Intermountain Health is deploying Layer Health’s artificial intelligence for clinical data abstraction to improve its quality reporting and clinical registry submissions in stroke, surgery and cardiovascular disease.
The Utah-based health system is also making a strategic investment in the company through its venture arm, Intermountain Ventures. The partnership will include collaboration with Intermountain subject matter experts and a setting to deploy new technologies.
Layer Health is building a core platform for AI chart review. Its algorithms can abstract structured and unstructured patient data to understand the type of care a patient received and their outcome. Such data are needed for quality reporting for payers and for participation in clinical registries, which can help improve the quality of care for patients and the effectiveness of treatments.
David Sontag, co-founder of Layer Health, estimates there are a billion hours worth of collective manual chart review performed by clinicians and staff across quality reporting and other clinical and administrative use cases. AI chart review could dramatically reduce the cost of chart review while providing comprehensive answers.
Intermountain Health brings a depth of experience in quality reporting, and the size of the system brings a new scale to Layer Health’s technology.
“Everyone that I've talked to in this space, they view registry abstraction and filling out these registries as basically a fixed cost, because it is so manual, it takes so much time, and it's just it is a cost we have to pay, or do, in order to be compliant, to be able to get reimbursements we need," Phillip Wood, Intermountain Ventures partnerships executive director, said in an interview. "To add to that, our team has worked very hard to be as efficient and effective as possible, and it is still manual. So having a tool like this to automate that and make it more efficient expands the aperture of the registries and the other things that they can do.”
Layer Health’s AI takes away the need for manual chart review for data submission to clinical registries. Its algorithms can parse structured and unstructured data from a patient’s chart and add relevant information—like the type of care the patient received and the outcome—to a clinic registry about the disease state.
Intermountain will deploy Layer Health at all of its 33 locations across Utah, Idaho, Nevada, Colorado, Montana and Wyoming. They plan to expand the use of the technology to other registries—they participate in more than 35—after collecting initial data points on the improvements that Layer Health makes to their quality reporting and contribution to clinical registries.
Intermountain has a large, centralized chart abstraction team. Because of the system’s focus on chart review, and the significant amount of human labor it takes to do it, its leaders have long been searching for a solution like Layer Health to aid the process.
Stroke, surgery and cardiovascular disease are influential areas in which to begin the chart review partnerships, and they offer opportunity for expansion of the tools, Sontag said.
“We want to demonstrate ... what the value of the technology is," Sontag explained. "The idea that, as we build trust in AI in these areas, as we develop the ROI, in terms of time savings and opportunity for expansion to more registries, more patients—then that will give not only Intermountain but other health systems that Layer will be working with in the future the proof points that they will need to expand much more broadly."
Clinical registries and chart abstraction are key to understanding what treatments work best for patients and to standardize clinical care, Sontag said in January. Teams of nurses survey charts at random for outcomes reporting and submitting data to clinical registries.
Not only does Layer’s AI remove the manual work of chart review, it has trained its algorithm to be able to look at a much larger swath of unstructured patient data than other LLMs of its kind. This allows the algorithm to answer questions that can be difficult without a comprehensive review of patient data.
The system has been interested in participating in more registries but did not previously have the capacity to do so.
“Layer Health’s technology reflects the kind of AI solution that will drive meaningful change for our system,” Cara Camiolo Reddy, M.D., chief quality and safety officer at Intermountain Health, said in a statement. “Our team manages more than 35 active registries and is constantly evaluating how to support more registries that demonstrate our clinical excellence. Layer offers a scalable approach that allows us to support existing registries more efficiently and finally move forward with new ones. Furthermore, with the support of AI, our team can focus more on driving meaningful improvements in patient care.”
Intermountain will work with Layer Health on two stages of validation as it rolls out the tool. The first phase will be retrospective validation. Layer Health will gather data on how Intermountain has been handling manual chart review and to get a baseline on its performance.
The second step, prospective validation, incorporates the teams of nurses who currently do manual chart review. The nurses will give their feedback on the algorithms’ responses to questions in the clinical registry with an option to correct the answer. The process demonstrates how well Layer Health works on the system’s data and how the AI and the clinical team can work together to achieve high-quality and trustworthy results.
Many AI vendors do not offer local validation in partnership with customers, but Sontag explained that, for Layer Health, the validation is part of the company’s ethics. He also explained how Layer Health’s local validation is more relevant to its customer than a third-party validation.
“AI is seldom used in a vacuum, in a way in which one could use a single number to quantify its performance,” Sontag explained. “This is a good example of how you can go the next step beyond what you could do in a more centralized validation. So here in the second stage of validation, where you have nurses using our AI software and bringing together their own clinical expertise with what AI is suggesting to come to a single right answer that requires a human in the loop. That's not something that you can easily do in a centralized way.”