Related Experiment Videos
Using classification tree and logistic regression methods to diagnose myocardial infarction.
C L Tsien1, H S Fraser, W J Long
1Laboratory for Computer Science, Massachusetts Institute of Technology, Cambridge, USA. chris@medg.lcs.mit.edu
Studies in Health Technology and Informatics
|June 29, 1999
Summary
Machine learning models, including FT Tree and logistic regression (LR), accurately predict myocardial infarction (MI) in emergency room patients. These models show comparable performance to existing methods, aiding early diagnosis.
Area of Science:
- Emergency Medicine
- Medical Informatics
- Machine Learning
Background:
- Early and accurate diagnosis of myocardial infarction (MI) is critical for patients presenting with chest pain to the Emergency Room (ER).
- Existing decision aids for MI diagnosis have not achieved widespread clinical adoption.
- Machine learning offers potential for developing accurate and simple decision support tools.
Purpose of the Study:
- To develop and evaluate machine learning models, specifically a classification tree (FT Tree) and a logistic regression (LR) model (FT LR), for predicting MI probability.
- To compare the performance of these novel models against established methods like the Goldman classification tree and Kennedy LR equation.
- To assess the models' diagnostic accuracy using data from distinct patient cohorts in Edinburgh, Scotland, and Sheffield, England.
Main Methods:
- Development of FT Tree and FT LR models using ER patient data with chest pain.
- Testing of FT Tree and FT LR models on independent Edinburgh and Sheffield datasets.
- Comparative analysis of FT Tree, FT LR, Goldman Tree, and Kennedy LR using Receiver Operating Characteristic (ROC) curve areas.
Main Results:
- On the Edinburgh test set, FT Tree, FT LR, and Kennedy LR demonstrated high and similar performance (ROC areas ~94%).
- The Goldman Tree showed significantly poorer performance (ROC area 84%) on the Edinburgh test set.
- On the Sheffield test set, FT Tree, FT LR, and Kennedy LR showed no significant performance differences, while FT Tree outperformed the Goldman Tree.
Conclusions:
- Classification trees, like FT Tree, can achieve diagnostic performance comparable to logistic regression models for MI detection.
- The developed FT Tree and FT LR models show promise as effective decision aids for MI diagnosis in the ER.
- Machine learning approaches offer a viable alternative to traditional methods for improving early MI diagnosis.