Development of an AI-Assisted Mobile Application for Improving Polypharmacy Management, Medication Adherence, and Patient Education in Outpatient Settings

Document Type

Conference Proceeding

Publication Date

5-9-2025

Abstract

Polypharmacy, the use of five or more medications by an individual, affects 65.1% (37.8 million) of US geriatric patients. The management of polypharmacy presents complex challenges that can lead to medication mismanagement and adverse health outcomes. Smartphone applications hold great potential to overcome these issues. Existing apps have features including reminder systems, adherence logging, and manual pill identification. Yet, there is room for advancement in areas such as individualized management and artificial intelligence integration to minimize errors in medication adherence. This study created a prototype smartphone application, Polly, to support polypharmacy management by combining traditional medication management tools with real-time machine vision verification. The app's goals include improving patient comprehension and adherence while reducing adverse drug events.

Polly was built using the Flutter framework for cross-platform compatibility, with a backend supported by FastAPI and hosted on Microsoft Azure. The machine vision capabilities were developed using the Ultranalytics YOLOv8 model that was trained on 697 images of pills taken from the National Institute of Health's Computational Photography Project for Pill Identification archive, publicly available images from Google, and photographs taken by the researchers. Validation testing was performed using 124 additional images.

Validation testing of the YOLOv8 model resulted in a precision of 84.49%, recall of 83.71%, and a mean Average Precision at an intersection-over-union (IoU) threshold of 0.5 (mAP50) of 81.24%. When evaluated across a range of IoU thresholds from 0.5 to 0.95 (mAP@50-95), the model scored 83.75%. Heuristic post-processing further increased real-world predictive performance. The Polly mobile application was successfully prototyped with several key features including a daily medication list, medication input/scheduling, medication and refill reminders, pill verification, adherence tracking, and as-needed pill identification. Each feature focuses on user-friendliness, accessibility, and practical utility for patients managing complex medication regimens.

The development and prototyping of Polly demonstrate the feasibility of combining traditional medication management tools with machine vision technology to create a comprehensive solution for polypharmacy management. The application successfully addresses several key limitations of existing medication management solutions, particularly through its innovative real-time verification and user-centric design for geriatric populations. The innovation behind Polly allows for increased patient care and safety standards while also promoting patient empowerment and education.

Comments

2025 Research Day Corewell Health West, Grand Rapids, MI, May 9, 2025.
Abstract 1649

This document is currently not available here.

Share

COinS