Predicting KRT or Death in Critically Ill Patients with Rhabdomyolysis Using Machine Learning: A Multicenter Study

Area of Science:

• Critical Care Medicine
• Nephrology
• Data Science in Healthcare

Background:

• Rhabdomyolysis is a serious condition with potential for severe outcomes including acute kidney injury (AKI), kidney replacement therapy (KRT), and mortality.
• Early identification of high-risk patients is crucial for timely interventions, monitoring, and specialist consultation.

Purpose of the Study:

• To develop and externally validate a machine-learning model for predicting in-hospital death or KRT in patients diagnosed with rhabdomyolysis.
• To assess the model's performance against established risk stratification tools.

Main Methods:

• Analysis of two large US intensive care unit databases (eICU and MIMIC).
• Inclusion criteria: Adults with creatine kinase (CK) > 5,000 IU/L and length of stay > 1 day, excluding pre-existing KRT or CK elevation from myocardial infarction.
• A random forest model was trained on the eICU cohort and validated on the MIMIC cohort, comparing its performance to the McMahon score.

Main Results:

• The study included 753 (eICU) and 768 (MIMIC) admissions; the composite outcome (death or KRT) occurred in 17% and 10% respectively.
• The final machine-learning model utilized 15 routinely available variables.
• External validation in MIMIC showed strong performance: AUC 0.89, sensitivity 80%, specificity 84%, outperforming the McMahon score.

Conclusions:

• A machine-learning model effectively predicts in-hospital death or acute KRT in critically ill rhabdomyolysis patients using common clinical data.
• The developed model demonstrated superior predictive accuracy compared to the McMahon score in independent cohorts.
• Further prospective validation is recommended prior to widespread clinical implementation.