- Published
- Updated
Assuming that every important reason for treatment selection was recorded is a strong claim in observational research. An unmeasured-confounding assessment should make visible how interpretation depends on additional assumptions rather than provide an automatic certificate of causal validity.
Define the question and assumptions
Identify plausible clinical reasons absent from the source. Justify scenarios for their relationships with exposure and outcome. Quantitative methods may help, but a single summary number does not identify the missing factor or establish clinical plausibility. Select the method with appropriate expertise and explain what mechanism it actually investigates.
Hypothetical study example
In a hypothetical comparison, functional status may influence treatment choice but is not recorded. The team discusses plausible distributions and relationships. It does not present a large sensitivity statistic as proof that every unmeasured cause has been excluded; it explains the assumptions examined and the limits of their scope.
Design review checks
- List unmeasured candidates using knowledge of the clinical decision process.
- Justify scenario magnitudes with available data or suitable external evidence.
- Identify bias mechanisms that the selected sensitivity method does not address.
Record the decision and its limits
Accepted interpretation explains the assumptions linking association to a causal claim. Data quality, selection and outcome measurement errors may remain separate problems. Do not examine only scenarios preserving the preferred result. Report the conditions under which conclusions change and what additional information is needed, without overstating the certainty supplied by the exercise.
Frequently asked questions
- Does sensitivity analysis measure the missing confounder?
- No. It explores potential influence under assumptions; the variable remains unobserved.
- Can one number exclude every bias?
- No. State the mechanism and scenario scope being assessed.