Objective: Development and validation of a critical care outcome prediction equation (COPE) using data that are collected routinely for administrative purposes.
Design: Retrospective observational study using multivariate logistic regression modelling. Calibration and discrimination were assessed by standardised mortality ratio (SMR), area under the receiver operating characteristic plot (ROC AUC), and Hosmer-Lemeshow contingency tables.
Setting: All intensive care units in the state of Victoria, Australia.
Participants: Consecutive adult hospital episodes between 1 July 2004 and 30 June 2006.
Results: 17 880 records (1 July 2004 - 30 June 2005) were used to derive the COPE model, which incorporated five variables (age, unplanned admission, mechanical ventilation, hospital category and admission diagnosis) and was validated on the 17 848 records from the following year (1 July 2005 - 30 June 2006). The 95% confidence interval of the SMR in the validation sample was 1.00-1.01, and for the ROC AUC was 0.83-0.84. The COPE model was validated in three major hospital categories (tertiary, metropolitan, and regional) and in five individual ICUs, and compared favourably to the APACHE III model (SMR = 0.83- 0.86; ROC AUC = 0.87-0.88).
Conclusion: The COPE model is a simple, robust, riskadjusted outcome prediction tool based on five fields from data that are routinely collected for administrative purposes.