Accepted answer
For weight loss the population dose-response curves in this class are clearly concave and still rising at the top rung — flattening, but not flat. The marginal return per milligram falls sharply as you climb, which means the top rung is real but expensive. Here is the data laid out so you can see the shape rather than take my word for it.
Published dose-response, weight change from baseline
| Agent and trial | Duration | Dose | Weight change | Marginal gain over previous rung | Gain per extra mg |
|---|
| Tirzepatide, obesity phase 3 | 72 weeks | placebo | −3.1 % | — | — |
| Tirzepatide | 72 weeks | 5 mg | −15.0 % | +11.9 pts | 2.38 pts/mg |
| Tirzepatide | 72 weeks | 10 mg | −19.5 % | +4.5 pts | 0.90 pts/mg |
| Tirzepatide | 72 weeks | 15 mg | −20.9 % | +1.4 pts | 0.28 pts/mg |
| Retatrutide, phase 2 | 48 weeks | placebo | −2.1 % | — | — |
| Retatrutide | 48 weeks | 1 mg | −8.7 % | +6.6 pts | 6.60 pts/mg |
| Retatrutide | 48 weeks | 4 mg | −17.1 % | +8.4 pts | 2.80 pts/mg |
| Retatrutide | 48 weeks | 8 mg | −22.8 % | +5.7 pts | 1.43 pts/mg |
| Retatrutide | 48 weeks | 12 mg | −24.2 % | +1.4 pts | 0.35 pts/mg |
| Survodutide, phase 2 | 46 weeks | placebo | −2.0 % | — | — |
| Survodutide | 46 weeks | 2.4 mg | −12.5 % | +10.5 pts | 4.38 pts/mg |
| Survodutide | 46 weeks | 3.6 mg | −15.0 % | +2.5 pts | 2.08 pts/mg |
| Survodutide | 46 weeks | 4.8 mg | −16.2 % | +1.2 pts | 1.00 pts/mg |
| Survodutide | 46 weeks | 6.0 mg | −18.7 % | +2.5 pts | 2.08 pts/mg |
| Semaglutide, obesity with T2DM | 68 weeks | placebo | −3.4 % | — | — |
| Semaglutide | 68 weeks | 1.0 mg | −7.0 % | +3.6 pts | 3.60 pts/mg |
| Semaglutide | 68 weeks | 2.4 mg | −9.6 % | +2.6 pts | 1.86 pts/mg |
Sources in order: the tirzepatide obesity phase 3 [1], the retatrutide phase 2 [2], the survodutide phase 2 [3], and the semaglutide trial in type 2 diabetes [4]. Durations and populations differ, so read down each block and not across them.
What the shape tells you
- The last rung is worth about a fifth of the second-to-last rung, per milligram. Tirzepatide 10 to 15 mg buys 1.4 points for 5 mg; 5 to 10 mg bought 4.5 points for the same 5 mg. Retatrutide 8 to 12 mg is the same pattern: 1.4 points for 4 mg after 5.7 points for the previous 4 mg.
- Most of the effect is in the first therapeutic rung. Tirzepatide 5 mg captured 11.9 of the eventual 17.8 points of placebo-adjusted effect — 67 % of the total at one third of the maximum dose. Retatrutide 4 mg captured 15.0 of 22.1 points, 68 %, at one third of the maximum dose. That is a strikingly consistent number across two different molecules and it is the single most useful thing in the table.
- The curves have not turned over. No arm in any of these programmes did worse at a higher dose on the weight endpoint. So there is no dose at which more became less; there is only a dose at which more became marginal.
- Survodutide is the exception that proves the noise floor. Its 4.8 mg arm underperformed the interpolation between 3.6 and 6.0 mg. In a phase 2 trial with arms of a few dozen participants that is unremarkable sampling variation, and it is a reminder that these dose-response curves are drawn through points with confidence intervals wide enough to hide a rung.
The three categories you asked about
Highest studied is the easiest and it is often above the label. Phase 2 programmes routinely test doses that never appear on a label, either because the increment was not worth the tolerability or because the sponsor chose a simpler ladder for commercial reasons.
Highest approved is a regulatory judgement about acceptable benefit-risk in a population, made once, for everyone. It is not a maximum tolerated dose in the oncology sense — nothing in this class was escalated to toxicity. It is the top of the range the sponsor filed on.
Maximum useful is not a single number, because it depends on what you are counting. On the weight endpoint the marginal-return columns above suggest the useful maximum sits below the label maximum for a lot of people — the 5 mg tirzepatide arm and the 4 mg retatrutide arm delivered two thirds of the effect. On glycaemic endpoints the curve flattens even earlier, because HbA1c has a floor and dose-response saturates against it: in the head-to-head diabetes trial, tirzepatide produced HbA1c reductions of roughly −2.01, −2.24 and −2.30 % at 5, 10 and 15 mg [5]. Tripling the dose bought 0.29 percentage points of HbA1c. The glycaemic curve is close to flat above 5 mg and the weight curve is not, which is a genuinely important asymmetry and the reason the same molecule has different label maxima for the two indications.
What follows practically is nothing, in the sense that where an individual should sit on that curve is a clinical judgement involving their tolerability, their comorbidities and their goals, and it is not something a marginal-return table decides. What the table does is dispose of the idea that the top rung is where the drug "properly" works.
edited 13 Mar 2026 by rota_site — tightened the wording; no substantive change
4Two thirds of the effect at one third of the dose, in two unrelated molecules. That is the number I will remember from this. – Dr_Elias_Weiss 9 months ago 3The glycaemic-versus-weight asymmetry explains the split label maxima and I had never connected the two. – ravenna_pace 8 months ago 6Marginal points per milligram is a much better way to present dose-response than the usual bar chart. – e_dziedzic 6 months ago add a comment