Two people can walk into two different clinics with the same knee. Same age, same X-ray grade, same stiff first ten minutes of every morning. Both get a course of low-level laser therapy. One of them notices a real difference. The other finishes the course and feels about the same as when they started.
The easy explanation is biology — different bodies respond differently. Sometimes that is genuinely the answer. But the published evidence points somewhere less mysterious and considerably more useful: a large share of that difference is the dose. Not whether a laser was used, but which wavelength, how much energy, delivered to how many points, how often, and over how long.
That is not a theory we invented. It falls out of the numbers in the largest meta-analysis on the question, and those numbers are specific enough to be worth walking through slowly.
What 22 trials found
Stausholm and colleagues, BMJ Open 2019 pooled 22 randomized placebo-controlled trials covering 1,063 patients with knee osteoarthritis. Every trial compared real laser against a sham device, so patients could not tell which one they had received.
Pain in these trials is measured on a visual analog scale: a 100-millimeter line, where 0 is no pain and 100 is the worst pain imaginable. You mark where you are. It is crude, but it is consistent, and it lets researchers add results across studies.
Across all 22 trials, average pain at the end of the treatment course fell by 14.23 mm (95% CI 7.31 to 21.14) more than sham.
Fourteen millimeters out of a hundred. That is a real effect, statistically speaking, and it is also a modest one. If you read that number and concluded that laser therapy for knee arthritis is a marginal intervention not worth rearranging your week for, you would be reading it reasonably. Plenty of clinicians stopped there.
The interesting part is what happens when you stop treating those 22 trials as if they were all doing the same thing.
The split
The World Association for Photobiomodulation Therapy (WALT) publishes dosage recommendation tables — specific guidance on wavelength, power output, energy delivered per treatment point, number of points, and treatment frequency. They are not laws of nature. They are a consensus attempt to write down the settings that the underlying research supports.
Stausholm’s team sorted the trials by whether the dose actually used fell within those recommendations. The pooled results separated:
- Trials using recommended doses: 18.71 mm (95% CI 9.42 to 27.99)
- Trials using non-recommended doses: 6.34 mm — and non-significant at follow-up
- Difference between the two subgroups: p = 0.02
That p-value is worth being precise about. It means the gap between the two groups is unlikely to be chance alone. It does not mean the question is settled. This is a subgroup analysis — the researchers divided existing trials after the fact rather than randomly assigning people to a good dose or a bad dose. Subgroup findings are weaker evidence than a trial built to test the question directly, and no such trial exists yet. What it does mean is that pooling every laser study into one average number produces something close to meaningless: you are averaging treatments that were delivered and treatments that were, in effect, only performed.
The 6.34 mm figure deserves its own sentence. Trials that used doses outside the recommendations did not just do less well. Their benefit was no longer statistically distinguishable from sham once follow-up came around. The wrong dose did not produce a smaller version of the effect. It produced something that faded into nothing.
The delayed peak, and why it should change how you judge your own results
Here is the finding that most changes how a patient should think about this.
In the recommended-dose trials, the effect did not top out on the last day of treatment. It kept climbing after the course ended, reaching a peak of 31.87 mm (95% CI 18.18 to 45.56) at 2 to 4 weeks after the final session.
Roughly double the group’s end-of-course result, arriving after the treatments had stopped.
Most people evaluate a treatment on the day it ends. You finish the last visit, take stock, and decide whether it did anything. If these trial averages reflect what happens in practice, that is the worst possible moment to decide. You would be judging on the way up.
So the practical version of this finding is: the assessment that matters is the one two to four weeks out, not the one on the final visit. A clinic that scores your pain at the end of the course and never contacts you again is not measuring the thing the research says is most informative — and neither are you, if you decide on the last day.
Two honest caveats. First, that confidence interval is wide — 18.18 to 45.56 — which tells you the true average is uncertain even within these trials. Second, and more important, this is a group average. It describes what happened to populations of patients in studies. It is not a schedule for your knee, and some people in those trials did not improve at all.
The limitation we are obligated to state
Heterogeneity across these trials was very high: I² of 93 to 95%.
I² estimates how much of the variation between study results comes from real differences between the studies rather than random noise. At 93 to 95%, the trials were substantially not measuring the same thing. Different lasers, different protocols, different patients, different outcomes.
That is a genuine problem, and it cuts in an interesting direction here. Dose variation is one plausible source of that scatter — which is part of the argument the paper is making. But heterogeneity stayed high even within the subgroups, so dose does not explain all of it. Anyone who tells you this literature is clean is not describing it accurately.
A second review, and a lower grade
Oliveira and colleagues, Physical Therapy 2024 looked at 10 studies covering 542 participants and found a benefit for pain at rest of SMD −0.7 (95% CI −1.1 to −0.2).
They graded their own certainty in that finding as very low.
Their conclusion is the most useful sentence in the paper, and we quote it directly: the evidence "does not allow to recommend its isolated use but may be used to complement other widely recommended therapies."
That is the frame. Not a replacement for anything. Something you can reasonably add to the treatments that already have stronger support.
Where knee osteoarthritis sits in the whole field
The broadest look available is Son and colleagues, Systematic Reviews 2025 — an umbrella review of 15 meta-analyses covering 204 randomized trials, more than 9,000 participants, and 35 health outcomes across 15 conditions.
Their certainty tally: 0 high (0%), 6 moderate (17.1%), 20 low (57.1%), 9 very low (25.7%). Their words: "No outcome was supported by high certainty of evidence."
We say that line often, because it is true and because it should be the starting point of any honest conversation about this modality.
Within that landscape, only five outcomes reached moderate certainty. Knee osteoarthritis disability was one of them, at eSMD 0.65 (95% CI 0.14 to 1.15). In a field where three quarters of outcomes are low or very low certainty, knee osteoarthritis is among the better-supported applications. That is a meaningful statement and a limited one at the same time.
The authors’ own recommendation was for "standardization of PBM protocols… before widespread clinical adoption." Which is, in different words, the same point Stausholm’s dose split was making.
What to ask any clinic, including ours
If dose is what separates 18.71 mm from 6.34 mm, then the only way to know what you are buying is to ask about dose. Reasonable questions:
- What wavelength is the device, and what is its power output?
- How many joules are delivered per point, and how many points per session?
- How many sessions, over how many weeks?
- Do those parameters correspond to the WALT recommendations for that specific wavelength? (The tables differ — the 780–860 nm table is largely per-treatment across a stated number of points, the 904 nm table is per-point. They should not be mixed.)
- Do you score my pain on the same scale at every visit?
- When do you re-check me after the course ends?
- What does "this isn’t working" look like, and at what point would you tell me to stop?
A clinic that cannot answer the first three questions cannot tell you which side of that line its protocol falls on. That is not a reason for suspicion. It is just a reason to ask.
What this is, and what it is not
Low-level laser therapy for knee osteoarthritis is adjunctive. It belongs alongside strengthening work, physical therapy, weight management where that applies, and whatever medication decisions you and your primary care physician or rheumatologist have made. The trials above measured pain and function. They did not measure cartilage, and nothing in them says anything about changing the structure of the joint or the course of the underlying arthritis.
What the evidence supports is narrower and still worth knowing: in trials where the dose followed the recommendations, groups of people with knee osteoarthritis reported meaningfully less pain than groups given a sham device, with the largest difference showing up a few weeks after treatment ended. That is the claim. Anything larger than that is not in the data.
This article is general education, not medical advice, and it is not a substitute for evaluation by a qualified clinician. It does not establish a physician–patient relationship. Individual results vary, and the studies described report group averages rather than what any one person should expect. Treatments discussed may not be FDA-cleared or approved for the conditions mentioned; where that is the case it is stated in the text. Talk to your own physician before starting, stopping or changing any treatment.