How CIOs are prioritizing AI investments for the next 5 years
In the first installment of Healthcare IT News’ newest feature series, five hospital and health system IT leaders describe how and where they’re planning to spend on artificial intelligence and machine learning.
While the pandemic is still raging, the chaos of the past 18 months has calmed a bit, and the dust is starting to settle. Now the time has come for healthcare CIOs and other health IT leaders to look forward and plan their IT investments – shaped, in no small part, by the lessons of the recent past.
According to new research from HIMSS Media, the average overall 2021 IT budget is nearly $13 million, with 15% on average being allocated to IT security. While that may be a lot of money, there are many technological areas yearning for more investment.
Today, Healthcare IT News launches a new feature article series, Health IT Investment: The Next Five Years.
We speak with health IT leaders, primarily CIOs, to learn the path forward through the priorities they set with their investments in six categories: AI and machine learning; interoperability; telehealth, connected health and remote patient monitoring; cybersecurity; electronic health records and population health; emerging technology; and other systems.
This first feature focuses on AI and machine learning. The top IT executives sharing their plans for the next five years in this first installment in the series include:
Dr. Shaun Grannis, vice president for data and analytics at Regenstrief Institute in Indianapolis.
Dr. J. Michael Kramer, chief medical informatics officer at OhioHealth in Dublin, Ohio.
Mike Mistretta, vice president and CIO at Virginia Hospital Center in Arlington.
B.J. Moore, CIO of Providence, which operates 52 hospitals across seven states – Alaska, Montana, Oregon, Washington, California, New Mexico and Texas.
Michael Restuccia, senior vice president and CIO at Penn Medicine in Philadelphia.
Everyone from board members and C-suite officers to IT managers and workers in the trenches will learn valuable information and guidance from their peers interviewed in these features.
Ambient, imaging, predictive analytics
Moore and his team at Providence have been making significant use of AI and machine learning. It’s an area that will see increased investment over the next five years.
“We use it pretty broadly, including ambient artificial intelligence in our partnership with Nuance,” said Moore. “That allows caregivers to practice their art of care delivery, while the ambient technology makes the updates, versus they’re having to type into the record. We’re evaluating artificial intelligence for imaging, particularly detection of cancer. There are a number of early detection pieces we’re doing with imaging.
“With predictive analytics, we’re using machine learning quite extensively,” he continued. “Everything from re-admittance, to no-shows, to acute care scenarios. That will help enhance patient delivery, decrease costs [and] allow us to more fully use our space.”
The provider organization has been using machine learning for predictive analytics extensively to meet surges and declines with COVID-19, and it plans to continue to invest in areas like that.
“And then we’re working with other companies like Truveta,” he added. “It’s a data consortium that we’re a founding member of, and that partnership involves using AI and machine learning on things like data normalization to help gain insights.
“There are a number of things we’re doing there that are confidential,” he noted. “We see machine learning as a really big-breakthrough set of technologies for us. We’ve been spending the last two-and-a-half years creating the foundation to allow us to do these things. My three strategic pillars apply here. They are: simplify, modernize and innovate.”
Moore does not foresee any issues selling more machine learning investments to the rest of the C-suite and the board.
“They know my background,” he stated. “I came in two-and-a-half years ago from 27 years at Microsoft. So people trusted they made the right hire. So when I talked about big data and machine learning and AI, they didn’t push back. Over time it went from, ‘Hey, we support B.J.,’ to, ‘Wow, this is a really big game-changer.'”
New to the technologies
Virginia Hospital Center has only just begun to work in the area of AI and machine learning.
“Honestly, I’m not sure what we will be doing in this space,” said Mistretta of Virginia Hospital Center. “We have our toe in the water, so to speak, with sepsis agents and some other predictive analytics currently, so we will see what the organization’s receptivity is. Our issue isn’t really selling investments to leadership. The challenges are below that level with adoption directly.
“There are two large challenges I see with this technology,” he continued. “First, you have to have a tremendous amount of data built up to be able to run the algorithm against, then have a method to validate the results. And second, are the users/clinicians ready to be challenged on their previous mental models of care delivery, and are they open enough to [considering] assistance from technology?”
For the first issue, Mistretta notes the organization has been running its Epic EHR for three years for various models “in the dark,” collecting enough data to have the proper use cases.
“Interestingly enough, when our users came forward looking for solutions, we were ready to activate these models and start validating results with them,” he said.
“Today we are producing dashboards for our different use cases to their respective stakeholders with what I consider mixed results. While leadership agrees for the most part with the findings, some have grabbed them and run, while others still need significant hand-holding to incorporate the AI into their workflow.”
For the second issue, Mistretta believes there is significant work to be performed convincing the clinical community the value of wherever AI is implemented to assist them.
“This technology has existed in imaging for years now – over-reading mammography, for example – and just becoming mainstream,” he observed. “We have a Philips EKG system that performs pre-reads for the cardiologists. They ignore it and have asked us to turn [it] off, regardless of the amount of tuning or education we provide.
“We are just, in the last three to six months, getting receptivity on our sepsis agent being used mainstream in patient care, even though it has been running for several years,” he added. “My gut tells me we are going to need the next five years to help the clinical community with the acceptance of AI and machine learning while we continue to work on improving it.”
Some clinicians simply require time to digest and understand data in-depth prior to accepting how it can impact their care delivery, he said.
Significant investments in AI
Regenstrief Institute, an internationally respected informatics pioneer and a key research partner to Indiana University School of Medicine, is making significant investments and advancements in applying AI to healthcare.
This area will continue to be a significant focus over the next five years for the organization, said Grannis of Regenstrief Institute. AI has tremendous potential to greatly improve the delivery of patient care, he added. Regenstrief focuses on machine learning, data mining and natural language processing.
“Current projects show encouraging proof of concepts, including developing and testing an award-winning machine learning system called Uppstroms,” he noted. “Evidence suggests that at least one in four adults, and possibly as many as one in two, have a need driven by social determinants of health.
“The project addresses patients’ socioeconomic, behavioral and financial needs by incorporating SDOH – factors such as accessibility to healthy foods or availability of safe, affordable housing,” he said. “Combining SDOH information with the EHR, the algorithm identifies primary care patients who may need wraparound services such as those provided by a social worker or counselor, allowing care providers to make referrals before the situation turns into a crisis.”
Employed in nine clinics within an extensive safety net health system in Indianapolis, Uppstroms can be integrated into EHRs and could be used in various healthcare settings to address SDOH, he added.
“In the same way, NLP can be used to tap into unstructured data within the EHR,” he explained. “Further, this technology provides a great tool for researchers, clinicians and healthcare administrators to identify cohorts and analyze trends to inform clinical and administrative decisions. For example, through NLP, Regenstrief research scientists created the largest chronic cough cohort to date, showing the promise of this approach.
“AI also can be used for clinical decision support – sorting through the plethora of available data and displaying only the most relevant,” he continued. “For example, an app developed by Regenstrief and partners called Health Dart is deployed in emergency departments of a university health system.”
The app sorts through the EHR and identifies relevant tests and information related to seven of the most common ED patient complaints: chest pain, abdominal pain, headache, weakness and dizziness, back pain, pregnancy, and heartbeat irregularities and trouble breathing. This novel search algorithm saves clinicians several minutes of clicks and searching, allowing them to spend more time with the patient and work more efficiently.
“These technologies have been demonstrated to work in real-world settings, and our teams will continue to refine and improve these tools and devise new ones,” Grannis said. “In addition, their demonstrated success will be used to encourage investment from health systems.”
Continuing to invest, but learning from experience
Penn Medicine has made a significant investment in the deployment of a common electronic health record to support all of its inpatient, ambulatory and home care operations. The initial goal of this objective was to ensure that all caregivers were providing services from the same system, so that patient data was centrally stored and easily available to all caregivers in all locations.
“With this goal achieved, a second objective emerged that focused on further improving patient care and efficiency by driving value out of the acquired patient data,” said Restuccia of Penn Medicine. “Our initial experience with AI and machine learning has been both fruitful and frustrating.
“In selected instances, we have been able to tie together a variety of data elements in order to provide insights regarding changes in patient clinical protocols, patient follow-up treatments and more timely disease recognition,” he continued. “These positive results have come with an equal amount of effort that [has] generated less than desired results.”
This experience typifies the state of artificial intelligence within the healthcare industry, he contended.
“Things are not as binary in healthcare as they may be in other industries,” he said. “As a result, we continue to invest and learn from our experiences. Our approach to learning – particularly in this area – is to use a hybrid model of self-developed algorithms, along with more generally available algorithms provided by industry vendor partners.
“This approach allows us to zero in on niche areas within the health system through our self-developed efforts, while leveraging vendor developed algorithms to address more broad areas of opportunity,” he added. “This combined approach ensures we are taking advantage of the skills that are resident within Penn Medicine, as well as across the industry.”
A three-year board goal
Kramer of OhioHealth reveals the health system is investing significantly in AI and machine learning. In fact, it is part of a three-year board goal.
“Starting in 2019, we made it a goal not just to turn on AI and predictive models, but to show they were used enterprise-wide and driving positive outcomes,” he explained.
“We looked at more than 50 different opportunities and models. Ultimately, we identified 17 workflows where AI could be effective and [was] likely to be successful in achieving positive outcomes. In addition, our board goal pushed us to develop the expertise and rigor needed to achieve and sustain positive outcomes from AI.”
In this work, staff learned three things, he said:
Workflow is key. AI will not solve for variation or be a solution when the other aspects of workflow are not considered.
Problems and models must be carefully selected to be successful. Models that were most successful reduced significant manual burden, as in automating documentation or review of complex data.
Managing AI requires new expertise and rigor. There is significant expertise needed designing, implementing and sustaining benefits. The tools and science are far more complex than logical decision support rules. The health system needed a central expert team and tools.
“OhioHealth built a hub-and-spoke model where there was a central data scientist team, clinical informatics, program management and ongoing monitoring,” he noted. “The various business units and clinical project teams led the business case, workflow design and change management.
“We now are in year three of our board goal and have 17 models moving down the pipeline,” he continued. “Nine are live. Of the 17, eight were provided by Epic, four were from non-Epic vendors and four were internally developed. We expect that at least 10 will have measurable positive outcomes by the end of the year. Some models are harder to measure in the short term, but the clinicians perceive the value of the AI in their decision-making processes.”
Source : Healthcare
