Drug Recommendation from Diagnosis Codes: Classification vs. Collaborative Filtering Approaches.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support

Background:

• Electronic health records (EHRs) contain vast clinical data, complicating patient-centered decision-making.
• Healthcare recommendation systems are needed to assist professionals in making faster, more accurate patient care decisions.
• Drug recommendation systems aim to match appropriate medications to patient diagnoses, a task challenging for patients with multiple comorbidities.

Purpose of the Study:

• To explore and compare approaches for drug recommendations in EHRs.
• To evaluate the performance of collaborative filtering and traditional machine learning classifiers for drug recommendations.
• To investigate the effectiveness of a hybrid model combining these approaches.

Main Methods:

• The study focused on elderly patients with diabetes, hypertension, and cardiovascular disease in primary care settings.
• Both collaborative filtering and traditional machine learning classifiers were examined for drug recommendation generation.
• A hybrid model integrating both approaches was developed and assessed.

Main Results:

• The hybrid model achieved a recall@5 of 76.61% and precision@5 of 46.20%.
• The macro-averaged area under the curve was 74.52%, with an average physician agreement of 47.50%.
• Collaborative filtering consistently underperformed traditional classification, showing sensitivity to class imbalances and bias towards popular drug classes.

Conclusions:

• Traditional machine learning classifiers are more effective than collaborative filtering for drug recommendations in EHRs, especially for complex patient populations.
• Challenges exist in developing robust recommendation systems for EHRs, particularly concerning data imbalance and class popularity.
• Further research is needed to address these challenges and improve the accuracy and utility of clinical decision support systems.