Session 2: Applied Model Building and Implementation: Predicting Acute Care Utilization Following Chemotherapy

Presenter Information/ Coauthors Information

McKenna Perrin, Avera Research InstituteFollow

Presentation Type

Invited

Student

No

Track

Epidemiology

Abstract

Acute Care Utilization (ACU), including emergency department visits and hospitalizations post-chemotherapy, poses a significant and often preventable burden. Existing predictive models rarely account for diverse populations or factors such as rurality and distance to care. This study leverages data from thousands of patients receiving chemotherapy at the Avera Cancer Institute to quantify differences among those experiencing ACU, validate existing models, and develop a tailored predictive model. Using real-world, multi-source data, we address challenges such as data integration, feature engineering, and missing data management. Additionally, we explore model building techniques, hyperparameter tuning, and the critical role of stakeholder input in shaping model design. Finally, we emphasize the importance of prospectively evaluating and refining the model to ensure its real-world applicability. This work highlights practical lessons and complexities in healthcare predictive modeling to improve ACU prevention strategies.

Start Date

2-7-2025 8:50 AM

End Date

2-7-2025 9:50 AM

This document is currently not available here.

Share

COinS
 
Feb 7th, 8:50 AM Feb 7th, 9:50 AM

Session 2: Applied Model Building and Implementation: Predicting Acute Care Utilization Following Chemotherapy

Dakota A & C (Room 250)

Acute Care Utilization (ACU), including emergency department visits and hospitalizations post-chemotherapy, poses a significant and often preventable burden. Existing predictive models rarely account for diverse populations or factors such as rurality and distance to care. This study leverages data from thousands of patients receiving chemotherapy at the Avera Cancer Institute to quantify differences among those experiencing ACU, validate existing models, and develop a tailored predictive model. Using real-world, multi-source data, we address challenges such as data integration, feature engineering, and missing data management. Additionally, we explore model building techniques, hyperparameter tuning, and the critical role of stakeholder input in shaping model design. Finally, we emphasize the importance of prospectively evaluating and refining the model to ensure its real-world applicability. This work highlights practical lessons and complexities in healthcare predictive modeling to improve ACU prevention strategies.