In Reply:—Dr. Desbiens makes several helpful suggestions and comments regarding alternate approaches to our analysis. We would like to specifically respond to a number of his comments.

First, we imposed stringent exclusion criteria to ensure that the changes in hemoglobin/hematocrit values seen were solely because of phlebotomy. Rather than controlling for these potential confounders as Dr. Desbiens suggests, we chose to restrict our sample to a more homogenous population. Practically, we would expect that the effect of phlebotomy in patients with gastrointestinal hemorrhage would be no different for patients with cellulitis.

Second, we disagree that pre-screening candidate variables with univariate analysis leads to over-fitting. This is particularly true because we had more than 400 observations and only 7 potential predictors. Repeating our analysis including all potential predictors provides identical results to those we obtained.

Third, whether or not to include a baseline measurement in analyses of change scores is controversial. We disagree that baseline hemoglobin is perforce correlated with hemoglobin change, although Dr. Desbiens is correct that it may be. We performed an unconditional analysis as suggested and found equivalent results.

Overall, we are confident that analytic methods described in our paper produced results that are robust in terms of the final conclusion: blood drawing for diagnostic studies in internal medicine inpatients causes anemia.