Interpretable Machine Learning for Personalized Medical Recommendations: A LIME-Based Approach.

Area of Science:

• Artificial Intelligence in Healthcare
• Medical Informatics
• Geriatric Medicine

Background:

• Chronic diseases pose significant threats to older adults' health and well-being.
• Hospitals possess vast health data crucial for understanding disease progression and personalizing treatment.
• Existing health recommendation systems often lack transparency, failing to provide explanations for their suggestions.

Purpose of the Study:

• To propose a novel interpretable recommendation system using deep learning and Local Interpretable Model-Agnostic Explanations (LIME).
• To enhance personalized patient-treatment recommendations for chronic diseases in older adults.
• To increase patient trust by elucidating the reasoning behind medical recommendations.

Main Methods:

• Applied a deep learning approach integrated with LIME for interpretable recommendations.
• Utilized six deep learning algorithms for interpretation after data preprocessing.
• Focused on two prevalent chronic diseases in older adults: heart disease and diabetes.
• Analyzed feature importance and contribution coefficients to explain model predictions.

Main Results:

• For heart disease, CholCheck, GenHith, and HighBP were identified as key predictive features.
• For diabetes, glucose, BMI, and age were found to be the most important features.
• LIME effectively approximated model predictions and determined feature importance for both datasets.
• The system successfully elucidated the impact of patient characteristics on recommendations.

Conclusions:

• The proposed LIME-based deep learning system provides interpretable health recommendations for chronic diseases.
• This approach enhances transparency and patient trust in medical recommendation systems.
• The findings have significant implications for improving decision-making in healthcare for older adults.