LASSO-Based Machine Learning Algorithm for Prediction of PICS Associated with Sepsis

Abstract

Introduction: This study aims to establish a comprehensive, multi-level approach for tackling tropical diseases by proactively anticipating and managing Persistent Inflammation, Immunosuppression, and Catabolism Syndrome (PICS) within the initial 14 days of Intensive Care Unit (ICU) admission. The primary objective is to amalgamate a diverse array of indicators and pathogenic microbial data to pinpoint pivotal predictive variables, enabling effective intervention specifically tailored to the context of tropical diseases.

Methods: A focused analysis was conducted on 1733 patients admitted to the ICU between December 2016 and July 2019. Utilizing the Least Absolute Shrinkage and Selection Operator (LASSO) regression, disease severity and laboratory indices were scrutinized. The identified variables served as the foundation for constructing a predictive model designed to forecast the occurrence of PICS.

Results: Among the subjects, 13.79% met the diagnostic criteria for PICS, correlating with a mortality rate of 38.08%. Key variables, including red-cell distribution width coefficient of variation (RDW-CV), hemofiltration (HF), mechanical ventilation (MV), Norepinephrine (NE), lactic acidosis, and multiple-drug resistant bacteria (MDR) infection, were identified through LASSO regression. The resulting predictive model exhibited a robust performance with an Area Under the Curve (AUC) of 0.828, an accuracy of 0.862, and a specificity of 0.977. Subsequent validation in an independent cohort yielded an AUC of 0.848.

Discussion: The acquisition of RDW-CV, HF requirement, MV requirement, NE requirement, lactic acidosis, and MDR upon ICU admission emerges as a pivotal factor for prognosticating PICS onset in the context of tropical diseases. This study highlights the potential for significant improvements in clinical outcomes through the implementation of timely and targeted interventions tailored specifically to the challenges posed by tropical diseases.

Keywords: LASSO regression; mortality; persistent inflammation immunosuppression catabolism syndrome; predictive model; sepsis.

Figure 1

The LASSO regression analysis identified variables predicting PICS. (A) Number of non-zero coefficients in the model. (B) Number of variables corresponding to different λ values. Six variables were selected by LASSO regression, and constituted the basic factors of the prediction model.

Figure 2

The predictive values of the model of machine learning algorithm. (A) The AUCs of training and validation sets. The AUC of training data to predict ICU mortality was 0.828, while that of validation group was 0.847. (B) The calibration of training set. (C) The calibration of validation set.

Figure 3

Decision curve analysis (DCA) of the prediction model. (A) The DCA curve of training set. (B) The DCA curve of validation set. Within this probability range of 0.15 to 0.80, the predictive performance of the model surpassed that of individual predictors within the cohort. The DCA curve of the validation data resembled that of the training data, further confirming the model’s favorable predictive effect.