Simplification of the SAPS by selecting independent variables

A series of 950 consecutive admissions to a multidisciplinary ICU were divided into 2 groups (group I = 450 and group II = 500). In the first group, 351 admissions were used to retrospectively create two scores: SAPS.R derived from the 14 original SAPS variables and SAPS.E derived from 22 variables (14 of SAPS plus 8 additional biological variables). Using a multiple logistic regression analysis only 5 variables were retained for SAPS.R and 7 for SAPS.E. In the second group, 446 patients were used to assess prospectively the discriminating power of SAPS, SAPS.R, SAPS.E by comparing the areas under the receiver operating characteristic (ROC) curves. There were no significant differences between the overall accuracies of these 3 scores. Adding biological variables did not improve the accuracy but the number of variables in the original SAPS could be greatly reduced.