Abstract
Candidemia is a life-threatening bloodstream infection associated with high mortality rates, particularly in critically ill patients. Accurate risk stratification is crucial for timely intervention and could improve patient outcomes. This study aimed to enhance the predictive performance of the sequential organ failure assessment (SOFA) score by developing a modified SOFA (mSOFA) score, which is specifically designed for candidemia patients. Using data from MIMIC-III, MIMIC-IV, and ICU-JN databases, we identified key prognostic variables through LASSO regression and integrated into the mSOFA_3 model. The model incorporated respiratory_SOFA, coagulation_SOFA, and circulatory_SOFA along with clinical biomarkers, including lactate, albumin, and blood urea nitrogen. The mSOFA_3 model demonstrated superior predictive performance across multiple machine learning algorithms, with the logistic regression-based model achieving the highest AUC of 0.826 in the internal validation cohort and 0.813 in the test cohort. Kaplan-Meier survival analysis further validated the model's utility in stratifying patients into high-risk and low-risk groups with distinct survival outcomes. These findings highlight the mSOFA_3 as a robust and clinically relevant tool for early risk stratification, offering potential for improved decision-making and therapeutic management in critically ill patients with candidemia.