Case Studies in Bayesian Statistical Modelling and Analysis

5

Adapting ICU Mortality Models for Local Data: A Bayesian Approach

Petra L. Graham,¹ Kerrie L. Mengersen² and David A. Cook³

¹ Macquarie University, North Ryde, Australia

² Queensland University of Technology, Brisbane, Australia

³ Princess Alexandra Hospital, Brisbane, Australia

5.1 Introduction

Risk adjustment is a method of controlling for casemix and severity of illness in a sample of patients using a statistical model to estimate the probability of death. For example, risk adjustment may be performed within a hospital to facilitate comparable care across a patient mix, or so that the quality of care given by different hospitals or units may be compared on an equal footing. In the intensive care setting models adjust for age, severity of illness, chronic health and reason for admission. Such variables are generally objective characteristics of the patient and not characteristics attributable to the unit involved (Zaslavsky 1997; Iezzoni 2001).

5.2 Case Study: Updating a known Risk-Adjustment Model for Local Use

The Acute Physiology and Chronic Health Evaluation III (APACHE ^® III) system¹ (Knaus et al. 1991, 1993; Wagner et al. 1994; Zimmerman 1989) is one of several available risk-adjustment tools and is used by medical practitioners to predict hospital or intensive care unit (ICU) mortality and to assess the severity of illness of their patients. It is a logistic regression model developed from a database of over 17 000 patients from 40 US hospitals. The APACHE III system ...