Early prediction of mortality upon intensive care unit admission

Abstract

Background: We aimed to develop and validate models for predicting intensive care unit (ICU) mortality of critically ill adult patients as early as upon ICU admission.

Methods: Combined data of 79,657 admissions from two teaching hospitals' ICU databases were used to train and validate the machine learning models to predict ICU mortality upon ICU admission and at 24 h after ICU admission by using logistic regression, gradient boosted trees (GBT), and deep learning algorithms.

Results: In the testing dataset for the admission models, the ICU mortality rate was 7%, and 38.4% of patients were discharged alive or dead within 1 day of ICU admission. The area under the receiver operating characteristic curve (0.856, 95% CI 0.845-0.867) and area under the precision-recall curve (0.331, 95% CI 0.323-0.339) were the highest for the admission GBT model. The ICU mortality rate was 17.4% in the 24-hour testing dataset, and the performance was the highest for the 24-hour GBT model.

Conclusion: The ADM models can provide crucial information on ICU mortality as early as upon ICU admission. 24 H models can be used to improve the prediction of ICU mortality for patients discharged more than 1 day after ICU admission.

Keywords: Critically ill; Intensive care; Mortality; Prediction.

Fig. 1

Flowchart of patient inclusion criteria and allocation to training and testing data sets. ADM admission, CORE Center of Outcome and Resource Evaluation critical care database, DL deep learning, GBT gradient boosting trees, ICU intensive care unit, LR logistic regression, MIMIC Medical Information Mart for Intensive Care database

Fig. 2

Performance of the ADM and 24 H models in the testing data sets. Performance measures are presented as values with corresponding 95% confidence intervals. The cutoff value for sensitivity, precision, and specificity was 0.04 for all models. ADM admission, APS acute physiology score of the acute physiologic assessment and chronic health evaluation (APACHE) III, AUPRC area under the precision-recall curve, AUROC area under the receiver operating characteristic curve, DL deep learning, GBT gradient boosting trees, LR logistic regression, SOFA sequential organ failure assessment

Fig. 3

Calibration plots of the ADM models. ADM admission, DL deep learning, GBT gradient boosting trees, LR logistic regression

Fig. 4

Feature importance of gradient boosting trees models. ADM admission, bmi body mass index, BUN blood urine nitrogen, cad coronary artery diseases, chf congestive heart failure, DBP diastolic blood pressure, ED emergency department, ESRD end stage renal disease, GBT gradient boosting trees, gcs Glasgow coma scale, hr heart rate, ICU intensive care unit, inv invasive ventilation, LOS length of stay, MAP mean arterial pressure, ph power of hydrogen, plat platelet, rr respiratory rate, SBP systolic blood pressure, wcc white cell counts