As healthcare experiences a rapid evolution driven by artificial intelligence, the sector faces the challenge of assessing the accuracy and performance of AI models and monitoring them for potential adverse effects downstream.
Epic, a prominent EHR provider, is set to unveil an AI validation software suite designed to empower healthcare institutions in evaluating AI models locally and overseeing their performance over time. Seth Hain, Epic’s Senior Vice President of R&D, shared insights on this initiative in an exclusive interview.
The innovative “AI trust and assurance software suite” developed by Epic streamlines data collection and mapping processes to deliver real-time metrics and analysis on AI models. By automating these tasks, the software ensures consistency and relieves healthcare organizations of the burden of manual data mapping, a traditionally arduous validation aspect.
The primary objective is to facilitate local AI testing and validation while enabling extensive ongoing monitoring capabilities. Hain emphasized the importance of merging local outcome data with AI model information to support evaluation and continuous monitoring within individual healthcare settings.
Epic advocates for AI validation standards to be tested on diverse patient populations at the local level, emphasizing the significance of ongoing monitoring. Different healthcare facilities cater to unique patient demographics and workflows, necessitating tailored validation processes. Epic aims to introduce this functionality within the next four to six weeks, with subsequent updates planned throughout the summer.
The AI software suite boasts user-friendly reporting dashboards that offer automated updates and detailed analyses categorized by various demographics such as age, sex, and race/ethnicity. Additionally, a standardized monitoring template and data schema are integrated to facilitate future expansions of the suite to encompass new AI models seamlessly.
Moreover, Epic intends to make the monitoring template and data schema publicly accessible upon release, enabling healthcare organizations to monitor both custom AI models and those procured from third-party vendors. This open-source framework aligns with evolving best practices in AI validation, fostering transparency and flexibility in analytical processes.
Hain envisions that tools like Epic’s AI validation software will instill trust in healthcare AI applications by integrating real-world evidence seamlessly with technological insights, enabling swift and confident decision-making. By anchoring these capabilities in the core of healthcare applications, organizations can adapt efficiently to evolving standards and enhance operational confidence.
In the realm of healthcare AI testing, organizations are swiftly deploying advanced models to streamline tasks like medical record summarization and clinical note automation. However, validating these AI models effectively to ensure accuracy, performance, and safety remains a paramount concern for early adopters.
Healthcare leaders view Epic’s forthcoming software suite as a pivotal step towards enabling local auditing of AI models. Dr. Christopher Longhurst, Chief Medical Officer and Chief Digital Officer at UC San Diego Health, anticipates leveraging Epic’s software suite to audit algorithm outcomes and integrate it into their AI governance framework.
UC San Diego Health has spearheaded healthcare AI deployment, collaborating with Epic to pilot a generative AI tool for patient message responses. Their recent study on using AI to identify sepsis infection risks in emergency departments demonstrated a 17% reduction in mortality, underscoring the pivotal role of local context in achieving meaningful clinical outcomes.
Aneesh Chopra, President of CareJourney, commends Epic’s AI validation software for promoting local AI model validation and emphasizing the importance of local outcomes resulting from AI utilization. He stresses the need for providers to prioritize responsible AI use and focus on delivering outcomes essential for the healthcare ecosystem.
The release of Epic’s AI validation tool coincides with ongoing discussions on national AI validation strategies and proposals to establish a network of AI assurance labs for algorithm testing. Notable entities in the health AI domain, including the Coalition for Health AI, and key regulators like the Office of the National Coordinator for Health IT and the FDA’s Digital Health Center of Excellence, endorse this initiative.
Dr. Longhurst asserts that while national AI assurance labs are crucial, local workflows and optimizations are equally vital for achieving desired outcomes. He advocates for shared responsibility between vendors and healthcare systems in ensuring governance and oversight of AI algorithms, emphasizing the need for unbiased algorithm development and local contextual alignment.
Epic and industry leaders align with White House guidelines advocating for AI testing conditions mirroring real-world deployment scenarios and continuous monitoring for adverse outcomes. This proactive approach fosters transparency, accountability, and local stakeholder engagement in AI governance processes.
Looking ahead, Dr. Longhurst proposes integrating responsible AI use into Medicare participation conditions to ensure robust data governance and oversight mechanisms within healthcare organizations. By monitoring AI implementations rigorously, healthcare providers can uphold patient safety and quality standards, aligning with evolving regulatory guidance and industry best practices.
As the healthcare landscape embraces AI innovations, initiatives like Epic’s AI validation software pave the way for a more transparent, accountable, and patient-centric approach to leveraging AI technologies effectively. These advancements herald a new era of data-driven healthcare delivery, empowering stakeholders to make informed decisions and enhance patient care outcomes.
