Enhance health evidence quality in classification tasks: A triangulation approach utilizing case-based reasoning and...

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Clinical Data Analysis

Background:

• Real-world health data for machine learning (ML) often contains biases and missing information due to multi-purpose collection.
• Healthcare applications of ML face challenges from ethical considerations and resource limitations.
• Accurate classification of patients, such as for ischemic heart disease (IHD) Diagnosis-Related Groups, is crucial for healthcare management.

Purpose of the Study:

• To propose an integrated approach to enhance health evidence quality for ML classification tasks.
• To improve the prediction of Medicare's Diagnosis-Related Groups for ischemic heart disease (IHD) patients.
• To address data quality issues in real-world healthcare data for ML.

Main Methods:

• Utilized the Medical Information Mart for Intensive Care IV (MIMIC IV) database for patient data.
• Employed six machine learning models for triangulation.
• Applied sequential triangulation using Local Process Mining (LPM) and Qualitative Comparative Analysis (QCA).

Main Results:

• Identified 8 health process features from 1545 IHD hospitalizations, showing higher correlations than non-process features.
• QCA identified 56 unique combinations, with 28 configurations having low raw coverage.
• Model performance improved, with misclassification rates decreasing by 47% after incorporating process features and reaching 0.0% after QCA.

Conclusions:

• The integrated approach enhances classification task quality through clinical relevance and improved performance.
• The method significantly reduces case-level error rates in health data analysis.
• Future work should focus on scalable QCA methods and broader applications of health process feature engineering.