Two Synthetic THC Medicines, One Evidence Review, Opposite Results for Pain

The 2026 update of the AHRQ evidence review on cannabinoids for chronic pain, published in Annals of Internal Medicine, found that products with high or comparable THC-to-CBD ratios produced small pain reductions, in the range of half a point to three quarters of a point on a 0 to 10 scale, with moderate or large increases in dizziness, sedation, and nausea.

Within the THC-only group, the two pharmaceutical products separated sharply. Nabilone reduced pain severity by a pooled 1.59 points. Dronabinol reduced it by 0.23 points, which is indistinguishable from no effect. Both are oral, synthetic, and prescription-only, which is why the gap is worth a careful look rather than a headline.

Most public argument about cannabis and pain treats the plant as one thing. This review does the opposite. It sorts interventions by THC-to-CBD ratio, by source, and by route, and the answers differ by category. That is closer to how a clinician reasons about any other drug class.

The practical consequence for a patient is unglamorous. A product low in THC relative to CBD, including CBD on its own, did not improve pain outcomes in these trials. The products that did move pain also moved dizziness, sedation, and nausea. There was no category in this review that delivered benefit without that trade.

Roger Chou and colleagues at the Pacific Northwest Evidence-based Practice Center at Oregon Health & Science University searched Ovid MEDLINE, PsycINFO, Embase, the Cochrane Library, and Scopus through 28 July 2025 and pooled 25 short-term randomized placebo-controlled trials covering 2,303 participants. Sixty-four percent of those participants had neuropathic pain. The work was funded by the Agency for Healthcare Research and Quality and registered with PROSPERO as CRD42021229579.

This is an update rather than a new project. The predecessor, published in the same journal in 2022 by Marian McDonagh and colleagues, pooled 18 randomized placebo-controlled trials with 1,740 participants plus seven cohort studies with 13,095. The 2026 version drops the cohort studies, adds randomized evidence, and keeps the categorization scheme that made the earlier version useful: cannabinoids grouped by THC-to-CBD ratio, by source, and by how they are taken.

Every trial pooled here ran between one and six months. Nothing in this review speaks to a year of use, to tolerance, or to what happens when a patient stops.

Oral synthetic or purified products with a high THC-to-CBD ratio, meaning THC alone, may slightly reduce pain severity, with a pooled difference of 0.78 points on a 0 to 10 scale. Oromucosal extracted products with a comparable THC-to-CBD ratio, the category that includes the 1:1 sprays, probably slightly reduce pain severity, with a pooled difference of 0.54 points.

Those two verbs are doing real work. In this review, as in AHRQ and GRADE practice generally, may signals low certainty and probably signals moderate certainty. The 1:1 oromucosal finding is smaller in size and stronger in confidence than the THC-only finding. A reader who scans for the bigger number gets the ranking backwards.

Both categories came with moderate or large increases in dizziness, sedation, and nausea. Products with a low THC-to-CBD ratio may not improve outcomes. Low-ratio mixed THC and CBD products may still increase dizziness, sedation, and nausea, while CBD by itself may not increase those harms.

The finding most likely to be misquoted is the split inside the THC-only group: nabilone moderately reduced pain severity at a pooled 1.59 points while dronabinol did not, at 0.23 points. The usual shorthand describing both as synthetic THC is where the confusion starts, because it is only half right.

Dronabinol is synthetic delta-9-tetrahydrocannabinol. It is the same molecule the plant makes, produced in a factory. Nabilone is a synthetic cannabinoid built to resemble THC rather than reproduce it, and its differences in structure, potency at the CB1 receptor, and oral behavior are real. Two drugs that act at the same receptor family can still differ in how much reaches the receptor and for how long, which is ordinary pharmacology rather than a paradox.

That said, a subgroup estimate is the most fragile number in any pooled analysis. Subgroups have fewer trials and fewer participants than the main comparison, and they were not what the trials were designed to answer. The honest reading is that this is a hypothesis about formulation worth testing, not a verdict that one drug works and the other does not.

A pharmacology-based systematic review and meta-analysis by Ainhoa Bilbao and Rainer Spanagel, published in BMC Medicine in 2022, analyzed 152 randomized controlled trials covering 12,123 participants across every indication where cannabinoids have been tested. It reported moderate-grade evidence for dronabinol in chronic pain, with a standardized mean difference of 0.31 in favor of treatment and a 95 percent confidence interval from 0.15 to 0.46. It reported moderate-grade evidence for nabiximols in chronic pain at 0.25.

Nabilone fares worse elsewhere. A Cochrane review of cannabinoids for fibromyalgia by Brian Walitt and colleagues found only two trials with 72 participants between them, both using nabilone at 1 mg at bedtime, and graded every efficacy outcome as very low quality. The reviewers concluded there was no convincing, unbiased, high-quality evidence that nabilone helps in fibromyalgia, and noted that tolerability was poor.

So the field does not speak with one voice on either drug. Different reviews use different inclusion rules, different outcome scales, and different pooling methods, and they land in different places. That is a reason to hold the nabilone number loosely, not a reason to dismiss the review that produced it.

The most common request in cannabis medicine is for something that helps pain without psychoactivity. This review says, at low certainty, that the products built around that request did not improve pain outcomes in randomized trials. That is worth saying plainly to a patient before they spend money on a CBD-dominant product for pain, and worth saying without contempt, because the same review found CBD alone may not increase the harms that THC-containing products do.

For a patient who has exhausted conventional options and is willing to accept sedation and dizziness, the categories with signal are THC-only oral products and 1:1 oromucosal sprays, and the expected benefit is under one point on a 10-point scale. That is the size of an adjuvant, not a replacement for a working regimen.

Nabilone occupies an odd position. It is a prescription medicine in the United States, approved for chemotherapy-induced nausea and vomiting rather than for pain, so any use for pain is off-label. It is unfamiliar to most prescribers and rarely covered generously. A single favorable subgroup estimate does not change that calculus by itself, but it is a reasonable thing to raise with a pain specialist when dispensary products have failed.

The method here is about as good as evidence synthesis gets on this topic. Dual review of data extraction, risk of bias, and strength of evidence; restriction to randomized placebo-controlled trials; a categorization scheme that keeps unlike products apart; and a federal funder with no commercial stake in the answer. When this team says probably, that word has been earned.

The evidence base underneath the method is thinner than the method deserves. Twenty-five short-term trials across many different products is not a large literature, and the review is limited by variability within categories, missing product details, unclear United States availability of several studied products, and restriction to English-language studies. The certainty ratings reflect those limits honestly rather than papering over them.

Subgroup findings deserve the most suspicion, and the nabilone result is a subgroup finding. It sits inside the THC-only category, rests on fewer trials and fewer participants than the category estimate, and was not the question any single trial set out to answer. Estimates like this are where pooled analyses most often mislead, and where later data most often regress toward the middle.

The categories themselves lump. High THC-to-CBD covers a wide span of actual ratios, and grouping synthetic with purified products puts pharmaceutical isolates next to preparations that may behave differently in the body. Publication and reporting patterns in cannabinoid research are shaped by commercial interest and regulatory friction on both sides, which makes the file drawer harder to estimate than in most drug classes.

This review says nothing about smoked or vaporized cannabis, about dispensary flower or edibles, or about any product a patient is likely to buy without a prescription. The interventions pooled here are pharmaceutical oils, capsules, and oromucosal sprays, which is a different exposure from what most people mean when they say they use cannabis for pain.

It also says nothing beyond six months. Tolerance, dependence, cognitive effects over years, the trajectory of benefit after the first few months, and whether any of these products reduce long-term opioid use are all outside what these trials measured. Chronic pain is a condition measured in years, and this is evidence measured in weeks.

The direction of travel in this field is away from asking whether cannabis works and toward asking which preparation, at what ratio, by what route. The 2022 review started that sorting and the 2026 update sharpens it. The uncomfortable part is that the sorting has so far narrowed the list of products with any randomized support rather than expanding it.

There is also a structural mismatch worth naming. The products with the best trial evidence are prescription medicines that most American patients cannot easily get for pain, while the products patients can buy freely are the ones this review found least supported. Evidence and access are pointing in opposite directions, and no amount of careful reading fixes that.

The number I keep coming back to is not 1.59. It is 0.54, the oromucosal 1:1 result, because that is the one the reviewers rated moderate certainty. A smaller effect you can believe is more useful in practice than a larger effect you cannot, and patients almost never read it that way. They see the bigger number and stop.

On nabilone, I would want a pain specialist involved before I went there. It is off-label for pain, it is sedating, it is expensive, and the evidence supporting it in this review comes from a subgroup. That is not a reason to ignore it for a patient who has tried everything else, but it is a reason to treat it as a considered specialist decision rather than a next step.

The conversation I have most often is with someone who wants CBD to be the answer. This review does not support that for pain, and I say so. What I add is that the same review found CBD on its own did not carry the dizziness and sedation burden the THC products did. If a patient wants to try it for other reasons, the downside is mostly the money and the delay.

Cannabinoid products containing meaningful THC produce small pain reductions, under one point on a 10-point scale, alongside real increases in dizziness, sedation, and nausea. Low THC-to-CBD products, including CBD alone, did not improve pain outcomes. The nabilone result is the most interesting number in the review and the least settled, and it belongs in a specialist conversation rather than a self-directed trial.

Carry forward the certainty language, not the effect sizes. The reviewers said probably once, about 1:1 oromucosal products, and may everywhere else. Carry forward the harms, which were consistent across the categories that helped. Do not carry forward the nabilone number as a fact about nabilone; carry it forward as a question about whether formulation matters more in this drug class than anyone has properly tested.