What Did Medicine Know About Cannabis by 2017?

A plain-English summary of MED01 Part 2 (v1.0 — SIGNED OFF by Danny 2026-10-12; P2 LOCK 2026-10-14; corrections #5–#16 applied 2026-10-13/14 after the cross-audit)

Based on: the technical synthesis for Part 2 (v1.0, signed off 2026-10-12) and the 45-claim ledger (C041–C085). The words are simpler, the uncertainty is the same. Claim numbers in brackets lead to the technical version and its sources. Covers: research published 2000–2017, as the evidence stood then, found through our search cutoff of 20 September 2026. Anything after 2017 is marked later. Not medical advice. This describes studies from 2000–2017. It does not say what cannabis does for you.

How we did this

Two AI researchers got the same frozen instructions and worked separately. A checker (the “verifier”) then:

• looked up every reference number to check it pointed to the right paper;
• checked each figure we use against the study’s summary (abstract). For most studies we could not read the full paper, and until a later audit we did not check whether journals had published corrections;
• re-ran the standard publication-count search;
• opened selected official records: regulator decisions, patents, court rulings and the 2017 US National Academies report.

Some details could not be settled and are marked as still being checked.

One researcher’s references were often wrong. Nearly half of its PubMed reference numbers pointed to unrelated papers, and it admitted it had estimated publication counts it said it had looked up. So where the two researchers agreed, we counted that as agreement on topics, never as proof of a number. Our own extra searches turned up important studies that neither researcher found. What survived is a 45-claim ledger. A few parts are still being checked and are marked that way.

How much research was there?

A lot, and growing. Our standard PubMed search finds about 30,700 cannabis papers from 2000 to 2017, with the fastest output in the last three years [C041]. Papers on medical cannabis grew about nine-fold over the period [C042]. These counts describe a database, not the whole world’s research.

The lab science — in cells and animals, not patients

Scientists learned that the body’s own cannabis-like chemicals carry signals across nerve connections. Mice lacking an enzyme that normally breaks one of these chemicals down had more of it, which helped make the enzyme a drug target [C044]. CBD was found to change how the main cannabinoid receptor responds, in cells in a dish [C045].

The idea that some illnesses might involve too little of the body’s own cannabinoids was proposed in 2004. Reviews found some indirect support, but our searches did not identify an independent test confirming it by 2017 [C046]. The “entourage effect”, the idea that plant compounds work better together, was also put forward as an idea [C047]. In the one patient trial comparing them, plant extract did no better than THC alone or placebo [C047].

What was tested in people

There were many controlled trials, but most were small and short. A 2015 review counted 79 trials with 6,462 people, and judged only four of them at low risk of bias [C066]. Results depended heavily on the product, the condition and what was measured.

• MS muscle stiffness. This was one of the most-tested areas. Patients often reported less stiffness, but doctors’ measurements did not consistently agree [C048]. The biggest early trial missed its main outcome, which was the doctors’ measurement [C048].
• Pain. Small, short trials of smoked or vaporised cannabis in nerve pain reported relief, perhaps for about one person in five or six [C050]. Four trials of the THC+CBD mouth spray were mixed: two positive, one negative, and one that missed its main outcome [C051]. The largest THC-pill trial in MS nerve pain found no difference from placebo [C049]. Nabilone, a man-made THC-like drug, did worse than a weak opioid painkiller [C052].
• Chemotherapy nausea. The evidence was mostly from trials done before 2000. In the one new comparison, dronabinol (a THC pill) and a modern anti-nausea drug had similar response rates for delayed nausea. The trial was small and not designed to show they were equal [C053].
• Appetite and weight loss in cancer. Two large trials found no advantage for dronabinol or cannabis extract [C054].
• Childhood epilepsy. CBD went from parent reports and open programmes to one randomized trial in 2017. In that trial of 120 children and young adults with Dravet syndrome, paid for by the drug’s maker, seizures fell more on CBD than on placebo, and side effects were more common on CBD [C055]. CBD also raised levels of another epilepsy drug in the blood [C055]. Later: the US approved this purified CBD medicine in 2018.
• Other conditions. Schizophrenia, Tourette syndrome, PTSD nightmares, dementia, Crohn’s disease and glaucoma each had only one or a few small studies [C056–C058]. In dementia, THC did not help on the main outcome. In Crohn’s disease, the trial missed its main outcome. In glaucoma, eye pressure fell for about two hours after one dose [C056–C058].
• Cancer. Our searches found no controlled human trial of whether cannabinoids treat cancer. The only treatment study found was an uncontrolled safety study in nine patients with recurrent brain cancer (glioblastoma), with THC delivered directly into the tumour. It was not designed to show whether the treatment worked [C059].
• Slowing disease. The one large trial, of THC in 498 people with progressive MS over three years, found no effect [C060].

Medicines and approvals

“Cannabis” in these studies meant many different things: a THC+CBD mouth spray, THC pills, nabilone, purified CBD, CBD oils and research-grade smoked cannabis. A result for one does not apply to the others [C061]. Approvals were for specific products and uses:

• Canada gave the spray conditional approval for nerve pain in MS (2005) and for pain in advanced cancer (2007) [C062]. Later: those approvals were removed in 2019 when follow-up studies did not confirm benefit [C062].
• The UK approved the spray for MS stiffness in 2010 [C062].
• The US approved a liquid THC medicine in 2016 [C063].

Drugs designed to act on the body’s own cannabis system went badly. A weight-loss drug was withdrawn over psychiatric risks, and in a 2016 early safety trial of an experimental drug, three of the four participants described in a published report developed a severe neurological illness [C064].

Were results always published?

Not always, as far as we could find. In one major trial registry, 132 completed cannabinoid trials had no posted results and no linked paper. Narrowing those step by step left 28 trials that gave a cannabinoid to patients as a treatment and could not be traced by trial number. We matched 17 of them to a paper; one, a 72-child trial that finished in 2017, was only published in 2020. For the other 11, no publication was identified in our searches [C067]. The 11 are what remained after this screening, not a share of all completed trials (the full sequence is in claim C067). “Not found” does not prove “never published”, and two other registries were not searched [C067].

The harms research was real too

• Psychosis. Heavier use went with later psychosis, and more use went with more risk. A 2017 national review called this link substantial. These studies show a link, not proof that cannabis causes psychosis [C070].
• Withdrawal. In a small study of users and former users, stopping heavy use was followed by withdrawal symptoms that began within 1–3 days, peaked around days 2–6 and mostly lasted 4–14 days [C072].
• Driving. Across nine observational studies, recent use went with roughly double the odds of a crash. In an 18-person driving-simulator study, blood THC levels of about 8 and 13 micrograms per litre were linked to lane-weaving similar to legal-limit alcohol. Neither study set a safe level for driving [C073].
• Lungs. The 2017 review found substantial evidence that long-term smoking goes with worse breathing symptoms, and moderate evidence of no link to lung cancer [C074].
• Pregnancy. Use in pregnancy went with lower birth weight [C075].
• Children. The 2017 review found moderate evidence of more child poisonings where cannabis was legal [C076].
• In treatment. A 2008 review found non-serious side effects were more common with medical cannabinoids in its 23 randomized trials. It found no difference in serious ones in those trials, but that evidence was too limited to rule out long-term harms. A 2015 review of a different, larger set of trials reported higher odds of serious side effects [C077].

What the official reviews concluded

These are their judgements at the time, not ours:

• A 2015 review in JAMA rated the evidence moderate for chronic pain and MS stiffness, and low for several other uses [C078].
• The 2017 US National Academies report judged the evidence conclusive or substantial for cannabis in adult chronic pain, and for oral cannabinoids in chemotherapy nausea and vomiting and in patient-reported MS stiffness. These were its ratings of the evidence at the time, not of every cannabis product. It found too little evidence for cancer, epilepsy and several other conditions [C079]. Its epilepsy rating came before the 2017 trial.
• Most expert (Cochrane) reviews of the period found too little evidence [C081].

The opioid question

A 2014 study found US states with medical-cannabis laws had about 25% fewer opioid-painkiller overdose deaths. That is a link at state level, not proof the laws caused it [C083]. No controlled trial tested cannabis as an opioid substitute [C083]. Later (2019): when researchers extended the analysis through 2017, the link reversed, and they judged the earlier link likely spurious [C083].

Eight things people often say, checked against 2000–2017 evidence

So: what was known by 2017?

From several controlled trials:

• patient-reported relief of MS stiffness [C048];
• short-term relief of nerve pain from smoked or vaporised cannabis in small trials [C050];
• relief of chemotherapy nausea, mostly from older trials [C053];
• no advantage for cannabinoids in cancer weight loss [C054].

From one controlled trial: fewer seizures with purified CBD in Dravet syndrome, with more side effects [C055].

Harms: a substantial link with psychosis, a real withdrawal syndrome, higher crash risk, and more non-serious side effects in treatment [C070, C072, C073, C077].

Limited or preliminary evidence: small trials and experiments looked at schizophrenia, Tourette syndrome, PTSD nightmares, fibromyalgia and glaucoma. In dementia, the trial’s main outcome was negative. In Crohn’s disease, the trial missed its main outcome. In cancer, there was a nine-patient safety study and case reports, with no treatment effect shown. The body-chemistry findings were lab results or unconfirmed ideas [C044–C047, C052, C056–C059].

Not established by the period evidence: cannabinoids treating or curing cancer in people; medical-cannabis laws causing fewer opioid deaths [C059, C083]. Tested, with a negative result: THC did not slow progressive MS in the one large three-year trial [C060]. Contradicted as worded: the claim that CBD has no meaningful risks, because studies reported side effects and drug interactions [C085].

What we may have missed

This was an evidence map, not an exhaustive systematic review. We searched medical databases, one major trial registry and selected official records, and checked every figure we use against the sources we could access. Our extra search for missed studies was targeted, not exhaustive. We did not search two other international trial registries or non-English journals well. For most studies we could read only the summary, not the full paper. A few details are still being checked. Anything we call missing means “not found by us”, not “doesn’t exist” [C084].