The regular professional use of artificial intelligence (AI) has grown increasingly common in the past few years, and AI tools in the healthcare sector are no exception. The clinical use of AI has incredible potential, but it also requires a strong cognizance of the unique needs of patients and healthcare providers.
To that end, the U.S. Department of Health and Human Services (HHS), together with the Assistant Secretary for Technology Policy and Office of the National Coordinator for Health Information Technology (ASTP/ONC), recently put out a request for information (RFI) on the advancement of AI use in clinical care. It asked what HHS can do to foster public trust and confidence in modern technology solutions, to reduce uncertainty that impedes AI innovation, and to align federal incentives so that AI is deployed in ways that enhance productivity, reduce burden, lower health care costs, and improve health outcomes for patients, caregivers, and communities.
The Computing Community Consortium (CCC), in conjunction with the Computing Research Association – Industry (CRA-I), recently assembled a response to this RFI. In it, computing experts and healthcare professionals come together to present their recommendations for how to make the most of AI advancements in a clinical setting while prioritizing continuing quality of care, patient privacy, and other essential concerns. Below are some of the prevailing themes from the response.
The Potential and Limitations of Clinical AI
One of the biggest barriers facing AI adoption is public hesitancy. Many Americans are reluctant to engage with, or have their healthcare providers engage with, AI tools in a clinical setting because of reliability and privacy concerns. Training AI models with more patient data is likely to improve their output’s usefulness; yet as of now, little assurance exists that that data will remain secure in a HIPAA-compliant way. Because of this, investing in making AI tools secure and trustworthy — and helping the public see this effort — is paramount.
At the same time, studies have found immense potential for AI to enhance clinical care, like assisting in recording patient visits, differential diagnosis, creating plans for therapy and alternatives, and administrative functions like appointment scheduling. Especially in cases where AI can assist providers in avoiding documentation-related burnout, it has incredible potential to enhance the doctor-patient relationship.
This response also highlights ways, moving forward, to improve the overall usefulness and usability of AI tools in healthcare, such as:
- Making sure to consult healthcare professionals on what would best support their work
- Keeping models up-to-date with local, not just aggregate, data
- Avoiding relying on autonomous models
- Investing in industry-academia partnerships across AI and clinical care
- Reforming regulations to improve access to anonymized health data and incentivizing health systems to collaborate with researchers to contextualize that data
- Creating and implementing standard safety evaluation frameworks
For these actions to have maximum impact, it is also crucial to address a growing “AI gap” between clinical settings in urban and rural communities. Devoting resources to supporting clinical care in states in the Established Program to Stimulate Competitive Research (EPSCoR) and Institutional Development Award (IDeA) programs will help to address this divide and promote quality care for all.
Future Research Directions
This RFI response recommends three key areas of research to advance clinical AI use:
- AI-Driven Clinical Decision Support: Developing new end-to-end models that include reasoning and inference to assist in clinical decision-making
- Human-AI Collaboration: Understanding how best to keep humans, including patients, caregivers, clinicians, and more, engaged as key stakeholders in AI tools
- Development and Implementation Trajectory Evaluation Frameworks: Ensuring the information produced by AI models is accurate, relevant, and understandable through the in-context evaluation of clinical AI use.
By investing in these research areas, we can move beyond experimental models and toward systems that are mindful, technologically sound, and seamlessly woven into the complex reality of modern healthcare.
Read the Full Response
For the full scope of CCC, CRA, and CRA-I findings and analysis on advancing AI in clinical care, access the full response below. You can also see more CCC responses to the community here.
Read the Full RFI Response Here
