Our article1 used a rich, clinically detailed database to offer a framework for understanding the epidemiology of the use of DNR orders. The work describes a set of patient and hospital characteristics that are associated with the use of DNR orders even after adjustment for clinical and nonclinical characteristics of the patient that previously were shown to predict death 30 and 180 days after admission.2 This is an important step forward, explaining about 20% of the variance in use of DNR orders. We agree that models that include measures of quality of life in the prediction of the use of DNR orders are likely to improve the prediction model. Moreover, our measure of sickness at admission did not predict length of life or serially reevaluate prognosis during the course of illness as might occur clinically. It is possible that an interaction between diagnosis and changes in sickness during