Predicting mortality index for ICU inpatients based on clinical data extracted from electronic health record

Abstract Predicting ICU inpatients mortality index needs to be improved to incorporate clinical data. It is also helpful to reflect the patient’s recovery and hospitals standards. In this research machine learning model LightGBM was trained and assessed. This study used a dataset for ICU admissions for adult patients from six countries. And a total of 130,000 patient records were included in the study. The final model achieved AUROC (95% CI) of 0.97, an accuracy of 0.95, and an F1 score of 0.81 on the dataset. Based on results, it is observed that machine learning models with the support of conventional mortality scoring indices can provide a successful and useful model for predicting the outcome of critical and severe cases in the ICU.