Explainable Machine Learning for Risk Prediction of Reduced Quality of Life in Hypertension

Area of Science:

• Machine learning applications in healthcare
• Cardiovascular disease research
• Quality of Life studies

Background:

• Hypertension significantly impacts patient quality of life (QoL).
• Explainable machine learning (ML) offers tools for risk stratification and identifying modifiable factors affecting QoL.

Purpose of the Study:

• To develop ML classifiers for QoL risk stratification in hypertension.
• To identify key determinants of low QoL using Shapley Additive Explanations (SHAP).

Main Methods:

• Analyzed data from 534 hypertensive individuals using various ML classifiers (Decision Tree, Gradient Boosting, XGBoost, AdaBoost, Random Forest, Naive Bayes).
• Employed 10-fold cross-validation for model evaluation and rank-based metrics for stability assessment.
• Applied SHAP to the Gradient Boosting model, identified as the most stable.

Main Results:

• Gradient Boosting and Random Forest demonstrated strong performance in identifying reduced physical QoL.
• Gradient Boosting and XGBoost excelled in classifying psychological QoL, with Gradient Boosting being the most stable.
• SHAP analysis revealed medication adherence and acceptance as primary shared risk drivers for both QoL domains.

Conclusions:

• Ensemble tree-based models, especially Gradient Boosting, effectively discriminate between reduced and good QoL in hypertensive patients.
• Medication adherence and patient acceptance are critical shared risk factors.
• Physical activity, age, and education influence domain-specific QoL variations.