SAPS III is superior to SOFA for predicting 28-day mortality in sepsis patients based on sepsis-3.0 criteria

Abstract Introduction The discrimination and calibration accuracy of prediction models tends to become poor over time. The performance of predictive models should be reevaluated periodically. We aimed to reassess the discrimination of the six commonly used models on predicting 28-day mortality in patients based on the sepsis-3.0 criteria. Methods Patient data were extracted from the fourth edition of the Medical Information Mart in Critical Care (MIMIC IV) database. The SIRS, SOFA, OASIS, LODS, SAPS II and SAPS III scores were calculated and collected. We used the area under the receiver operating characteristic curve (AUROC) to compare the discrimination abilities of the models using non-parametric Wilcoxon statistics. The Delong method was used to compare the AUROCs of the models pairwise. Multiple subgroup analyses of age, BMI, and sex with regard to the 28-day mortality prediction of the models were performed. Results A total of 12691 patients were included. The mean age of the patients was 65.97±15.77 years, and 7673 patients (60.50%) were male. The mean SIRS, SOFA, OASIS, SAPS II, LODS and SAPS III scores were higher in the non-survival group than in the survival group. The discrimination for 28-day mortality with the SAPS III (AUROC = 0.812; 95% CI, 0.802–0.822) and the LODS (AUROC = 0.804; 95% CI, 0.743–0.765) models was superior to that of the SIRS (AUROC = 0.575; 95% CI, 0.562–0.589), SOFA (AUROC = 0.612; 95% CI, 0.598–0.626), OASIS (AUROC=0.753; 95% CI, 0.742–0.764), and SAPS II (AUROC=0.754; 95% CI, 0.743–0.765) models. The Youden index of the SAPS III model was 0.484, which was the highest among the models. Subgroup analysis showed similar results to the overall results. Conclusions The discrimination for 28-day mortality with the SAPS III and LODS models was superior to that of the SIRS, SOFA, OASIS, and SAPS II models. The SAPS III model showed the best discrimination capacity of 28-day mortality compared with the other models.