Cannabis eases chronic pain in trials, but is tied to psychosis and car crashes

Cannabis is more available than at any point in modern history, both as a recreational drug and as a licensed medicine. Claims about it run in opposite directions, from an effective treatment for pain and epilepsy to a driver of psychosis. Individual pooled analyses each cover one slice of this picture, so their conclusions are easy to cherry pick. This umbrella review set out to grade the credibility of the entire evidence base, benefits and harms together, across both randomised trials and studies that followed people in the real world.

The authors searched PubMed, PsychInfo and Embase up to 9 February 2022 for systematic reviews with meta-analyses covering cannabis, cannabinoids or cannabis based medicines. They included 101 meta-analyses, 50 of observational evidence and 51 of randomised trials. Observational findings were graded on a scale from convincing down to not significant, trial findings were graded with the GRADE system, and the quality of each review was assessed with a standard appraisal tool. Sensitivity analyses tested whether the conclusions held.

On the benefit side, with high certainty, cannabidiol reduced seizures in epilepsy and cannabis based medicines reduced chronic pain by 30% across different conditions. Moderate certainty evidence supported improvements in spasticity and pain in multiple sclerosis, quality of life in inflammatory bowel disease, and sleep in cancer. On the harm side, convincing observational evidence linked cannabis to psychosis in the general population, to car crashes in drivers, and to small for gestational age babies and low birth weight in pregnancy. Trials also showed increased adverse events affecting the nervous system, psychiatric symptoms and sleepiness, and tolerability was poor across conditions.

This is the broadest credibility assessment of cannabis evidence to date, and it cuts through the habit of quoting single studies on either side of the debate. It confirms genuine medical value in specific conditions, including epilepsy, multiple sclerosis, chronic pain, inflammatory bowel disease and palliative care, while hardening the case on psychosis, driving and pregnancy. Much of the harm evidence comes from following users rather than experiments, so causation is not fully settled. How risk varies with dose, potency and product type remains an open question.

The picture is genuinely two-sided rather than a simple good or bad verdict. For people with the specific conditions studied, cannabis based medicines can work, but rarely without side effects. For everyone else, the authors conclude the convincing evidence supports avoiding cannabis during adolescence and early adulthood, in anyone prone to mental health problems, in pregnancy, and before or while driving. That is a conclusion about risk patterns in populations, not a prediction about any one person.