and 1% on insulin.[1]
What SURPASS-3 did not establish
It adjudicated no cardiovascular outcome and reported no event composite. It ran 52 weeks and no longer, so nothing in it speaks to durability past a year. It was open-label, because an injected insulin titrated to a target cannot be masked against a weekly injection, so every patient-reported measure on this page — satisfaction, perceived hypoglycemia, symptom reporting — was collected by people who knew what they were taking.
It compared tirzepatide with insulin and with nothing else: no GLP-1 comparator, no sulfonylurea, no placebo. It enrolled only insulin-naive people on metformin with or without an SGLT2 inhibitor, so it says nothing about adding tirzepatide to existing insulin. And the imaging substudies describe a selected subset — entry required a fatty liver index of 60 or above, and 246 of 1,444 participants produced the muscle result. That is a well-instrumented minority, not the trial.
What was in the pen
Every figure above belongs to branded, FDA-approved tirzepatide supplied free, escalated under protocol at 2.5 mg every four weeks to a fixed assigned dose of 5, 10 or 15 mg, injected weekly for 52 weeks on top of metformin, in people with type 2 diabetes who had never used insulin. 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 15.2. It is the distance between two arms, one of which was taking a drug that causes weight gain, in a diabetes population on metformin with a glycated hemoglobin above 7%. The figure that describes what tirzepatide did is 12.9 kg over 52 weeks, and the scanner that produced this trial’s best evidence also recorded that roughly half a liter of thigh muscle went with it.