Bias where treatment choice correlates with underlying patient risk
Confounding by indication is a bias seen in observational studies where the reason a patient was given a particular treatment is itself linked to their prognosis. Clinicians don't allocate treatments at random; they choose based on how severe, complex, or high-risk a patient appears, so the treatment groups being compared were never equivalent to begin with.
When appraising a study, this bias means any raw difference in outcomes between treatment groups may reflect who was selected for each treatment rather than the treatment's actual effect. It is the central reason non-randomized comparisons of surgical techniques or interventions can mislead, since the group that looks worse may simply have been sicker or more complex at baseline, not worse treated.
It shows up as an absence of randomization combined with plausible clinical reasons why one group would differ systematically from another in prognosis, such as more severe pathology, older age, or more comorbidities being steered toward a particular procedure. Reviewers look for whether authors described the treatment decision process, reported baseline characteristics of each group, and used statistical adjustment such as multivariable regression, propensity score matching, or restriction to balance known prognostic differences between groups.
Statistical adjustment can only correct for confounders that were actually measured and recorded; unmeasured or unrecorded factors that influenced treatment choice will still bias the result even after careful analysis. Readers should be wary of studies that adjust for a few baseline variables and then claim the comparison is now equivalent to a randomized trial, since adjustment reduces but does not eliminate confounding by indication.
This guide was auto-drafted and is pending editorial review.