Patient Risk for Medication Non-Adherence (Implementation of CLASSify)

Stroke is a leading cause of mortality and disability in the United States. Stroke survivors are typically prescribed chronic medications to prevent recurrence; however, medication non-adherence is highly prevalent, with up to 50% of patients not taking their medications as prescribed. This non-adherence is a complex, multifactorial issue—many of its barriers are modifiable. Early identification of patients at risk for non-adherence could enable timely intervention, optimal resource allocation, and ultimately reduce the risk of recurrent stroke.

While previous studies have used traditional statistical models to explore factors associated with non-adherence among stroke survivors, few have leveraged advanced predictive analytics to develop robust prediction algorithms. Dr. Mansoor’s work addresses this gap by using machine learning to predict early post-stroke medication non-adherence with improved accuracy.

The CLASSify tool played an instrumental role in this work, allowing us to build and compare prediction algorithms across multiple machine learning models. Several models demonstrated good performance. Based on these results, Dr. Mansoor is currently drafting two manuscripts. Notably, one of the projects was selected for a podium presentation by the NIH at the Building Interdisciplinary Research Careers in Women’s Health (BIRCWH) annual meeting.