Editorial Standards

How We Grade Evidence

Not all studies are created equal. Understanding how evidence is graded — and the specific challenges of cannabis research — is essential for interpreting what the science actually shows.

Our Evidence Labels

Every finding on this site carries one of four evidence labels. Here's exactly what each means.

Well-Studied

Multiple high-quality RCTs or a Cochrane-level systematic review with consistent findings. Effect size and direction are reliably established. Sufficient evidence to inform clinical practice.

Examples in cannabis research:

· CBD in Dravet/LGS epilepsy (FDA-approved)
· Dronabinol for CINV (FDA-approved)
· Cannabinoids for neuropathic pain (Cochrane review)
· THC acute cardiovascular effects

Emerging

At least one well-designed RCT or multiple consistent observational studies. Direction of effect is reasonably clear but effect size, optimal dosing, or long-term safety are not fully established.

Examples in cannabis research:

· CBD for anxiety disorders
· Cannabis for PTSD symptoms
· CBD for cannabis use disorder
· CBD blood pressure reduction

Limited

Primarily preclinical data, case reports, or small pilot studies. Biological plausibility exists but human evidence is insufficient to draw clinical conclusions. Hypothesis-generating only.

Examples in cannabis research:

· CBD for intestinal permeability
· Cannabinoids for autoimmune conditions
· CB2 agonists for cardiac protection
· Cannabis and gut microbiome

Contested

Multiple studies exist but findings are inconsistent or contradictory. Active scientific debate about effect direction, magnitude, or mechanism. Requires careful interpretation.

Examples in cannabis research:

· Cannabis anti-tumor effects
· Long-term cognitive effects of moderate use
· Cannabis and opioid use disorder outcomes

The Evidence Hierarchy

Different study designs answer different questions with different levels of confidence. Here's how they rank, with cannabis-specific examples.

Systematic Reviews & Meta-Analyses

Strongest

Systematically identify, appraise, and synthesize all available evidence on a specific question. Meta-analyses pool data from multiple studies to produce a single quantitative estimate of effect size.

Cannabis example: Cochrane Review of cannabinoids for chronic neuropathic pain (2023) — pooled 36 RCTs, NNT=11 for ≥30% pain reduction.

Strengths

+ Reduces bias from individual study limitations
+ Largest effective sample size
+ Identifies consistency across studies
+ GRADE framework assesses overall evidence quality

Limitations

− Only as good as the included studies
− Publication bias can skew results
− Heterogeneity between studies can limit pooling
− May lag behind current evidence by years

Randomized Controlled Trials (RCTs)

Strong

Participants are randomly assigned to treatment or control groups, minimizing selection bias. Double-blinding (neither participant nor researcher knows group assignment) further reduces bias.

Cannabis example: Devinsky et al. NEJM 2017 — Phase 3 RCT of Epidiolex in Dravet syndrome, 120 patients, 14-week follow-up.

Strengths

+ Random allocation controls for known and unknown confounders
+ Allows causal inference
+ Pre-registration reduces outcome reporting bias
+ Gold standard for efficacy evidence

Limitations

− Expensive and time-consuming
− Blinding is difficult with psychoactive substances
− Short follow-up periods common in cannabis trials
− Exclusion criteria limit generalizability

Cohort Studies

Moderate

Follow a group of people over time, comparing outcomes between those exposed and unexposed to a factor. Can be prospective (forward-looking) or retrospective (using existing records).

Cannabis example: Di Forti et al. Lancet Psychiatry 2019 — prospective cohort of 1,087 first-episode psychosis patients across 11 European sites.

Strengths

+ Can study rare outcomes and long-term effects
+ Establishes temporal sequence (exposure before outcome)
+ Can examine multiple outcomes simultaneously
+ More feasible than RCTs for long-term questions

Limitations

− Cannot fully control for confounding
− Loss to follow-up introduces bias
− Exposure measurement may be imprecise
− Cannot prove causation

Case-Control Studies

Moderate

Compare people with a condition (cases) to similar people without it (controls), looking backward to identify differences in prior exposure. Efficient for studying rare conditions.

Cannabis example: Studies examining cannabis use and myocardial infarction risk in young adults using hospital admission records.

Strengths

+ Efficient for rare diseases
+ Relatively quick and inexpensive
+ Can examine multiple exposures
+ Good for hypothesis generation

Limitations

− Recall bias (cases remember exposures differently)
− Selection of appropriate controls is challenging
− Cannot calculate incidence rates
− Susceptible to confounding

Cross-Sectional Studies & Surveys

Limited

Measure exposure and outcome at a single point in time. Useful for prevalence estimates and hypothesis generation but cannot establish temporal relationships.

Cannabis example: Patient surveys on cannabis use in IBD — consistently show symptom improvement but cannot determine if cannabis caused improvement.

Strengths

+ Quick and inexpensive
+ Good for prevalence estimates
+ Can identify associations for further study
+ Large sample sizes feasible

Limitations

− Cannot establish causation or temporal sequence
− Prevalence-incidence bias
− Self-report bias common
− Snapshot in time may not reflect typical patterns

Case Reports & Case Series

Weakest

Detailed descriptions of individual patients or small groups. Valuable for identifying rare adverse events and generating hypotheses, but cannot establish causation.

Cannabis example: Case reports of warfarin-CBD interaction with INR elevation — prompted formal pharmacokinetic studies confirming CYP2C9 inhibition.

Strengths

+ Identifies rare or unexpected effects
+ Generates hypotheses for formal study
+ Valuable for adverse event surveillance
+ Can document novel presentations

Limitations

− No control group
− Cannot establish causation
− Selection bias (unusual cases reported)
− Not generalizable

Preclinical Studies (Animal & In Vitro)

Mechanistic only

Laboratory studies using cell cultures or animal models. Essential for understanding mechanisms and safety before human trials, but results frequently do not translate to humans.

Cannabis example: CB2 agonist cardioprotection in murine ischemia-reperfusion models — compelling mechanistic data, no human RCT evidence yet.

Strengths

+ Controlled conditions allow mechanistic insight
+ Can study effects impossible to test in humans
+ Essential for drug development pipeline
+ Identifies safety signals before human exposure

Limitations

− Poor translation to humans (>90% of preclinical findings fail in human trials)
− Animal models imperfectly replicate human disease
− Cannot establish clinical efficacy
− Publication bias toward positive results

Biases Specific to Cannabis Research

Cannabis research faces unique methodological challenges beyond the standard limitations of clinical research. Understanding these is critical for interpreting results.

Blinding Bias

Cannabis's psychoactive effects make true double-blinding nearly impossible in THC trials. Participants often correctly guess their group assignment, inflating placebo response and treatment effect estimates.

Publication Bias

Positive studies are more likely to be published than null results. Cannabis research may be particularly affected due to historical regulatory barriers that limited negative trial publication.

Confounding by Indication

People who use cannabis medicinally often have more severe conditions than non-users, making observational comparisons misleading. Symptom improvement may reflect regression to the mean.

Dose Heterogeneity

Cannabis trials use widely varying doses, formulations, and routes of administration, making cross-study comparisons difficult and meta-analyses potentially misleading.

Short Follow-Up

Most cannabis RCTs have follow-up periods of 4–12 weeks, insufficient to assess long-term efficacy, tolerance development, or safety outcomes that may take months or years to manifest.

Self-Report Bias

Many cannabis studies rely on self-reported use, which is subject to recall bias, social desirability bias, and imprecise dose estimation — particularly for illicit or unregulated products.

Our Editorial Process

Source Selection

We only cite peer-reviewed studies published in indexed journals. Preprints, press releases, and industry-funded studies without independent replication are flagged or excluded.

Evidence Grading

Each finding is graded by our scientific team using a modified GRADE framework, considering study design, sample size, effect size, consistency across studies, and directness of evidence.

Expert Review

Topic pages and clinical resources are reviewed by at least one subject-matter expert (physician, pharmacist, or PhD researcher) before publication and updated when significant new evidence emerges.

Browse Research Library Our Editorial Team Ask a Researcher