DHDIP: An interpretable model for hypertension and hyperlipidemia prediction based on EMR data

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Predictive Modeling

Background:

• Traditional hypertension and hyperlipidemia prediction models face limitations due to data heterogeneity, small sample sizes, and inconsistent standards.
• These limitations hinder the clinical applicability of existing predictive models.

Purpose of the Study:

• To develop DHDIP, an interpretable prediction model for hypertension and hyperlipidemia utilizing electronic medical record (EMR) data.
• To address the shortcomings of existing models by leveraging large-scale, unstructured EMR data.

Main Methods:

• A pre-processing algorithm was developed for massive, high-dimensional, unstructured EMR data to enable machine learning application.
• XGBoost, CatBoost, and RandomForest models were evaluated to identify optimal algorithms for prediction.
• The SHAP framework was integrated into the DHDIP model to identify key contributing factors for hypertension and hyperlipidemia, enhancing model interpretability.

Main Results:

• The DHDIP model achieved a Mean Squared Error (MSE) of 0.0285 and a LOSS value of 0.0054.
• These performance metrics surpassed those reported in previous studies.

Conclusions:

• The DHDIP model effectively balances predictive performance with interpretability.
• Multi-objective learning enhances disease analysis and prediction, reducing costs and aiding clinical diagnosis.
• The study provides accessible datasets and source code for further research and application.