Epic launched an open source tool on Wednesday to enable healthcare organizations to test and monitor artificial intelligence models.
The AI validation software suite is free and available to the public on GitHub, Corey Miller, vice president of research and development at Epic, told Fierce Healthcare. Health systems can download the code to their electronic health record systems, he said.
Health systems can use the tool to validate AI models that integrate with EHR systems, including models developed by Epic as well as other those developed by other organizations. As AI best practices are developed, the open-source framework will enable organizations to bring in those standards and practices alongside the AI validation capabilities, Epic executives said.
Miller said it marks Epic's first open-source tool.
"By publishing on GitHub, it's truly available to everyone; it's not behind any lock or key that we control. We're excited to dive into this world. It's fitting that a tool intended to ensure the equity of health AI is going to be publicly available and open to contributors from around the globe," he said in an interview.
In early April, the EHR giant announced plans to release an AI validation software suite to enable healthcare organizations to evaluate AI models at the local level and monitor those systems over time.
The AI software suite, what Epic calls an "AI trust and assurance software suite," automates data collection and mapping to provide near real-time metrics and analysis on AI models, Seth Hain, Epic senior vice president of R&D, said back in April. The automation creates consistency and eliminates the need for healthcare organization data scientists to do their own data mapping—the most time-consuming aspect of validation, according to Hain.
The key is to enable AI testing and validation at a local level and allow ongoing monitoring at scale, Hain noted.
The current version of the open-source tool does not validate the performance of generative AI models, Miller said this week, but Epic plans to expand it to more AI models in the future.
The Health AI Partnership (HAIP), a collaboration of organizations including Duke Health, Mayo Clinic and Kaiser Permanente, creates and disseminates best practices for AI product use in healthcare. Two HAIP sites, Duke Health and University of Wisconsin Health, will use Epic's AI trust and assurance software suite to conduct a study to generate evidence around use of the tools to locally test and monitor AI models.
HAIP will share results of the study with Epic to improve usability of the tools among community and rural settings that participate in the HAIP Practice Network technical assistance program.
Epic also plans to join HAIP and the University of Wisconsin to study the use of a predictive model using the validation tool, Miller noted.
"We will be able to learn quickly how the tool needs to evolve and develop," he said. "We are excited that HAIP decided to use the tool because that is exactly why we created it so that individual health systems and these third-party collaboratives could use the tool and adopt it to further the use of responsible AI to benefit the entire healthcare system."
Healthcare organizations have been using predictive AI models and machine learning for almost a decade. But large language models (LLMs) and generative AI tools present a different challenge.
Health systems are moving quickly to deploy LLMs and gen AI to tackle tasks like summarizing medical records and automating clinical notetaking. But these early adopters are still working through the best methods to validate AI models to feel confident about the technology's accuracy, performance and safety.
Hain said in April the AI software suite includes intuitive reporting dashboards that are updated automatically. Users get analysis broken down by age, sex, race/ethnicity and other demographics. The software features a common monitoring template and data schema to make it easier to extend the suite to new AI models in the future, he noted.
The open source tool enables healthcare organizations run AI validation in their EHR system on their own patient populations and workflows, Epic executives said.
"It will pull in downstream outcomes and interventions. You can slice and dice the data across different patient cohorts. When you're thinking about AI equity, you could filter to certain protected classes, like age, sex, race or ethnicity, to ensure that your AI model is working appropriately across all the different patient cohorts. Looking at these downstream workflows is going to be critical to ensure fairness and equity," Miller said. "The tool is like a funnel of all this data and it's putting it into visualizations that are easy to understand. With single points and clicks, you can dig deeper into any certain piece of data."
He added, "The tool is targeted primarily at data scientists and clinicians, but we hope that it will be easy enough to understand that a clinician without a data science background can also dig in and learn where the tools are fair and equitable, and where they might need to change the way they are delivering care to improve equity."
Some stakeholders have raised concerns that Epic, a health IT vendor with a massive influence in the market, developing tools to validate AI models represents a conflict of interest.
“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.”