- May 30, 2024
- Posted by: Thomas Anderson
- Categories:
On Wednesday, Epic introduced an open-source tool designed to help healthcare organizations test and monitor artificial intelligence models.
Corey Miller, vice president of research and development at Epic, told Fierce Healthcare that the AI validation software suite is free and publicly accessible on GitHub. He mentioned that health systems can download the code to integrate with their electronic health record systems.
Health systems can use the tool to validate AI models that integrate with EHR systems, including those developed by Epic and other organizations. Epic executives stated that as AI best practices evolve, the open-source framework will allow organizations to incorporate these standards and practices alongside the AI validation capabilities.
Miller noted that this is Epic’s first open-source tool.
In publishing on GitHub, it’s truly accessible to everyone; it isn’t restricted by any locks or keys under our control. We’re excited to venture into this domain. It’s fitting that a tool designed to ensure the equity of health AI will be publicly available and open to contributors worldwide,” he said in an interview.
In early April, the EHR giant announced plans to release an AI validation software suite, allowing healthcare organizations to evaluate AI models locally and monitor these systems over time.
Back in April, Seth Hain, Epic’s Senior Vice President of R&D, explained that the AI software suite, referred to as an “AI trust and assurance software suite” by Epic, automates data collection and mapping. This automation enables near real-time metrics and analysis on AI models. According to Hain, the automation ensures consistency and removes the necessity for healthcare organization data scientists to conduct their own data mapping—a process deemed the most time-consuming aspect of validation.
Hain emphasized the importance of enabling AI testing and validation at a local level while facilitating ongoing monitoring at scale.
Miller stated this week that the current version of the open-source tool does not validate the performance of generative AI models. However, Epic intends to expand its scope to encompass more AI models in the future.
Miller mentioned that the Health AI Partnership (HAIP), comprising organizations like Duke Health, Mayo Clinic, and Kaiser Permanente, intends to utilize Epic’s AI trust and assurance software suite to locally validate AI models for its health system members.
He noted that Epic also intends to collaborate with HAIP and the University of Wisconsin to explore the utilization of a predictive model using the validation tool.
“We’ll swiftly gather insights on how the tool should evolve and improve,” he stated. “We’re thrilled that HAIP chose to utilize the tool because that’s precisely why we developed it: to enable individual health systems and collaborative third parties to adopt and enhance responsible AI usage for the greater benefit of the healthcare system as a whole.”
For nearly a decade, healthcare organizations have employed predictive AI models and machine learning. However, the emergence of large language models (LLMs) and generative AI tools poses a distinct challenge.
Health systems are swiftly adopting LLMs and generative AI to address tasks such as summarizing medical records and automating clinical notetaking. However, these early adopters are still navigating the optimal approaches to validate AI models, ensuring confidence in the technology’s accuracy, performance, and safety.
In April, Hain mentioned that the AI software suite incorporates user-friendly reporting dashboards that update automatically. Users receive analysis categorized by age, sex, race/ethnicity, and other demographics. He highlighted that the software includes a standardized monitoring template and data schema, facilitating seamless extension of the suite to new AI models in the future.
According to Epic executives, the open-source tool empowers healthcare organizations to conduct AI validation within their EHR systems, utilizing their own patient populations and workflows.
Miller explained, “The tool will integrate downstream outcomes and interventions, allowing users to analyze data across various patient cohorts. When considering AI equity, filters can be applied to specific protected classes, such as age, sex, race, or ethnicity, to ensure the AI model’s effectiveness across diverse patient groups. Examining these downstream workflows is crucial for ensuring fairness and equity.” He added, “The tool acts as a conduit for this data, presenting it in easily understandable visualizations. Users can delve deeper into specific data points with simple clicks.”
He further remarked, “While the tool is primarily aimed at data scientists and clinicians, we aim for it to be accessible enough that clinicians without a data science background can also explore it. This way, they can identify areas where tools demonstrate fairness and equity, as well as areas where adjustments in care delivery might be necessary to enhance equity.
Certain stakeholders have expressed concerns about the potential conflict of interest arising from Epic, a major player in the health IT market, developing tools to validate AI models.
“This is completely agnostic to any of our code. It’s designed to work with predictive models that we created as well as predictive models we had no hand in creating,” Miller said. “At Epic, the motto we like to live by is ‘Do good, have fun, make money’ and we feel like this one really fits in that ‘Do good’ part. It feels like an opportunity to leverage our expertise in this space to further the global community’s ability to evaluate AI and use it safely.”