Supplementary tables of sepsis machine learning
Context: Non-thyroid disease syndrome (NTIS) occurs in various serious illnesses, with sepsis being a common cause. Early identification of NTIS risk in sepsis patients is crucial.
Objective: This study aims to investigate the effectiveness of machine learning algorithms in predicting NTIS occurrence and mortality in sepsis patients.
Methods: Clinical data, including biochemical markers, physiological parameters, and treatment information, were collected from sepsis patients to train and test eight machine learning models: eXtreme Gradient Boosting (XGBoost), generalized linear model (GLM), logistic regression, Poisson regression, random forest, support vector machine (SVM), decision tree, and Lasso regression. The area under the receiver operating characteristic curve (ROC AUC), accuracy, precision, recall, and balanced accuracy were used to assess the accuracy and efficacy of each model.
Results: XGBoost showed the highest accuracy and sensitivity in predicting NTIS occurrence and mortality. Lower levels of T3, FT4, kappa/lambda (KAP/LAM) ratio, and anti-thymocyte globulin (ATG) were associated with increased NTIS risk. Mechanical ventilation, age, adrenaline, norepinephrine, lower CD3 counts, and lower fibrinogen (FIB) were indicative of worse outcomes.
Conclusions: Machine learning algorithms, especially XGBoost, effectively predict NTIS occurrence and mortality in sepsis patients. XGBoost outperformed other models in accuracy and sensitivity, highlighting its potential in early identification and risk assessment of NTIS in critically ill patients.
