Aug-SAPS II improved mortality prediction calibration and discrimination with an AUC of 0.866 compared to AUC of 0.841 for SAPS II, and reduced hospital-level random-effects variance by 35.3%.
Cohort (n=2,156,332)
Yes
Does an augmented SAPS II model incorporating comorbidities improve in-hospital mortality prediction and benchmarking in adult ICU patients compared to the original SAPS II?
Augmenting the SAPS II score with routinely collected administrative comorbidity data significantly improves mortality prediction calibration and discrimination while reducing unexplained between-hospital variability for ICU benchmarking.
Effect estimate: AUC increased from 0.841 to 0.866 (95% CI 0.865-0.866)
Absolute Event Rate: 19.8% vs 17.1%
Severity scoring systems such as the Simplified Acute Physiology Score II (SAPS II) are widely used for benchmarking intensive care unit (ICU) performance, but their calibration deteriorates over time, leading to systematic overestimation of mortality in contemporary populations. We aimed to evaluate recalibrated SAPS II (r-SAPS II) and augmented SAPS II (Aug-SAPS II) models for mortality prediction using routinely collected national data. This nationwide retrospective study used the French national database of hospital stays. Adult ICU stays from 2015 to 2023 constituted the derivation cohort; 2024 data were used for temporal external validation, and 2025 data for sensitivity analyses. Demographic characteristics, SAPS II score, and ICD-10–derived comorbidities defined according to the Elixhauser classification, were used to construct aug-SAPS II model. Model calibration was assessed using Cox calibration regression and standardized mortality ratios (SMRs), defined as the ratio of observed to predicted in-hospital deaths. Discrimination and between-hospital heterogeneity were quantified using the area under the receiver operating characteristic curve (AUC) and random-effects variance in SMRs from mixed-effects Poisson regression. In the derivation cohort (2 156 332 ICU stays), discrimination improved, with the AUC increasing from 0.841 (95% CI, 0.840–0.841) for SAPS II to 0.866 (95% CI, 0.865–0.866) for aug-SAPS II. Calibration improved for both r-SAPS II and aug-SAPS II, with calibration-in-the-large shifting from − 1.10 to approximately 0.00 and calibration slope from 0.72 to approximately 1.00 at both the hospital level and across clinical subgroups. In the validation cohort (242 465 stays), calibration remained close to ideal and SMRs were near 1; AUCs were 0.835 (95% CI, 0.833–0.837) for SAPS II and 0.862 (95% CI, 0.860–0.864) for aug-SAPS II. Hospital-level random-effects variance was unchanged with r-SAPS II but decreased by 28.7% with aug-SAPS II. Findings were consistent in 2025 data. Compared with the original SAPS II, aug-SAPS II improved calibration and discrimination and reduced between-hospital variability without requiring additional data beyond routinely collected national administrative information.
Kamel et al. (2026) conducted a cohort in ICU stays mortality prediction (n=2,156,332). Augmented Simplified Acute Physiology Score II (Aug-SAPS II) vs. Simplified Acute Physiology Score II (SAPS II) was evaluated on In-hospital mortality prediction (AUC increased from 0.841 to 0.866, 95% CI 0.865-0.866). Aug-SAPS II improved mortality prediction calibration and discrimination with an AUC of 0.866 compared to AUC of 0.841 for SAPS II, and reduced hospital-level random-effects variance by 35.3%.