The Reviewer Asked: Address Confounding

Begin with the question the model is meant to answer

“Adjust for confounders” is not an instruction to add every available column. In an observational comparison, the relationship between exposure, outcome, and candidate covariates matters. Variables measured after exposure may play a different role from baseline characteristics.

Write down the intended contrast and the timing of each candidate variable. Discuss the assumed relationships with someone who understands the clinical setting. A causal diagram can make disagreements visible; it does not make the assumptions true by itself.

A fictional example

Suppose a team studies an association between treatment initiation and later hospital use. Baseline illness severity could influence both treatment choice and hospital use. A post-treatment biomarker may instead lie on a pathway through which treatment operates. Including both without explaining their roles can change the question being answered.

This example does not establish a correct adjustment set for a real treatment study. That requires substantive knowledge, measurement details, and a defined estimand.

What to check before revising

• Why was each adjustment variable selected?
• Was it measured before the exposure or intervention?
• Are there missing measurements or poorly represented categories?
• Do the comparison groups have enough overlap for the proposed analysis?
• Which relevant factors were unavailable?

Changing the model may change the estimate and its uncertainty. Report the revised methods and results consistently, including sensitivity analyses that were actually performed. Adjustment does not automatically eliminate residual confounding or justify causal claims.

Respond with reasons, not a longer variable list

Explain the selection rationale, identify the revised manuscript sections, and acknowledge the limits. If a requested variable is a mediator or otherwise inappropriate for the stated question, explain the concern and discuss an appropriate alternative.

Open the Reviewer Comment Decoder, download the response tracker, or request revision support.

Reporting reference: STROBE. Reporting completeness and causal validity are separate judgments.