Most articles about low-level laser therapy for neck pain cite one study: a 2009 meta-analysis in the Lancet. It is a strong paper in a very good journal, and its results are favorable. It is also not the last word — four years later, a different group of reviewers looked at overlapping evidence and concluded close to the opposite.
That disagreement is more useful to you than either paper alone. Here is both, and what a clinician actually does with two credible reviews that do not agree.
What Chow 2009 found
Chow, Johnson, Lopes-Martins and Bjordal (2009, Lancet) pooled 16 randomized controlled trials covering 820 patients with neck pain, comparing low-level laser therapy (LLLT) with placebo or other treatment.
The headline numbers:
- Acute neck pain: relative risk of improvement 1.69 (95% CI 1.22–2.33). Patients receiving laser were about 1.7 times as likely to report improvement as patients receiving placebo.
- Chronic neck pain: relative risk 4.05 (95% CI 2.74–5.98).
- Pain intensity: a 19.9 mm reduction on a 100 mm visual analog scale (95% CI 10.0–29.7) compared with placebo.
- Duration: the benefit persisted up to 22 weeks after treatment ended.
Two features made the paper influential. First, none of those confidence intervals cross the line of no effect, so these are not marginal findings within their own dataset. Second, the 22-week persistence: many pain treatments produce a difference that evaporates the week you stop, and a benefit outliving the treatment course is more interesting than one measured on the last day of therapy. For scale, 19.9 mm on a 100 mm scale is about the difference between calling your neck a 7 out of 10 and calling it a 5.
What Kadhim-Saleh 2013 disputes
Kadhim-Saleh, Maganti, Ghert, Singh and Farrokhyar (2013, Rheumatology International) asked the same question and returned a different answer.
Their meta-analysis included 8 RCTs and 443 patients — a narrower slice, reflecting stricter inclusion criteria. Pooled improvement in pain on a visual analog scale was 10.54 (95% CI 0.37–20.71), heterogeneity I² = 65%.
Read that confidence interval carefully; it is the whole argument. The lower bound is 0.37, a hair above zero. The result is technically significant, yet the data are entirely compatible with an effect too small for any patient to notice.
The authors concluded that the evidence is "inconclusive because of significant between-study heterogeneity and potential risk of bias," and they made a further point that matters more than the p-value: the benefit they measured does not meet the threshold of minimal clinically important difference.
This is not a fringe objection. It is a methodologically conventional review in a peer-reviewed rheumatology journal, applying standard appraisal tools to a set of trials that largely overlaps Chow’s. Any honest account of this treatment has to carry it.
What "minimal clinically important difference" actually means
This is the crux.
Statistical significance answers one question only: is this difference bigger than chance alone would likely produce? It says nothing about whether the difference is big enough to matter to a human being. With enough participants, a trial can detect half a point on a 100-point scale and report it as highly significant — a result both real and useless.
The minimal clinically important difference (MCID) fixes that gap. It is the smallest change on a scale that a patient would actually notice and consider worthwhile — established by asking patients after treatment whether they feel meaningfully better, then looking at how far their scores moved. It converts "the numbers differ" into "people can tell."
So there are two questions, and they can have different answers:
- Did pain scores drop more with laser than with placebo? (Both reviews say yes.)
- Did they drop far enough that a person with a sore neck would notice and care? (Chow’s 19.9 mm suggests yes. Kadhim-Saleh’s 10.54 says no, and the authors said so explicitly.)
The two pooled estimates differ by roughly a factor of two, and they land on opposite sides of a clinically meaningful threshold. That is the disagreement in one sentence.
A third paper reinforces the caution. Tehrani and colleagues (2022, Lasers in Medical Science) pooled 13 RCTs and 556 patients: pain mean difference −1.29 (95% CI −2.36 to −0.23) favoring laser, but disability did not improve significantly. Pain relief and functional improvement are not the same outcome, and it is function that determines whether you can turn your head to check a blind spot or get through a day at a desk. (A footnote for anyone reading that abstract: it contains a p-value for the disability outcome that is arithmetically impossible — a reminder to read papers rather than summaries.)
How a clinician reconciles two reviews that disagree
The instinct is to pick the better journal and move on. A few things actually explain the gap.
They pooled different trials. Chow included 16 RCTs, Kadhim-Saleh 8, Tehrani 13. Meta-analyses are only as good as their inclusion criteria, and different but defensible criteria produce different answers from the same literature. A stricter review that excludes weaker trials usually reports a smaller effect — often a sign the smaller number is closer to the truth.
The trials disagreed with one another. Kadhim-Saleh’s I² of 65% means roughly two-thirds of the variation in results across trials reflects real differences between the studies rather than chance. When heterogeneity runs that high, the pooled average describes no particular trial well, and the right response is to ask what differed between the trials rather than to trust the average.
And there is a strong candidate for what differed: dose. In knee osteoarthritis, Stausholm and colleagues (2019, BMJ Open) pooled 22 trials and 1,063 patients and found that trials using doses consistent with published recommendations produced an 18.71 mm reduction (95% CI 9.42–27.99), while trials using non-recommended doses produced 6.34 mm, non-significant at follow-up. The difference between those subgroups was significant, p = 0.02. Different body part, same lesson: a review that happens to include more under-dosed trials reports a smaller effect, and two reviewers end up appearing to disagree about a treatment when they are really disagreeing about a sample.
And neither review is the last word. The 2025 umbrella review by Son and colleagues (Systematic Reviews) — 15 meta-analyses, 204 RCTs, more than 9,000 participants, 35 outcomes — found that no outcome anywhere in the field was supported by high-certainty evidence. Six of the 35 outcomes reached moderate certainty, and neck pain was not among the five the review names. That is the honest ceiling on this whole discussion.
The working conclusion a careful clinician draws is not "it works" or "it doesn’t." It is narrower: for neck pain, LLLT at appropriate parameters is a reasonable adjunct with a modest average effect that may or may not clear the threshold you personally would notice, and the evidence is not strong enough to promise more than that. Anyone who tells you differently is picking one review.
What this means if your neck hurts
A few practical implications.
Treat it as an addition, not a substitute. Laser belongs alongside strengthening and endurance work for the neck and shoulder girdle, not instead of it. Nothing in either review supports it as a stand-alone plan, and a course of laser that displaces exercise is a step backward.
Ask for the parameters before you start. Wavelength, output in milliwatts, joules per point, number of points, and how many sessions. This is the difference between the trials that showed an effect and the trials that did not. It is a fair question and a clinician who cannot answer it has not read this literature closely.
Set a decision point in advance. Agree with your clinician how many sessions you will try and what change would count as worth continuing. Write down your pain score and something functional — how far you can rotate before it catches, whether you sleep through the night — then compare honestly. Most acute neck pain improves regardless of what is done to it, which is exactly why placebo-controlled trials exist.
Be willing to stop. If a defined course produces nothing you would call a real change, the useful conclusion is that you are in the group that did not respond, and the money is better spent elsewhere. For a treatment with this evidence profile, that is a normal outcome rather than a failure.
Keep the ceiling in view. The most favorable review found a benefit lasting up to 22 weeks. The most skeptical found an effect too small to meet the minimal clinically important difference. Both are legitimate readings of an imperfect body of evidence, and we would rather hand you both than pretend the question is closed.
For the wider picture on how this modality is studied, where the evidence is stronger, and where it is plainly negative, see What Photobiomodulation Actually Is.
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.