Comparison of weight loss interventions in overweight and obese adults with knee osteoarthritis: A systematic review and network meta-analysis of randomized trials

Shahid A, Thirumaran AJ, Christensen R, Venkatesha V, Henriksen M, Bowden JL, Hunter DJ · Osteoarthritis and Cartilage

Separates losing weight from feeling better, and the two come apart differently on each outcome. Four of the seven strategies beat control on weight, from −11.2 kg (95% CI −16.0 to −6.5) for psychological + diet + exercise down to −4.7 kg (−6.7 to −2.7) for diet + exercise; Mediterranean diet (−4.3, −8.9 to 0.3), pharmacological (−4.0, −9.1 to 1.1) and exercise alone (−0.4, −2.7 to 1.9) did not, which the abstract states the other way round. On pain, only diet + exercise beat control (−2.2, 95% CI −4.1 to −0.2); the strategy that took off the most weight did not (−3.9, −8.4 to 0.5), and no two strategies differed from each other. On physical function that same strategy was the strongest in the network (−15.8, −22.2 to −9.3), with diet + exercise the only other to separate (−3.2, −6.3 to −0.2). Nothing improved quality of life. Meta-regression infers a threshold of at least 7% body-weight loss for pain and 8.3% for physical function, at roughly 0.20% and 0.26% a week, whatever the route.

Population

2,800 participants across 13 randomized trials in overweight and obese adults with knee osteoarthritis

Intervention

Seven strategies networked against each other — diet, exercise, diet + exercise, pharmacological, psychological, psychological + diet + exercise, and Mediterranean diet

Comparator

Control comparators and each other

Limitations

An Expression of Concern was published in Osteoarthritis and Cartilage in November 2024 (doi 10.1016/j.joca.2024.10.001, indexed against this record at PubMed), and the version of record circulating in repositories does not carry it. Twelve of its thirteen trials are rated high risk of bias, 8% of them blinded participants, and the abstract's claim that every strategy beat control on weight is contradicted by its own results section. The thresholds are indirect inferences from meta-regression on aggregate study-level data, a method whose pitfalls the paper's own reference list cites. Network meta-analysis compares interventions that were never randomised against each other, across seven nodes and thirteen trials.