.jpg){kind=link}
- Published
- Updated
Clinical treatment selection in routine care is not necessarily random. Disease severity or anticipated risk can influence both treatment choice and outcomes. Consequently, an unadjusted difference between groups is not automatically the causal effect of an intervention.
Define the question and assumptions
Map clinical reasons for treatment selection starting from the research question. Assess whether recorded variables capture the decision moment and necessary detail. A diagnosis code does not establish that severity or the clinician rationale was fully measured. Justify the analytical approach alongside source limitations rather than treating a familiar model name as evidence of causality.
Hypothetical study example
In a hypothetical observational comparison, more severely affected people receive one intervention more often and subsequently have poorer outcomes. The team does not immediately conclude that the intervention causes harm. It examines selection reasons, baseline risk and how well available data can account for those differences.
Design review checks
- List factors that could influence both treatment choice and the outcome.
- Check pretreatment timing and sufficient detail in potential confounder measurements.
- Discuss possible unmeasured confounding explicitly when interpreting the estimated association.
Record the decision and its limits
An accepted design explains comparator selection and the limits of its adjustment strategy. Do not conceal unavailable variables behind a statement that the analysis was adjusted. Present crude and appropriately adjusted summaries with context and report remaining confounding concerns. Research associations should not automatically become individual treatment-selection recommendations without further clinical and scientific assessment.
Frequently asked questions
- Does a multivariable model remove all confounding?
- No. Measurement quality, unavailable variables and assumptions remain important.
- Is balance in age sufficient?
- No. Other clinical reasons may influence treatment selection.