What Is GRADE?
GRADE (Grading of Recommendations, Assessment, Development, and Evaluation) is a transparent methodology for assessing evidence quality and recommendation strength. It separates these two dimensions: high-quality evidence can support a weak recommendation, and low-quality evidence might warrant a strong recommendation if effects are large.
Starting Points and Downgrades
Each study type begins at a different evidence ceiling. Randomized controlled trials start as high quality; observational studies start as low quality. Evidence is then downgraded for: risk of bias (inadequate randomization, blinding, or follow-up), inconsistency (heterogeneity, conflicting results), indirectness (population, intervention, or outcome differs from question), imprecision (wide CIs, small sample size), and publication bias (missing negative studies). Each factor can lower evidence by one or two levels.
Upgrades and Strength of Evidence
Observational studies can be upgraded for large effects (RR >2 or <0.5), dose-response gradients, or when plausible residual confounding would increase effect estimates. Microbiome probiotics for C. difficile, for example, showed such large effects that observational data reached moderate quality. The final rating: high, moderate, low, or very low.
Recommendation Strength
Strength (strong vs. conditional) reflects whether benefits clearly outweigh harms. A strong recommendation applies to most patients; a conditional recommendation acknowledges trade-offs and uncertainty. For probiotic supplementation in diarrhoea, evidence quality (e.g., moderate) and recommendation strength (e.g., conditional, due to heterogeneity across strains) are stated separately.
Practical Interpretation for Patients
When reading guidelines, high-quality strong recommendations offer confidence for action. Moderate-quality conditional recommendations suggest discussing individual circumstances. Very-low-quality recommendations are expert opinion; stronger evidence is needed. Understanding GRADE helps clinicians and patients calibrate confidence and recognize where reasonable people disagree due to evidence gaps, not evidence against.