Construction of an associative model for prolonged intensive care unit stay in sepsis patients combined with myocardial injury.

objectiveThis study aimed to identify the factors contributing to Prolonged Length of Stay (PLOS) in intensive care units for sepsis patients combined with Myocardial Injury (MI) and to construct an associative model.

methodsData were from the Medical Information Mart for Intensive Care IV database. Variables were selected using the Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis. The performance of five Machine Learning (ML) models established based on key factors, including the Logistic model, XGBoost, LightGBM, AdaBoost, and RandomForest, was compared by 10-fold nested cross-validation. The optimal associative model performance was validated by 10-fold cross-validation repeated 5-times.

resultsAmong 1792 sepsis patients combined with MI, 448 patients developed PLOS. LASSO regression analysis indicated that the Sequential Organ Failure Assessment score, potassium, age, heart rate, systolic blood pressure, red blood cell, acute kidney injury, vasopressor, mechanical ventilation, and continuous renal replacement therapy might be factors related to PLOS. Combining the results of 10-fold nested cross-validation, the Logistic model, which included the 10 variables, was more stable than the other four ML models. The mean Area Under the Curves (AUCs) for the training and validation sets by 10-fold cross-validation repeated 5-times were 0.852 (0.849‒0.856) and 0.848 (0.837‒0.857). The AUC of the test set was 0.846 (0.796‒0.890).

conclusionPLOS in sepsis patients combined with MI involved multiple influences. Early identification of high-risk factors and intensive multidisciplinary treatment can help to shorten LOS and reduce the risk of complications.