Predicting mortality in hemodialysis patients using machine learning analysis.

Area of Science:

• Nephrology
• Medical Informatics
• Machine Learning

Background:

• Traditional logistic regression is commonly used for predictive modeling.
• Machine learning techniques, such as random forest, offer alternative non-parametric approaches.
• Evaluating advanced methods for predicting outcomes in haemodialysis patients is crucial.

Purpose of the Study:

• To assess the efficacy of random forest in developing mortality prediction models for haemodialysis patients.
• To compare the predictive accuracy of random forest models against logistic regression models.
• To identify key variables utilized by each model for mortality prediction.

Main Methods:

• Utilized data from incident haemodialysis patients (1995-2015).
• Developed random forest and logistic regression models to predict mortality at 6 months, 1 year, and 2 years.
• Compared model accuracy using area under the curve (AUC), with data collected at 30, 60, and 90 days post-initiation of haemodialysis.

Main Results:

• Included 1571 incident haemodialysis patients with a mean age of 62.3 years and a Charlson comorbidity index of 5.99.
• Random forest models demonstrated adequate accuracy (AUC 0.68-0.73) and were superior to logistic regression models (ΔAUC 0.007-0.046).
• Both random forest and logistic regression identified distinct sets of variables for mortality prediction.

Conclusions:

• Random forest is an adequate and superior method for generating mortality prediction models in haemodialysis patients.
• The findings support the use of machine learning for improved prognostic accuracy in this population.
• Random forest offers a valuable alternative to logistic regression for clinical decision-making.