Decide against pooling when clinical diversity, risk of bias, or heterogeneity conflicts with the pooling criteria you specified before extraction, and record that decision in a synthesis-method memo. The memo states the criteria applied and the narrative structure chosen, so reviewers can see why a pooled estimate was withheld.
Set the criteria before you extract data
Write the conditions under which you would pool results: which populations, interventions, comparators, and outcomes count as similar enough, which effect measure you would use, and what level of statistical heterogeneity you would accept. Cochrane meta-analysis guidance covers effect measures, fixed and random effects, heterogeneity, missing data, and sensitivity analyses, so the criteria can name the specific quantity each check will produce.
Run three checks against those criteria
- Compare clinical diversity. List the differences in participants, interventions, comparators, and outcome definitions across studies. If a study falls outside the prespecified similarity conditions, record it as a candidate for exclusion from the pooled estimate and note the reason.
- Assess risk of bias for each result you plan to pool. RoB 2 judges a specific randomized-trial result across bias domains using signalling questions and written justifications, producing low risk, some concerns, or high risk. Record the judgement per result and decide in advance how a high-risk result affects its inclusion.
- Examine heterogeneity. Report the heterogeneity statistics your protocol named and compare them with the threshold you set. When the observed heterogeneity exceeds that threshold, the prespecified decision is to withhold the pooled estimate.
Write the decision memo
State the criteria applied, the evidence for each check, and the outcome: pool, pool with a stated sensitivity analysis, or move to narrative synthesis. Then describe the narrative structure you chose, such as grouping studies by population or by intervention type, and say which outcomes each group addresses. GRADE's outcome-level certainty judgements give the memo a consistent way to report how confident readers should be in each conclusion.
Choose the narrative structure to fit the conflict
Group by the dimension that caused the conflict. If populations differ, organise by population; if risk of bias drives the decision, organise by risk-of-bias judgement and describe the pattern across groups. Keep effect estimates visible in the text so readers can see the underlying results even without a pooled figure.