Brian J. Piper, PhD Associate Professor of Neuroscience Geisinger Commonwealth School of Medicine Scranton PA 18411

Prior studies claiming to find a lethal cannabis-immunotherapy drug interaction riddled with errors Interview with:

Brian J. Piper, PhDAssociate Professor of Neuroscience Geisinger Commonwealth School of Medicine Scranton PA 18411

Dr. Piper

Brian J. Piper, PhD
Associate Professor of Neuroscience
Geisinger Commonwealth School of Medicine
Scranton PA 18411 What is the background for this study?

Response: Many cancer patients use marijuana to treat pain, nausea, or anxiety, often without communicating this with their health care providers. Two observational studies (1, 2) from a single institution in Israel purporting to find a dangerous drug interaction between medical cannabis and immunotherapy have been cited hundreds of times, including by clinical practice guidelines.

The cannabinoid CB2 receptor is found on immune tissues so it is biologically possible that marijuana could make immunotherapies like nivolumab less effective. However, there were anonymous reports on PubPeer (3-5) of many irregularities in the data-analysis. If there were unappreciated differences on other important variables at baseline besides subsequent cannabis use, this could change the interpretation of these influential reports (1, 2). This investigation involved attempting to repeat and verify the data-analysis. What are the main findings?

Response: This report (6) identified many omissions, irregularities, and mistakes in the prior studies

(1, 2). The cannabis users were six years younger than non-users which was erroneously reported as a non-significant group difference. There was also a tendency (p = .06) for the cannabis users to be more likely to be tobacco smokers in the retrospective report (1). Less than five percent of the statistics could be verified in the prospective study (2) and smoking status, a major variable in oncology studies, was not reported. Liver metastases were previously described as equivalent in cannabis users and non-users (p = .89) but re-analyses found they were more common in the cannabis users at baseline (p < .0001). There were also many irregularities in rounding and occasionally even difficulties in calculating percentages. For example, brain metastasis (8 out of 34 total patients) was reported as 13.2% but calculated as 23.5%. What should readers take away from your report?

Response: The conclusions of the prior non-randomized studies (1, 2) should be viewed skeptically as there were other confounds like smoking, age, and other baseline differences that could account for the observed differences between cannabis users and nonusers. The readers trust in evidence-based medicine is tested when the methods list one statistic (1,2) but the reported p-value indicates that another statistic was completed instead. Further, the methods claimed that two-tailed statistics were run but the p-value in the results suggests that some were one-tailed. These are clear deviations from accepted research practice. In general, this publication generally verified the PubPeer reports (3-5).

Medical marijuana is a controversial topic. Some previous researchers have confused correlation with causation. A clear example of this is the “gateway” model of a sequence in drug use from readily available substances like alcohol and cigarettes, to marijuana, and later to heroin. Many have claimed that marijuana caused the subsequent heroin use, even though that is now recognized as a non-causal temporal sequence. What recommendations do you have for future research as a result of this study?

Response:  There are many practices that could decrease the likelihood that erroneous findings are published. This includes registering the methods in advance and making the raw data available as a supplemental material. Many journals do not have statisticians that complete the peer review (7). Just like software is used to assist journals in detecting plagiarism, software could be developed to detect issues like were identified in these reports.

Future research could involve sending manuscripts with various levels of mathematical (addition and division) and statistical errors (e.g. running statistics that do not meet the assumptions of that test) to journals with low vs high-reputation or basic vs clinical science publications to determine how robust they are in detecting similar irregularities. Is there anything else you would like to add?

Response: PubPeer is a freely available tool that provides post-publication peer review. Everyone that regularly uses Pubmed should consider using PubPeer (8).


  1. Taha T, et al. Cannabis impacts tumor response rate to nivolumab in patients with advanced malignancies. Oncologist 2019, 24, 549-554.
  2. Bar-Sela G, et al. Cannabis consumption used by cancer patients during immunotherapy correlates with poor clinical outcome. Cancers 2020, 12, 2447.
  3. Available online: (accessed on 13 Mar. 2024).
  4. Available online: (accessed on 13 Mar. 2024).
  5. Available online: (accessed on 13 Mar. 2024).
  6. Piper BJ, et al. Immunotherapy and cannabis: A harmful drug interaction or Reefer Madness? Cancers 2024, available at:
  7. Hardwicke TE, et al. How often do leading biomedical journals use statistical experts to evaluate statistical methods? The results of a survey. PLoS One 2020, 15, e0239598.
  8. com

Last Updated on March 23, 2024 by Marie Benz MD FAAD