Body weight & the knee

If this section ranked topics by evidence per dollar spent, body weight would win by a distance. It is the only foods topic with a large randomized trial behind it, and the only one where independent datasets converge on actual numbers you can aim at.

Two mechanisms, both demonstrated in humans

Weight works on the knee twice. Mechanically, each kilogram multiplies through joint compressive load — and in the IDEA trial, diet-induced weight loss measurably reduced knee compressive forces compared with exercise alone (2487 versus 2687 newtons). Metabolically, fat tissue is inflammatorily active: the same trial found weight loss lowered plasma IL-6, one of the inflammatory mediators that drive cartilage catabolism. Most entries in this guide would kill for one mechanism demonstrated in humans; this one has two, from a single randomized trial.

The randomized evidence

IDEA randomized 454 overweight and obese adults aged 55 and up with radiographic knee osteoarthritis to 18 months of intensive diet, exercise, or both. The diet-plus-exercise group lost a mean 10.6 kg and beat exercise alone on pain, function and walking, and beat diet alone on the same three.

Its two prespecified primary endpoints were not those. They were IL-6 and knee compressive force, and there the picture is narrower: IL-6 was lower in both diet arms than in exercise alone, while compressive force separated only for diet against exercise. Adding exercise to the diet did not lower joint load further. Diet alone matched diet plus exercise on both of those mechanisms and delivered half the pain reduction, which the authors report and say they cannot explain. A meta-analysis of four weight-loss RCTs points the same way, with modest pooled effects: 0.23 for disability and a borderline 0.20 for pain — and a useful threshold, with disability improving reliably once loss exceeded about 5 percent of body weight.

The largest and longest test arrived in 2024, and it used a drug. STEP 9 randomized 407 adults with a mean BMI of 40.3 and moderate knee osteoarthritis to once-weekly semaglutide or placebo for 68 weeks, both arms receiving activity and diet counselling. Weight fell 13.7 percent on semaglutide against 3.2 percent on placebo, and WOMAC pain fell 41.7 points against 27.5 on a 0–100 scale (p < 0.001), with physical function moving the same way.

Read the placebo arm before the headline. Those participants lost 3.2 percent of their body weight on counselling alone and their pain still fell by 27.5 points — about two-thirds of the semaglutide arm's improvement. What the trial demonstrates cleanly is that the relationship holds at scale in a properly blinded design; what it adds on top of lifestyle advice is real and smaller than the headline number. It was funded by Novo Nordisk, with company employees among the authors, and it has no imaging endpoint at all.

A network meta-analysis of 13 randomized trials in 2,800 people put seven weight-loss strategies against each other and found the two halves of this entry coming apart. Four of the seven beat control on weight, from −11.2 kg for psychological-plus-diet-plus-exercise to −4.7 kg for diet-plus-exercise; a Mediterranean diet, drug treatment and exercise alone did not. On pain, only diet-plus-exercise beat control — the approach that removed the most weight did not, and no two approaches differed from each other at all. On physical function that same approach was the strongest thing in the network, and quality of life moved for nobody. What predicted relief across the network was the amount lost: a threshold around 7 percent of body weight for pain and 8.3 percent for function, whatever route took it off.

Read that one at arm's length. An Expression of Concern was published on it in November 2024. Twelve of its thirteen trials are rated at high risk of bias. And its own abstract says every strategy beat control on weight, where its results section says four did — which is the sentence most likely to be quoted onward. The thresholds agree with the older literature, which is why they are repeated here, and they are flagged rather than leaned on.

The dose-response

Three independent datasets stack into a consistent gradient. In a 1,383-person community program, every additional increment of weight lost improved every knee symptom subscale, with roughly 7.7 percent loss needed for a clinically important functional gain. In 640 Osteoarthritis Initiative participants followed with MRI for four years, cartilage degeneration scores progressed 1.0 with more than 10 percent loss, 1.6 with 5–10 percent, and 2.3 with stable weight — though when that study asks the simpler question of whether a knee got worse at all, only the group above 10 percent separates from stable weight, and the middle stratum does not. And in 111 obese adults losing weight through surgery or diet, about 9 percent loss was associated with preserved medial femoral cartilage thickness and improved proteoglycan quality on dGEMRIC imaging, with an estimated threshold around 7 percent.

The working numbers: more than 5 percent of body weight for reliable symptom benefit, 7–8 percent for clinically important function gains and measurable cartilage effects, above 10 percent for the largest differences. And in IDEA, diet plus exercise beat either alone on how people felt and moved, though not on the two mechanisms the trial was designed around.

The limits of the weight-loss trials

The cartilage-structure evidence is entirely observational — people who lose weight differ from people who don't, and no randomized trial has shown weight loss changes cartilage structure. "Slower degeneration" is also not regrowth: the imaging studies show preservation, never thickening.

A systematic review makes that limit precise. Fourteen articles from six cohorts, only two of them from randomized trials, were assembled to ask what weight loss does to imaging in hip and knee osteoarthritis. Direct cartilage thickness measurements conflicted with each other. Compositional dGEMRIC and T2 mapping in early knee OA showed trends toward less degeneration and no more than trends. Joint space width on radiographs did not change. Synovitis and effusion were unaffected. Two things did move: the infrapatellar fat pad got smaller, and DXA showed bone loss at the hip. The authors' conclusion is that no consistent evidence of a structural effect exists — and that the field has not agreed on what to measure.

That is the boundary of this entry's tier. Symptoms, function and loading are established by randomized evidence. Structure is not.

The randomized pain effects, while real, are modest — weight loss helps a painful knee; it does not cure one. And the evidence has boundaries: IDEA was single-blind in adults 55 and older with established OA, the dose-response cohort had no control group, and rapid bariatric-surgery loss may not generalize to dieting. For younger post-injury knees, everything here is extrapolation.

Practical notes

Weight loss in older adults costs muscle and bone as well as fat — one reason IDEA paired the diet with exercise, and a good reason to copy that design rather than dieting alone. Very-low-calorie and surgical approaches need medical supervision. And the advice has a floor: for normal-weight people there is no evidence that losing more helps a knee.

There is also a route to a knee that is not about load at all. In 615 people at risk of knee osteoarthritis but free of pain and radiographic disease, an ultra-processed diet tracked with fatty infiltration of thigh muscle on MRI — and thigh muscle is what unloads a knee. Body composition, not just body mass, may be part of what these trials are moving.

The open question the field is now testing is whether GLP-1-scale weight loss changes cartilage structure in a randomized design. STEP 9 established the symptom half at 13.7 percent weight loss and collected no imaging; running the same design with MRI would settle the one claim this entry cannot yet make.