We agree with and appreciate several of the thoughtful comments offered by the authors of these letters. Most of these comments have been mentioned in the Discussion section of our paper. We would like to make some important points for clarification.

Recalibration was necessary for the severity model. For this reason, we used an expanded SAPS II for severity adjustment in our study. We believe that Figure 1 in our article, along with the reported value for the Hosmer–Lemeshow goodness-of-fit statistic, demonstrates the usefulness of the expanded SAPS II for predicting actual mortality in the population used in our study.

Patients on CCM did not receive any care from non–critical care physicians. The category "managed entire stay by critical care physicians" excluded any patients managed by non–critical care physicians in the first 24 hours. Because Project IMPACT records dates but not times of management by a critical care physician, and in order to eliminate potential confounding by this uncertainty, we excluded all patients managed partially by critical care physicians from analysis in the study. We agree that the ability to detect unrecognized confounding by severity is limited by the information contained in the Project IMPACT database.

As stated in our Discussion section, it is true that the Project IMPACT database does not allow for the separation of "full-time intensivist" management from management by board-certified critical care physicians who are not necessarily on site. Therefore, the "being there" factor was not evaluated in our study and neither was the impact or role of management by trainees.

Although the published literature suggests that full-time, on-site intensivists are more likely to write for do-not-resuscitate orders for critically ill patients, do-not-resuscitate orders did not account for the differences in mortality among the groups evaluated in our study.

Patients in coronary care units were excluded from analysis in our study.

The study by Diringer and colleagues (1) included patients from institutions outside of the Project IMPACT database and therefore does not have relevance to the current study.

The Project IMPACT database does not describe the "culture" of the ICUs included in the database. It is possible that the ICUs in which there is no management by critical care physicians are not representative of the typical open ICU in the United States.

We welcome all the comments included in these letters as part of a new, ongoing dialogue aimed at improving the care provided to critically ill patients.