Intra-articular ozone

Ozone is oxygen with a third atom attached. It is unstable — about forty-five minutes of half-life at room temperature — so it cannot be bottled and shipped like a drug. A generator in the clinic makes it from pure oxygen and it goes into the joint immediately, which is why the delivered concentration is a property of the room rather than of a label.

That fact shapes the whole literature. There is no manufacturer, no licensed product and no registration dossier, so the trials are investigator-run, small, and spread across concentrations from 10 to 35 micrograms per millilitre with no agreement on schedule. What there is, unusually for something in this position, is a blinded trial against a sham injection.

The blinded trial

Ninety-eight patients aged 60 to 85, with Kellgren-Lawrence grade II or III knee osteoarthritis, were randomised to one intra-articular injection a week for eight consecutive weeks: ozone at 20 micrograms per millilitre in a 10 millilitre volume, or 10 millilitres of air. Both patients and assessors were blinded. Follow-up ran to sixteen weeks, eight weeks past the last injection.

At eight weeks the ozone group was ahead on the visual analogue pain scale (mean difference 2.16, 95% CI 0.42–3.89), on the Lequesne index (4.05, 95% CI 1.10–7.00), on the Geriatric Pain Measure (18.94, 95% CI 3.43–34.44), on all three WOMAC subscales, and on every SF-36 domain the trial measured. The Timed Up and Go test — the one performance measure rather than questionnaire — showed no difference between the groups.

At sixteen weeks WOMAC pain was still separated (median of differences 14.999, 95% CI 0.000–25.000, P = 0.005). WOMAC stiffness had converged by then.

Two details sit only in the full paper. The allocation was not balanced: 63 patients to ozone and 35 to placebo, a ratio of 1.75 to 1, because the total in each group was never controlled during assortment — the authors say so, and accept the loss of statistical efficiency. And a correction published in 2024 swaps the placebo and treatment columns of the WOMAC table, which had been printed the wrong way round for seven years.

Against a corticosteroid injection

Seven randomised trials and 409 patients compare ozone with intra-articular corticosteroid — the comparison that matters most in a clinic, because corticosteroid is one of the two injections international guidelines actually recommend.

Pain favoured ozone at every horizon: short term SMD −0.30 (95% CI −0.53 to −0.08, I² = 0%), medium term SMD −0.93, long term SMD −1.05. Removing the single high-risk trial with the largest effect brought the medium-term figure down to SMD −0.61 and it stayed significant.

Function does not follow pain here, and the split is worth carrying. There was no difference between ozone and corticosteroid on function in the short term (SMD −0.02, 95% CI −0.40 to 0.37) and none in the long term (SMD −0.87, 95% CI −2.15 to 0.40). Only the medium term separated. The reviewers grade short-term pain as moderate evidence and medium-term function as low.

Read the heterogeneity alongside the effect sizes: the short-term pain analysis was homogeneous at I² = 0%, and every medium- and long-term estimate ran between I² = 83% and 92%. The largest effects come from the least consistent pools.

How long it lasts

One synthesis pooled sixteen trials and 1,172 participants by follow-up point instead of averaging across them, and it is the most useful thing in this literature.

At one month WOMAC total improved (SMD −0.63). At two months WOMAC pain (SMD −1.39), WOMAC function (SMD −1.27) and the pain scale (SMD −1.25) all improved, and that is the peak. At three months nothing differed on any outcome. At twelve months the ozone arm scored worse than its controls, on WOMAC total (SMD 0.63) and on pain (SMD 1.74).

That last number is the one most likely to be quoted out of context, and the authors take it apart themselves. The long-term control arms were almost entirely hyaluronic acid or platelet-rich plasma — treatments chosen because they last. Strip the active comparators out and the single placebo-controlled trial available at six months favours ozone strongly (SMD −1.79, 95% CI −2.52 to −1.06), against a pooled estimate of +0.57 with them left in. Their reading is that the twelve-month result reflects the disadvantage of one course of a short-acting treatment against long-acting ones, not harm from the ozone.

A single course works for about two months. Whether repeating it sustains anything is untested — no trial in that review evaluated repeat dosing.

Against hyaluronic acid, and added to PRP

A level I meta-analysis of 424 patients found no difference in pain between ozone and hyaluronic acid at final follow-up, over a window restricted by design to between four and six months.

A network meta-analysis of eighteen trials and 1,376 participants asked whether adding something to platelet-rich plasma helps. Hyaluronic acid did, on both function and pain. Ozone did not separate from PRP alone — and the authors are explicit that this is not a finding: the entire ozone comparison rests on one small trial, the confidence intervals are wide, and they write that the absence of statistical significance cannot be read as evidence of no effect. That is a missing trial rather than a negative one, and the difference matters.

What the trials have not measured

No study behind this entry measured cartilage, joint space width, or any imaging or turnover endpoint.

That is not an oversight anyone has hidden. The blinded trial's authors name the absence of imaging as their first limitation. The time-course review reports that its sixteen trials carried no imaging progression, no biomarkers and no long-term safety data. The corticosteroid comparison reports no structural endpoint at all. So every claim on this page is about how a knee feels and what it can do, over weeks to a few months.

The same review raises the question from the other side, and nobody has answered it: what repeated oxidative stress does to cartilage over time.

Quality of the evidence base

Worth stating plainly, because it is the reason this entry does not sit higher.

Of the seven trials in the corticosteroid comparison, four were rated unclear risk of bias and three high — none low, with the problems concentrated in selection, performance and detection bias. A separate systematic review of eleven trials and 858 patients found that none of them reached its standard for good quality: two were fair and the rest poor.

The time-course review could not assess publication bias at all, because fewer than ten studies contributed at every follow-up point, and it flags that as a limitation of its own work. It also names the small-study effect directly — a mean of about 73 participants per trial — and points out that its one-to-two month conclusion rests on placebo-controlled trials that all came out positive, so unpublished negative ones would move it.

Safety

Three adverse events across the 98 patients of the blinded trial, all puncture accidents, and two of those were in the placebo arm. The eleven-trial review reports no major complications or serious adverse events after intra-articular ozone in any included study.

The contraindications the trial authors list are acute alcohol intoxication, recent myocardial infarction, haemorrhage from any organ, pregnancy, hyperthyroidism, thrombocytopenia and ozone allergy. They state there is no restriction on use in elderly people, which is the population their trial enrolled.

One piece of standing worth knowing beside the safety record: the only intra-articular injections international guidelines recommend for knee osteoarthritis are corticosteroids and hyaluronic acid. Ozone is used widely and is not among them.

What would change the tier

A placebo-controlled trial long enough to separate two possibilities the current evidence cannot: that the effect of ozone ends at three months, or that a single course does. Every long-term comparison so far runs against an active treatment, and at twelve months no placebo-controlled study exists at all.

For a claim about the joint rather than the symptom, the study is one built on an imaging or turnover endpoint from the start. None has been run.