Low-dose naltrexone: what the evidence actually shows

Low-dose naltrexone is one of the strangest objects in modern medicine: a fifty-year-old generic, taken at roughly a twentieth of its licensed dose, for conditions it was never designed to treat, in a form no manufacturer makes. It has a devoted patient community, a plausible and genuinely interesting mechanism, and an evidence base that is much smaller than either of those would suggest. The useful question is not “does LDN work?” — that is unanswerable as posed — but which indication, at what effect size, and against what quality of trial. In fibromyalgia, the best-studied use, the honest answer turns on a single uncomfortable contrast: the small early trials were encouraging, and the larger, better-designed one was not.

What “low dose” actually means

Naltrexone is a pure opioid antagonist, a synthetic congener of oxymorphone with no agonist activity of its own. In the United States the only approved oral product is a 50 mg film-coated tablet, indicated for opioid use disorder and alcohol use disorder; the only other approved form is a monthly extended-release injection. At that 50 mg dose the label describes near-complete absorption, peak plasma levels within an hour, and blockade of the effects of 25 mg of intravenous heroin for as long as 24 hours — doubling the dose extends blockade to roughly 48 hours.^[2]

LDN is that same molecule at 1.5–4.5 mg once daily, usually at bedtime — between a tenth and a thirtieth of the licensed dose. The clinical trials cluster tightly: 4.5 mg nightly in the Stanford fibromyalgia work, 4.5 mg in the Crohn’s and MS trials, 6 mg in the largest fibromyalgia trial, and 1–2 mg in the long COVID cohorts. Two pharmacokinetic facts from the label do real work in the rationale: naltrexone’s own elimination half-life is about 4 hours (13 hours for the active metabolite 6-β-naltrexol), and blockade is competitive and therefore surmountable.^[2] A small nightly dose is, in other words, deliberately transient — and that transience is the entire premise of the older of the two mechanistic stories.

Two mechanisms, and why the second one is the interesting one

The opioid story (the Bihari hypothesis). A brief, partial blockade of opioid receptors overnight is proposed to trigger a compensatory response as the drug clears: more endogenous opioid peptides, and upregulated receptors, producing a net increase in endorphin tone over the following day. It is a rebound argument, and it depends entirely on the dose being small and short-acting.^[1] Direct human support is thinner than the popularity of the idea implies, but it is not zero: serum [Met⁵]-enkephalin, an endogenous opioid, was found to be reduced in people with multiple sclerosis and to rise with low-dose naltrexone in one small study,^[5] and β-endorphin concentrations rose over six months in an open-label pilot in primary progressive MS.^[16] Two small, uncontrolled-or-nearly-so measurements are a data point, not a demonstration.

The glial story (TLR4), which has better evidence. Naltrexone antagonises toll-like receptor 4 on microglia and macrophages, blunting the innate-immune signalling cascade that keeps neuroinflammation running — the mechanism proposed to underlie “centralised” pain states such as fibromyalgia.^[1] The decisive detail is that this action appears largely non-stereoselective. Naltrexone is used clinically as the levo isomer, which is what binds opioid receptors; the mirror-image (+)-naltrexone has essentially no opioid-receptor activity, and it still reverses neuropathic pain in rodents via TLR4.^[3] Detailed pharmacology in microglia and macrophages showed (+)-naltrexone and (+)-naloxone acting as TRIF–IRF3-biased TLR4 antagonists, suppressing LPS-induced nitric oxide, TNF-α and reactive oxygen species without touching NF-κB, p38 or JNK signalling.^[4]

Why that matters practically: if TLR4 is the route, then LDN is best understood as an atypical anti-inflammatory drug that happens to be an opioid antagonist, not as an opioid drug. It predicts benefit in glia-driven, centrally sensitised conditions rather than in nociceptive pain, and it explains why the same 4.5 mg tablet keeps being trialled in conditions with nothing else in common. The honest caveat is that essentially all of the TLR4 work is in cell culture and rodents, at concentrations well above what 4.5 mg produces in a human brain. The closest human echo is a small crossover trial in eight women with fibromyalgia, which found plasma concentrations of a long list of pro-inflammatory cytokines — IL-1β, IL-2, IL-6, IL-17A, TNF-α among them — reduced after eight weeks of LDN, alongside a 15% reduction in pain and 18% in overall symptoms.^[10] Eight participants, single-blind. Suggestive; not proof. For the wider problem of translating an anti-inflammatory mechanism into a clinical result, our curcumin evidence review is the cautionary case study.

Fibromyalgia: where the small trials and the big trial disagree

This is the indication with the most data, and the one where reading the literature in chronological order is genuinely instructive.

2009, Stanford, n = 10. A single-blind crossover pilot in ten women, using 4.5 mg daily for eight weeks, reported a greater than 30% reduction in symptoms over placebo, with improved mechanical and heat pain thresholds. Its most-quoted finding was a hypothesis-generating one: baseline erythrocyte sedimentation rate predicted over 80% of the variance in response — the people with more measurable inflammation improved the most.^[6]

2013, Stanford, n = 31. The replication was randomised, double-blind and counterbalanced-crossover, again at 4.5 mg. Pain fell 28.8% from baseline on LDN versus 18.0% on placebo (p = 0.016); the responder rate was 32% versus 11% (p = 0.05); life satisfaction and mood improved, fatigue and sleep did not. Tolerability was rated equal to placebo. The authors themselves called it preliminary and asked explicitly for parallel-group randomised trials.^[7]

2024, Denmark, n = 99 — the FINAL trial. That parallel-group trial arrived. Ninety-nine women with fibromyalgia were randomised 1:1 to 6 mg of naltrexone or identical placebo for twelve weeks, with participants, investigators, assessors and statisticians all masked, no losses to follow-up, and funding from the Danish Rheumatism Association and academic foundations rather than an interested party. Mean pain fell 1.3 points on an 11-point scale on naltrexone and 0.9 points on placebo — a between-group difference of −0.34 (95% CI −0.95 to 0.27; p = 0.27; Cohen’s d 0.23). Adverse events occurred in 84% of the naltrexone group and 86% of the placebo group.^[8]

That contrast is the most valuable thing on this page, so it is worth stating plainly. A 0.34-point difference on an 11-point pain scale is not a small positive result; it is a null result. Conventional thresholds for a minimum clinically important difference in chronic pain sit near a full point, or a 30% reduction. The trial that was designed to settle the question — larger, parallel-group, independently funded, no crossover carry-over to argue about — did not find an effect on pain. Its one positive suggestion was on memory symptoms, which the authors flagged as a question for future trials rather than a finding.^[8]

Pooling does not rescue it as much as it first appears. A 2024 meta-analysis of four randomised trials totalling 222 patients found a significant pooled reduction in pain scores (mean difference −0.86, 95% CI −1.20 to −0.51) and a higher pressure-pain threshold, but no significant difference on the Revised Fibromyalgia Impact Questionnaire — the measure of how much the illness actually interferes with life — or on pain catastrophising. Vivid dreams and nausea were more common on LDN; serious adverse events were not.^[9] A pooled 0.86-point pain reduction, sitting just under the usual clinical-importance threshold, with the function measure null and 222 patients in the entire randomised literature, is a fair description of a modest and inconsistent intervention. It is not a fair description of the treatment that patient forums describe.

What the evidence looks like across every other indication

Crohn’s disease

The Crohn’s story is the one where the headline numbers and the systematic review say noticeably different things, and the difference is worth understanding. An open-label pilot in 17 patients reported that 89% responded and 67% achieved remission on 4.5 mg for twelve weeks.^[11] The follow-up randomised placebo-controlled trial in 40 adults found 88% versus 40% achieved a 70-point drop in Crohn’s Disease Activity Index (p = 0.009) and 78% versus 28% achieved an endoscopic response (p = 0.008).^[12] Those are the figures that circulate.

Cochrane, pooling the whole randomised literature — that adult trial plus a 12-patient paediatric safety study, 46 participants in total — put the same data through its own primary outcome, clinical remission, and got a very different picture: 30% versus 18% remission, RR 1.48 (95% CI 0.42 to 5.24), not statistically significant. The 70-point response (RR 2.22, 1.14 to 4.32) and endoscopic response (RR 2.89, 1.18 to 7.08) did favour LDN, but endoscopic remission did not (RR 8.05, 0.47 to 138.87 — a confidence interval that tells you everything about the sample size). GRADE rated the overall evidence low, for serious imprecision, and the authors’ conclusion is unambiguous: “insufficient evidence to allow any firm conclusions regarding the efficacy and safety of LDN” in active Crohn’s disease.^[13] One further disclosure the trial itself makes and the enthusiasm rarely repeats: the senior investigators hold a patent on the use of naltrexone in inflammatory bowel disease.^[12]

None of that makes LDN useless in refractory Crohn’s. It makes it an add-on with a signal worth a properly powered trial, not a substitute for biologics or immunomodulators. For a comparison with another anti-inflammatory agent that is popular in gut-health circles on much weaker human footing, see our KPV peptide evidence review.

Multiple sclerosis

The MS literature is the clearest example of LDN improving how people feel without touching what the disease does. A UCSF crossover trial enrolled 80 patients for eight weeks of 4.5 mg nightly and reported significant gains on mental-health quality-of-life measures — a 3.3-point improvement on the SF-36 Mental Component Summary (p = 0.04), six points on the Mental Health Inventory, and smaller gains on pain-effects and perceived-deficits scales — while noting frankly that dropouts and data-management errors had substantially reduced its statistical power.^[14] A larger 17-week Iranian trial in 96 patients using the MSQoL-54 found no meaningful difference in pain, energy, emotional well-being, cognitive function or overall quality of life, and concluded that LDN was relatively safe but that its efficacy was “under question.”^[15] A six-month open-label pilot in 40 patients with primary progressive MS found it safe and well tolerated, with a measured reduction in spasticity and rising β-endorphin levels — but it was uncontrolled.^[16]

The negative claim here needs stating carefully, because it is the one that matters clinically. We searched specifically for the opposite result — a randomised trial of LDN reporting relapse rate, MRI lesion burden or EDSS progression — and found none; the randomised MS literature is quality-of-life endpoints, and the disability data comes from an uncontrolled pilot in which “neurological disability progressed in only one patient” over six months, with no comparator to judge that against.^[16] LDN has never been shown to modify the course of multiple sclerosis, and it is not a substitute for disease-modifying therapy.

CRPS, long COVID and ME/CFS — where the evidence runs out

In complex regional pain syndrome, the published basis is two case reports in which dystonic spasms and fixed dystonia remitted after LDN was added to failed conventional therapy.^[18] A systematic review that set out specifically to assess LDN in CRPS found so few CRPS-specific articles that it had to build its analysis largely out of other chronic-pain conditions, and concluded that prospective and interventional studies in CRPS are still needed.^[17] Two cases is a reason to try something in a person who has run out of options. It is not evidence of efficacy.

In long COVID, the position is unusually well documented, because a 2026 systematic review searched PubMed, Embase, the Cochrane Library, ClinicalTrials.gov and the WHO trial registry through May 2026 and reported the negative finding explicitly: no randomised controlled trials were identified. What exists is four uncontrolled pre-post studies totalling 155 patients, at doses from 1 to 6 mg, which pooled to moderate effects on fatigue (Hedges’ g −0.74) and brain fog (−0.53) and large effects on pain and daily functioning (−0.93 each) — with certainty of evidence rated low.^[19] The largest of them, a Dublin cohort of 52 patients given 1 mg then 2 mg over two months, found improvement in six of seven self-reported parameters and no safety signal, and its own authors called for randomised trials.^[20] The design limit is not a technicality: an uncontrolled pre-post study in a fluctuating, spontaneously remitting condition cannot separate drug effect from natural recovery, regression to the mean and expectation. A double-blind randomised trial of LDN for post-COVID fatigue syndrome is under way in British Columbia with a published protocol, which is where the real answer will come from.^[21]

In ME/CFS, we could find no randomised trial at all. The mechanistic case is specific and testable — TRPM3 ion-channel function is impaired in natural killer cells from patients with ME/CFS and post-COVID syndrome, and naltrexone restores it in vitro and ex vivo, which has been proposed as the explanation for the moderate clinical effects clinicians report.^[22] Testable is not tested. Across chronic pain as a whole, a scoping review of 47 relevant studies found the literature dominated by case reports, case series and narrative reviews, with wide variability in dosing and in time to response — a fair summary of the field as it stands.^[23]

Getting it: compounded only, and what that means for you

Because the smallest approved oral naltrexone product is a 50 mg tablet,^[2] there is no licensed way to take 4.5 mg. Every LDN prescription in the US is dispensed by a compounding pharmacy, which makes capsules or an oral liquid to order. That has four practical consequences worth understanding before you start, and it is also the reason the bill is hard to predict — we break the components down in what low-dose naltrexone actually costs.

• It is prescription-only, and it is off-label. A clinician must write it, and no regulator has evaluated naltrexone at this dose for pain, autoimmune disease or fatigue. Off-label prescribing is legal and routine; it simply means the evidence bar was never formally cleared.
• Compounded drugs are not FDA-approved products. They are not reviewed for safety, efficacy or manufacturing quality before they reach you, and potency, dissolution and content uniformity depend on the individual pharmacy’s practices. The same regulatory logic we set out for compounded versus branded medicines applies here: choose a pharmacy that does third-party potency testing and will tell you what is in the capsule.
• The filler is not a trivial detail. Pharmacies compound LDN into a filler — commonly avicel or calcium carbonate — and slow- or sustained-release fillers are generally avoided, because the endorphin-rebound rationale depends on a short exposure that clears within hours.^[1][2] That is a mechanistic rationale rather than a trial finding, but it is the reason a “time-release LDN” product is a contradiction in terms. Lactose fillers are also worth flagging if you are lactose-intolerant.
• Splitting or dissolving a 50 mg tablet is the cheap route, and it is imprecise. The approved tablet is film-coated and made in a single strength.^[2] Cutting it into roughly an eleventh, or dissolving one in a fixed volume of water and measuring out a fraction, has no assurance of dose accuracy, uniform distribution or stability of the resulting solution. People do it because the tablet itself is cheap — a matter of a couple of dollars a month at pharmacy acquisition cost — against a compounded preparation that insurance rarely covers for an off-label use of an unapproved formulation. What that compounded preparation actually costs varies enough by pharmacy that it is worth getting quotes; we break the components down in what low-dose naltrexone actually costs. It is a real trade-off, not a free one.

On dosing practice: prescribers typically start low — 0.5 to 1.5 mg at night — and step up every one to two weeks toward 3–4.5 mg, mostly to blunt the early sleep disturbance. That titration is convention rather than trial protocol; the randomised trials used a fixed 4.5 mg nightly,^[7][12][14] or 6 mg.^[8] Response, where it happens, is usually described over weeks rather than days, and the scoping review specifically flagged how variable reported time-to-response is across studies.^[23] A sensible plan is a defined trial — eight to twelve weeks at a target dose, with a symptom measure you wrote down at baseline — and a decision to stop if nothing has moved.

Safety, side effects, and the one interaction that genuinely matters

Set against that, the day-to-day tolerability is genuinely good, and this is the strongest thing the evidence base actually supports. In the largest fibromyalgia trial, adverse events occurred in 84% of the naltrexone group and 86% of the placebo group, with discontinuations for adverse events at 8% versus 6% — barely distinguishable.^[8] The meta-analysis found significantly more vivid dreams and nausea on LDN, with no difference in serious adverse events, headache, diarrhoea or dizziness.^[9] Cochrane’s pooled Crohn’s data found no significant difference in sleep disturbance, unusual dreams, headache, appetite, nausea or fatigue, and no serious adverse events in either trial.^[13] The 2013 fibromyalgia crossover rated LDN equally tolerable as placebo.^[7]

The pattern most people encounter is vivid dreams and disturbed sleep in the first one to two weeks, which typically settle; taking the dose earlier in the evening, or moving it to the morning, is the usual workaround, and the 2009 pilot described these effects as rare, minor and transient.^[6] Sleep disturbance was the most common complaint in the Crohn’s pilot too, in 7 of 17 patients.^[11]

On the liver: the naltrexone label carries a Hepatotoxicity warning describing cases of hepatitis and clinically significant liver dysfunction, plus transient asymptomatic transaminase elevations, seen during development and post-marketing — often in patients with other contributing causes such as pre-existing alcoholic liver disease or hepatitis B or C.^[2] Two pieces of context matter. That experience is at the 50 mg dose and above, in a population with a high background rate of liver disease; and the current US label carries hepatotoxicity under Warnings rather than as a boxed warning, which is where it used to sit. At 1.5–4.5 mg the exposure is a fraction of that, and no liver signal has emerged in the LDN trials — the Crohn’s pilot reported no laboratory abnormalities at all.^[11] It is still a sensible reason to mention existing liver disease to your prescriber rather than to self-start.

The genuine unknown is duration. The randomised trials ran eight to seventeen weeks. There is no long-term controlled safety data for LDN taken for years, which is how most people in the patient community actually take it.

The honest bottom line

LDN occupies an unusual position: an intervention where the risk side of the ledger is genuinely reassuring and the benefit side is genuinely thin. It is cheap, it is a well-characterised old molecule, its adverse-event profile in trials is close to placebo, and it has a mechanism — TLR4 antagonism on microglia, demonstrated non-stereoselectively — that is more interesting than most repurposing stories ever manage.^[3][4]

What it does not have is a trial that convincingly demonstrates a clinically meaningful effect on anything. In fibromyalgia, its best-studied use, the pattern is the one that should always make you cautious: encouraging small crossover pilots from an invested group, followed by a larger, independent, better-designed trial that found a 0.34-point difference on an 11-point scale.^[7][8] In Crohn’s, Cochrane looked at the whole randomised literature — 46 patients — and said firm conclusions are not possible.^[13] In MS, quality of life sometimes improves and the disease does not change.^[14][15] In long COVID and ME/CFS, there is no randomised trial to discuss at all.^[19]

The defensible use, then, is narrow and honest: a low-cost, low-risk, time-limited trial in someone with refractory centralised pain or an inflammatory condition who has already worked through the treatments that do have evidence behind them — run for eight to twelve weeks against a symptom score you recorded beforehand, and stopped if nothing moves. The two things it should never be are a reason to delay disease-modifying therapy, and a reason to expect the transformation the internet promises. The gap between the enthusiasm and the evidence is not proof that LDN does nothing. It is a measure of how little we have actually asked.

This article is research information, not medical advice. Low-dose naltrexone is available only on prescription and only as a compounded preparation — compounded medicines are not FDA-approved and are not reviewed for safety, efficacy or manufacturing quality before dispensing, and naltrexone has no FDA approval at low dose for fibromyalgia, Crohn’s disease, multiple sclerosis, complex regional pain syndrome, long COVID, ME/CFS or any other indication discussed here. Naltrexone blocks the effect of opioid analgesics at any dose and will precipitate withdrawal in anyone physically dependent on opioids, including tramadol; tell every clinician, dentist, surgeon and emergency provider that you take it, and plan around elective surgery with your prescriber. Discuss existing liver disease, current medications and any planned procedure with a licensed clinician before starting, and do not stop or delay an evidence-based treatment in order to try it.