I thank Drs. Vandenbroucke and Vandenbroucke-Grauls for their provocative letter. It underscores why I wrote the article, calling it "a cautionary tale." Faced with varying methodologic options, perhaps all legitimate within their own contexts, it becomes critically important to open black boxes and look inside-to understand what was done and its implications. Otherwise, misconceptions can arise.

Context is critical. Drs. Vandenbroucke and Vandenbroucke-Grauls adopt a classic epidemiologic perspective, in which exposure per unit of time is the fundamental quantity that informs inferences. In that context, their assertions are reasonable. My comments were framed within the current enthusiasm for publishing "report cards" on hospitals. In this setting, "mortality rates" purportedly shed insight into hospital quality-the belief of Nightingale and Farr. The exposure of interest is an entire "care package" (for example, coronary artery bypass graft surgery) provided during a hospitalization. The public is more interested in outcomes at the end of that care than in deaths per unit of time in the hospital.

In 1864, Farr's critics furnished numerous examples of how his calculation was misleading, given the common English meaning of "death rate." I offer one example from my own hospital. Last year, there were 366 myocardial infarction admissions; 29 ended in death. On an average day, roughly eight patients who had had myocardial infarction were hospitalized. By using Farr's approach, our mortality rate is 362% [29/(8 x 100)], or 3.62 deaths per patient-year. Not only is a 362% death rate confusing, but it also does not address the public's real concern-what is the outcome at the end of care for a heart attack? By using the approach of Farr's critics, our mortality rate is 7.9% [29/(366 x 100)]. Suppose a competing teaching hospital had an identical 366 admissions for infarction and 29 deaths but kept its patients twice as long, with an average daily census of 16 patients. With Farr's method, the mortality rate of this hospital is 181% [29/(16 x 100)], or 1.81 deaths per patient year. Is that hospital's heart attack care twice as good as ours?

As I indicated in my paper, the approach of Farr's critics certainly fails to hold constant the window of observation. A current solution is examining mortality 30 days after admission rather than in-hospital deaths, data permitting. In the heated letter exchange between Farr and his critics, however, Bristowe made a key point: "If Dr. Farr had made his calculations about Hospitals in a tentative spirit, with the object of ascertaining whether they were likely to lead to any useful results, he would have acted in a way to which no exception could have been taken" [1]. Even today, regardless of how hospital death rates are calculated, their true meaning often remains elusive.