RECOMED: A comprehensive pharmaceutical recommendation system.

Area of Science:

• Artificial Intelligence in Medicine
• Computational Pharmacology
• Health Informatics

Background:

• Accurate drug recommendations are crucial for effective patient treatment and require sophisticated systems.
• Existing methods often lack the ability to integrate diverse data sources for personalized recommendations.

Purpose of the Study:

• To develop a high-accuracy pharmaceutical recommendation system using comprehensive datasets.
• To leverage machine learning and natural language processing for improved drug suggestions to physicians and patients.

Main Methods:

• Designed a pharmaceutical recommendation system integrating patient, disease, and drug features from databases.
• Employed natural language processing for sentiment analysis and neural networks with recommender system algorithms.
• Developed matrix factorization models and a deep learning model trained on patient data, incorporating drug interaction rules.

Main Results:

• The proposed model demonstrated significant improvements in accuracy (26%), sensitivity (34%), and hit rate (40%) compared to conventional matrix factorization.
• Achieved average improvements of 31% in accuracy, 29% in sensitivity, and 28% in hit rate over other machine learning methods.
• The system successfully recommends acceptable medicine combinations aligned with real-world prescriptions.

Conclusions:

• The developed deep learning model offers superior performance over traditional machine learning and matrix factorization techniques for drug recommendation.
• While promising, further improvements in clinical accuracy and sensitivity can be achieved with larger datasets.
• The open-sourced implementation facilitates further research and development in AI-driven pharmaceutical recommendations.