The FDA has now cleared more than 500 healthcare Artificial Intelligence (AI) algorithms and the applications for AI in healthcare are growing each day.1 As healthcare systems embrace this new technology, AI Governance is emerging as an essential, critical role. “While we believe AI has the potential to transform healthcare,” says Dr. Ryan Lee, Chair, Department of Radiology, Einstein Healthcare Network, Philadelphia, “there are important criteria that needs to be standardized for deployment, installation, and collaboration with other specialties. The role of AI governance is essential for seeking recommendation and guidance across the enterprise, and the time to develop a roadmap is now.”
According to panelists at a recent AIMED Global Summit panel discussion, Moving from Data Governance to AI Governance, participants discussed how their healthcare organizations are working to ensure AI implementations are safe, protected, and realize their potential. Some shared their experiences in developing AI Governance Frameworks with oversight by AI Governance Council.
If your healthcare organization is considering developing an AI Governance program, here are three first steps to get the process started according to the panelists.
Step One: Develop an AI Governance Program to Drive Value and Deployment
The first step is to understand the key objectives and stakeholders to include in your AI Governance strategy and identify the key elements to incorporate. According to Dr. Orest Boyko, Chief Scientific Officer at the VA and Founding Member, American Board of Artificial Intelligence in Radiology, the Department of Veteran Affairs (VA) uses its Mission Statement as its guidance to build robust capacity in AI to develop, apply innovative AI solutions, and transform the VA by facilitating a learning environment that supports the delivery of world-class benefits and services to our Veterans. Dr. Boyko said, “The Executive Orders (EO 13960) are serving as frameworks for AI policy development and implementation. To date, the VA has identified 41 AI use cases including machine learning (ML) for enhanced diagnostic error detection and ML classification of protein electrophoresis text.”
When designing, developing, acquiring, and using AI, the Dr. Boyko said the VA adheres to nine principles:
- lawful and respectful of our Nation’s values,
- purposeful and performance-driven,
- accurate, reliable, and effective,
- safe, secure, and resilient,
- understandable,
- responsible and traceable,
- regularly monitored,
- transparent,
- accountable.
“With a well-developed governance program in place, healthcare organizations can better ensure transparency, expandability and monitoring to help them drive success with AI applications,” said Hortense Allison, Vice President, Head of Regulatory Digital, Bayer Radiology.
Step Two: Establish an AI Governance Council
AI governance need not be separate from an existing leadership body of authority. If a strategic leadership council for a data and analytics program already exists, then this is the most obvious fit.
Whether through an existing or separate council, successful AI governance includes four pillars:
- Legal, regulatory and compliance review to decide what happens and who is held accountable when an AI output causes harm,
- Clinical and scientific verification and valuation to confirm that the AI algorithm has been tested on a valid data set,
- Ethical evaluation and usage guidelines to determine whether or to what extent patients are informed about the role AI is playing in their diagnosis and treatment,
- Organizational deployment and change management for training staff on what is expected and the correct actions to take when using AI.
“Another aspect for hospital governance is monitoring the AI performance over time,” added Allison. “AI systems can change in performance over time due to data drifts such as changes in image acquisition device, disease prevalence, virus mutation, and other items. Hospital governance should consider tools for monitoring the AI system performance and for communicating degradation in the AI performance to the developer.”
David Hilderbrand, Senior Vice President of Corporate Development, Blackford Analysis, noted “AI applications perform differently in different environments. Degrees of success are usually attributed to the clinical applications chosen and how they’re deployed. In a situation where success is not realized, usually applications from other providers can be used for evaluation to pursue optimal results,” he said. Recommendations were for quarterly or biannual review against false positives using notification and best practices for measuring success.
Step Three: Consider the Role of Radiology
Most of AI medical algorithms available in the U.S. are related to medical imaging.3 As such, many healthcare organizations have involved Radiology heavily in the governance process. “At our organization, our governance committee has determined that the Radiologist is the final say,” said Dr. Lee. Radiology is heavily involved because radiologists have the expertise to evaluate and deploy imaging algorithms since that is the basis of the specialty, he added.
Looking to the future, AI tools will likely evolve to handle more varied data, become integrated into consolidated workflows, become more transparent, and ultimately more useful for increasing efficiency and improving patient care. If you are interested in learning more about AI Governance, visit here.
1. Source: FDA has now cleared more than 500 healthcare AI algorithms (healthexec.com)
2. Source: See the complete list of FDA-cleared algorithms here
3. Source: Artificial Intelligence and Machine Learning (AI/ML)-Enabled Medical Devices | FDA