Insights / Topic · GRADE & certainty
Applying GRADE to a review: the five reasons to rate down, the reasons to rate up, summary of findings tables, and writing conclusions that match certainty.
A pooled estimate answers how large the effect appears to be. GRADE answers how much confidence that estimate deserves. Many journals now expect a certainty rating for each main outcome, and a review that reports precise-looking effects without one invites the reviewer to supply their own, usually less generous, assessment.
Certainty is rated per outcome on four levels: high, moderate, low and very low. Evidence from randomised trials starts high; evidence from observational studies traditionally starts low, or high when risk of bias is assessed with ROBINS-I. It is then rated down for five reasons: risk of bias, inconsistency, indirectness, imprecision and publication bias. Observational evidence can be rated up for a large effect, a dose-response gradient, or when plausible residual confounding would only reduce the observed effect.
Each judgement needs a stated reason. Imprecision is judged against a threshold that matters to decisions, not only against the null; inconsistency draws on the heterogeneity analysis; indirectness asks whether the populations, interventions and outcomes match the question. The results are presented in a summary of findings table, usually produced in GRADEpro GDT, with footnotes explaining every downgrade. Network meta-analyses apply GRADE to each comparison.
Common reviewer objections are worth anticipating. Downgrading with no stated reason, rating down twice for the same problem, judging imprecision only by whether the confidence interval crosses the null, or presenting certainty for some outcomes and not others. Each of these weakens trust in the whole assessment, which is why a short, consistent set of footnotes matters more than a long narrative justification placed elsewhere in the paper.
The notes in this topic work through each domain with examples, show how to set imprecision thresholds, explain how to build a summary of findings table that a reviewer can check, and cover how to word conclusions so that a low-certainty finding is not reported as if it were settled.
Explainer
Imprecision is the GRADE domain most often applied by intuition. The two checks that make it defensible, and why RevMan now asks you to write the reason down.
