The NSAID problem

Ibuprofen after every run. Naproxen before bed. It's the default answer to joint pain — yet a line of trials and cohort studies suggests some NSAIDs may accelerate cartilage loss in osteoarthritic joints, while other analyses find little structural effect at all. The key to reading this literature: "NSAIDs" is not one exposure. Indomethacin, diclofenac, ibuprofen and the COX-2-selective agents come apart from each other in the same experiments, and the class-wide question and the drug-specific one have different answers.

Where the harm signal comes from

The most consistent human harm data belong to one drug: indomethacin. In a 1989 Lancet trial of 105 hip-OA patients, those on indomethacin — a strong prostaglandin-synthesis inhibitor — lost hip joint space faster than the contralateral hip and reached arthroplasty sooner, while patients on azapropazone, a weak inhibitor, showed no such asymmetry. The larger LINK study randomized 812 knee-OA patients: among completers, radiographic deterioration hit 40 of 85 on indomethacin (25 mg three times daily) versus 19 of 85 on placebo — more than double, p=0.009. A 1979 hip series had already flagged the same pattern, the original "indomethacin hip."

The population-based Rotterdam Study extended the concern to diclofenac: long-term users progressed faster on radiographs than non-users. The finding was drug-specific — most other NSAIDs examined showed no such association — which fits a 12-year human-explant program showing NSAIDs split into three classes by effect on cartilage matrix synthesis: stimulatory, neutral, and inhibitory (naproxen, ibuprofen, and indomethacin sit in the inhibitory group). Notably, tiaprofenic acid — "neutral" in the explant data — did not differ from placebo in LINK, and celecoxib, a COX-2-selective agent, actually reversed cytokine-induced proteoglycan loss in explants.

The modern cohorts

The modern large-cohort data point the other way. An Osteoarthritis Initiative new-user analysis of 1,846 participants found sustained prescription NSAID use trending, non-significantly, toward less joint-space narrowing. A 2024 analysis of 4,197 people (8,394 knees) across the OAI, MOST and CHECK cohorts found long-term NSAID users had worse symptoms and three-fold higher knee-replacement odds; on radiographic deterioration its main analysis found 1.4-fold odds of a worse grade, and the propensity-matched analysis the authors lead with did not reach significance. And the 2025 OAI imaging analysis of serial knee MRIs — 142 new users against 707 matched non-users over four years — found no association with synovitis on any measure, and cartilage lesions progressing significantly more slowly in the NSAID users, more so in those with an inflammatory imaging phenotype. Its "limited effects" title is about the synovitis endpoints, not the cartilage one.

So the harm case rests on two small 1980s–90s trials of one drug plus observational diclofenac data — LINK lost over half its patients before one year, making its completer analysis fragile — while the class-wide question comes up null in every modern cohort.

The fundamental problem with every observational study in this entry: people who take more NSAIDs have more painful knees, and more painful knees progress faster. Confounding by indication can manufacture a harm signal out of nothing — or mask a real one. It cuts the other way too: better analgesia can hasten elective surgery, mimicking harm on time-to-arthroplasty endpoints.

What I take from it

Occasional use for a flare is a different exposure from a daily habit — every harm signal in this literature comes from sustained, full-dose daily use, and no study indicts a short course. The class question is genuinely unresolved (no RCT of a modern OTC ibuprofen regimen with cartilage MRI endpoints exists), but the direction of prudence is clear: treat NSAIDs as a short-term tool, not maintenance. If a daily NSAID is truly needed and cartilage is the concern, both the trial data and the explant hierarchy point away from indomethacin, and COX-2-selective or "neutral" agents carry no human harm signal — though no proven structural benefit either. And if you need them daily, that is information about the joint that deserves a workup rather than a standing prescription. The established GI, cardiovascular, and renal risks of chronic use argue the same way, independently of cartilage; injected corticosteroids are a separate precaution covered under injections.