Related Experiment Video
Updated: Jun 14, 2025
12:18
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
7.5K
A Comparison of Interpretable Machine Learning Approaches to Identify Outpatient Clinical Phenotypes Predictive of
Matthew Hodgman1, Cristian Minoccheri1, Michael Mathis2
1Department of Computational Medicine and Bioinformatics, University of Michigan, Ann Arbor, MI 48109, USA.
Diagnostics (Basel, Switzerland)
|August 29, 2024
Summary
Predicting first-time acute myocardial infarctions is crucial. Temporal computational phenotyping of electronic health records, using interpretable machine learning, identified key risk factors like back pain and high blood pressure.
Area of Science:
- Biomedical informatics
- Machine learning in healthcare
- Cardiovascular disease research
Background:
- Acute myocardial infarction (AMI) poses a significant threat to patients and healthcare systems.
- Most AMIs are initial events occurring outside hospitals, often without prior cardiac conditions.
- The underlying pathophysiology of AMI is not fully understood, with limited use of explainable AI for early risk identification.
Purpose of the Study:
- To develop and evaluate interpretable machine learning models for predicting first-time acute myocardial infarction.
- To identify predictive clinical phenotypes from longitudinal electronic health record data.
- To compare the efficacy of temporal computational phenotyping against other feature extraction methods.
Main Methods:
- Extracted outpatient electronic health record data from 2641 AMI cases and 5287 matched controls.
- Compared six interpretable feature extraction approaches, including temporal computational phenotyping.
- Trained seven interpretable machine learning models to predict AMI onset within six months.
Main Results:
- Temporal computational phenotyping significantly enhanced model performance over alternative methods.
- Achieved a mean cross-validation area under the receiver operating characteristic curve of 0.674.
- Identified back pain, cardiometabolic syndrome, family history of cardiovascular disease, and hypertension as key predictors.
Conclusions:
- Computational phenotyping of health records improves classifier performance and identifies predictive clinical concepts.
- Interpretable machine learning models can enhance AMI risk assessment.
- Prioritizes key risk factors for further investigation and validation.
