AUTHOR=Yang Boshen , Zhu Yuankang , Lu Xia , Shen Chengxing 

TITLE=A Novel Composite Indicator of Predicting Mortality Risk for Heart Failure Patients With Diabetes Admitted to Intensive Care Unit Based on Machine Learning

JOURNAL=Frontiers in Endocrinology

VOLUME=Volume 13 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/endocrinology/articles/10.3389/fendo.2022.917838

DOI=10.3389/fendo.2022.917838

ISSN=1664-2392

ABSTRACT=Abstract：
Background: Heart failure (HF) patients with diabetes faced higher mortality risk, especially for those in intensive care unit. So far, there is no precise risk stratification tool for this kind of patients.
Method: We compared nine machine learning models and selected the best one to establish a mortality-risk prediction model in Medical Information Mart for Intensive Care IV (MIMIC-IV) population, which consisted of more than 40000 patients in ICU between 2008 to 2019 at Beth Israel Deaconess Medical Center. Major indicators were identified based on a visualization method developed for machine learning. A novel composite indicator ASL was established using logistics regression based on three major indicators. This indicator performed well in MIMIC-IV cohort and was external validated using the eICU Collaborative Research Database. 
Results: The random forest model outperformed among nine models with AUC=0.92 after hyper-parameter optimization. Twenty indicators associated with hospital mortality were identified using Shapley Additive Explanations (SHAP) method. APS III, SOFA, Max lactate were selected as major indicators and a composite indicator was developed, which was named as ASL. In MIMIC-IV population, ASL had great risk discrimination ability with AUC higher than 0.80 in both low and high-risk group compared with existing tools. In external validation, AUC and DCA curve indicated that this indicator also had respectable clinical value. In addition, this indicator had a good risk stratification ability when the patients were divided into three risk levels.
Conclusion: Based on machine learning model, we developed a novel composite indicator for HF patients with diabetes admitted to intensive care unit, which was validated in internal and external cohort.
Keywords：Heart failure, diabetes, machine learning, hospital mortality, indicator.