Recent strides in artificial intelligence strategies call for closer attention to specific facets like ethics and data modernization. With recently outlined efforts and in light of executive orders, the Department of Veterans Affairs is further aligning its AI direction to ultimately improve care for veterans.
“Taking the trustworthy AI executive order to the next step in terms of applying it to the VA — we wrote a paper, which basically takes some of those points and sees how those those trustworthy AI principles specifically can apply in VA settings," Gil Alterovitz, VA’s director of the National Artificial Intelligence Institute (NAII), told GovernmentCIO Media & Research. "We're really looking at this as a multi-pronged approach to look at trustworthy AI, AI aspects, data aspects, and aspects of what is common to all agencies versus what are specific things that we have to look at in the VA directly.”
The agency’s July 2021 AI strategy formalized how it will develop, use and deploy AI capabilities. Alterovitz noted that this strategy has detailed action points with four primary goals:
- Use existing AI capacities to better deliver health care and benefits to veterans
- Develop these existing AI capacities
- Increase veteran and stakeholder trust in AI
- Build upon partnerships with industry and other government agencies.
VA is developing an AI checklist to better enable researchers, developers and clinicians to proactively identify gaps and challenges with AI capabilities, Alterovitz said. The checklist will provide considerations to have a better quality, more robust AI model.
"There are really a number of different ways that AI, natural language processing and machine learning can be used to discover and collect data,” Alterovitz said. “We can, for example, link documents based on common topics that can make it easier to find those documents. We can leverage it to help to de-identify data so that it could be processed. There are ways to classify notes ... where once you have those categories and it's easier to search and to retrieve that information.”
VA has several initiatives centered around leveraging veteran data to improve health care delivery. These efforts will impact various veteran programs, like Electronic Health Record Modernization (EHRM) and the Million Veteran Program.
With longitudinal data, like medical records, having many contributing factors that influence data analysis and results, VA is leveraging automation and AI to improve the accuracy of predictions and discover trends. The agency is also integrating these technologies to summarize large data sets, such as research papers and clinical notes, to identify data gaps and effectively communicate findings.
"Making summaries of large sets of data about documents, often common themes or summary points, is another area that we're seeing emerge, and it's an important area,” Alterovitz said.
Especially with patient data, VA is continuing to make health data accessible, discoverable, interoperable and transferable — ongoing efforts dating as far back as a 2019 executive order.
Earlier this year, VA developed a COVID-19 prediction tool, which uses AI to calculate the risk of a COVID-19 patient dying within 120 days of diagnosis. The tool has the potential to improve patient treatment and is currently being piloted at 13 VA medical centers.
“What it does is it lets people see the underlying factors for prediction via explainable AI. So, it makes that data as potentially more accessible in some ways and more understandable and interpretable,” Alterovitz said.
Looking ahead, VA is collaborating with its federal and industry partners on AI advancements. The agency also holds AI tech sprints to make different data sets available and then collaborates with industry to develop more tools.
“We've actually seen a number of those winners go on to develop programs and pilots and initially go into production. That's another way that AI is making things more accessible,” Alterovitz said.