In our systematic review of studies investigating THC-containing cannabis products for the treatment of IBD, we identified four RCTs and four non-interventional studies that met inclusion criteria, constituting the most up to date and inclusive synthesis on this topic. Several noteworthy findings emerged. First, among interventional studies and IBD outcomes assessed more than once, no single outcome demonstrated consistent improvement with cannabis use. Remission, examined in three trials, showed no significant differences between cannabis and control groups, suggesting with moderate confidence that cannabis may not enhance remission rates in patients with IBD. The moderate evidence quality of the heterogenous findings among studies examining disease activity scores and QoL indicate that evidence is insufficient to conclude whether cannabis might reduce disease activity scores and improve QoL in patients with IBD. While some outcomes assessed once demonstrated improvement (e.g., bloating and appetite), most carried low-quality evidence ratings, leading to high uncertainty in derived effect estimates. Weight change, assessed in a single small trial, was the only outcome rated as high-quality evidence and showed no effect of cannabis, yet the study might have been underpowered to capture small weight changes.

A second noteworthy finding was the incorporation of results from four recently published non-interventional studies with mixed findings regarding cannabis exposure on various IBD outcomes. Though, critical methodological limitations undermined confidence in their estimates for relevant clinical outcomes, hindering potential to draw sound conclusions based upon their findings, and the ability to provide robust actionable evidence in this field of research. We note that methodological constraints – including not only limitations in exposure measurement but also the heterogeneity and partial unblinding inherent to the cannabis interventions — reduced the quality of evidence across both interventional and non-interventional research. In light of these findings, this review identified no conclusive evidence of moderate or higher certainty supporting the effectiveness of THC-containing cannabis products in patients with inflammatory bowel disease. Accordingly, the current evidence does not support the use of these products as a substitute for approved therapies with established efficacy in reducing inflammation and improving symptoms.

Placing our findings in the context of previous systematic reviews reveals consensus for some outcomes and notable divergences for other IBD outcomes. Regarding clinical remission, our results concur with earlier analyses showing no consistent benefit of cannabis, a pattern similarly reflected in inflammatory biomarker findings. (Doeve et al. 2021; Kumar et al. 2024) QoL outcomes mostly align with earlier meta-analyses documenting improvements or mixed effects, though discrepancies may arise from our emphasis on adjusted analyses, which prior meta-analyses did not consider (e.g., preferred unadjusted trial results), and our exclusion of one study reporting in-vitro findings. (Doeve et al. 2021; Kumar 2024; Kang et al. 2024) We did not identify consistent improvement in disease activity scores, which is a departure from prior reviews. This may be attributable to our inclusion of more recent data, our inclusion of adjusted results, and consideration of different disease activity scoring systems under a single outcome type, rather than separating by disease entity or score type. (Doeve et al. 2021; Kumar et al. 2024; Kang et al. 2024; Vinci et al. 2022).

Confirming earlier systematic review findings, we observed no significant improvement in endoscopic outcomes after the addition of more recent studies. (Kang et al. 2024) Additional findings for clinical symptoms—such as nausea, appetite, and general well-being—partially differs from a previous meta-analysis, likely due to our inclusion of more recent studies. (Doeve et al. 2021) The modest improvement in bowel movements for ulcerative colitis reported by previous studies aligns with our own observations if separating our results between ulcerative colitis and Crohn’s disease. (Kumar 2024) In contrast, however, a previous meta-analysis suggested that cannabis exposure might provide pain relief; though, after including more recent studies and removing critically high risk of bias non-interventional studies, we did not consistently identify an improvement in pain among cannabis users when compared to controls. (Doeve et al. 2021).

Methodological challenges underscored significant barriers to deriving reliable effect estimates. In the non-interventional studies, severe risk of bias arose from insufficient control of confounders, including the absence of detailed data on concomitant medication use. Exposure misclassification further hampered interpretability, as the included studies relied on self-reported “ever” cannabis use or on diagnosis codes for cannabis dependence and cannabis use disorder, preventing an accurate quantification of current (acute) cannabis exposures, quantities of exposure, use duration and patterns of use, use modes, or cumulative cannabis exposures. Cross-sectional designs, which do not adequately capture the temporal sequence of exposure, outcome, and confounders, also precluded causal inferences in non-interventional studies. To address these issues, future non-interventional investigations need to adopt robust cannabis exposure measurement strategies as well as causal inference frameworks—such as target trial emulation—and integrate data sources that permit precise longitudinal exposure assessments. (Hernán et al. 2022; MMJ Outcomes 2024).

Included RCTs exhibited lower risk of bias, but limitations such as small sample sizes, reporting inconsistencies due to protocol deviations, challenges with blinding, and limited generalizability constrained the overall quality of evidence. Partial unblinding contributed to higher participant dropout rates across study arms, leading to issues with differential attrition in RCTs, and increased uncertainty in effect estimates specifically for subjective endpoints like pain and overall well-being. Three of the four RCTs deviated from prespecified protocols in terms of outcomes reported and analysis strategies, which could affect the transparency of reported findings. We observed limited generalizability since most trials did not include patients with moderate-to-severe IBD or complex comorbidities. Furthermore, the dosage and composition of cannabis interventions administered in trials often diverged from cannabis exposures in real-world settings, particularly in terms of THC content and frequency of use (e.g., THC content among medicinal and recreational programs is on average higher than 19%, whereas in included studies the THC content ranged from 0.6% to 16% THC). (Cash et al. 2020; Jugl et al. 2023) While the potential rescheduling of cannabis to a less restrictive category could reduce barriers to research and enable larger, more pragmatic RCTs, future studies also need to account for methodological challenges such as blinding in the presence of THC’s psychoactive effect. Ultimately, addressing these methodological shortcomings in both non-interventional and randomized research is imperative for a more definitive understanding of cannabis’s potential as a therapeutic in IBD.

Key strengths of this systematic review lie in its comprehensive search strategy, which encompassed both randomized trials and non-interventional studies, and thorough risk-of-bias and quality-of-evidence assessments. However, several limitations must be acknowledged. First, because few studies met inclusion criteria, our qualitative evidence synthesis pooled findings from heterogeneous populations, interventions, and control groups, prohibiting us from further exploring the effect of specific cannabis products (e.g., products with the same THC content, or the same cannabis product type) on reported outcomes or differences in effect sizes across specific subsets of patients. The preponderance of studies originating from the same research group raises questions regarding external validity, underscoring the need for replication by other investigators. High risks of bias resulted in substantial uncertainty and reduced quality of evidence for many outcomes. In addition, we opted not to incorporate findings from highly biased non-interventional studies into our quality-of-evidence ratings to avoid reliance upon very uncertain effect sizes. Moreover, although we excluded studies that solely focused on federally legal hemp or FDA-approved cannabinoid products—as they are not classified under Schedule I of the Controlled Substances Act (CSA)—in all but one non-interventional study, it was not feasible to distinguish between botanical cannabis, synthetic cannabinoids, and hemp due to the administrative nature of the data. Moreover, we did not evaluate FDA-approved or synthetic cannabinoid medicines (e.g., Epidiolex or dronabinol, and synthetic THC analogues such as nabilone). Because these products typically contain one cannabinoid (or a small, fixed set), whereas botanical cannabis comprises > 100 phytocannabinoids and hundreds of additional constituents (e.g., terpenes), our findings from cannabis products that contained more than one cannabinoid may not be generalizable to pharmaceutical cannabinoids. (André et al. 2024; Christensen et al. 2023) Moreover, since our search concluded in February 2023, we did search for any newly published RCTs between February 2023 and February 2026, which did not yield any new RCT, confirming timeliness of this review (French et al. 2005). Finally, the wide array of heterogenous outcome measures precluded the generation of meta-analytic summary estimates; instead, our ordinal summaries capture adjusted analyses and provide a useful framework to guide future research and policy. However, a limitation of this approach is that this synthesis relies largely on p-values, which can obscure clinically relevant effects when studies are heterogeneous and variably powered, ignores the magnitude and direction of observed effects, and invites misinterpretation when multiple comparisons are made.