STEP 9 is the only dedicated randomized trial of a GLP-1 receptor agonist in knee osteoarthritis, and it reported two primary endpoints at once: how much weight came off, and how much pain went with it. The weight result behaves like every other trial in the program. The pain result is the interesting one, because of how wide the interval around it is and how little the trial’s one behavioral measure moved. What the broader evidence says about this class and this joint — including the threshold at which a pain change becomes noticeable, and the lifestyle programs that compete with it — belongs to the osteoarthritis article. This page is about the trial.
What was run
A 68-week, double-blind, placebo-controlled trial at 61 sites in 11 countries enrolled 407 participants with a body-mass index of 30 or above and a clinical and radiologic diagnosis of moderate knee osteoarthritis with at least moderate pain. Randomization was 2:1 to once-weekly subcutaneous semaglutide 2.4 mg or placebo, with counseling on physical activity and a reduced-calorie diet in both arms. Mean age was 56, mean body-mass index 40.3, mean WOMAC pain score 70.9, and 81.6% of participants were women.[1]
The posted results give the split as 271 and 136, with 269 and 135 in the safety set. Completion ran 246 of 271 and 122 of 136. The reasons for leaving are small and scattered: withdrawal by participant 7 and 8, lost to follow-up 7 and 2, physician decision 2 and 1, failure to meet randomization requirements 2 and 1, and site closure 7 and 2 — a trial that ran from October 2021 losing nine participants to sites that shut.[2]
A body-mass index of 40.3 is worth holding on to. This is not a trial of people at the lower threshold for these drugs; it is a trial of severe obesity with an arthritic knee, and four fifths of it is women.
Two accounts of the same two endpoints
The published abstract and the posted results use different analyses and return different numbers, and both belong to this trial. The abstract reports weight change of −13.7% against −3.2%, and WOMAC pain change of −41.7 points against −27.5, each at P < 0.001, with SF-36 physical-function improvement of 12.0 points against 6.5.[1]
The registry reports weight change of −14.2% (SD 8.6) against −2.5% (SD 5.6), and WOMAC pain change of −43.7 (SD 25.3) against −26.2 (SD 25.0).[2] The pattern is the familiar one: the treated arm looks a little better and the placebo arm a little worse when the analysis conditions on staying in treatment. Quote one set or the other, never a figure from each.
The interval the abstract does not print
The abstract gives P-values and no confidence intervals. The registry gives both, and they are the most informative numbers in the trial. The estimated treatment difference in body weight was −10.48 percentage points (95% CI, −12.34 to −8.63). The estimated treatment difference in WOMAC pain was −14.14 points (95% CI, −19.98 to −8.30), each at p < 0.0001.[2]
Those two intervals are not the same shape. The weight interval spans 3.7 percentage points around a 10.5-point estimate — tight, and every value inside it tells the same story. The pain interval spans 11.7 points around a 14.1-point estimate. At its upper bound the drug removes a fifth of the scale beyond what the control arm removed; at its lower bound it removes eight points, which is little more than half the headline figure. A trial of 407 people cannot say which end it is closer to, and the abstract’s P-value conveys none of that.
The reason for the difference in precision is visible in the standard deviations. Pain change carried an SD of roughly 25 points in both arms, against 8.6 and 5.6 for weight. Pain varies enormously between people given the same treatment; weight does not.
The subscales all move together
Every self-reported measure in the trial separated in the same direction and by a similar amount. WOMAC physical function fell 43.4 points against 25.8, stiffness 45.4 against 27.6, and the WOMAC total 43.8 against 26.0. On the SF-36, physical functioning rose 12.7 against 6.4, bodily pain 12.8 against 7.7 and the physical component summary 13.2 against 6.9. Waist circumference fell 13.3 cm against 5.9.[2]
One scale stayed put. The SF-36 mental component summary changed by 1.9 points on semaglutide and 1.1 on placebo — in a trial where every physical scale moved by more than ten.[2] Losing a seventh of body weight and reporting substantially less knee pain registered almost nothing on the mental-health half of the instrument the trial itself selected.
The one column that records behavior
Everything above is what participants said on a questionnaire, in a trial where one arm was losing 14% of its body weight and the other was not, so blinding was under strain by the end. The trial carried one measure of what participants did about pain: use of permitted rescue analgesics during the washout before assessment.
It did not separate. 4.9% of the semaglutide group and 5.1% of the placebo group used rescue analgesics during washout, and the amounts used were 224.2 and 170.1 units against standard deviations of 1,756.4 and 1,100.3 — dispersions many times the means themselves, which is what a column driven by a handful of heavy users looks like.[2] No treatment difference or significance test is posted against either figure.
That is one secondary outcome and it should not be inflated into a refutation. Washout use is a narrow window, the counts are small, and a questionnaire administered under protocol is a legitimate endpoint. But it is the only place in STEP 9 where reported pain and acted-on pain can be compared, and they do not agree.
Side effects, and a reversal in the joint column
Adverse events leading to permanent discontinuation of the trial regimen occurred in 6.7% of the semaglutide group against 3.0% on placebo, gastrointestinal disorders being the most common reason, and the incidence of serious adverse events was similar in the two groups.[1] The posted table puts serious events at 27 of 269 against 11 of 135, which is 10.0% against 8.1%.[2]
The symptom counts are the usual ones: nausea 59 of 269 (21.9%) against 12 of 135 (8.9%), constipation 32 against 7, diarrhea 21 against 7, and vomiting 21 against 1.[2] What those feel like in practice, and how long they last, is covered in the side effects article.
One line runs the other way. Arthralgia — joint pain recorded as an adverse event rather than as a questionnaire score — was reported by 8 of 269 participants on semaglutide (3.0%) and 13 of 135 on placebo (9.6%).[2] Those are small counts with no test attached, and they are the only place in the trial where the safety table and the efficacy table point the same way about the joint.
How the weight result compares with the rest of the program
The weight half of STEP 9 is unremarkable, which is itself informative. In the pivotal obesity trial of the same drug at the same dose over the same 68 weeks, 1,961 adults without an arthritis requirement lost 14.9% against 2.4%, an estimated treatment difference of −12.4 percentage points (95% CI, −13.4 to −11.5), with 4.5% discontinuing for gastrointestinal events against 0.8%.[3] That trial is covered in the STEP 1 article, and the dose ladder behind both is in the semaglutide dose article.
So a population with a mean body-mass index of 40.3 and painful knees lost essentially what the general obesity population lost. The joint did not make the drug work less well on weight, and the extra weight did not make it work better. What the drug did for the knee has to be read out of the pain columns alone.
What STEP 9 did not establish
It ran no imaging endpoint. Cartilage, joint space and bone were assessed at entry to confirm the diagnosis and never remeasured as an outcome, so the trial says nothing at all about whether the joint itself did better, held steady or continued to deteriorate. A 68-week pain improvement and a structural effect are separate claims and only one of them was tested.
It ran no active comparator: no anti-inflammatory, no physiotherapy program, no intra-articular injection, no supervised diet and exercise arm. Both groups got counseling; neither got a competing treatment. It also ran no arm that lost weight without the drug, so nothing in the trial separates the mechanical effect of a lighter body from anything else semaglutide does. And it did not withdraw treatment, so whether pain returns when the injections stop is untested here — what is known about stopping is in the discontinuation article.
Nor does it generalize easily. The trial enrolled severe obesity with moderate radiographic disease and at least moderate pain, four fifths of it women, at a mean age of 56. Someone with a body-mass index of 31 and early knee pain is outside the enrolled range, and no regulator has approved any drug in this class for osteoarthritis.
What was in the injection
Every figure above belongs to branded, FDA-approved semaglutide supplied free, titrated under protocol to 2.4 mg weekly for 68 weeks alongside structured diet and activity counseling, in a population selected by radiographic knee disease as well as by weight. Sellers listed on the compounded semaglutide 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 number to be careful with here is 41.7, or 43.7 depending on the analysis. Neither is the drug’s effect. The drug’s effect is 14.14 points with an interval running from 19.98 down to 8.30, and the remainder belongs to a control arm that received counseling, attention and a placebo injection for sixty-eight weeks.