Using Machine Learning Methods to Predict the Lactate Trend of Sepsis Patients in the ICU

Abstract

Serum lactate levels are considered a biomarker of tissue hypoxia. In sepsis or septic shock patients, as suggested by The Surviving Sepsis Campaign, early lactate clearance-directed therapy is associated with decreased mortality; thus, serum lactate levels should be assessed. Monitoring a patient's vital parameters and repetitive blood analysis may have deleterious effects on the patient and also bring an economic burden. Machine learning and trend analysis are gaining importance to overcome these issues. In this context, we aimed to investigate if a machine learning approach can predict lactate trends from non-invasive parameters of patients with sepsis. This retrospective study analyzed adult sepsis patients in the Medical Information Mart for Intensive Care IV(MIMIC-IV) dataset. Inclusion criteria were two or more lactate tests within 6 h of diagnosis, an ICU stay of at least 24 h, and a change of >= 1 mmol/liter in lactate level. Naive Bayes, J48 Decision Tree, Logistic Regression, Random Forest, and Logistic Model Tree (LMT) classifiers were evaluated for lactate trend prediction. LMT algorithm outperformed other classifiers (AUC = 0.803; AUPRC = 0.921). J48 decision tree performed worse than the other methods when predicting constant trend. LMT algorithm with four features (heart rate, oxygen saturation, initial lactate, and time interval variables) achieved 0.80 in terms of AUC (AUPRC = 0.921). We can say that machine learning models that employ logistic regression architectures, i.e., LMT algorithm achieved good results in lactate trend prediction tasks, and it can be effectively used to assess the state of the patient, whether it is stable or improving.