In Reply: The purpose of our Commentary was to illustrate the use of restriction to minimize bias in observational studies. Drs Turakhia and Heidenreich chose to illustrate how the use of restriction might introduce bias. Their example focuses on a cohort study of low-risk patients with atrial fibrillation treated either with warfarin or aspirin.
Why laboratory testing would be used as the method of restriction remains unclear in their example. Presumably, they want to exclude or censor patients who discontinue the use of warfarin. If so, then the optimal method of assuring exposure status is not laboratory testing; it would be some combination of warfarin prescription refills available from a computerized pharmacy database plus medical record reviews or patient interviews that capture information about changes in exposure that may have occurred between refills. The details provided by Turakhia and Heidenreich fail to address how they would assess the use of aspirin, an over-the-counter drug, for which high-quality information may not be available in any form in the medical record or the pharmacy data.
Psaty BM, Siscovick DS. Using Restriction to Minimize Bias in Observational Studies—Reply. JAMA. 2010;304(21):2359-2360. doi:10.1001/jama.2010.1728