Designing Human-Centered AI to Prevent Medication Dispensing Errors: Focus Group Study With Pharmacists.

Area of Science:

• Pharmacy Practice
• Human-Computer Interaction
• Artificial Intelligence in Healthcare

Background:

• Medication dispensing errors pose significant global health risks, impacting patient safety and healthcare costs.
• Current verification methods like barcode scanning have limitations.
• Artificial intelligence (AI) offers potential for improved accuracy and efficiency in pharmacy verification.

Purpose of the Study:

• To gather pharmacists' feedback for designing a user-centered AI prototype for medication dispensing verification.
• To inform the initial user interface and iterative design of the AI system.

Main Methods:

• A multidisciplinary team engaged 8 pharmacists in 3 focus group rounds.
• A Bayesian neural network was used to predict National Drug Codes (NDC).
• Content analysis of transcribed discussions guided by human-machine teaming frameworks informed AI design.

Main Results:

• Pharmacists favored a hybrid AI-pharmacist teaming model for verification.
• Key design needs included enhanced AI interpretability (e.g., checkmarks, probability scores) and simplicity.
• The final prototype featured visual aids for comparison, probability histograms, and clear decision options (accept, reject, unsure).

Conclusions:

• A human-centered AI prototype for dispensing verification was developed in partnership with pharmacists.
• The design emphasizes AI interpretability, trust, and human-machine collaboration, positioning AI as an assistive tool.
• This study demonstrates a process for creating user-centric AI in pharmacy, focusing on collaboration and confidence visualization.