Why -41.7 and -4.5 Are the Same Size

What the five items are

The pain subscale asks about pain in five specific situations, each scored on a five-point scale from none to extreme. Knowing the items matters, because it tells you what the instrument can and cannot register.

• Walking on flat ground
• Going up or down stairs
• At night while in bed
• Sitting or lying
• Standing upright

The two scoring conventions, and the confusion they cause

On the Likert version each item scores 0 to 4, giving a subscale range of 0 to 20. Many trials instead normalise to 0 to 100. Both are 'the WOMAC pain score' and both are correctly described that way in a paper. The problem arises when results from trials using different conventions are quoted side by side, which is exactly what happened with the two large incretin osteoarthritis trials. STEP 9 reported semaglutide on the 0-100 scale: mean baseline 70.9, change -41.7 points with semaglutide and -27.5 with placebo. TRIUMPH-4 reported retatrutide on the 0-20 Likert scale: mean baseline 6.0, change -4.5 and -4.4 points on the two active arms and -2.4 on placebo.

Why the raw comparison is wrong twice over

Line up -41.7 and -4.5 and semaglutide appears to have relieved pain roughly nine times more effectively. It did not. The first correction is the scale: a factor of five separates 0-100 from 0-20, so -4.5 on the Likert scale corresponds to about -22.5 on the normalised scale. The second correction is the baseline, and it is larger than the first. STEP 9 required at least moderate pain and enrolled a population averaging 70.9 out of 100. TRIUMPH-4 set entry criteria on pain duration and frequency - more than 12 weeks, on more than 15 days of the previous month - plus Kellgren-Lawrence grade 2 or 3 radiographic change, with no pain-severity threshold, and enrolled a population averaging 6.0 out of 20, which is 30 out of 100. The two trials studied populations with roughly twice the difference in baseline pain, and a scale can only fall as far as its floor allows.

Relative change is the least bad common currency

Expressing each result as a percentage of its own baseline removes both problems at once, though not perfectly. On that basis STEP 9's semaglutide arm improved 58.8% and its placebo arm 38.8%; TRIUMPH-4's retatrutide arms improved 75.8% and 74.3% and its placebo arm 40.3%. The residual distortion is that floor effects bite harder at low baselines - it is easier to lose 75% of 6 points than 75% of 70.9 - so relative change flatters the milder population. It is still far better than comparing raw point changes, which is not an approximation but an error.

The placebo response is the most reproducible number here

The single most striking thing in that paragraph is easy to miss: the two placebo arms improved by 38.8% and 40.3% of their own baselines, in different populations, on different scales, in trials run by different sponsors on different continents years apart. Osteoarthritis pain trials produce large and remarkably consistent placebo responses, driven by regression to the mean, the natural fluctuation of osteoarthritis symptoms, the attention and structure of trial participation, and the well-documented responsiveness of self-reported pain to expectation. Any pain trial without a randomised control arm is close to uninterpretable, and any single-arm pain claim should be read against a 40% expectation.

What counts as a meaningful change

Thresholds for minimal clinically important difference in WOMAC vary by population, by version and by how they were derived, and no single number applies everywhere. A commonly used rule of thumb treats a change of around 10% of the scale as the smallest that a person is likely to perceive, with higher thresholds where the question is whether an intervention justified a major procedure. The general point is more useful than any specific number: a statistically significant change on a patient-reported scale is not automatically a change a person would notice, and the two questions have to be asked separately.

Quick reference