Mass General, Mayo Clinic, & More Share AI, Analytics Plans for Next Year

MASS GENERAL BRIGHAM

Keith Dreyer, DO, PhD, chief data science officer and chief imaging information officer at Mass General Brigham, highlighted the system’s current AI and data analytics efforts that are over half a decade in the making.

“[In 2016,] we realized that not only do we have massive amounts of data, but we have the ability to create AI… We realized we also need the ability to validate those and then, if we want to, to be able to deploy them internally and [commercialize them],” he said in a Zoom interview.

These realizations led to the creation of the Center for Clinical Data Science, which has since become the Data Science Office (DSO). The DSO currently serves as Mass General Brigham’s home for end-to-end large-scale AI product development infrastructure and expertise.

Over the past six years, researchers have been working with industry and other stakeholders to develop and commercialize healthcare AI algorithms, efforts which became more formalized within the last year. Dreyer noted that as of December, most of Mass General Brigham’s AI algorithms are related to medical imaging and applicable across multiple healthcare specialties.

In 2022, the health system focused on the deployment of these algorithms within the Mass General Brigham network, which is a key step to making AI systems commercially available for other health systems to use, Dreyer explained.

Moving forward, Mass General plans to continue developing and deploying algorithms internally for validation purposes, alongside deploying algorithms created outside the organization and establishing governance around the use of AI.

“We're going to put more algorithms in place because we have all the infrastructure to do it, [and] we're going to develop more algorithms because we now have the capability to do that at scale much faster,” Dreyer noted. “All of that’s going to continue; all of that's going to ramp up.”

Beyond these efforts, the health system is also looking to launch new data analytics and AI-related projects, some of which will support AI creation pipelines and infrastructure.

One of the efforts to improve these foundational areas is concerned with data integrity.

“A lot of this data is not well defined at the time of use. So, when you have an inference at runtime, you need to know exactly what the integrity of the data [is],” Dreyer explained. “If the algorithm runs on this data, is it going to be what's called 'out of distribution'?”

If the data is out of distribution, the AI may not recognize the information, resulting in a ‘wrong answer’ or inaccurate insights.

“You almost need AI to control the AI,” Dreyer continued. “And so we're developing a lot of [tools] with industry to do that. It speaks to the immaturity of the market today that a lot of this isn't in place.”

The health system will also work to build an “AI factory” for healthcare-friendly machine learning-based software and algorithm creation while making efforts to improve commercialization and offer products directly to other health systems in 2023, Dreyer said.

MAYO CLINIC

Ajai Sehgal, Mayo Clinic’s chief data and analytics officer, shared that many of Mayo Clinic’s AI and analytics priorities for 2023 leverage the health system’s growing expertise in machine learning (ML).

One of the focus areas that will leverage ML is hospital management, which Sehgal noted includes financial and capacity management projects. Much of this innovation came in response to the COVID-19 pandemic, during which Mayo Clinic created a predictive analytics model to forecast ICU capacity as beds were filling up.

The health system has pivoted to use the same predictive technology in 2022 to address the RSV outbreak.

“Most recently, we fine-tuned [the model] to address the Pediatric and Children's Center because RSV is now dominating, and our ICUs are filling up,” Sehgal explained. “So, we need to be able to predict when those children are going to be released so that we can also coordinate pediatric surgeries, [and] complex surgeries that require those same ICU beds in the [pediatric ICU].”

On the clinical care side, the health system has also used ML to detect various health conditions and improve patient outcomes. Sehgal highlighted recent work by Mayo Clinic researchers in cardiology, which showcased how an established 12-lead ECG AI algorithm was adapted to use single-lead ECG signals from Apple Watches to identify weak heart pumps.

Mayo Clinic is working on getting the 12-lead ECG algorithm approved by the FDA and ready for commercialization, a process Sehgal noted is one of the biggest challenges for healthcare AI development and deployment. To address this, his team is working to streamline the process.

“I have a team that specializes in software-as-a-medical device regulatory, and we're building out a remote process automation application that will provide guidance for our physician researchers from the moment that they have the idea for what they want to do in machine learning,” he explained. “Right away, they fill out our questionnaire for the project, which is called T-REX, [or] translational regulatory expert system.”

T-REX is designed to help researchers determine if their tool will need FDA approval. If the product is likely to be subject to approval, the system can help Mayo Clinic transfer administrative burdens off individual researchers by providing documentation and data quality support.

Moving into next year, Sehgal stated that his team will focus on building out automated clinical translation workflows and breaking down data silos.

“If we can reduce the 80 percent of the time that's spent sourcing data by having a single data source that is well governed, well documented, well cataloged, and people can just very easily discover data, we're going to really exponentially increase the amount of machine-learning algorithms that are out there,” Sehgal stated.

“Both regular analytics and machine learning suffer from the same problem: where's the data and where's the data that I need to do what I want?” he added. “So, we're focusing a lot of effort into building out those data sources, unifying our silos into a single data mesh using data virtualization and making it easier to develop the models.”

MOUNT SINAI HEALTH SYSTEM

Thomas J. Fuchs, DSc, dean of artificial intelligence and human health and co-director of the Hasso Plattner Institute for Digital Health at Mount Sinai, noted that much of the health system’s work in 2023 will continue to focus on health equity and ethical AI.

As the largest health system in New York City, Mount Sinai serves a substantial and diverse patient population. According to Fuchs, taking this into account is critical for healthcare AI innovation.

“[A diverse patient population is] specifically important if you want to build AI that's not only safe and effective, as it should be in a regulated setting, but that's also equitable,” he explained. “That [AI] is used for all patients with all kinds of backgrounds, and for that, it's important that the systems are built, trained and tested, and validated on a very diverse patient population.”

Mount Sinai’s department of artificial intelligence and human health was launched in 2021 as part of this effort with the mission to lead the AI-driven transformation of healthcare. Fuchs stated that the health system has plans to invest in the department further by renovating a building on the Mount Sinai campus to house it, which is set to open next year.

This AI hub will help foster collaboration among computer scientists and clinicians to support AI, ML, and computational breakthroughs using Mount Sinai’s large patient datasets. This work is taking place alongside the health system’s AI ethics initiative that incorporates expertise from not only computer scientists and clinicians but also bioethicists, Fuchs said.

“A lot of AI questions, ethics questions, in our space actually have been addressed in healthcare and are just unknown to the computer scientists who build AI,” he explained. “So, we want to draw on these decades of experience in bioethics and healthcare ethics, and then these experts can, together with the computer scientists, expand these settings for AI and the use of AI in healthcare.”

To foster this research and collaboration across its network, Mount Sinai will focus on data access in 2023 using a cloud-based platform known as AI Ready Mount Sinai. The platform aims to link patient data generated from different clinical departments across the health system, which can be heavily siloed and difficult to access.

This work aims to improve patient care and reduce clinician burden.

“What Mount Sinai aims to do with all these initiatives is to be a fertile ground to build many more of these AI systems in many of these areas that then are truly safe and effective, and work for everyone, so that we get more AI in the hands of the physician,” Fuchs said. “Physicians and nurses are completely overworked, and we need these tools urgently to help them to just keep up with the amount of work that's necessary.”

CLEVELAND CLINIC

Rohit Chandra, PhD, Cleveland Clinic’s chief digital officer, noted that a large part of the health system’s AI and data analytics focus has centered on predictive analytics and AI as an assistive technology for clinicians.

In 2022, a significant portion of these efforts has been operationalizing predictions that can be applied within the context of clinical care. An example of this is Cleveland Clinic’s ongoing work to predict the onset of sepsis to enable early intervention, Chandra stated.

Looking ahead, the health system’s plans for AI in 2023 are three-fold: infrastructure, expertise, and prioritizing patient safety and clinical care quality.

“[No.] 1, we want to invest in technology so that leveraging AI becomes easier and easier. And that's a question of do you have the data? Do you have the infrastructure? Is it made available to AI researchers with some confidence that the data is quality represented? So…think of it as technology infrastructure to do AI more easily,” Chandra said.

Category No. 2 involves building on that infrastructure and developing AI in-house by partnering with health technology companies or other stakeholders with AI expertise through a streamlined process, he continued.

“The third is that AI is obviously a very broad capability and has value in a variety of settings,” he added. “The third area that we [will] look at is how do we prioritize and fit the [AI] areas that [have] direct line of sight to patient safety and quality of clinical care. And starting with those kinds of problems is sort of the most exciting because then with those, you can test, refine, and have an impact sooner rather than later.”

The opportunity that these efforts can create to improve healthcare is a major driver for Chandra’s work at Cleveland Clinic.

Despite their varying approaches, health system leaders shared excitement for what the next year of AI and analytics innovation holds for healthcare.

“What gets me excited is there's a win-win to be had for the industry,” Chandra said. “I think that if I look at the state of AI in other industries, it's much further along. Healthcare is different because, obviously, the risk profile is different… But sitting in my shoes at the Cleveland Clinic, I actively solicit win-win partnerships where there are companies whose core competence is AI, whereas our core competence is providing the best clinical care. And I think that that's where a win-win is to be had, where we can together solve problems that…it's harder for us to [solve] just by ourselves.”

Editor's note: This article was updated on Dec. 29, 2022, to clarify that Mayo Clinic is readying its 12-lead ECG algorithm for FDA approval.