Artificial intelligence to assist decision-making on pharmacotherapy: A feasibility study

Area of Science:

• Medical Informatics
• Computational Medicine
• Pharmacotherapy

Background:

• Artificial intelligence (AI) offers potential for analyzing complex healthcare data.
• Its application in pharmacotherapy decision-making is limited by patient-specific and dynamic factors.

Purpose of the Study:

• To investigate the utility of AI in guiding pharmacotherapy decisions.
• Utilized clinical data including diagnoses, lab results, and vital signs.

Main Methods:

• Adapted data from a prior medication therapy optimization study.
• Employed decision trees with R and tidymodels, splitting data into 74% training and 26% testing sets.
• Used bootstrapping and hyperparameter tuning to prevent overfitting; calculated performance metrics like accuracy.

Main Results:

• The AI model achieved prediction accuracies from 38% to 100% for cardiovascular drug classes in 101 elderly patients.
• Laboratory and vital sign data interpretation by the model was unclear.
• AI lag time issues were addressable by manual decision tree adjustments.

Conclusions:

• The AI model demonstrated potential for personalized medication recommendations.
• Obstacles in AI model development were largely overcome.
• Future studies should incorporate drug effects alongside drug data for better lab interpretation; human oversight is essential for AI pharmacotherapy support.