Is there an honest apples-to-apples comparison of the STEP and SURMOUNT headline results?

Every comparison I find is either a marketing graphic putting -20.9% next to -14.9% with no caveats, or a methodology piece telling me cross-trial comparison is invalid and then refusing to say anything further. Both are useless to me.

I accept that indirect comparison is weaker than a head-to-head. What I want is a table that actually lays out the trials with the features that differ - sample size, dose, duration, population, what the placebo arm was doing - so that I can see for myself which comparisons are nearly fair and which are not.

Specific things I do not understand yet: SURMOUNT-1 ran 72 weeks and STEP 1 ran 68, which is a 4-week difference on a curve that has not fully flattened, and I have no idea whether that matters. STEP 2 and SURMOUNT-2 both enrolled people with type 2 diabetes and both show smaller losses than the non-diabetes trials, which I assume is real, but the size of that penalty differs between programmes. And I have read that SURMOUNT-5 is a direct head-to-head, which if true should settle the primary question and make the indirect arithmetic mostly moot.

Can someone lay the programme out properly and say which numbers can be compared and which cannot?

A structural point about the table above that is easy to miss: the two programmes did not use the same titration philosophy, and this leaks into the headline numbers in a way that has nothing to do with receptor pharmacology.

Semaglutide in STEP has one obesity dose. You escalate to 2.4 mg weekly over 16 weeks and that is the dose, with the protocol allowing you to sit at a lower dose if you cannot tolerate it. Tirzepatide in SURMOUNT-1 randomised participants to three fixed maintenance doses and reported them separately, and SURMOUNT-3, -4 and -5 used maximum tolerated dose instead.

Consequences for reading the numbers:

• Quoting -20.9% for tirzepatide against -14.9% for semaglutide compares the top of a dose-response curve against a single licensed dose. That is a legitimate practical comparison but it is not a comparison at equipotent exposure, and nobody knows what semaglutide at a genuinely higher exposure does in this population.
• The 5 mg tirzepatide arm at -15.0% and semaglutide 2.4 mg at -14.9% are indistinguishable. Any story that has tirzepatide categorically superior has to explain why its lowest maintenance dose ties.
• Maximum-tolerated-dose designs bake tolerability into the efficacy endpoint. If a drug is easier to tolerate, more participants reach the top dose, and the arm mean rises for a reason that is real but is not potency.

So the honest summary is a dose-response statement, not a ranking: across the published range, tirzepatide reaches higher mean weight loss than the licensed semaglutide obesity dose, and the head-to-head confirms that at maximum tolerated dosing. Whether that reflects GIP co-agonism, higher achievable exposure, or both, the trials as designed cannot separate.

Two housekeeping items for anyone building this table themselves.

First, baseline weight is not constant across the programme and percentages hide that. STEP 6 enrolled an East Asian population with a lower BMI entry threshold of 27 and a substantially lower mean baseline weight, so its -13.2% is a smaller absolute number of kilograms than a -13.2% in STEP 1 would be [1]. The same applies to the later East Asian trial in the programme [2]. If you care about kilograms rather than percentages, record baseline weight as its own column and do the multiplication yourself.

Second, waist circumference, HbA1c and blood-pressure changes are reported as confirmatory secondary endpoints in most of these trials, and they are frequently more comparable across programmes than weight is, because they are less sensitive to the lifestyle background. If two trials disagree on weight but agree on waist and HbA1c, that is a hint that the disagreement is about trial conduct rather than about the drugs.

Third, check whether a trial reported body composition. Very few in this programme did, and the ones that used DEXA reported it in a subset rather than the full population. Since the fraction of lost mass that is lean tissue is one of the live questions about this class, a table of percentage weight loss with no composition column is silently treating 15% loss from one agent as equivalent to 15% from another, and nothing in the published data establishes that.

Fourth, note the run-in and screening attrition. Several of these trials screened substantially more people than they randomised, and the withdrawal designs additionally required completing an open-label lead-in on active drug. That means the randomised population in STEP 4 and SURMOUNT-4 consists of people who already tolerated the drug for months, which is a tolerability-selected group. Their regain figures are therefore estimates of what happens to tolerators who stop, not to a general population.

Neither point changes the ranking. All of them change how much confidence you should attach to small differences, and the last one changes how you should read the withdrawal trials specifically.