Establishment of prediction model for predicting the death risk in patients with sepsis in 30 days / 中华急诊医学杂志

ABSTRACT

Objective:To predict the sepsis patients with bad outcomes in short term and help clinical physicians to take intervention measures to reduce the mortality.Methods:A total of 900 patients with sepsis who were hospitalized in the Dongyang Peoples’ Hospital between 1st Jan 2013 and 30th Mar 2021 had been involved in this study. Information including gender, age and examination results of first time within 24 hours following hospitalization were collected. Independent risk factors of death in 30 days were screened by logistic regression analysis and further confirmed by stepwise regression analysis. Based on the screened variables, nomogram prediction model was established. Finally, the prediction model was evaluated for its prediction power by the area under the curve of receiver operating characteristic (AUC), calibration accuracy by GiViTI calibration curve and clinical effectiveness by decline curve analysis (DCA). The established prediction model was validated by using bootstrap assay.Results:Stepwise regression analysis results showed that B-type natriuretic peptide, lactic acid, albumin, oxygenation index, mean artery pressure, hematocrit and heart rate within 24 hours after hospitalization were significantly associated with death in 30 days among patients with sepsis. The AUC of prediction model was 0.846, with P of 0.886 in calibration curve, calibration slope of 1.0, R2 of 0.385, brier scaled value of 0.092 and DCA curve above the two extreme curves. In validation using bootstrap, the prediction model owned an AUC of 0.854, a P of 0.994 in calibration curve, a brier scaled value of 0.090, a calibration slope of 1.0 and a R2 of 0.389. Also, its DCA curve was above the two extreme curves. Conclusions:B-type natriuretic peptide, Lactic acid, albumin, oxygenation index, mean artery pressure, hematocrit and heart rate within 24 hours after hospitalization were independent risk factors of death in 30 days among patients with sepsis. The established prediction model in this study owned good prediction power of sepsis patients who owned high risk of death in 30 days.