Low-Dose Naltrexone: A Small Dose, A Large Set of Claims

Naltrexone was approved decades ago for one job: blocking opioid receptors at a standard 50-milligram dose, so that people dependent on opioids or alcohol could not get the effect they were trying to chase. Low-dose naltrexone — LDN, in the shorthand every patient forum uses — is the same molecule at roughly 1 to 4.5 milligrams a day, sometimes up to 6, usually prepared by a compounding pharmacy because no manufacturer sells it at that strength. It is prescribed off-label for fibromyalgia, long COVID, ME/CFS, multiple sclerosis and a wider circle of chronic pain conditions. Nothing about that arrangement is secret, and nothing about it is FDA-approved: the prescribing is legal, the indication is not.

LDN occupies a particular corner of the chronic illness world, where desperate patients, cautious specialists and enthusiastic telehealth prescribers all meet. The proposed mechanisms are genuinely interesting. The clinical record, read carefully, is a story about how a small early trial can launch a movement that later, larger trials spend years catching up with.

Why it should work, in theory

Two mechanisms carry most of the argument. The first is glial: laboratory work suggests naltrexone can interfere with Toll-like receptor 4 signaling, the pathway by which microglia — the brain’s immune cells — pump out pro-inflammatory signals. A widely cited 2014 review laid out this rationale, and it is the reason LDN is sometimes described as an “anti-inflammatory” rather than a painkiller. The second is endorphin: a brief, low-dose opioid-receptor blockade is proposed to produce a compensatory rebound in endogenous opioid activity or receptor expression, a kind of hormonal exercise.

Both stories are plausible and both remain hypotheses. Laboratory evidence that a drug interferes with a signaling pathway does not establish that a 4.5-milligram oral tablet suppresses brain inflammation in a living patient — no trial has demonstrated that — and the “endorphin boost” framing, with its talk of precise bedtime dosing windows, runs well ahead of anything measured in humans. It is worth holding this in mind before reading a single trial result, because the mechanism story is what most patients hear first.

What holds up

Fibromyalgia is the one indication with a genuine randomized-trial record. The field’s founding result came from Jarred Younger’s group in 2013: a small, double-blind, placebo-controlled crossover trial in 31 women using 4.5 milligrams a day, which found 28.8 percent baseline pain reduction on LDN versus 18.0 percent on placebo, a difference the authors reported as statistically significant, with 32 percent of participants responding on the drug versus 11 percent on placebo. Mood and life satisfaction improved; fatigue and sleep did not.

A 2024 meta-analysis in the Korean Journal of Pain pooled four randomized trials totaling 222 participants and reported a statistically significant pooled benefit for pain — a mean difference of −0.86 on pain scales, in plain terms a modest but real average edge over placebo across the trials combined. And the practical case for continuing to study the drug is legitimate: it is cheap — a 2014 review estimated roughly $35 a month with compounding, a historical figure rather than a guaranteed current price — the side effects recorded in trials were mild (vivid dreams, insomnia, nausea, headache), and no serious adverse events appeared in the fibromyalgia and MS trials.

What doesn’t hold up

The trouble starts with the later trials. A 2023 randomized crossover trial led by Lucas Bested gave 4.5 milligrams daily to 58 randomized patients with fibromyalgia in roughly three-week treatment periods and found no clinically relevant analgesic effect: the between-treatment difference on the Fibromyalgia Impact Questionnaire was not statistically significant, and neither was the difference in pain ratings. Then the largest trial to date — a Danish parallel-group study published in The Lancet Rheumatology in December 2023, with 99 women taking 6 milligrams daily for twelve weeks — failed to demonstrate superiority over placebo for its primary pain outcome. Larger, longer, and better powered than the founding trial, it found what the founding trial did not.

Then came 2026’s uncomfortable correction. The 2024 meta-analysis’s responder analysis — the headline that a meaningful share of patients improved on LDN versus placebo — had double-counted patients from one trial, folding the same subjects into two response thresholds. A January 2026 re-analysis by the Danish team corrected the arithmetic: the trial-specific responder odds ratio fell from a compelling 2.82 (95 percent confidence interval 1.25 to 6.39) to 1.96 (0.84 to 4.60) — an interval that now includes no difference. This does not automatically void the pooled pain estimate, which rests on different endpoints. But it removes the analysis most often quoted to patients, and it is a case study in how enthusiasm can leak into data extraction.

The largest positive claim in the LDN literature rests on a 31-person crossover trial. The largest negative finding is a 99-person trial most patients have never heard of.

For long COVID, the evidence base is thinner still. A July 2026 systematic review in BMJ Open searched the literature through May 5, 2026, screened 397 records, and found no eligible randomized controlled trials at all — only four observational before-and-after studies totaling 155 participants, using 1 to 6 milligrams daily. Before-and-after designs cannot separate a drug effect from natural recovery, regression to the mean, expectation, or everything else a person was doing at the same time. A properly randomized, placebo-controlled trial is running in British Columbia — 160 planned participants, titrating from 1 milligram toward 4.5, registered as NCT05430152 and listed as completed in February 2026 — but as of this writing no published randomized result has verified the drug’s benefit there. A protocol is not an efficacy result, and neither is a registry status.

The other conditions have their own versions of the same pattern. The often-quoted claim that roughly 74 percent of ME/CFS patients respond to LDN comes from a 2019 retrospective review of clinical records — 218 patients, no control group, no blinding — which cannot attribute that response rate to the drug. In multiple sclerosis, a 2010 pilot found small improvements in mental quality-of-life scores, not fewer relapses, not less MRI activity, and not slower disability — none of which supports using LDN in place of an approved disease-modifying therapy.

What a reader should take from this

For fibromyalgia, LDN is one of the few unapproved options with randomized evidence behind it at all — and the evidence points in more than one direction. The founding trial was small and positive; the two later trials, including the largest, were negative; the pooled estimate is modest and its most-quoted component was corrected in 2026. Someone trying LDN for fibromyalgia is making a defensible bet on an unresolved question, not following settled evidence.

For long COVID and ME/CFS, the honest description is “awaiting a randomized answer.” The Canadian trial is the first properly powered test; when it reports, it will deserve more attention than any before-and-after story that preceded it.

The practical cautions are real even at low doses. Naltrexone blocks opioid analgesia and can precipitate withdrawal in opioid-dependent people — label guidance calls for an opioid-free interval of at least seven to ten days, and longer transitions from methadone or buprenorphine require medical supervision. The label also carries hepatic warnings, and pregnancy and very long-term use remain poorly characterized by these small trials. Anyone prescribed LDN should know it is off-label, compounded, and priced outside insurance coverage more often than not — a modest recurring cost for a benefit that, for most of the conditions it is sold for, has not yet been demonstrated in a randomized trial.

This article is general information, not medical advice. See our Editorial Standards.