A large 2025 observational study reports that chronic pain patients using medical cannabis had fewer emergency department and urgent care visits than cannabis-naive patients. However, the study was conducted by employees of the cannabis certification company whose customer data were analyzed, and the comparison groups differed so dramatically at baseline that the findings cannot reliably support causal conclusions.

Chronic pain affects an estimated 50 million Americans and drives enormous healthcare costs through emergency department visits, urgent care encounters, and hospitalizations. Millions of patients now use medical cannabis as part of their pain management, yet payers and policymakers have very little rigorous evidence on whether cannabis use actually reduces downstream healthcare utilization. Any large study that appears to quantify this relationship will be influential in clinical, insurance, and legislative discussions, making it critical that the evidence is appraised carefully before it shapes decisions.

Chronic noncancer pain remains one of the most common reasons patients seek medical cannabis certification, yet relatively little large-scale evidence exists on whether cannabis use actually reduces healthcare system encounters such as emergency department visits, urgent care visits, or hospitalizations. This study drew data from Leafwell, a major commercial telehealth platform facilitating cannabis certifications across 36 states, and applied Targeted Maximum Likelihood Estimation with SuperLearner ensemble machine learning to compare healthcare utilization between experienced cannabis users renewing their certifications and first-time cannabis-naive applicants. The mechanistic rationale rests on the hypothesis that effective pain management through cannabis may reduce the need for acute care encounters driven by uncontrolled pain.

After adjustment, the study estimated a 2.0 percentage point reduction in urgent care visits, a 3.2 percentage point reduction in ED visits, and 3.52 fewer unhealthy days per month among cannabis-exposed patients. Hospitalization rates trended lower but did not reach statistical significance. However, the unexposed group had dramatically worse baseline health: 85% reported bothersome or high-impact pain compared to 59% in the exposed group, and 76% reported three or more unhealthy weeks per month compared to 52%. All outcomes were self-reported and unverified against medical records or insurance claims. The authors acknowledge that randomized controlled trials are needed, but the study’s framing as a causal inference analysis may overstate the confidence warranted by these findings.

Fewer ER Visits Among Cannabis Users With Chronic Pain: Association, Not Causation

A company that sells medical cannabis certifications just published a study showing that their customers use the emergency room less than people who haven’t bought a cannabis certification yet. The methods are genuinely sophisticated. The conflict of interest is real. And the truth, as usual, is somewhere in between. I want to be clear about what this paper gets right before discussing what it gets wrong. The authors identified a genuine evidence gap: we have remarkably little large-scale, adjusted data on whether cannabis use actually changes healthcare utilization patterns in chronic pain populations. Their choice of TMLE with SuperLearner is methodologically sound and represents best practice for causal estimation from observational data. The validated pain severity and quality-of-life measures reflect real methodological care. This is not a sloppy study. But the central problem is not statistical. It is conceptual. Imagine studying whether gym memberships improve health by comparing people who have been going to the gym for a year and are renewing their membership against people who just signed up for the first time. The renewing members are already healthier, not necessarily because of the gym, but because the people who felt worse or saw no benefit already dropped out. That is exactly what happened here. The recertifying cannabis users are survivors: they tolerated cannabis, found it helpful enough to continue, and returned to renew. The first-time applicants are a fundamentally different population, with dramatically worse pain and quality of life at baseline. No statistical method, however precise, can reconstruct the counterfactual these groups were meant to represent.

The conflict of interest deserves direct acknowledgment. Three of four authors work for Leafwell, the company whose proprietary data formed the entire study. This does not mean the data are fabricated or the analysis was dishonest. But it means the research question, the framing, the analytical choices, and the publication decision all passed through a commercial filter. Asking a coffee company’s research team to study whether coffee reduces fatigue using the company’s own customer data is simply not the same as an independent university investigation, even if the statistical methods are impeccable. Readers and policymakers should weigh this paper accordingly. The absence of pre-registration further limits confidence that outcome selection and analytical decisions were not influenced by results that emerged during analysis.

What would I say to a patient who brings me this study? I would say it is consistent with a growing body of evidence that medical cannabis may help some people with chronic pain, and that it is biologically plausible that better pain control could reduce emergency care needs. But I would be honest: this particular study cannot prove that cannabis caused fewer ER visits, because the groups being compared were never truly comparable, and the people who ran the study have a financial stake in the answer. To a colleague, I would say the TMLE work is competent and the sample size is unusually large, but the design is fundamentally compromised by survivor selection and healthy user bias. To a policymaker, I would say this is not the evidence base on which to build coverage or integration decisions. The field urgently needs this question answered. It also needs the answer to come from independently funded, pre-registered, prospective research with outcomes verified against claims data. Until then, this study is best understood as a well-executed argument for why that research should happen, not as evidence that the answer is already known. Methodological sophistication, however genuine and well-intentioned, cannot compensate for a flawed comparator design or resolve conflicts of interest inherent in industry-conducted research on the industry’s own customers. The lesson for the field is that the urgent need for cannabis research evidence should not lower the bar for what counts as credible evidence.

This study sits at an early stage in the research arc for cannabis and healthcare utilization. While meta-analyses of randomized controlled trials have established modest efficacy for cannabinoids in chronic pain, the downstream effects on healthcare encounters remain almost entirely uncharacterized by rigorous designs. This analysis is among the first to apply advanced causal inference methods to this specific question at scale, but its observational design, structural comparator problems, and industry authorship place it firmly in the hypothesis-generating category rather than the evidence-confirming category.

From a pharmacological perspective, the study captures no information about cannabis dose, formulation, THC-to-CBD ratio, route of administration, or concurrent pain treatments, all of which are critical variables for understanding any utilization effect. Drug interactions, tolerability, and safety signals cannot be assessed from these data. Clinicians should note that the absolute effect sizes, even if taken at face value, are modest: 2 to 3 percentage point reductions in visit rates. The most actionable recommendation is to discuss cannabis as one component of multimodal chronic pain management while being transparent with patients that the evidence for healthcare utilization reduction is preliminary and requires independent confirmation.

This is a retrospective observational cohort study using proprietary administrative data from a single commercial telehealth platform. Despite employing Targeted Maximum Likelihood Estimation, a method designed for causal inference, the study occupies a position well below randomized controlled trials in the evidence hierarchy. The single most important inference constraint is that the exposed and unexposed groups are structurally non-comparable populations, meaning the no-unmeasured-confounding assumption required for valid causal estimation is not credibly satisfied.

The findings are directionally consistent with the broader body of evidence suggesting that cannabinoids provide modest benefits for chronic pain. The 2017 National Academies of Sciences report concluded there was substantial evidence for cannabis efficacy in chronic pain, and subsequent meta-analyses of randomized trials have generally supported this, though effect sizes are moderate. However, the specific claim that cannabis reduces healthcare utilization is far less established. Most prior studies examining this question have used ecological designs or claims analyses with their own significant limitations. This study extends the literature by applying more sophisticated causal methods, but it does not resolve the fundamental challenge of confounding that has plagued this entire research area. Independent replication with verified outcomes remains the critical next step.

The most consequential analytical choice was the definition of the comparator group. Had the authors compared cannabis users against a matched cohort of chronic pain patients not seeking cannabis certification at all, drawn from an independent data source such as insurance claims or electronic health records, the results might have looked substantially different because the comparison would not have been contaminated by the selection pressures unique to a certification platform. Additionally, the absence of an E-value or Rosenbaum bounds sensitivity analysis means we cannot quantify how strong an unmeasured confounder would need to be to explain away the observed associations. Given the enormous baseline differences between groups, such an analysis would likely have revealed that only modest unmeasured confounding would be needed to eliminate the observed effects entirely.

The most likely overinterpretation is that the use of TMLE makes this study equivalent to a randomized trial, or that it has proven cannabis causes fewer ER visits. TMLE is a powerful tool for improving estimation efficiency and reducing reliance on parametric model assumptions, but it cannot eliminate unmeasured confounding or fix a structurally flawed comparator design. The exposed group in this study was composed of experienced cannabis users with milder baseline pain who chose to continue their treatment, while the unexposed group was made up of first-time applicants with dramatically worse health. These are fundamentally different populations, and the observed differences in healthcare utilization may simply reflect that difference rather than any treatment effect.

This study contributes the largest adjusted observational estimate to date on medical cannabis and healthcare utilization in chronic pain, using genuinely advanced statistical methods. It does not establish that cannabis causes reduced emergency department or urgent care visits, because the comparison groups were structurally non-comparable and all outcomes were self-reported by the cannabis certification company’s own customers. For now, these findings are best treated as hypothesis-generating rather than practice-changing, and they underscore the urgent need for independently funded, prospective, pre-registered research with verified outcomes.

Does this study prove that medical cannabis reduces ER visits for chronic pain patients?

No. The study found an association between cannabis use and fewer self-reported ER visits, but the two groups being compared were very different to start with. People who had already been using cannabis for a year had milder pain at baseline than first-time applicants. The observed differences may reflect those pre-existing differences rather than any effect of cannabis itself. A randomized controlled trial would be needed to establish causation.

Should I bring this study up with my doctor if I have chronic pain?

It is always reasonable to discuss new research with your doctor, but this study alone should not be the basis for a treatment decision. Its findings are preliminary and come with significant limitations, including the fact that the researchers work for the company that sells cannabis certifications. Your doctor can help you weigh this alongside stronger evidence about cannabis and chronic pain management.

Why does it matter that the study was conducted by Leafwell employees?

Leafwell is a cannabis certification platform, meaning the company has a direct financial interest in findings that make medical cannabis look beneficial. Three of four study authors are Leafwell employees who analyzed the company’s own customer data. This does not mean the findings are wrong, but it means the research question, analytical decisions, and publication framing were all made by people with a commercial stake in the outcome. Independent replication is essential before these findings can be considered reliable.

What would it take to answer this question more definitively?

Stronger evidence would come from prospective studies where patients are followed over time, ideally with random assignment to cannabis versus other pain management strategies. Outcomes should be verified using medical records or insurance claims rather than self-report, the research should be funded independently, and the study protocol should be registered in advance so analytical decisions are transparent.

Have thoughts on this? Share it: