A study of risk analysis and prognosis models for the mortality of sepsis based on real-world data in China

Abstract

AbstractBackgroundThe study aimed to explore the factors associated with the mortality of sepsis and to develop prognosis models for predicting outcomes based on real-world data in China. Methods: Data regarding sepsis patients’ medical records were extracted from the hospital information systems in four hospitals. The data included general information, laboratory tests, score systems, and supportive treatment for sepsis. In total, 507 medical records with complete data were available for data analysis. Multiple variable regression (MR) analysis used to explore associations, and to develop prognosis modelsResultsThe mortality of sepsis was 0.3124 in the total sample. A univariate analysis indicated 23 variables significantly associated with the mortality of sepsis (p <0.05 for all). The MLR analysis showed independent and significant variables of age, GCS, SOFA, shock, breath rate, TBIL, CHE, BUN, LAC, OI, HCO3-, IMV, and ALB (P <0.05 for all). Prognosis models have a high predictive performance (AUC = 0.885, 95% CI: 0.854–0.917 in model2). Conclusion: The study showed evidence of independent and significant factors associated with the mortality of sepsis, including age, GCS, SOFA, septic shock, breath rate, TBIL, CHE, BUN, LAC, OI, HCO3-, IMV, and ALB. Prognosis models with a high performance were developed. (Trial registration: ClinicalTrials.gov Identifier: NCTNCT03883061 retrospectively registered 19 Mar 2019.)