Leg strengthening

The quadriceps are the knee's shock absorbers. Weakness increases the rate of loading on cartilage, and the cleanest prospective evidence that this matters comes from asking how much of a known risk gap runs through muscle. In the Osteoarthritis Initiative, 5,126 knees with no radiographic osteoarthritis at baseline were followed eight years; women reached lateral radiographic OA at 5.9% against 3.4% in men. Knee extensor strength accounted for about a third of that difference and flexor strength about the same — a modifiable variable carrying a third of a risk gap. Two qualifications travel with it: this is observational, so strength may be a marker as much as a cause, and two-thirds of the difference runs through something else entirely. Strengthening improves load distribution and neuromuscular control — and does so without any imaging or biomarker signal of cartilage harm.

The pooled trial evidence

The Cochrane review of 54 RCTs found land-based exercise reduces knee OA pain by a standardized mean difference of 0.49 — 12 points on a 0–100 scale (95% CI 10 to 15), against a control group sitting at 44 — which its authors call comparable with published estimates for simple analgesics and NSAIDs taken for knee pain. Physical function improved on moderate-quality evidence, about 10 points, and benefits persisted 2–6 months after programs ended. People were no more likely to drop out of the exercise arms than the control arms: 14% against 15%.

The largest comparison of exercise types puts that in context. Across 217 randomised trials and 15,684 people, strengthening exercise probably produces a large improvement in function at twelve weeks (SMD 0.86, 95% CI 0.53 to 1.18, moderate certainty) — and aerobic exercise ranked ahead of it on pain, function, gait and quality of life, with the highest probability of being the best modality overall. Strengthening is one strong route, not the only one, and on this evidence not even the strongest.

How hard, and how much

A network of 46 resistance-training trials and 3,463 people compared seven ways of doing it. All seven beat control on pain. The one that came out ahead is not the one most programmes prescribe: high-speed training — concentric contractions performed at maximum speed — led on pain (SMD 1.35), stiffness (1.26) and function (1.70), and beat every other type on stiffness in the head-to-head comparisons. Its dose analysis puts the optimum for pain and function at a moderate load, 43–47% of one-repetition maximum, sustained a long time: 35–37 weeks at 610–640 repetitions a week. Stiffness responded to the opposite prescription, more repetitions over a shorter block.

Two cautions about that. The dose figures come from modelling across trials rather than from anyone randomising people to different doses, so they describe an association between study characteristics and effect size. And speed is not the same instruction as load: high-speed training is still resistance training, performed with intent, not thrown.

A wording note that matters: the trials behind this tier used progressive resistance, without tempo prescriptions, and pain-monitoring evidence permits acceptable pain — up to about 5/10, provided it settles by the next morning — rather than requiring a strictly pain-free range. "Controlled tempo, tolerable range" is what the evidence supports; "slow tempo, pain-free only" is not what was tested. Isometrics can serve as the entry ramp when joint irritability limits range.

One adjunct has been measured against the programme rather than beside it: 11 randomized trials and 2 systematic reviews added low-level laser to an exercise programme in knee osteoarthritis, and the combination came out generally superior to exercise alone, though some of the systematic reviews disagree. The endpoints are pain, function and range of motion over weeks, so what the light adds is to how the knee feels, not to what the cartilage does.

The structural and symptom endpoints

The MOST cohort supplies two null results that have to be read together. Across 2,275 knees with no frequent symptoms at baseline, neither knee-extensor strength nor the hamstring-to-quadriceps ratio predicted the onset of frequent knee pain or stiffness at 15 or 30 months. A second MOST analysis used a stricter endpoint — a new Kellgren-Lawrence grade of 2 or more together with frequent symptoms — and there the strongest quadriceps tertile carried well under half the odds of the weakest (OR 0.43, 95% CI 0.31–0.61). Against the radiographic grade on its own, strength did nothing; body mass index was the only consistent predictor.

Those results sit together rather than in conflict. Strength tracked who crossed into diagnosable symptomatic disease — not who simply started hurting, and not who accumulated radiographic change. No trial shows strengthening changes cartilage structure. Strength is protective for how the knee feels and functions — it is not a proven shield for the cartilage itself, and it is not a complete shield for symptoms either.

Two further caveats: participants can never be blinded in exercise trials, so effect sizes include expectation effects; and the largest network comparison finds strengthening is not required for symptom benefit — aerobic exercise outranked it on every symptom outcome measured.

Safety

Exercise therapy shows no cartilage harm on imaging across RCTs, and biomarkers of cartilage breakdown do not rise with structured programs. The one signal worth respecting: in an RCT using heavy 80% 1RM training in knee OA, 4 of 16 participants dropped out with exercise-induced knee pain, versus none in the low-load blood-flow-restriction arm. Heavy loading works, but it needs a graded introduction in painful knees — start lower, progress by the load-management rules, and use isometrics or BFR as on-ramps when direct heavy loading flares symptoms.

Open questions

Whether actually achieving the modeled 30–40% strength gain causally delivers the predicted symptom benefit, and whether strength gains slow structural progression over horizons longer than five years — the trials to answer both have not been run.