LASSO-Based Identification of Risk Factors and Development of a Prediction Model for Sepsis Patients.

Abstract

Objective: The objective of this study was to utilize LASSO regression (Least Absolute Shrinkage and Selection Operator Regression) to identify key variables in septic patients and develop a predictive model for intensive care unit (ICU) mortality.

Methods: We conducted a cohort consisting of septic patients admitted to the ICU between December 2016 and July 2019. The disease severity and laboratory index were analyzed using LASSO regression. The selected variables were then used to develop a model for predicting ICU mortality. AUCs of ROCs were applied to assess the prediction model, and the accuracy, sensitivity and specificity were calculated. Calibration were also used to assess the actual and predicted values of the predictive model.

Results: A total of 1733 septic patients were included, among of whom 382 (22%) died during ICU stay. Ten variables, namely mechanical ventilation (MV) requirement, hemofiltration (HF) requirement, norepinephrine (NE) requirement, septicemia, multiple drug-resistance infection (MDR), thrombocytopenia, hematocrit, red-cell deviation width coefficient of variation (RDW-CV), C-reactive protein (CRP), and antithrombin (AT) III, showed the strongest association with sepsis-related mortality according to LASSO regression. When these variables were combined into a predictive model, the area under the curve (AUC) was found to be 0.801. The AUC of the validation group was 0.791. The specificity of the model was as high as 0.953. Within the probability range of 0.25 to 0.90, the predictive performance of the model surpassed that of individual predictors within the cohort.

Conclusion: Our findings suggest that a predictive model incorporating the variables of MV requirement, HF requirement, NE requirement, septicemia, MDR, thrombocytopenia, HCT, RDW-CV, CRP, and AT III exhibiting an 80% likelihood of predicting ICU mortality in sepsis and demonstrates high accuracy.