Healthcare Analytics, Population Health Management, Healthcare Big Data


Artificial Intelligence Promises a New Paradigm for Healthcare

Artificial intelligence is poised to overhaul the healthcare industry, bringing success to those who can adapt quickly to the new care delivery paradigm.

Source: Thinkstock

The speed with which artificial intelligence has become an inescapable component of every new technology offering is either highly encouraging or deeply troubling, depending on the observer’s zeal or trepidation around the notion of integrating data-rich algorithms into the complex and highly personal practice of caring for patients.

Unease over AI is still common, and perhaps somewhat justified as researchers start to turn well-controlled pilots into commercialized deployments of diagnostic tools, clinical decision support systems, and workflow optimization aids.

Many of these offerings must still earn the trust of clinicians, especially those who question the underlying integrity and potential biases of the data upon which these algorithms were trained.

Yet many more are showing truly astonishing results that are leaving providers clambering for access, eager to take advantage of more intuitive workflows while gaining the ability to harness the knowledge of thousands of their colleagues and experiences of millions of patients.

Regardless of where any individual or institution falls along the enthusiasm spectrum, it is becoming increasingly clear that nothing is going to stay the same once the healthcare industry hits its AI event horizon – and that moment may be coming very soon.

“We’re seeing an ever-increasing number of cases that have the potential to truly transform the way providers care for patients,” said Samuel Aronson, Executive Director of IT at Partners Personalized Medicine.

“And there is a great deal of activity around open-source health innovation platforms that are focused on making it cheaper and more efficient to build, secure, validate, deploy, share, and ultimately network these applications across institutions.”

“What we’re really focused on now is the last-mile problem associated with enabling the clinical process transformations required to get algorithmic-enabled care more broadly rolled out in healthcare.”

A staggering number of those potentially game-changing innovations were on display at the World Medical Innovation Forum (WMIF) on Artificial Intelligence, presented by Partners HealthCare. 

From interpreting lab tests, simplifying in-vitro fertilization, and personalizing cancer care to monitoring surgical video in real-time or aiding in the diagnosis of pneumothorax, the creativity and ingenuity evident in the flourishing AI research community is both astounding and heartening.

“We’re seeing an ever-increasing number of cases that have the potential to truly transform the way providers care for patients.”

If brought to scale and implemented appropriately, each of these concrete solutions for specific use cases is likely to save dozens, hundreds, or thousands of lives over time. 

Such accomplishments will certainly be nothing to sneeze at.  But artificial intelligence promises the healthcare system something that cannot be counted on the scale of individuals.

Machine learning, deep learning, neural networks, natural language processing, and all of the other components of the AI ecosystem are poised to bring about a complete change in the paradigm, from how doctors are trained to how they make decisions to how they deliver care.

The challenges and opportunities of bringing AI to healthcare

The AI revolution is already well underway in other industries, panelists at the three-day event pointed out repeatedly. 

Mentions of autonomous vehicles popped up in nearly every session, while references to how consumer technology giants like Apple, Amazon, Uber, and Google integrate AI into their smartphones, search functions, apps, and other offerings were nearly as frequent.

"Reflecting on the Impact of AI at the Bed and the Bench: Chairs Roundtable"

Source: Partners HealthCare

Innovation in these sectors is moving at lightning speed due to high demand and fierce competition to become the lifestyle platform of choice for consumers craving convenience and on-demand services.

“In the consumer space, if you pull out your smartphone and search for an object inside your pictures, you can identify thousands of different features,” explained Keith Dreyer, DO, PhD, Chief Data Science Officer at Partners. “We don’t think of the fact that artificial intelligence is doing that.  We just see a smarter phone.  When AI works, you don’t call it AI anymore.”

Healthcare is uniquely positioned to take advantage of what AI has to offer, and many members of the care continuum are eager to see data-driven changes to the way they are now delivering care. 

“When AI works, you don’t call it AI anymore.”

Few clinicians argue with the assertion that the recent explosion of health-related data has turned every decision into minefield of uncertainty.  Providers must agonize over what they might be missing among the gigabytes of information they’re supposed to consider. 

And no matter how much data seems to be available, it is difficult to be sure that the information on the screen is complete, up-to-date, and accurate.

Coupled with outcomes-based reimbursement structures, increasing pressures on their time and attention from documentation requirements, and the ever-present threat of legal action when a decision goes awry, it seems a natural fit to bring some algorithmic intelligence into the process.

But the healthcare industry is structured much differently than the environment in which a handful of purely profit-driven behemoths now roll out comprehensive AI-driven updates at their user conferences and trade shows. 

Healthcare’s mission is different, and its tech culture is, too.

There is no such thing as a “healthcare 3.0” operating system that can be force-installed into every workflow during an overnight system update.

The HHS and the Office of the National Coordinator have made some attempt to bring such an approach to the industry through the Certified EHR Technology (CEHRT) program, which requires participants in meaningful use – now known as Promoting Interoperability – to adopt the same set of core health IT functionalities from approved vendors.

But that process has taken years, and the version known as the 2015 Edition won’t even start to be required until 2019 – a lag in adoption that would be nearly unthinkable in any other industry.

“Getting to the point of complete [artificial intelligence] ubiquity is going to take a long time in healthcare,” Dreyer predicted, and he wasn’t alone in expressing reservations about the speed at which AI will be able to take hold. 

Many healthcare organizations perennially lack the “activation energy” to adopt new workflows, said Aronson, due to the daily demands of simply getting through the day.

“Everyone is running flat out,” he said.  “When you talk about bringing AI or any other technology into the organization, clinicians have to work with you to define the new workflow, and they have to validate it – often site by site – before they implement it.” 

“If they are working totally flat out, it’s difficult for them to find the time to invest in that, even if they know it’s going to save them time after the process is complete.”

Free time isn’t the only problem, added Neil de Crescenzo, CEO of Change Healthcare.

“In addition to the activation issue, there are some institutions that just don’t have the resources,” he said.  “Even if they could free up the time, they simply don’t have the expertise in-house to make it happen.” 

“At the same time, we’ve seen an incredible surge in the size of institutions over the past few years,” de Crescenzo continued.  “Many [health systems] now have fifty, seventy-five, or a hundred ambulatory offices all over the community.” 

Those newly expanded networks may still be struggling to create cohesion among disparate communities that could be using a dozen or more different electronic health records.  

“It’s difficult to find the time to invest, even if they know it’s going to save them time after the process is complete.”

Working to unify that fragmented environment is typically the top priority, and many organizations are simply not prepared yet to take the next step.

“In order to start thinking about how to deploy AI and machine learning consistently across all of those component entities, you need to be a sophisticated, mature organization.  Right now, it’s not practical for a lot of these institutions.”

Many of these organizations could benefit from partnering with technology companies that offer out-of-the-box services and comprehensive platforms that require little to no additional development from the client, he added. 

“What we’re seeing is that they’re asking us to start fishing for them instead of teaching them to fish,” he said.  “Maybe only for a little while – maybe in five years or ten years, they’re going to want to take it in-house.”  

“But right now, if they’re going to take advantage of what AI can bring them, they’re going to have to give it to someone else to do the things they don’t have the time or skill to do right now.”

Outsourcing the heavy lifting to a dedicated business partner could be a promising way for many organizations to reap the benefits of artificial intelligence without investing in becoming miniature innovation factories.

"AI in Hospital Environments: The Learning Provider"

Source: Partners HealthCare

The convergence of value-based reimbursement, changes in consumer attitudes, and an epidemic of clinician burnout has left providers who can’t develop their own solutions desperate to invest in tools and solutions that will help organizations recharge their “activation energy” batteries and restore a sense of joy to patient care.

Market analysts predict enormous growth opportunities for artificial intelligence products: a recent Research and Markets report forecast a 60 percent compound annual growth rate (CAGR) for the healthcare AI sector until 2022.

In the first quarter of 2018 alone, data analytics and clinical decision support tools, many of which include a machine learning or AI component, attracted nearly $1.2 billion from investors eager to pounce on the next “unicorn” company to rear its head above the herd.

In such a feverish environment, as non-traditional players enter the market and nearly every health IT vendor and developer under the sun seems to be offering an AI-driven product, it can be difficult for organizations to work their way through the promotion, the hype, and the exuberant marketing.

A healthy skepticism when evaluating new tools and platforms is essential, but so too is the recognition that it’s only a matter of time before the majority of health IT tools have at least some sort of AI thread running through them.

That future state of affairs is nothing to dread, the experts stressed.  Physicians and other care providers shouldn’t harbor too much trepidation about being made obsolete by omniscient, infallible robot colleagues who never take vacation days.

Sometimes we do talk in extremes when we’re discussing AI and machine learning,” said Dan Burton, CEO of Health Catalyst. 

“It’s either ‘we’re not going to have any physicians at all anymore – it’s just going to be the machines,’ or we swing the other way towards ‘it’s all horrible and we should never use machines.’  Of course, the truth is almost always somewhere in the middle.”

Uncertainty, and even fear, is a natural response to a concept as fundamentally revolutionary as artificial intelligence is promising to be, agreed Anne Klibanski, MD, Chief Academic Officer at Partners.

“But if we look back in history to the first Industrial Revolution, the exact same themes were there,” she observed.  “Machines will ruin everything; jobs will go away; people will never be the same.”

“And when we started exploring genetics and genomics, there were many, many people who voiced considerable concerns about what it would mean to be human when we had that kind of knowledge.”

“If we look back [at] the Industrial Revolution, the exact same themes were there. Machines will ruin everything; jobs will go away; people will never be the same.”

The human experience has certainly changed since the Age of Steam, and most would argue that the progress has been, on the whole, a positive development. 

Even in the few short years since the mapping of the genome, science has accelerated drastically – not necessarily to fundamentally alter what it means to be alive, but to offer new hope for patients and families around dozens of previously incurable or undiagnosable diseases.

Using AI to cut billions in spending, improve the patient experience

Within the workflows of physicians and care teams, artificial intelligence brings similar opportunities to augment, not supplant, the irreplaceable value of the clinical mind.

And in the administrative and operational spheres, hopes are incredibly high that AI can smooth out some of the most jagged pain points afflicting providers, patients, payers, and the care delivery process.

The trick to achieving a positive AI outcome will be to develop trustworthy, accurate algorithms that can be widely deployed to address major shortcomings in the status quo.

Many of those problem areas generate huge amounts of financial waste and leave patients with significant gaps in their care, said A.G. Breitenstein, Partner at Optum Ventures.

“AI is already out there to some degree, but it’s not great AI.  It’s being used in all sorts of consumer contexts with variable success.  We’ve all seen the problem of patients going to Doctor Google and scaring themselves with what they find, for example.”

“That’s part of a $50 billion problem of overutilization of the ED and underutilization of primary care,” she pointed out. 

“The solution to that isn’t necessarily telling people they should go to primary care instead of the ED.  The solution is intercepting the patient at the first point of contact – when they sit down to their computer – and using better algorithms to give them more clinically validated, more rigorous data in the places where they’re looking for answers.”

Using AI to get upstream of developing problems is a common theme across the research and development ecosystem. 

Preventing adverse drug events, alerting providers to patient deterioration, stratifying patients by risk for population health management programming, supporting chronic disease management, identifying metastasizing cancers, and predicting kidney function decline are just a small handful of the hundreds of use cases for predictive AI.

In the life sciences arena, artificial intelligence has the potential to carve out a piece of the $500 billion spent on wasted interventions, said Colin Hill, CEO of GNS Healthcare, to during the WMIF event.

“There’s at least half a trillion dollars wasted when patients aren’t getting the right drug, or they’re not in the right care management program, or they didn’t get the right procedure or medical device,” he said.

“That’s a staggering cost, and it’s a massive opportunity to change health outcomes by getting down to the fundamentals of slowing disease progression, reducing hospitalizations, and optimizing therapeutic effectiveness.”

Precision healthcare will be “the most important application” of AI, he asserted. 

“If we can better match health interventions to individual patients, we can improve outcomes and lower the total cost of care.  Part of that requires the discovery and development of new drugs.  And it also means discovering the biomarkers that can help stratify patients into responding and non-responding subpopulations.”

“We need to understand what works for whom, what the underlying mechanism is, how a biological system operates, and how an intervention impacts that.  Machine learning is critical for that process, because it helps us advance from grappling with the little pieces of what we think we know and presents more insights than humans can typically uncover.”

Genomic data and electronic health record data will be critical for that process, added Noga Leviner, CEO of Picnic Health.

“Genomics and the EHR are a perfect use case for AI,” she said.  “You’ve got huge volume, but you also have some weak signals within that.  Humans are just not going to be as good as algorithms are at teasing those signals out.” 

“As we layer more and more types of data on top of that, it’s only going to become more obvious that doctors using their brains alone aren’t going to be as good as brains combined with algorithms.” 

AI could cut even more costs and free up human brain power on the administrative side of the industry, said Change Healthcare’s de Crescenzo.

“It’s only going to become more obvious that doctors using their brains alone aren’t going to be as good as brains combined with algorithms.”

“I’m not sure anyone knows exactly how many people work in utilization management for payers and providers,” he said, “but no one has ever argued if I estimate it’s at least 100,000 people.  They’ve got an average salary of about $80,000 a year.  That’s $8 billion spent on managing utilization.”

“What if we could take a conservative estimate of 30 percent of those people and use AI to free them up to work on other things?  Many of them are nurses or other clinicians or caregivers – that’s $2.4 billion in direct savings, not to mention the savings from bringing more clinical expertise back to patients without adding net-new personnel.”

"Smart EHRs: AI for All"

Source: Partners HealthCare

Routine tasks like patient scheduling, supply chain management, and booking operating rooms or testing equipment could also benefit from some automation, said Katherine Andriole, PhD, Director of Research Strategy and Operations at the MGH & BWH Center for Clinical Data Science (CCDS).

“An MRI scanner is a very expensive piece of equipment, and letting it sit idle is like throwing money down the drain,” she said. 

“There are more occasions than we’d like where a patient shows up to have an MRI, but it turns out they’re contraindicated. Right now, the process for gathering that information before the appointment is to call them.  Sometimes the patient doesn’t know, or gets something wrong, and sometimes we don’t actually connect with them at all – but we still keep that appointment booked.”

“Why can’t we use data to start identifying the patients likely to have a contraindication before we put them through that entire process?  If we can use AI to predict a problem before we waste resources and time on both sides of the patient-provider relationship, we’re going to create better experiences for everyone.”

Improving the patient experience is an imperative that healthcare providers cannot afford to ignore, agreed Breitenstein.

“If we can use AI to predict a problem before we waste resources and time, we’re going to create better experiences for everyone.”

Leaving a positive impression on consumers will be especially as healthcare moves out of the hospital and into the ambulatory setting, the retail clinic, the smartphone, or the home.

“The original hospitals were places where people went to die.  They were the last stop.  We started to build care around that, which is at the root of a lot of our systemic problems,” she explained. 

“As we start to think about turning the hospital model into one that centers on managing health over time, that need for the physical plant changes dramatically.  The central role of the hospital is starting to break apart, which enables an entirely new wellness environment – an entirely new driver for preemptive care.”

“We don’t have to wait for the patient to get sick and present themselves anymore.  Now, we can intervene before they end up in the ED or go to see the specialist.  Illness is usually detectable to an algorithm before it is detectible to a patient.  The fact that we wait long enough for someone to acknowledge that they should get some help is an artifact of the traditional notions we have about healthcare.”

Flexible care delivery and the ability to get further and further upstream of costly diseases that reduce quality of life for patients may help to salvage some of the consumer goodwill that has drained away due to cumbersome interactions with their providers and mounting costs.   

“One of the places we find the most stored-up kinetic energy is around consumer frustration,” Breitenstein said.  “To be honest, satisfaction with the healthcare system is about on part with the prison system.  In addition to that, we have people paying $1000 or $1200 out of pocket every time something goes wrong.”

“When you have a very information-aggressive cohort of young people starting to age into their first experiences with illness on that scale, it’s going to drive a lot of change.  We need to move towards a more on-demand system that allows for easy payment, access to scheduling, and more of a self-guided experience of the healthcare system.  AI can help us enable that.”

Charting a new path towards success with artificial intelligence

As artificial intelligence becomes more deeply integrated into these areas of opportunity, healthcare providers will have to make at least a few changes to the way they interact with their technologies and their patients, the experts cautioned.

“Remember that only a few short years ago, biostatistics and data analytics were considered new skills for doctors to learn,” said Klibanski. 

“AI is going to be a fundamental core curriculum that physicians are going to need to know.  They need to know what to trust and what not to trust.  And they need to really understand what they’re being asked to do.  Without those capabilities, they’re either going to trust everything or deny everything, and we need to be somewhere in between.”

Finding the middle ground between faith and discernment will be critical for ensuring that clinical decision support tools are most effective. 

“They need to know what to trust and what not to trust.  And they need to really understand what they’re being asked to do.”

When users of these tools find that optimal balance, the impact could be staggering, said Burton of Health Catalyst.

“Some of the most interesting products coming out are clinical decision support tools that inform the perspective of someone like a radiologist,” he said.

Every diagnostician consumes information slightly differently based on their unique experiences and training, which could lead to disagreements in complex cases or the need to consult colleagues when an unusual presentation appears, he explained.

“How much could those conversations be improved with input from algorithms that are trained on 100,000 patients over here or 200,000 over there – and then 50,000 patients that had a slightly different set of characteristics that present a third option you might not have thought of?” Burton queried. 

“To me, that is something that marries the decision-making strength of a clinician with the scale that technology can bring.”

Integrating AI results into consults with colleagues might only be the first step in changing the foundations of clinical practice.

Klibanksi even hinted that the entire medical education system might be in line for a makeover as artificial intelligence starts to reveal novel associations between genetics, biological systems, and the interventions applied to patients.

“We have traditionally trained primary care doctors, specialists, and subspecialists as different areas and different disciplines,” she said. “Everything is very organ-focused or disease-specific.”

“But once we have a very broad and perhaps different understanding of diseases and disease pathways, we might actually think about training physicians in a whole different way. Some of the traditional disciplines may not accurately match with the way we’re going to approach diseases in five or ten or twenty years.”

Healthcare providers won’t just have to interact differently with their data, their textbooks, and their colleagues, said Joshua Gluck, VP of Global Healthcare Technology Strategy at Pure Storage. 

Whether developing their own tools or purchasing platforms or services, healthcare organizations will need to reevaluate their partnerships with legacy technology providers to achieve the level of scale required to succeed.

“It’s important to find a strategic partner that can really support these discoveries and do it in an innovative way,” Gluck said.  “Having an infrastructure and a data platform to build upon and scale to the magnitude of data points and disparate datasets that they need to combine to generate those insights is vastly important.”

“Some of the traditional players in that market space are taking the technology they’ve developed over the past several years and trying to get that to scale out.”

“You can do that for a short amount of time, but just the amount of data and the amount of systems that have to be integrated to support some of these workflows and pipelines, especially in the genomics space…if you don’t rethink the way these structures work, they just won’t scale.”

Organizations that commit to overhauling their infrastructure in addition to embracing innovative workflows and evolving relationships, are likely to be among a select group of entities walking along the path to success, said Aronson from Partners HealthCare. 

“There are two different futures in front of us,” he said.  “In one future, we take these tools that are incredibly powerful, and we develop them based on traditional business models.  That leads us to new infrastructure that does represent a substantial jump forward.”  

“But once that happens, I believe we will just wind up at a different plateau based on new proprietary systems that will replace the old proprietary systems.  Innovation will slow down again.”

“The other future depends on everyone putting in the significant amount of effort required to fund and deploy open business models.”

Those models may be more complex, and Aronson acknowledged that they may be risker for businesses to engage in. 

However, “I do believe they will also be much more profitable,” he said.

“If we can figure that out, then we wind up with a scenario where we will all be able to innovate on top of each other’s work,” he envisioned.  “We’ll be able to use as much open data as possible, and we’ll start to be able to think about a truly continuously learning healthcare system.”

Healthcare stakeholders have the opportunity, right now, to decide which future will win out, he stressed.

“The future depends on who’s in the room and what they decide,” he said. 

“It depends on as many of us as possible standing up and advocating for open, innovative models.  They might require more time to put in place, and they were require a lot of funding, but I truly believe they could bring some amazing benefits to humanity.”

This article was originally published on May 10, 2018.


Join 25,000 of your peers

Register for free to get access to all our articles, webcasts, white papers and exclusive interviews.

Our privacy policy

no, thanks

Continue to site...