Using methods from the data-mining and machine-learning literature for disease classification and prediction: a case...

Area of Science:

• Cardiology
• Data Science
• Machine Learning

Background:

• Physicians classify patients based on disease presence and etiology.
• Classification trees are common but can lack accuracy.
• Advanced data-mining techniques offer alternative classification methods.

Purpose of the Study:

• To compare the performance of advanced classification methods against conventional classification trees.
• To evaluate these methods for classifying heart failure (HF) subtypes: HF with preserved ejection fraction (HFPEF) and HF with reduced ejection fraction (HFrEF).
• To assess the predictive accuracy of these methods for the probability of HFPEF compared to logistic regression.

Main Methods:

• Comparison of conventional classification trees with data-mining methods including bootstrap aggregation (bagging), boosting, random forests, and support vector machines.
• Application of these methods to classify patients with heart failure into HFPEF and HFrEF subtypes.
• Evaluation of predictive performance for HFPEF probability using logistic regression.

Main Results:

• Flexible tree-based methods demonstrated substantial improvements in predicting and classifying HF subtypes compared to conventional trees.
• Conventional logistic regression exhibited superior performance in predicting the probability of HFPEF compared to the data-mining methods evaluated.
• Tree-based methods showed enhanced accuracy for HF subtype classification.

Conclusions:

• Tree-based methods provide superior performance for predicting and classifying HF subtypes in a population-based sample.
• These advanced methods do not significantly improve upon logistic regression for predicting the presence of HFPEF.
• The findings suggest a nuanced approach to selecting classification methods based on the specific clinical prediction task.