Extract arm-level counts, denominators, and timepoints into fixed fields, then flag every value that required calculation or imputation so a second extractor can verify each derived entry.
- Create one row per outcome per trial arm. Include columns for trial ID, outcome name, arm label, number of participants in the arm (denominator), number of events or mean value, timepoint, and source location (table or page). This structure keeps the raw reported data separate from any derived values.
- Record the reported values exactly as they appear in the trial report or results registry. For each value, note whether it is a count, a proportion, a mean, or another measure. If the trial reports a proportion without the underlying counts, leave the count field blank and mark the row as incomplete.
- Flag any value that you calculated or imputed. For example, if you compute a risk difference from arm-level counts, add a note in a dedicated 'derived' column stating the formula and the source numbers. If you impute a missing standard deviation using a method from the Cochrane Handbook, record the method and the assumed value. This flag tells a second extractor which numbers need checking against the original source.
- Cross-check each row against the trial's registration entry on ClinicalTrials.gov to confirm the outcome definition and timepoint match the reported results. If the registry lists a different timepoint or denominator, record both and note the discrepancy.
- Have a second extractor independently review the table, focusing on flagged derived values and any rows marked incomplete. Resolve differences by returning to the source document or registry entry.
The Cochrane Handbook provides guidance on effect measures and handling missing data, which can inform how you document imputed values. ClinicalTrials.gov data element definitions describe the structured fields for outcomes and reported results, helping you align your table with registry terminology.