Bayesian Approach to Predict Hospital Mortality of Intensive Care Readmissions during the Same Hospitalisation

Abstract

No specific prognostic model has been developed for patients readmitted to the intensive care unit (ICU) during the same hospitalisation. This study assesses the performance of the Acute Physiology and Chronic Health Evaluation (APACHE) II predicted mortality measured at the time of ICU readmission and whether incorporating information prior to the readmission will improve its performance to predict hospital mortality of patients readmitted to ICU during the same hospitalisation. A total of 602 readmissions during the same hospitalisation between 1987 and 2002 were identified. The first admission APACHE II predicted mortality was significantly associated with the hospital mortality only in the subgroup of patients readmitted within seven days of ICU discharge (odds ratio 1.16, 95% confidence interval 1.01 to 1.34; P = 0.035). In the subgroups of patients readmitted within seven days of discharge, the readmission APACHE II predicted mortality was also significantly better than the first admission APACHE II predicted mortality in discriminating between survivors and non-survivors (area under the receiver operating characteristic curve: 0.785 vs. 0.676, z statistic = 2.93; P = 0.003). Incorporating the first admission APACHE II predicted mortality to the readmission APACHE II predicted mortality, either by multilevel likelihood ratios or logistic regression, did not significantly improve its discrimination (area under the receiver operating characteristic curve: 0.792 vs. 0.785, z statistic = 0.52; P = 0.603). Our results suggested that information on prior ICU admission during the same hospitalisation is not as important as the severity of illness measured at the time of readmission in determining the mortality of intensive care readmissions during the same hospitalisation.