Nomogram and Machine Learning Models Predict 1-Year Mortality Risk in Patients With Sepsis-Induced Cardiorenal...

Area of Science:

• Critical Care Medicine
• Nephrology
• Cardiology

Background:

• Sepsis-induced cardiorenal syndrome (CRS) presents a significant challenge for predicting long-term patient outcomes.
• Early and accurate prognostication is crucial for effective clinical management.

Purpose of the Study:

• To develop and validate predictive models for 1-year mortality in patients with sepsis-induced CRS.
• To identify key risk factors influencing long-term survival in this patient population.

Main Methods:

• Retrospective analysis of 340 patients with sepsis-induced CRS (discovery cohort) and external validation with 103 patients.
• Development of a nomogram using logistic regression and machine learning (ML) models (SVM, Random Forest, Gradient Boosted Decision Tree).
• Evaluation of model performance using Area Under the Receiver Operating Characteristic Curve (AUC) and accuracy.

Main Results:

• The nomogram identified age, SOFA score, serum myoglobin (MYO), vasopressor use, and mechanical ventilation as independent risk factors.
• The nomogram achieved an AUC of 0.855, outperforming the SOFA score (0.756).
• The Random Forest ML model demonstrated high accuracy (0.765) and AUC (0.854), with key features including age, MYO, SOFA score, vasopressor use, and creatinine. Both models performed well in external validation (AUCs 0.877 and 0.863).

Conclusions:

• Nomogram and ML models effectively predict 1-year mortality in sepsis-induced CRS.
• Key predictors include age, SOFA score, serum MYO, and vasopressor use.
• These models offer valuable tools for assessing long-term prognosis in patients with sepsis-induced CRS.