What a sensitivity analysis can tell you about a quitting review

Identify the decision being tested

A review may vary how uncertain eligibility, missing information, or a statistical method is handled. Read the stated purpose of the sensitivity analysis before comparing its numbers. An analysis with a new outcome or a different target population may answer another question rather than test the original one.

Cochrane describes sensitivity analyses as repeating an analysis with alternative decisions for arbitrary or unclear choices. The reading map here concerns that purpose. It does not instruct you to remove studies based on their favorable or unfavorable results, or to perform a new meta-analysis without statistical expertise.

Cochrane: Cochrane Handbook, Chapter 10: Analysing data and undertaking meta-analyses

Make a small paired reading note

Write three lines: the primary decision, the alternative decision, and what changed in the interpretation. Keep effect units and uncertainty visible on both sides. Looking only at whether a significance threshold was crossed can hide a small numerical change or a substantial change in precision.

In an invented example, a review includes a study whose eligibility information is incomplete, then repeats its analysis without that study. The second result uses largely the same evidence. If a headline calls the two analyses “two studies confirming the effect,” it overcounts independent research and loses the purpose of the comparison.

Distinguish robustness from certainty

If the conclusion changes materially under a plausible assumption, that dependence belongs in the summary. It does not automatically prove that the primary analysis is wrong. If the conclusion is similar, you have learned about that particular assumption, not established that all missing evidence, measurement problems, or conflicts of interest are harmless.

Also distinguish the exercise from a subgroup analysis. A sensitivity check may restrict a dataset to test a decision without estimating an effect for the removed group. Do not claim that the intervention works better for a demographic merely because a restricted sensitivity result looks stronger. That would require a different question and appropriate analysis.

Retain the tested scope in the claim

Your conclusion could be: “The estimate was similar under this stated alternative, but these other limitations remain.” If the alternative decision or result is not reported clearly, leave it unresolved. Avoid calling a review robust solely because it contains a heading named “Sensitivity analyses.” Read what was actually done.

For treatment-related decisions, ask a qualified professional about the full evidence. An introductory note cannot settle all analytical choices or establish your personal outcome. Practical tracking can help describe your own experience, but it cannot supply the study’s missing data or prove that an assumption was correct. The useful result is a transparent comparison of one analytical choice and its consequence.