Probability Model of Hospital Death for Severe Trauma Patients Based on the Simplified Acute Physiology Score I

Background: We evaluated whether or not the Simplified Acute Physiology Score (SAPS) I is a suitable audit system for trauma patients admitted to general intensive care units (ICUs), A probability model for SAPS I was retrospectively assessed on trauma patients admitted to general ICUs from 1990 to 1992. Because it was determined that SAPS did not fit the data well, we developed a customized probability model of SAPS I for trauma patients and validated it prospectively on an independent data set (patients admitted to general ICU in 1993-1994). Measures of calibration (goodness of fit) and discrimination (receiver operating characteristic curve) were adopted to assess the performance of the model. Methods: A multicenter study was performed involving general ICUs in northern Italy that participated in a project of standardized data collection. This project is known in Italy under the acronym of ARCHIDIA (Diagnostic Archive), All patients consecutively admitted to the ICU were registered in the data base from January 1, 1990, to December 31, 1994 (16,767 patients), For the SAPS analysis, we considered only patients who fulfilled the following criteria: older than 17 years, length of stay longer than 24 hours, and SAPS I correctly computed, Inclusion criteria resulted in 12,156 patients analyzed; there were 1,936 patients with trauma. Results: Trauma patients represented more than 15% of all patients admitted to ICUs during the period considered, Compared with the general population, they were younger (41.2 vs. 56.7 Sears), stayed in ICUs for longer periods (12.4 vs. 9.5 days), and had a lower hospital mortality rate (17.8 vs. 28.7%). Because the original probability model did not perform well among patients with trauma, a new model was developed and prospectively validated only for trauma patients (customization). Measures of calibration and discrimination showed a good performance both in the developmental (goodness of fit: chi(2) = 8.47; p = 0.38; area under the curve = 0.94 +/- 0.01) and in the validation (goodness of fit: chi(2) = 12.75; p = 0.12; area under the curve = 0.85 +/- 0.01) data sets. Conclusion: Customization of SAPS I for trauma patients has shown good calibration and high discriminatory power in Italian ICUs and when applied to an independent data base, The advantage of customization would be the collection of the same set of variables for all patients admitted to ICUs against the use of specific scoring systems.