When Bayes meets von Neumann and Morgenstern

Physicians' decisions at the ICU

Abstract

This paper presents and empirically tests a simple model addressing the behavior of physicians treating terminally ill persons. The patient related data include the survival probabilities as predicted by the physicians 24 and 48 hours after ICU admittance, the daily costs of treatment and the length of stay as well as the survival status at the time of their hospital discharge. The empirical evidence vindicates the two main hypotheses of the model: the physicians' spending at the ICU is i) for high survival probabilities negatively correlated with the predicted survival probability, as an increase of the predicted survival chance renders the treatment of the patient outside the ICU more attractive, and ii) positively correlated with the amount of new information either bad or good, a behavior consistent with Bayesian learning.