Medical authorities have asserted the importance of observing a patient's clinical course over time. Distinguished committees have suggested that changes over time in physicians' prognostic estimates should influence decisions to transfer patients out of intensive care units (ICUs). This study evaluated how the opportunity to observe patients over time affected physicians' prognostic estimates for a cohort of 269 critically ill patients sequentially admitted to a medical-surgical ICU in a teaching hospital. As soon as possible after admission and again 48 hours later, the authors obtained a quantitative estimate of the probability of survival through hospital discharge from each patient's house officer and primary attending physician, and the critical care attending physician on duty. They independently determined each patient's survival. From this population they analyzed 181 pairs of judgments made by the same house officers, 211 pairs by the same primary attendings, and 172 pairs by the same critical care attendings. The physicians' 48-hour estimates were little changed from their previous estimates for the same patients. The correlation coefficient for the house officers' paired estimates was 0.84 (p less than 0.0001); for the critical care attendings' estimates 0.84 (p less than 0.0001), and for the primary attendings' estimates, 0.90 (p less than 0.0001). Forty-eight hours did not substantially reduce the disagreements present between estimates made by different physicians for the same patient. No group of physicians substantially improved the reliability or the discriminating power of its later estimates. The physicians in the study could not take advantage of sequential clinical information over time. These results point out the need to teach physicians how to better integrate and process sequential clinical data.