• Chung-Liang Chao.
• Chutung General Hospital, Department of Health, The Executive Yuan, 52 Jshan Road, Judung Jen, Taiwan.
• J Formos Med Assoc. 2002 Jun 1;101(6):406-15.

Background And PurposeThis study compared the goodness-of-fit of six prognostic prediction systems.MethodsA total of 1,170 patients in the medical and surgical intensive care units of a public tertiary care hospital were included in this study. Data from the Acute Physiology and Chronic Health Evaluation (APACHE) II, APACHE III, Simplified Acute Physiology Score II (SAPS II), Condition Index Score (CIS), and Mortality Probability Models II (MPM II; at admission MPM IIadm and 24-hours later MPM II24 h) prediction systems were collected. The scores or death-risks from these systems were compared between survivors and non-survivors. Lemeshow-Hosmer (LH) statistics, calibration curve, classification matrix, and receiver operating characteristic (ROC) curve for the whole study group were determined. Mortality ratio, LH statistics, and ROC curve for individual disease categories and operative states were compared. A model with good fit was defined as having an LH chi-square value close to the degrees of freedom and yielding a corresponding p value of greater than 0.05, conformation to the ideal calibration diagonal (low deviation value), high classification matrix overall accuracy, a large area under the ROC curve (> 0.8), and a mortality ratio close to 1.ResultsLH chi-square value was 12.37 for APACHE II and 8.61 for MPM II24 h. The calibration curve deviation value was 1.5069 for APACHE II and 1.4308 for MPM II24 h. Classification matrix overall accuracy was 78.88% for APACHE II and 79.54% for MPM II24 h. The area under the ROC curve was 0.779 for APACHE II and 0.790 for MPM II24 h. The mortality ratios of trauma were 1.00 for both APACHE II and MPM IIadm. The mortality ratios for respiratory diseases and rare diseases were both 1.02 for MPM II24 h. The other systems studied did not provide good-fit data.ConclusionIn this study, MPM II or APACHE II had the best predictive model of the tested systems for critically ill patients.

13 keywords...

hide…