Assessment of the performance of the SAPS 3, SAPS II, and APACHE II prognostic models in a surgical ICU

Introduction

The aim of this study was to assess the comparative performance of the SAPS 3 score with that of the APACHE II and SAPS II scores in surgical ICU patients.

Methods

In a cohort, observational study in a 51-bed postoperative ICU of a university hospital, we included all consecutive patients admitted to the ICU between August 2004 and December 2005. The SAPS 3 score was retrospectively calculated from prospectively collected data. The probability of ICU mortality was calculated for SAPS II, APACHE II, adjusted APACHE II (adjAPACHE II), SAPS 3, and customized SAPS 3 for West Europe (C-SAPS3 (Eu)) using standard formulas. A first-level customization was performed using logistic regression on the original scores, and the corresponding probability of ICU death was calculated for the customized scores (C-SAPS II, C-SAPS 3, and C-APACHE II).

Results

The study group constituted 1,851 patients; 1,173 males (63.4%) and 678 females (36.6%), mean age 62 years. Patients were mostly admitted after cardiac surgery (26.4%). Gastrointestinal, neurosurgery, vascular, and trauma surgeries contributed to 18.7%, 8.1%, 5.7%, and 7.5%. The overall ICU and hospital mortality rates were 6.4% and 9%. Hosmer and Lemeshow (H–L) statistics showed poor calibration for SAPS II, APACHE II, adjAPACHE II, SAPS 3, and C-SAPS 3 (Eu) (H–L C-statistics and H-statistics: P > 0.05), whereas C-SAPS II, C-APACHE II, and C-SAPS 3 showed good calibration. Discrimination was generally good for all models (area under the receiver operator curve (aROC) ranged from 0.78 (C-APACHE II score) to 0.89 (C-SAPS 3). APACHE II and C-APACHE II scores had significantly lower aROC compared with other scores. C-SAPS 3 score appears to have the best calibration curve by visual inspection.

Conclusion

In this group of surgical ICU patients the performance of the SAPS 3 score was similar to that of the APACHE II and SAPS II scores. Customization improved calibration of all prognostic models.