As artificial intelligence (AI) in health systems moves past the hype and closer to becoming the standard of care, what are the implications for large organizations considering rolling out AI across their networks? Dr. Rick Abramson, former VP of Radiology Service Line at HCA Healthcare, shared his insights with Aidoc CEO Elad Walach in a recent webinar, where they discussed how large healthcare systems are thinking about implementing AI. Both foresee that the use of AI is at an inflection point that may propel health care systems, and radiology departments in particular, along a new trajectory.
Abramson’s role was to build and lead an enterprise service line for radiology at HCA, whose network encompasses more than 185 hospitals across 22 U.S. states, with a few in the UK. This gave him a superb vantage point for insight into and a granular understanding of corporate health enterprise needs, across all dimensions of imaging (including informatics and data science), to radiology professional contracting, quality and safety, and operations growth strategy, as well as the ROI in AI.
What are current attitudes to AI?
There are a range of attitudes toward AI, which are rapidly changing as its adoption becomes more widespread. One early concern among physicians–that AI will replace radiologists and thus save money–no longer matches up with what most in the field think. “This is a misperception that you still hear a lot of,” says Abramson, “and it’s something that we have to actively speak against.” Where AI is utilized, it has not been shown to take away radiologists’ jobs, but rather to act as a partner in improving care quality, mitigating risks and reducing burnout in radiology departments by reducing image-reading times. Similarly, there are now fewer who look at AI as a new toy and are more driven by the understanding that AI is becoming the new standard of care with measurable value. Also, the cohort of individuals who believe AI is overhyped is becoming smaller as attention turns to measuring ROI, according to Abramson. “These perceptions are fading away the longer these tools are around and the more integrated they become in clinical practice.”
Abramson points out another attitude in play: “There’s this idea that AI is going to lead to operational efficiencies; it’s going to improve workflows and reduce costs by optimizing patient throughput.” In fact, there are current examples where AI is doing just this: Research findings published by the Radiological Society of North America (RSNA) show that when AI was introduced into the clinical workflow, and used to prioritize interpretation of head CT with intracranial hemorrhage, it reduced wait time and overall turnaround time. “Data showing decreased ICU lengths of stay and hospital admission lengths of stay effects is really going to move the needle,” Abramson says, and this too has been tested: 4Ways Healthcare reported a 2.8-day reduction in inpatient length of stay and 10.4% reduction in ED length of stay whilst using Aidoc’s “Always-on AI” solutions. These detect acute abnormalities, automatically highlighting them directly in the radiology workflow.
Alongside these operational efficiencies, health system leaders are looking at AI as a quality improvement tool, says Abramson. Here the efficacy of AI is measured against clinical metrics in urgent pathways, such as time-to-treatment and pulmonary embolism (PE) response times. “Organizations are starting to track PE response times as a kind of metric for quality improvement, so there are certainly opportunities for workflow enhancement through AI, right down the list on the quality side.” This is borne out by research at the Department of Diagnostic Imaging at Sheba Medical Center in Israel. It showed that AI for detection of PE demonstrates high sensitivity, specificity and accuracy compared with the gold standard and concluded that AI in PE detection might provide solutions for assisting with quicker physician diagnosis and reducing rates of missed detections.
What are the challenges for advocates of new imaging technology?
Health leaders, C-suite executives in particular, do not spend their time thinking about radiology. They’re often focused on urgencies – payer negotiations, contracts, shifting market dynamics, the competitive landscape, managing staff, optimizing operations, new operating models – it’s a lot. None of it has been made easier by the COVID-19 pandemic. “Medical imaging often occupies a somewhat peripheral position on the health system radar,” says Abramson. “That’s not to say that health systems don’t feel that imaging is important, and in fact it’s quite the opposite, but there’s a general perception that medical imaging takes care of itself. It’s off the side kind of humming away.” This kind of “if it ain’t broke, don’t fix it” thinking can create a challenge for those who are trying to advocate for the introduction of new imaging technology. What decision makers need, says Abramson, is to be show proven value in areas that align with an institution’s specific strategic priorities.
What are the priorities around AI adoption in large health systems?
Advocates of new imaging technology, Abramson says, need to start with the system’s high-level priorities, and then show what the technology can do to advance them. These strategic objectives, set quarterly or annually, are often articulated and promulgated throughout the organization. The AI value proposition should be tied to these specific metrics. “Health systems will invest in product in quality improvement but the caveat is that the quality improvement initiative has to be tied to measurable performance objectives that are of interest to the system.”
It’s here that radiologists can articulate specific, use cases that are measurably aligned with set metrics aimed squarely at system leaders’ objectives. “This is where radiologists can be especially helpful, because there’s really no field that’s better positioned to offer specific examples where deploying a technology like AI is going to improve patient care or enhance efficiency or save money.” For example, a goal may be to improve quality of care through specified improvements in door-to-needle time for stroke patients. Showing how stroke-care AI contributes to this key metric, with specific measurable outcomes, is what hospital leaders need to hear.
Radiologists can birth AI into its value pathway
Radiologists are well-placed to articulate use cases but, says Abramson, they can also play a key role in proactively validating the technology and advocating for its adoption. “AI should be ours to own, but we’re going to have to work for it. We have to be participating in the investigations and the validation studies of the new technologies. We have to be articulating the use cases and demonstrating the value proposition. And we have to be front and center in approaching health systems operators and leaders to get the technology adopted. It’s a matter of adopting the right mindset and getting to work in really championing the technology.”
