What the FDA Knows About CBD and Cancer: Regulatory Evidence, Expert Commentary, and Public Health Failure

 Background.

Cannabidiol (CBD) now dominates consumer health markets worldwide. Manufacturers market it predominantly on the basis of perceived safety. However, CBD carcinogenicity FDA evidence, generated during the pharmaceutical approval process for purified CBD (Epidiolex, GW Research Ltd, NDA 210365), raises substantive concerns that regulators have not communicated to the public, prescribers, or policymakers.

Evidence reviewed. This commentary analyses three bodies of evidence: (1) an independent statistical review of a 104-week oral carcinogenicity study in CD-1 mice, conducted by the FDA’s Division of Biometrics VI (Reference ID: 4896227, November 2021); (2) a systematic review and meta-analysis of 25 studies examining cannabis use and cancer risk, published in JAMA Network Open (Ghasemiesfe et al., 2019); and (3) a correspondence in Gastroenterology (Reece and Hulse, 2023) proposing a biologically grounded hypothesis linking cannabis to rising pancreatic ductal adenocarcinoma incidence.

Principal findings. The FDA’s independent statistical review identified dose-dependent, significant increases in hepatocellular adenoma in male mice given purified CBD (p = 0.0001, high-dose vs water control). Expert commentary by Professor Albert Stuart Reece, a recognised researcher in cannabinoid genotoxicity, indicates these findings likely understate human carcinogenic risk. Mice resist cancer more readily than humans do. Additionally, a parallel two-year rat carcinogenicity study, described in the FDA submission as ongoing, appears absent from the public domain. Human epidemiological data show a significant association between sustained cannabis use and testicular germ cell tumours (OR 1.36; 95% CI 1.03 to 1.81). Emerging evidence also implicates cannabis in rising pancreatic cancer rates among younger women.

Conclusions. Animal toxicology, basic genotoxicity science, and human epidemiological data converge on a coherent carcinogenic signal. Current regulatory communication has not addressed this adequately. Therefore, this commentary calls for public disclosure of existing carcinogenicity data, clarification of the missing rat study, and prioritised funding of longitudinal cohort studies in cannabis-only users.

1. Introduction

Cannabidiol (CBD) is the principal non-psychoactive phytocannabinoid derived from Cannabis sativa. Following the 2018 FDA approval of Epidiolex for treatment-resistant epilepsy, CBD gained pharmaceutical legitimacy. Consequently, it expanded rapidly into over-the-counter consumer markets across North America, Europe, and Australasia. By the early 2020s, CBD products appeared in pharmacies, health food retailers, and online platforms, promoted for anxiety, insomnia, chronic pain, and systemic inflammation (Corroon and Phillips, 2018).

This commercial normalisation brought with it a persistent public narrative of safety. CBD’s non-psychoactive profile, its distinction from delta-9-tetrahydrocannabinol (THC), and its regulatory approval for a paediatric epilepsy indication together reinforced the perception that it poses minimal toxicological risk. Consumer behaviour, clinical practice, and regulatory posture all reflect this perception widely.

CBD carcinogenicity FDA evidence challenges this assumption directly. As a condition of Epidiolex’s approval, GW Research Ltd submitted pre-clinical long-term carcinogenicity data to the FDA. The FDA’s Division of Biometrics VI then conducted an independent statistical review of those data in November 2021. That review identified statistically robust carcinogenic signals in the animal model studied. These findings exist in the public domain under NDA 210365. Yet they have attracted minimal scientific or public attention.

This commentary brings together that regulatory evidence with the parallel bodies of basic genotoxicity science and human epidemiological data on cannabinoid carcinogenicity. The aim is to evaluate the coherence and strength of the overall carcinogenic signal and to identify the regulatory and research actions this evidence warrants.

2. The FDA Carcinogenicity Submission: Study Design and Regulatory Context

2.1 Study Design

To secure approval for Epidiolex, GW Research Ltd submitted a 104-week oral carcinogenicity study in Crl:CD1(ICR) mice. Researchers administered purified CBD by oral gavage at doses of 30 mg/kg/day (low), 100 mg/kg/day (mid), and 300 mg/kg/day (high). Concurrent vehicle control and water control groups each contained 60 animals per sex. This design conforms to standard international regulatory requirements under ICH guideline S1B.

The FDA’s Division of Biometrics VI then conducted an independent statistical review of the submitted data. Dr Hepei Chen served as Statistical Reviewer and Dr Karl Lin as concurring Team Leader. They completed and signed the review in November 2021 (Reference ID: 4896227).

2.2 The Absent Rat Carcinogenicity Study

ICH S1B requires carcinogenicity assessment across two rodent species, typically the rat and the mouse, to provide independent corroborating or contradictory evidence of tumourigenic potential. The FDA submission document explicitly noted that a parallel two-year carcinogenicity study in rats was ongoing at the time of filing.

Expert commentary by Professor Reece indicates that this rat study appears never to have been filed with the FDA. If that assessment is accurate, it constitutes a significant gap in the regulatory submission package. Under standard pharmaceutical toxicology principles, negative findings in one species carry equal scientific weight to positive findings in another. Regulatory bodies ordinarily scrutinise incomplete carcinogenicity packages.

Two possibilities exist. First, GW Research may have completed the rat study and filed the data at a later date not reflected in publicly accessible records. Second, the study may not have been completed or submitted at all. In either case, the absence of these data from the public scientific record is notable, particularly given the positive findings in the mouse model. Accordingly, regulatory bodies and relevant parliamentary health committees should seek formal clarification of the rat study’s status.

3. Independent Statistical Review: Principal Findings

3.1 Methodological Considerations

The FDA submission contained two statistical analyses: the sponsor’s own analysis and the FDA’s independent review. GW Research used the Peto test for occult tumours and the Wilcoxon and log-rank tests for palpable tumours, consistent with standard industry practice. By contrast, the independent FDA reviewer applied the Poly-k method, a mortality-adjusted approach that modifies the effective animal count at risk. Specifically, it accounts for animals dying before study termination without having developed the tumour in question.

The Poly-k adjustment is especially important in 104-week bioassays. Without it, intercurrent mortality can confound tumour incidence calculations. The method is widely regarded as the more statistically robust approach for long-duration rodent carcinogenicity studies (Portier and Bailer, 1989). Furthermore, the reviewer applied significance thresholds of alpha = 0.005 for common tumours and alpha = 0.025 for rare tumours, forming a conservative multiple-testing framework. Findings that surpassed these thresholds carry strong statistical confidence.

Professor Reece, in his expert commentary on this submission, identifies the independent reviewer’s methodology as the more powerful and more reliable of the two analytical approaches. Both are valid; however, the Poly-k method is better suited to the study’s duration and structure.

3.2 Hepatocellular Adenoma in Male Mice

The independent reviewer found dose-dependent increases in hepatocellular adenoma across all treated male groups compared with both the vehicle control and the water control groups. These increases reached statistical significance throughout. The table below summarises the principal p-values from the independent analysis:

These findings held statistical significance regardless of whether the tumour was common or rare under the multiple-testing framework. A p-value of 0.0001, under a conservative analytical threshold, does not describe a marginal result. Rather, it describes a highly robust carcinogenic signal.

3.3 Additional Tumour Signals in Male and Female Mice

The sponsor’s own analysis added further signals in male mice. Malignant fibrosarcoma and the combined fibrosarcoma/sarcoma NOS category in skin and subcutaneous tissue both showed significant dose-response trends. Wilcoxon test p-values were 0.0093 and 0.0027 respectively.

In female mice, the independent reviewer found a dose-response trend toward increased hepatocellular adenoma at p = 0.0208. Although this did not clear the conservative common-tumour threshold of 0.005, it points in the same direction as the male findings and therefore warrants note.

Taken together, the independent review and the sponsor’s analysis converge on a pattern of hepatic and soft-tissue tumour formation in CBD-treated mice. That pattern is dose-dependent and statistically robust across both analytical approaches.

4. Species Extrapolation: Why Mouse Findings May Understate Human Risk

4.1 The Standard Limitation Argument and Its Limits

Animal carcinogenicity data routinely attract the qualification that rodent findings do not automatically translate to human risk. Doses used in regulatory bioassays typically exceed realistic human exposures substantially. Moreover, inter-species physiological differences complicate direct extrapolation. These are legitimate scientific cautions.

However, Professor Reece cautions that this standard framing may misrepresent the direction of uncertainty when genotoxic mechanisms drive the carcinogenic effect. His expert assessment is that mice resist cancer more readily than humans do. Consequently, carcinogenic signals observed in mice are likely to understate, rather than overstate, the risk to people. Three biological features of the mouse support this assessment.

4.2 Telomere Biology and Chromosomal Instability

Mice carry substantially shorter absolute telomere lengths than humans, despite showing higher telomerase activity. Their telomere dynamics across the replicative lifespan also differ fundamentally (Calado and Dumitriu, 2013). These differences alter how chromosomal instability develops under sustained genotoxic pressure. They also affect the rate at which tumour suppressor pathways become compromised. In longer-lived organisms with greater cumulative tissue replication, sustained genotoxic exposure has more replicative cycles across which to accumulate oncogenic mutations.

4.3 Metabolic Enzyme Profiles and Reactive Metabolite Generation

Murine drug-metabolising enzyme systems differ in composition and activity from human equivalents. As a result, the reactive metabolites that cannabinoid biotransformation generates, and therefore the DNA adduct formation, oxidative DNA damage, and epigenetic disruption that follow, may differ meaningfully between species. There is no empirical basis for assuming human metabolic processing of sustained cannabinoid exposure produces a lower genotoxic burden than the murine equivalent.

4.4 Genotoxic Pathway Kinetics in Long-Lived Organisms

The genotoxic mechanisms cannabinoids activate, including DNA demethylation, CTCF boundary disruption, reactive oxygen species generation, and proto-oncogene overexpression, are not species-specific. However, they operate over time. Organisms with longer lifespans and greater total replicative tissue turnover accumulate more oncogenic damage per unit of exposure. The human case is therefore not simply analogous to the mouse case; it is likely to be more severe.

4.5 Implications for Interpreting the Regulatory Framework

The FDA’s regulatory framework treats statistically significant rodent findings as signals warranting further investigation. Importantly, it does not treat them as ceiling estimates of human risk. This distinction is scientifically critical. Yet public discussion of the CBD carcinogenicity FDA evidence consistently underweights it.

5. Mechanistic Basis: Cannabinoid Genotoxicity in the Basic Science Literature

5.1 Convergence With Pre-existing Genotoxicity Evidence

The carcinogenic signals in the FDA mouse study do not arise in isolation. They align precisely with a substantial and growing body of basic science literature on the genotoxic and epigenotoxic effects of cannabinoids across independent experimental systems (Reece et al., 2021; Reece et al., 2023). This alignment is scientifically significant. It transforms the animal findings from a potentially anomalous result into the predicted biological consequence of a compound class with a documented genotoxic profile.

5.2 DNA Demethylation and Epigenomic Disruption

CBD and related cannabinoids induce widespread DNA demethylation in laboratory models (Reece et al., 2021). DNA methylation is a foundational epigenetic mechanism governing gene expression, cellular differentiation, and oncogenic pathway suppression. When normal methylation patterns break down, programmes controlling cell proliferation, apoptosis, and lineage identity become dysregulated. This creates permissive conditions for neoplastic transformation.

5.3 CTCF Boundary Disruption

CTCF is a master regulator of three-dimensional chromatin architecture. It maintains topologically associating domain boundaries that separate distinct cell lineage gene expression programmes and constrain enhancer elements. Research has linked cannabinoid exposure to disruption of these CTCF-maintained boundaries (Reece et al., 2021). When boundary integrity weakens, oncogenic superenhancers gain access to proto-oncogene promoters that CTCF would ordinarily insulate. The consequences for cell growth regulation can be significant.

5.4 Reactive Oxygen Species and Direct DNA Damage

Cannabis combustion products and cannabinoid metabolites generate reactive oxygen species (ROS). Researchers have long established ROS as mediators of direct DNA damage, including strand breaks, base oxidation, and chromosomal rearrangements (Melamede, 2005). ROS-mediated genotoxicity is a recognised mechanism of chemical carcinogenesis across numerous compound classes.

5.5 Proto-oncogene Expression in Human Bronchial Epithelium

The 2019 JAMA Network Open systematic review (Ghasemiesfe et al., 2019) noted that several included studies found proto-oncogene overexpression in the bronchial epithelium of cannabis-only smokers at frequencies exceeding those in tobacco-only smokers. Researchers also observed squamous cell metaplasia, mitotic figures, and cellular disorganisation. These are established histopathological markers of pre-malignant change in human airway tissue, not theoretical projections.

Reece and colleagues characterise the overall cannabinoid genotoxic profile as implicating teratogenesis, carcinogenesis, and accelerated cellular ageing (Reece et al., 2023). Therefore, the FDA mouse study reflects not a surprising finding but the expected outcome of giving a compound with a documented genotoxic profile to a mammalian model across a two-year study window.

6. Human Epidemiological Evidence

6.1 Systematic Review Methodology and Evidence Grading

Ghasemiesfe and colleagues (2019) conducted a systematic review and meta-analysis of 25 studies on marijuana use and cancer development, published in JAMA Network Open. The review applied GRADE methodology for evidence quality assessment and used random-effects meta-analytic models where studies showed sufficient homogeneity to permit pooling.

6.2 Testicular Germ Cell Tumours: A Significant Association

The review identified a statistically significant positive association between more than ten years of regular cannabis use and testicular germ cell tumours (OR 1.36; 95% CI 1.03 to 1.81). The non-seminoma subtype showed a stronger association still (OR 1.85; 95% CI 1.10 to 3.11). Notably, neither result showed significant heterogeneity between contributing studies. This supports the reliability of the pooled estimates despite their modest magnitude.

Researchers graded this evidence as “low strength” due to methodological limitations: imprecise exposure quantification, inadequate separation of cannabis from tobacco use, probable recall bias, and follow-up durations too short to capture the full latency of solid tumour development. However, these limitations describe structural constraints on effect detection. They do not constitute evidence against a carcinogenic association. Consequently, studies that are underpowered or methodologically constrained cannot logically serve as exculpatory evidence for an effect they were not designed to detect at full magnitude.

6.3 Lung, Head, Neck, and Other Malignancies: Evidence Gaps

For lung cancer, head and neck cancer, oral cancer, and the remaining malignancy types examined, the review returned insufficient evidence classifications. Here, insufficient evidence and absence of evidence are not equivalent. The authors noted specific methodological deficiencies, including the predominant co-use of tobacco in cannabis-using populations. They explicitly called for large-scale longitudinal cohort studies in cannabis-only users with rigorous prospective exposure measurement. To date, researchers have not conducted those studies.

6.4 Pancreatic Cancer: A Biologically Grounded Emerging Hypothesis

Reece and Hulse (2023) published a correspondence in Gastroenterology responding to epidemiological data on rising pancreatic ductal adenocarcinoma incidence in the United States. The data showed annual percentage increases approximately four times higher in women than in men. The largest differential appeared in the 15 to 34 year age cohort. On the basis of this pattern, the authors propose cannabis as a significant missing environmental carcinogen contributing to this trend.

6.5 Biological Mechanism and Epidemiological Alignment

The biological mechanism Reece and Hulse outline rests on established molecular pathology. Pancreatic carcinogenesis characteristically proceeds through cycles of acinar cell inflammation and dedifferentiation, producing pre-malignant field changes that render the pancreatic epithelium susceptible to oncogenic KRAS mutation. Cannabis exerts pro-inflammatory effects on pancreatic tissue, induces the DNA demethylation and CTCF disruption described above, and thereby plausibly lowers the mutational threshold for this carcinogenic sequence.

The epidemiological signal reinforces this concern. Daily or near-daily cannabis use among women aged 18 to 25 increased by approximately 89% between 2001 and 2019, compared with approximately 43% among age-matched males (Substance Abuse and Mental Health Services Administration data, as cited in Reece and Hulse, 2023). This differential uptake pattern temporally and demographically mirrors the disproportionate rise in pancreatic cancer among younger women. The authors propose a testable causal hypothesis. It remains a hypothesis requiring dedicated prospective epidemiological evaluation. Nevertheless, the mechanistic underpinning and epidemiological alignment are sufficient to justify urgent formal investigation.

7. Regulatory Transparency and the Communication Gap

7.1 What the FDA Record Contains

CBD carcinogenicity FDA evidence exists in the public domain. Regulators generated it under conditions designed to protect public health. Yet they have not communicated it to the populations who use the product it concerns. The FDA’s own independent statistical review confirmed dose-dependent hepatic tumourigenesis in male mice given purified CBD at p = 0.0001. That finding is not ambiguous. On the standard framework for interpreting animal carcinogenicity data, it is a positive carcinogenicity signal in the studied species.

Hundreds of millions of consumers purchase CBD products globally. In most cases, the clinicians who recommend CBD to patients lack knowledge of this regulatory finding. Similarly, policymakers who have progressively liberalised CBD’s regulatory status operate without reference to it. This gap between the regulatory record and public and professional knowledge represents a failure of regulatory communication, irrespective of how one interprets the clinical significance of the underlying data.

7.2 The Missing Rat Data and Its Significance

The absence of the parallel rat carcinogenicity data from public scientific discourse deepens this concern substantially. Standard pharmaceutical carcinogenicity assessment requires two rodent species precisely because findings in one species may not replicate in another. The pattern of concordance or discordance between species informs the overall weight of evidence. Therefore, if researchers completed the rat study, its findings are scientifically necessary for interpreting the mouse data in full context. If they did not complete or file it, the regulatory basis for that decision requires scrutiny.

Historical precedent in pharmaceutical regulation demonstrates the consequences of incomplete species coverage in carcinogenicity assessment. As a result, the FDA, the EMA, and relevant national agencies should formally answer whether researchers completed, submitted, and reviewed the two-year rat carcinogenicity study for purified CBD and, if not, why not.

8. Evaluation of Counter-Arguments

Three principal objections to the significance of the animal carcinogenicity data arise routinely. Each deserves careful consideration.

8.1 The High Dose Objection

The doses in the mouse study (30, 100, and 300 mg/kg/day) substantially exceed typical human CBD exposures from consumer products. This is a legitimate methodological consideration. However, regulatory bioassays routinely use high doses to detect carcinogenic signals within the statistical power constraints of two-year studies. The relevant scientific question is not whether the doses were high. Rather, it is whether the mechanisms producing tumours at high doses would operate, at lower magnitude, at human-relevant exposures. The mechanisms involved, including DNA demethylation, CTCF disruption, and ROS generation, do not show threshold-dependent biological activity. Consequently, dose cannot serve as a categorical exculpatory factor.

8.2 The Species Difference Objection

As discussed in Section 4, expert assessment indicates the direction of this uncertainty points toward greater, not lesser, human susceptibility. The species difference argument therefore does not straightforwardly support reassurance.

8.3 The Incomplete Human Evidence Objection

The incompleteness of human epidemiological evidence reflects structural limitations of existing study designs, not an investigated and negative finding. The appropriate response to evidential incompleteness is research investment to resolve it. Confident reassurance on the basis of absent evidence is not scientifically justified.

8.4 The Precautionary Principle

Taken together, these considerations support applying the precautionary principle. Where credible pre-clinical carcinogenicity evidence exists alongside a mechanistic basis for human relevance, that principle supports protective regulatory action and transparent public communication, pending resolution of the evidential gaps.

9. Recommendations

The weight of available evidence points to five specific actions.

Public disclosure of carcinogenicity findings. Regulatory agencies should issue accessible plain-language summaries of the GW Research mouse carcinogenicity data and the FDA’s independent statistical review. Furthermore, these summaries should appear in product information for both pharmaceutical-grade and consumer CBD products.

Clarification of the rat carcinogenicity study. The FDA, EMA, and relevant national agencies should formally clarify whether researchers completed, reviewed, and filed the parallel two-year rat carcinogenicity study for purified CBD and what its findings were. Parliamentary health committees should pursue this clarification through formal inquiry processes.

Prioritised funding for longitudinal cohort research. Governments and research funding bodies should designate large-scale longitudinal cohort studies of cannabis-only users as a public health research priority. Moreover, these studies require prospective and rigorous exposure quantification. The methodological limitations preventing definitive conclusions from existing human studies are correctable with appropriate study design and funding.

Integration of cannabis exposure into cancer surveillance. National cancer registries, surveillance datasets, and major population cohort studies should systematically capture cannabis use as an exposure variable. Sufficient granularity to support dose-response and latency analysis is essential.

Investigation of the pancreatic cancer hypothesis. The hypothesis Reece and Hulse (2023) advance regarding cannabis and rising pancreatic ductal adenocarcinoma rates deserves formal epidemiological evaluation. Specifically, researchers should apply prospective cohort methods and space-time analytical approaches within established causal inference frameworks.

10. Conclusion

This commentary does not assert that CBD or cannabis is a proven human carcinogen. Current evidence does not support that categorical conclusion. However, CBD carcinogenicity FDA evidence, taken together with the mechanistic literature on cannabinoid genotoxicity and the available human epidemiological data, constitutes a coherent and concerning carcinogenic signal. Current regulatory communication has failed to reflect this signal adequately.

The FDA’s independent statistical review of the GW Research mouse carcinogenicity study found highly significant, dose-dependent hepatic tumourigenesis at p = 0.0001. Expert commentary indicates this finding likely underestimates human carcinogenic risk. Furthermore, a parallel two-year rat study appears absent from the public record. The 2019 JAMA Network Open meta-analysis found a significant association between sustained cannabis use and testicular germ cell tumours (OR 1.36 to 1.85 across subtypes). Reece and Hulse identified biologically grounded epidemiological signals consistent with a cannabis contribution to rising pancreatic cancer rates in younger populations. Moreover, the basic genotoxicity literature predicts all of these findings through well-characterised mechanisms.

CBD carcinogenicity FDA evidence is not hidden. It is publicly accessible. What has been absent is the regulatory and scientific communication required to bring it to the attention of clinicians, policymakers, and the public who rely on those systems to identify and communicate risk. Closing that communication gap is a prerequisite for informed individual decision-making and proportionate public health policy on cannabinoids.

References

1. Corroon J, Phillips JA. A cross-sectional study of cannabidiol users. Cannabis and Cannabinoid Research. 2018;3(1):152-161. doi:10.1089/can.2018.0006
2. Ghasemiesfe M, Barrow B, Leonard S, Keyhani S, Korenstein D. Association between marijuana use and risk of cancer: a systematic review and meta-analysis. JAMA Network Open. 2019;2(11):e1916318. doi:10.1001/jamanetworkopen.2019.16318
3. Reece AS, Hulse GK. Cannabis could be the missing environmental carcinogen hiding in plain view [correspondence]. Gastroenterology. 2023. doi:10.1053/j.gastro.2023.02.050
4. Chen H (Statistical Reviewer), Lin KK (Concurring Reviewer). Statistical review and evaluation: carcinogenicity study, NDA 210365 (purified CBD). 104-week oral carcinogenicity study in CD-1 mice. US Food and Drug Administration, Center for Drug Evaluation and Research, Division of Biometrics VI; Reference ID: 4896227. Signed 30 November 2021.
5. Reece AS, Hulse GK, Norman A, Bower C. Perturbation of epigenomic landscapes in the genotoxicity of cannabis. Scientific Reports. 2021;11:13892-13912. doi:10.1038/s41598-021-93212-6
6. Reece AS, Hulse GK. Multiple genotoxic and epigenotoxic effects of cannabinoids and implications for public health. International Journal of Environmental Research and Public Health. 2023;20:3360-3383. doi:10.3390/ijerph20043360
7. Portier CJ, Bailer AJ. Testing for increased carcinogenicity using a survival-adjusted quantal response test. Fundamental and Applied Toxicology. 1989;12(4):731-737. doi:10.1016/0272-0590(89)90004-3
8. Calado RT, Dumitriu B. Telomere dynamics in mice and humans. Seminars in Hematology. 2013;50(2):165-174. doi:10.1053/j.seminhematol.2013.03.030
9. Melamede R. Cannabis and tobacco smoke are not equally carcinogenic. Harm Reduction Journal. 2005;2:21. doi:10.1186/1477-7517-2-21

Expert commentary in this article came from written communications with Professor Albert Stuart Reece, Division of Psychiatry, University of Western Australia, and School of Medical and Health Sciences, Edith Cowan University. Professor Reece’s published research on cannabinoid genotoxicity is cited separately in the reference list above.

The GW Research Ltd carcinogenicity submission and FDA independent statistical review are publicly accessible under NDA 210365 via the FDA’s online submission database. The JAMA Network Open systematic review is published under open access (CC-BY licence). The Gastroenterology correspondence is cited as published.

This commentary was prepared for the Dalgarno Institute. It does not constitute medical advice. Clinicians and researchers should consult primary sources and exercise independent professional judgement. Members of the public seeking guidance on individual health decisions should consult a qualified healthcare professional.

(Source: WRD News)