Choose One Eligible Outcome Result Without Cherry-Picking

Identify the choices hidden inside an outcome

A review outcome such as pain at three months can correspond to several instruments, visits and analyses within one study. Selecting whichever result is easiest to extract can introduce an unplanned preference. Selecting the most favorable result is worse. Write a hierarchy before reading the comparative estimates wherever possible.

Cochrane’s data-collection guidance emphasizes clear outcome and result information. Make the hierarchy specific enough that two reviewers can independently choose the same result from a report containing several candidates.

Separate the decision dimensions

Define the outcome construct first. Then specify acceptable instruments and any preference for validated or commonly used measures. Define the target follow-up and an eligible window, including what happens when two visits are equally close. Finally specify the analysis population, adjusted versus unadjusted estimate policy, and how multiple models will be handled.

These rules require scientific judgment. A hierarchy should not mechanically prefer a model adjusted for a post-treatment variable or an instrument measuring a different construct. Document exceptions with their reason and timing, and retain every candidate result so the selection can be audited.

Apply a hypothetical hierarchy

Suppose the target is symptom burden at 12 weeks, with eligible visits from 10 through 14 weeks inclusive. The plan prefers instrument X over instrument Y, then the visit closest to 12 weeks, then the prespecified primary model.

A study reports X at weeks 8 and 13, Y at week 12, and three models for X at week 13. Under this hierarchy, X at week 13 is selected; week 8 is outside the window. The existence of a week-12 Y result does not override the instrument preference. Choose the prespecified primary model only after checking that its estimand and covariate set remain compatible with the review question.

Make exclusions visible

Record each candidate’s instrument, time, population, model and source location. Add a selection status and the exact hierarchy rule that determined it. If the study’s reporting does not identify a primary model, apply the review’s fallback rule or seek clarification rather than guessing.

Test the hierarchy on a few diverse studies before full extraction. A rule that fails repeatedly may need a protocol amendment; explain the change and whether results were already visible. Sensitivity analyses can assess reasonable alternative selections, but they should answer a declared robustness question. The final review should reveal how multiple reported results became one synthesis entry instead of concealing that choice inside the spreadsheet.

Source verification: 6 November 2026. Examples are hypothetical.