Short Trials Found Almost Nothing for MS Spasticity. Longer Trials Found More. The Numbers Need Care.

A systematic review and meta-analysis in Clinical Therapeutics pooled nine clinical trials enrolling 2,544 people with multiple sclerosis and reported improvement in spasticity with cannabis-based therapies, with larger effects in trials of longer duration. Short-term studies produced a pooled mean difference of 4.53 with a confidence interval running from -0.06 to 9.12, which does not exclude no effect.

The review also reports heterogeneity of 100 percent for its overall analysis and funnel plot asymmetry consistent with publication bias. Those two findings together mean the pooled effect estimates should be read as a direction of travel, not as quantities.

Spasticity is one of the more disabling features of multiple sclerosis, and the standard options, baclofen, tizanidine, and dantrolene, carry sedation, weakness, and cognitive dulling that some patients find as limiting as the stiffness itself. Nabiximols, a standardized oromucosal spray, is approved for this indication in the United Kingdom, Germany, Spain, Canada, and a number of other countries, and is not approved by the FDA in the United States.

The practical question patients ask is how long to give it before deciding. This review is one of the few pieces of evidence that addresses duration directly, even though it does so imperfectly. That makes it worth reading carefully rather than either quoting uncritically or ignoring.

The review, Assessing the Role of Cannabis in Managing Spasticity in Multiple Sclerosis: A Systematic Review and Meta-Analysis, was led by Yazan AlHabil with colleagues at An-Najah National University in Nablus, the Arab American University, Al-Quds University, Hebron University, and the ASAB-Palestinian Neurology Nexus. It appeared online in Clinical Therapeutics in August 2025 and in the January 2026 print issue.

Nine clinical trials contributed data, enrolling 2,544 patients with multiple sclerosis. The trials ran between 2003 and 2021 across several countries. The interventions were not uniform: whole-plant extracts, standardized oils, and smoked cannabis, containing delta-9-tetrahydrocannabinol, cannabidiol, or both. Spasticity was measured with the Ashworth scale in some trials, a visual analog scale in others, and a numeric rating scale in others.

That last detail is the hinge on which the whole analysis turns. The Ashworth scale is a clinician-rated ordinal measure of resistance to passive movement. A numeric rating scale is a patient-reported number. These instruments measure related but different things on entirely different ranges, and combining them into a single pooled figure requires assumptions that this review does not make visible.

Trials of shorter duration produced a pooled mean difference of 4.53, with a 95 percent confidence interval of -0.06 to 9.12. That interval includes zero, so the short-term result did not reach statistical significance. Longer trials produced a pooled mean difference of 75.81, with a confidence interval of 66.39 to 85.22.

Read as a contrast rather than as two quantities, this is a real and clinically familiar pattern. Cannabinoid effects on spasticity in multiple sclerosis have never behaved like a dose of a muscle relaxant, where benefit is evident within hours. The randomized literature on nabiximols has generally used an enrichment period of several weeks before randomizing responders, which is itself an acknowledgment that a short look at an unselected group tends to show little.

What this review cannot tell you is why. Longer trials differ from shorter ones in more than duration. They tend to enroll patients who tolerated the drug, to have selected for early responders, and to lose the people for whom it did nothing. Attrition alone can manufacture a larger apparent effect over time without any change in underlying biology.

The overall pooled result is reported as a standardized mean difference of 39.19, with a confidence interval of 34.32 to 44.05. A standardized mean difference expresses an effect in units of standard deviation, and values above about 0.8 are conventionally described as large. A value of 39 is not a large effect size. It is a number that indicates raw scores from incompatible scales have been pooled without standardization, or that the label has been applied to something else.

The subgroup figures reinforce the concern. The Ashworth scale subgroup is reported as a mean difference of 20.36 with a 95 percent confidence interval of 20.35 to 20.37. A confidence interval two hundredths of a point wide, drawn from a handful of trials, is not a plausible result. The numeric rating scale subgroup, reported as 1.18 with an interval of 1.16 to 1.21, has the same problem in miniature.

Heterogeneity is reported as 100 percent for the overall analysis and for the Ashworth subgroup, and 91 percent for the numeric rating scale subgroup. An I-squared of 100 percent means essentially none of the variation between trials is attributable to chance. In that situation, a pooled point estimate describes no population that exists. The authors flag the heterogeneity honestly and call for standardized protocols in future work.

Cannabinoids reach spasticity by a different route than the standard drugs. Baclofen is a GABA-B agonist, tizanidine an alpha-2 adrenergic agonist, and dantrolene acts directly on skeletal muscle. THC and CBD act on CB1 and CB2 receptors distributed through the central nervous system and on several non-cannabinoid targets, modulating neurotransmitter release and excitatory tone in spinal circuits.

That difference is a reasonable explanation for why some patients who fail conventional antispasmodics respond to cannabinoids. It is a weaker explanation for a time course, and it should not be offered as one. Suggesting that the endocannabinoid system needs weeks to engage is a plausible story, not a finding, and this review provides no mechanistic data to support it.

The more defensible account of why longer treatment looks better is mundane: titration takes time, tolerance to the sedating and cognitive effects develops over weeks, and patients who stay in a trial are the ones for whom the trade is working.

The usable content here is about expectations, not about effect sizes. A patient starting a cannabinoid for spasticity should be told that a two-week verdict is premature, that dose adjustment is expected, and that the first weeks often bring side effects before they bring benefit. Setting that expectation is one of the higher-yield things a clinician can do in this conversation.

It should also be said plainly that the evidence base does not support a promise. Nabiximols, the most studied preparation, produces modest average benefit in its randomized trials, and a meaningful fraction of patients discontinue. Telling a patient that longer use produces a large effect, on the strength of this pooled analysis, would be a misrepresentation.

Reported adverse events in the included trials were generally mild, mostly dizziness and dry mouth. That is a reassuring signal and not a safety assessment. The authors themselves call for comprehensive safety evaluation in future trials, which is a fair reading of what these nine studies did and did not capture.

A systematic review sits high in the evidence hierarchy only when the trials it pools are similar enough to pool. This one reports 100 percent heterogeneity for its main analysis, which is the statistical statement that its trials are not similar enough. The underlying trials, several of which are well conducted, remain useful individually. The pooled numbers built on top of them are not.

The direction of the finding is more robust than its magnitude. Nine trials over roughly two decades, in more than two thousand patients, pointing toward benefit, is a signal worth respecting. The size of that benefit is better estimated from the individual nabiximols trials and from prior network meta-analyses than from these pooled figures.

The reported effect sizes are internally implausible. A standardized mean difference of 39.19 is not interpretable as a standardized effect, and an Ashworth subgroup confidence interval spanning 20.35 to 20.37 is not a credible result from a small number of trials. Whatever these figures represent, they are not quantities a clinician can carry into a consultation.

Pooling clinician-rated ordinal scores with patient-reported numeric ratings across studies that used different products, doses, and routes is the source of most of the problem. Whole-plant extract, standardized oil, and smoked cannabis are not one intervention, and the review does not resolve them into comparable exposures.

The duration comparison is observational across trials, not a randomized comparison of treatment length. Long trials differ from short trials in selection, attrition, and design as well as in duration, and any of those can produce the observed gap.

Funnel plot asymmetry suggests small negative trials may be missing from the literature, which would bias the pooled estimate upward. The true average effect is likely smaller than these figures imply.

This review does not show that continuing cannabis for longer produces a larger benefit in an individual patient. It compares trials of different lengths, which is not the same question.

It does not identify which preparation, THC to CBD ratio, dose, or route works best, and it does not tell a clinician which patients are likely to respond.

It does not establish the safety profile of long-term cannabinoid use in multiple sclerosis. Reporting that adverse events across the included trials were mostly dizziness and dry mouth is not evidence of an absence of cognitive, motor, or dependence-related effects, and the authors explicitly ask for comprehensive safety assessment in future work.

Cannabis for multiple sclerosis spasticity is among the oldest and best-populated questions in cannabinoid medicine. Nabiximols has been approved for the indication outside the United States for well over a decade on the strength of randomized trials showing modest average benefit, with responder-enriched designs that concentrate the effect in a subgroup.

Recent syntheses of this literature have tended to land in the same place: a real but moderate effect, wide individual variation, low to moderate certainty of evidence. This meta-analysis points the same direction with weaker arithmetic. Read alongside the larger network meta-analyses, it adds a duration question and little else.

The clinical observation behind this paper is one I would defend without the paper. Patients who try a cannabinoid for spasticity and judge it at two weeks are judging the titration period, not the treatment. Most of what happens early is side effects and dose finding. If there is going to be benefit, it usually becomes visible somewhere in the second month, after the dose has settled.

The arithmetic in this review, though, I cannot defend. A standardized mean difference of 39 is not a thing. A confidence interval two hundredths wide on an Ashworth pooled estimate is not a thing. When I see numbers like that in a meta-analysis, I stop reading them as measurements and start reading them as a signal that the pooling was done across scales that should not have been combined. The authors are candid about their heterogeneity, which I respect, but candor about a problem does not remove the problem.

So what I take from this is one sentence I will use with patients and no numbers I will quote to them. Give it six to eight weeks at a stable dose before deciding, and expect the first two weeks to be the worst of it. That is honest, it matches the direction of this evidence, and it does not borrow authority the underlying data cannot supply.

Pooled short-term results in this review did not reach statistical significance while longer-term results looked substantially better, which supports counseling patients to allow weeks rather than days before judging a cannabinoid trial for spasticity. Do not quote the effect sizes. Heterogeneity of 100 percent and internally implausible confidence intervals make them unusable as quantities.

Take away the duration message and the honesty of the limitations section. Leave behind every number in the results. The distinction between a finding worth acting on and a figure worth citing is the whole lesson of this paper, and it applies well beyond multiple sclerosis.