The CMLA score: A novel tool for early prediction of renal replacement therapy in patients with cardiogenic shock

Abstract

Background: Early identification of cardiogenic shock (CS) patients at risk for renal replacement therapy (RRT) is crucial for improving clinical outcomes. This study aimed to develop and validate a prediction model using readily available clinical variables.

Methods: A retrospective cohort study was conducted using data from 4,133 CS patients from the MIMIC and eICU-CRD databases. Patients from MIMIC databases were randomly divided into 80 % training and 20 % validation cohorts, while those from eICU-CRD constituted the test cohort. Feature selection involved univariate logistic regression (LR), LASSO, and Boruta methods. Prediction models for RRT were developed using stepwise selection by LR and five machine learning (ML) algorithms (naive bayes, support vector machines, k-nearest neighbors, random forest, extreme gradient boosting) in the training cohort. Model performance was evaluated in both validation and test cohorts. A nomogram was constructed based on LR model. Kaplan-Meier survival analysis assessed 28-day mortality.

Results: The incidence of RRT was approximately 13 % across all cohorts. Ten variables were selected: age, anion gap, chloride, bun, creatinine, potassium, ast, lactate, estimated glomerular filtration rate (eGFR), and mechanical ventilation. Compared with ML models, the LR model showed superior predictive performance with an AUC of 0.731 in the validation cohort and 0.714 in the test cohort. Four variables that best predicted the need for RRT (age, lactate, mechanical ventilation, and creatinine) were used to generate the CMLA nomogram risk score. The CMLA model showed better predictive accuracy for RRT in the test cohort compared to the previous CALL-K model (AUC: 0.731 vs. 0.699, DeLong test P < 0.05). Calibration curves and decision curve analysis (DCA) indicated that the CMLA model also had good calibration (Hosmer-Lemeshow P=0.323) and clinical utility in the test cohort. Kaplan-Meier analysis indicated significantly higher 28-day mortality in the high-risk CMLA group.

Conclusions: A clinically applicable nomogram with four key variables was developed to predict RRT risk in CS patients. It demonstrated good performance, promising enhanced clinical decision-making.

Keywords: Cardiogenic shock; Machine learning; Nomogram; Prediction model; Renal replacement therapy.