Scientific Reports (Dec 2022)

An improved prognostic model for predicting the mortality of critically ill patients: a retrospective cohort study

  • Xianming Zhang,
  • Rui Yang,
  • Yuanfei Tan,
  • Yaoliang Zhou,
  • Biyun Lu,
  • Xiaoying Ji,
  • Hongda Chen,
  • Jinwen Cai

DOI
https://doi.org/10.1038/s41598-022-26086-1
Journal volume & issue
Vol. 12, no. 1
pp. 1 – 17

Abstract

Read online

Abstract A simple prognostic model is needed for ICU patients. This study aimed to construct a modified prognostic model using easy-to-use indexes for prediction of the 28-day mortality of critically ill patients. Clinical information of ICU patients included in the Medical Information Mart for Intensive Care III (MIMIC-III) database were collected. After identifying independent risk factors for 28-day mortality, an improved mortality prediction model (mionl-MEWS) was constructed with multivariate logistic regression. We evaluated the predictive performance of mionl-MEWS using area under the receiver operating characteristic curve (AUROC), internal validation and fivefold cross validation. A nomogram was used for rapid calculation of predicted risks. A total of 51,121 patients were included with 34,081 patients in the development cohort and 17,040 patients in the validation cohort (17,040 patients). Six predictors, including Modified Early Warning Score, neutrophil-to-lymphocyte ratio, lactate, international normalized ratio, osmolarity level and metastatic cancer were integrated to construct the mionl-MEWS model with AUROC of 0.717 and 0.908 for the development and validation cohorts respectively. The mionl-MEWS model showed good validation capacities with clinical utility. The developed mionl-MEWS model yielded good predictive value for prediction of 28-day mortality in critically ill patients for assisting decision-making in ICU patients.