Prediction algorithm for ICU mortality and length of stay using machine learning

Abstract

Machine learning can predict outcomes and determine variables contributing to precise prediction, and can thus classify patients with different risk factors of outcomes. This study aimed to investigate the predictive accuracy for mortality and length of stay in intensive care unit (ICU) patients using machine learning, and to identify the variables contributing to the precise prediction or classification of patients. Patients (n = 12,747) admitted to the ICU at Chiba University Hospital were randomly assigned to the training and test cohorts. After learning using the variables on admission in the training cohort, the area under the curve (AUC) was analyzed in the test cohort to evaluate the predictive accuracy of the supervised machine learning classifiers, including random forest (RF) for outcomes (primary outcome, mortality; secondary outcome, length of ICU stay). The rank of the variables that contributed to the machine learning prediction was confirmed, and cluster analysis of the patients with risk factors of mortality was performed to identify the important variables associated with patient outcomes. Machine learning using RF revealed a high predictive value for mortality, with an AUC of 0.945 (95% confidence interval [CI] 0.922-0.977). In addition, RF showed high predictive value for short and long ICU stays, with AUCs of 0.881 (95% CI 0.876-0.908) and 0.889 (95% CI 0.849-0.936), respectively. Lactate dehydrogenase (LDH) was identified as a variable contributing to the precise prediction in machine learning for both mortality and length of ICU stay. LDH was also identified as a contributing variable to classify patients into sub-populations based on different risk factors of mortality. The machine learning algorithm could predict mortality and length of stay in ICU patients with high accuracy. LDH was identified as a contributing variable in mortality and length of ICU stay prediction and could be used to classify patients based on mortality risk.

Figure 1

Predictive accuracy and key variables for intensive care unit mortality in the test cohort. (a) ROC curves and AUCs for ICU mortality were obtained from machine learning methods (Random Forest, XGBoost and Neural Network) and logistic regression. (b) Relative importance of variables for ICU mortality in Random Forest. Lac, LDH, and PLT had the highest importance for the precise prediction of ICU mortality. ROC receiver operating characteristic, AUC area under the curve, CI confidence interval, ICU intensive care unit, APACHE acute physiology and chronic health evaluation, SOFA sequential organ failure assessment, Lac lactate, LDH lactate dehydrogenase, PLT platelet count, NBPs non-invasive systolic blood pressure, UN urea nitrogen, cBase (Ecf) standard base excess, NBPd non-invasive diastolic blood pressure, ALP alkaline phosphatase, cBase (B) actual base excess, CRE creatinine, HR heart rate, AST aspartate aminotransferase, PT-PER prothrombin time (%), PR pulse rate, WBC white blood cell, RR (impedance) impedance respiratory rate, PT-SEC prothrombin time (s).

Figure 2

Clustering analysis based on mortality risk in the intensive care unit in the test cohort. (a) A clustering with UMAP for ICU patients was performed based on the risk of ICU mortality and the distribution of each variable. The analysis classified the patients into five clusters. (b) The top three variables (Lac, LDH, and PLT) contributed to predicting ICU mortality and the other two factors (Diagnosis and Department) characterized each cluster. UMAP uniform manifold approximation and projection, Lac lactate, LDH lactate dehydrogenase, PLT platelet count.

Figure 3

Predictive accuracy and key variables for the length of intensive care unit stay in the test cohort. (a,b) ROC curves and AUCs for the short (a) and long (b) length of ICU stays were derived from the machine learning methods using Random Forest and logistic regression. (c,d) Relative importance of variables for the short (c) and long (d) length of ICU stays in Random Forest. Elective surgery, HR, and LDH had the highest importance for the precise prediction of short length of ICU stay; LDH, HR, and UN had the highest importance for the precise prediction of long length of ICU stay. ROC receiver operating characteristic, AUC area under the curve, CI confidence interval, ICU intensive care unit, APACHE acute physiology and chronic health evaluation, SOFA sequential organ failure assessment, HR heart rate, LDH lactate dehydrogenase, UN urea nitrogen, PR pulse rate, PLT platelet count, CRP C-reactive protein, CRE creatinine, RR (impedance) impedance respiratory rate, NBPs non-invasive systolic blood pressure, MetHb methemoglobin, RR respiratory rate (count), COHb carboxyhemoglobin, D-Bil direct bilirubin, AST aspartate aminotransferase, ALT alanine aminotransferase, PT-SEC prothrombin time (in seconds), AMY amylase, PT-PER prothrombin time (%), PT-INR prothrombin time (international normalized ratio), Alb albumin, GGT gamma-glutamyltransferase, CPK creatine phosphokinase.