Related Experiment Videos
Developing and Implementing Predictive Models in a Learning Healthcare System: Traditional and Artificial
David Atkins1, Christos A Makridis2, Gil Alterovitz2
1Office of Research and Development, Department of Veterans Affairs, Washington, DC, USA;
Annual Review of Biomedical Data Science
|May 24, 2022
Summary
Predictive models using electronic health records enhance healthcare decisions. This review shares 10 years of experience implementing national risk prediction models at scale within the Veterans Health Administration.
Area of Science:
- Health Informatics
- Machine Learning in Healthcare
- Clinical Decision Support
Background:
- Electronic health record (EHR) data, linked with demographic, socioeconomic, and geographic information, has transformed predictive modeling in healthcare.
- Advances in computing power and artificial intelligence (AI) methods enable complex data analysis for clinical risk prediction.
- Despite progress, ongoing debates exist regarding the development, reporting, validation, evaluation, and implementation of these models.
Purpose of the Study:
- To review over a decade of experience in developing, testing, and implementing large-scale clinical risk prediction models.
- To share lessons learned from the Veterans Health Administration's (VHA) experience with national risk prediction models.
- To propose a future research agenda for clinical predictive modeling.
Main Methods:
- Retrospective analysis of implementation experiences with national risk prediction models.
- Case study focusing on the Veterans Health Administration (VHA) as a large integrated healthcare system.
- Synthesis of lessons learned across model development, testing, and scaled implementation.
Main Results:
- The VHA has developed and implemented numerous national risk prediction models leveraging EHR data and advanced analytics.
- Successful large-scale implementation requires careful consideration of model reporting, validation, evaluation, and integration into clinical workflows.
- Significant challenges and lessons were identified regarding the practical application of predictive models in a real-world healthcare setting.
Conclusions:
- Clinical risk prediction models are increasingly vital for informed healthcare decision-making, from treatment to resource allocation.
- The VHA's experience provides valuable insights into the complexities of deploying predictive models at a national scale.
- Further research is needed to address the ongoing challenges in the lifecycle of clinical predictive models to optimize their impact on patient care.