In Reply: Falsification hypotheses can help adjudicate whether observational associations are robust or whether they reflect selection bias among patients who receive an intervention. Dr Groenwold thoughtfully expands on a prerequisite to proper falsification analyses: the falsification hypothesis must test a putative mechanism of bias. Consider, for example, the question posed by Groenwold of whether beta-blocker therapy reduces perioperative cardiac risk. Even after adjusting for observed patient demographics and claims-based clinical characteristics, an estimated association between beta-blocker therapy and cardiac risk may still be confounded by unobserved patient and physician characteristics, including unmeasured socioeco-nomic characteristics.
展开▼
