Research

Mortality prediction using SAPS II: an update for French intensive care units

Jean Roger Le Gall¹ , Anke Neumann² , François Hemery³ , Jean Pierre Bleriot⁴ , Jean Pierre Fulgencio⁵ , Bernard Garrigues⁶ , Christian Gouzes⁷ , Eric Lepage⁸ , Pierre Moine⁹ and Daniel Villers¹⁰

¹Professor, head of the unit of Medical intensive, Hôpital Saint Louis, Paris, France

²Statistician, Délégation à l'Information Médicale et Epidémiologie, AP-HP, Paris, France

³Statistician, center of Biostatistique Médicale, Hôpital Henri Mondor, Créteil, France

⁴Delegate to the Ministère de la Santé, Paris, France

⁵Department of Anesthésie Réanimation, Hôpital Tenon, Paris, France

⁶Professor, head of the unit of multidisciplinary internsive care, Centre hospitalier du Pays d'Aix, Aix en Provence, France

⁷Epidemiologist, Information Médicale, Hôpital de Nimes, Nimes, France

⁸Professor, Head of the Délégation à l'Information Médicale et Epidémiologie, AP-HP, Paris, and of the center of Biostatistique Médicale, Hôpital Henri Mondor, Créteil, France

⁹Department of Anesthesiology, University of Colorado Health Science Center, Denver, Colorado, USA

¹⁰Professor, Head of the unit of Medical intensive care, Hôpital de l'Hotel Dieu, Nantes, France

author email corresponding author email

Critical Care 2005, 9:R645-R652doi:10.1186/cc3821

Published:	6 October 2005

See related commentary http://ccforum.com/content/9/6/636

Abstract

Introduction

The standardized mortality ratio (SMR) is commonly used for benchmarking intensive care units (ICUs). Available mortality prediction models are outdated and must be adapted to current populations of interest. The objective of this study was to improve the Simplified Acute Physiology Score (SAPS) II for mortality prediction in ICUs, thereby improving SMR estimates.

Method

A retrospective data base study was conducted in patients hospitalized in 106 French ICUs between 1 January 1998 and 31 December 1999. A total of 77,490 evaluable admissions were split into a training set and a validation set. Calibration and discrimination were determined for the original SAPS II, a customized SAPS II and an expanded SAPS II developed in the training set by adding six admission variables: age, sex, length of pre-ICU hospital stay, patient location before ICU, clinical category and whether drug overdose was present. The training set was used for internal validation and the validation set for external validation.

Results

With the original SAPS II calibration was poor, with marked underestimation of observed mortality, whereas discrimination was good (area under the receiver operating characteristic curve 0.858). Customization improved calibration but had poor uniformity of fit; discrimination was unchanged. The expanded SAPS II exhibited good calibration, good uniformity of fit and better discrimination (area under the receiver operating characteristic curve 0.879). The SMR in the validation set was 1.007 (confidence interval 0.985–1.028). Some ICUs had better and others worse performance with the expanded SAPS II than with the customized SAPS II.

Conclusion

The original SAPS II model did not perform sufficiently well to be useful for benchmarking in France. Customization improved the statistical qualities of the model but gave poor uniformity of fit. Adding simple variables to create an expanded SAPS II model led to better calibration, discrimination and uniformity of fit, producing a tool suitable for benchmarking.