Personalized Medication for Chronic Diseases Using Multimodal Data-Driven Chain-of-Decisions

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Medicine
• Pharmacogenomics

Background:

• Personalized medication is crucial for chronic disease management but faces challenges integrating diverse patient and drug data.
• Current models often rely on limited data types (clinical or molecular), hindering comprehensive patient-medication relationship modeling.
• A sequential decision-making process is inherent in clinical practice for determining optimal medication regimens.

Purpose of the Study:

• To propose a novel Multimodal Data-Driven Chain-of-Decisions (MDD-CoD) framework for personalized medication.
• To integrate multimodal clinical phenotype data, multi-attribute medication data, and expert insights into a coherent decision-making process.
• To improve the accuracy and interpretability of personalized medication recommendations for chronic diseases.

Main Methods:

• Developed a three-stage deep learning framework (MDD-CoD) mimicking expert clinical decision-making.
• Incorporated multimodal patient data (phenotypes) and medication attributes (macro- and micro-level).
• Validated the framework on five datasets across four chronic diseases (CKD, MN, RA, CRC, KOA) from multiple hospitals.

Main Results:

• The MDD-CoD framework demonstrated superior predictive performance in personalized medication decision-making compared to baseline models.
• Achieved enhanced accuracy by integrating individual patient characteristics with comprehensive medication properties.
• The model showed improved generalization and interpretability in cross-disease personalized decision-making tasks.

Conclusions:

• The MDD-CoD framework offers a scalable and effective solution for personalized medication in chronic diseases.
• This foundational model advances clinical decision support by leveraging multimodal data and a chain-of-decisions approach.
• The framework holds promise for improving patient outcomes through more precise and individualized treatment strategies.