An important source of systematic error, which in our eyes was not sufficiently acknowledged by the authors, is confounding by indication. The physician’s decision to allocate patients to warfarin therapy is influenced by many factors, including patient characteristics, social contexts, prognostic expectations, the physician’s traditions/experience, reimbursements, and hospital policies. Treatment and nontreatment groups therefore differ in their probability distribution for their (hoped as well as feared) clinical outcomes to the point that meta-analyses comparing results from randomized trials vs nonrandomized trials can find opposite results. Such bias may be magnified if only unadjusted (crude) treatment effects are meta-analyzed. Although various statistical approaches may be used in an attempt to mitigate this bias, adjustment for administrative records are unlikely to fully reflect the prescription behavior of physicians.