How do I extract outcome data when a trial reports only medians and interquartile ranges?

Record the reported median and interquartile range exactly as published, then flag the outcome as needing conversion or an assumption before it enters a pooled analysis. The extraction note should state which transformation you applied and the assumption behind it, so a second reviewer can check the step.

Extraction sequence

1. Copy the median, the two quartiles, the sample size per arm, and the outcome scale into the extraction form without altering the values. Expected result: the source table and your form match line by line.
2. Mark the outcome as reported in median and interquartile range and set its pooling status to pending. Expected result: the outcome is visibly separated from outcomes already in a poolable format.
3. Decide whether the review needs a mean and standard deviation for this outcome. If the analysis plan calls for a mean difference, a conversion or a stated distributional assumption is required; if the plan can handle the reported format, record that decision instead. Expected result: a documented choice with its rationale.
4. Write the transformation note: the method used, the assumption it rests on, and the resulting values. Expected result: a second reviewer can reproduce the numbers from the note alone.
5. Have the second reviewer check the note against the source table. Expected result: agreement on the values and on whether the assumption is acceptable for this outcome.

What the note should contain

State the reported statistics, the target statistic, the conversion or assumption applied, and any condition that would change the choice. Cochrane Handbook Chapter 10 covers effect measures, missing data, and sensitivity analyses, and is the relevant guidance to consult when deciding how a converted outcome enters the analysis. Chapter 15 covers interpretation of results, including confidence intervals and certainty, which matters when a converted estimate carries extra uncertainty.

Boundary

A median and interquartile range describe the middle and spread of a skewed distribution. Treat the converted value as an estimate tied to the stated assumption, and record that assumption so the review team can test how much the pooled result depends on it through a sensitivity analysis.