Temporal versus probabilistic approach to survival estimation: Physician’s accuracy over time.

Abstract

e19584 Background: Physicians are inaccurate in estimating survival. Both Temporal and Probabilistic approaches have been used for prediction. Serial prognostication might improve accuracy. In this study we compare these two approaches and assess whether serial prognostication improves accuracy. Methods: Physicians prognosticated survival daily for cancer patients admitted to palliative care units in two hospitals until death/discharge, using two prognostic tools: Temporal (What is the approximate survival of this patient in days?) and Probabilistic (What is the probability that this patient will be alive in 24 hrs?, 0% to 100%) (Hui et al. Oncologist 2011). Temporal prognosis was accurate if it fell within ± 33% of the actual survival. Probabilistic prognosis was accurate if the clinician selected a survival probability ≥70% and the patient survived in ≤24 hours or the clinician endorsed a survival probability ≤30% and the patient died. We compared physicians’ accuracy with each method at specific time points using Mc Nemar’s test. We assessed accuracy for each method over time comparing the accuracy at each time point with day -14 using a test of proportions. Results: Baseline characteristics of 306 patients included mean age 58 (18-88), female 69%, Caucasian 76%. Physicians were significantly more accurate in prognosticating survival with the Probabilistic v/s Temporal approach (p<.001 at each time point). The accuracy of Temporal approach was stable over time (Table). The accuracy of Probabilistic approach decreased close to death (Table). Conclusions: Physicians were more accurate at estimating probability of survival at 24 hrs than at estimating days to death. The accuracy of the probabilistic approach decreases as patient approaches death. Serial prognostication does not improve accuracy of survival estimation. [Table: see text]