Applied AI: Shaping the Future of Personalized Health
The explosion of clinical, genomic and imaging data, combined with breakthroughs in machine learning, is creating new opportunities to personalize health care and improve the patient experience.
Personalized medicine is an approach to patient care in which physicians make treatment decisions based on the synthesis of genetic information and risk factors specific to a patient. Data analytics and artificial intelligence (AI) are important components of personalized medicine because they can support physicians to tailor treatments based on evidence.
Recently, during a roundtable sponsored by GE Healthcare at the American Hospital Association's Executive Forum in Chicago, leaders from four prominent hospital systems gathered to share their initial forays into AI for personalized medicine and perspectives on the technology’s potential to improve health care. They discussed the implications of AI on today’s care delivery system and how it could benefit, disrupt and challenge the physician-patient relationship.
The term AI has become a buzzword that is often overhyped. But in ways large and small, it is quickly gaining acceptance in health care. AI is a computer’s ability to mimic such human characteristics as learning, problem-solving and acquiring knowledge. As health care organizations take advantage of rapid advancements in AI and access to many different types of data, cloud technologies and automation, they could transform health care delivery in the near future.
The opportunity: Six areas in which AI could rapidly transform health care
Hospital executives at the forum identified at least six areas of opportunity for AI to transform processes and interventions. They are:
- Inpatient care quality to reduce readmissions and “never-events” such as falls.
- Patient triage and care pathway support.
- Outpatient monitoring and support.
- Click-based data entry tasks associated with the electronic health record (EHR).
- Revenue-cycle and supply-chain management.
Carilion Clinic, an eight-hospital system based in Roanoke, Va., is already using a predictive analytics algorithm to more accurately identify patients at high risk for readmission within 30 days of discharge and those at high risk for inpatient falls. Steve Morgan, M.D., senior vice president and chief medical information officer at Carilion Clinic, said metrics for both readmissions and falls have improved with the implementation of these tools.
Atlantic Health System, a nonprofit system based in Morristown, N.J., created a Center for Business Intelligence that brings all claims into a central database to study ways to leverage AI to improve patient care and efficiencies. Atlantic Health is in discussions with Amazon to explore how Amazon’s cloud technology could help support AI applications, said Tom Kloos, M.D., vice president of Atlantic Health System.
Kloos is also president of Atlantic Health System’s accountable care organizations (ACOs). The system has about 400,000 beneficiaries in three ACOs, with 2,700 participating physicians at multiple locations. Since 2012, when the system formed its first ACO, it has used data analytics to maximize the information collected through ACO participation, he said. With Medicare claims data from participation in the Medicare Shared Savings Program, Atlantic Health would like to hone in on the best way to integrate inpatient and outpatient services to improve patient retention, the care experience and outcomes, and AI could be useful in this respect, Kloos said.
Morgan of Carilion Clinic agreed that AI could provide enhanced care-decision support to more accurately direct patients to care pathways inside the organization.
“We have a lot of variability in the way we approach things within the ambulatory and inpatient side,” Morgan said, citing the example of developing personalized algorithms for patients with heart failure to guide them to care pathways inside the health system depending on the severity and level of support needed.
Rob Nelson, global marketing director for AI and analytics solutions at GE Healthcare, said an area of interest is to focus on prioritizing the review of critical care cases. One example would be pneumothorax, where an AI algorithm is embedded in the X-ray system and is designed to speed the detection of a collapsed lung.1 “We have many use cases in the imaging workflow, as well. We have monitoring devices, and we see that it becomes a part of the workflow, it makes things easier, takes steps out, makes it more accurate,” Nelson said.
Stephen Scogna, CEO of Northwest Community HealthCare, a suburban Chicago hospital and medical group based in Arlington Heights, Ill., said his organization is interested in using AI to conduct triage in primary care. Northwest Community has expanded its ambulatory care in recent years, Scogna added.
“We continue to strive to be value-based and very cost-efficient in the way that we deliver our care,” Scogna said. “We've been very successful, but we're at the point now where we need to become more aggressive and more creative in the way that we get to that cost savings, and so we see AI as a significant opportunity.”
“Alexa, call my doctor.”
All four health system executives who participated in this discussion see great promise in using voice-controlled smart speakers to improve support of providers and patients in a variety of ways. The Amazon Echo and Google Home, two of the biggest players, allow users to converse with AI-enabled platforms (known as Alexa and Google Assistant, respectively) to ask questions and perform such tasks as setting alarms and making phone calls. Alexa and Google Assistant are possible because of machine learning and natural language processing (NLP), where computers learn to interpret and respond to human commands and speech. While still imperfect, Alexa and Google Assistant are learning all the time, and with that learning will come more ways to apply the technology, including in health care.
A big opportunity for smart speakers could be in medication adherence, said Eric Wenke, corporate vice president of enterprise transformation at Baptist Health South Florida in Coral Gables. Today, frequent patient interactions that require a nurse or other clinician can be cost-prohibitive, Wenke said. But these interactions “would be cheaper and easier … with an AI-enabled world,” he added.
Smart speakers could be used to expand home-monitoring programs by reminding and prompting patients to step on a smart scale, take their blood pressure and conduct other daily health-monitoring tasks at home, Wenke said. Medication adherence is an aspect of AI-supported home care. Data gathered by the home speakers potentially could be uploaded automatically to the patient’s EHR, without human intervention, he said.
GE Healthcare’s Nelson said the company is focused on finding ways to network many smart devices at the point of care for more seamless and secure information sharing and personalization.
Atlantic Health’s Kloos said he sees smart speakers or a similar AI-enabled device being instrumental in untethering physicians from click-based tasks driven by today’s EHR design, which requires a tremendous amount of data entry during the patient care visit. “I think that, frankly, there is a huge opportunity to bring that to fruition,” Kloos said.
Morgan of Carilion Clinic agreed, suggesting that in the next few years there could be “a kind of Google Assistant or Alexa sitting in the physician’s office and able to discern a conversation, create a note and apply that feedback.” He added that using smart speakers in this way could improve productivity and reduce burnout in family medicine.
Other new types of data collection enabled by AI that don’t require traditional data entry also hold potential to improve the patient care experience and conduct personalized medicine, including smart watches like the Apple Watch, said Wenke. He noted that the Apple Watch already has the capability to detect irregular heart rhythms.
Who will pay for AI-enabled care?
Beyond the headline-grabbing, consumer-facing AI applications, the executives saw an opportunity to use AI to streamline administrative and other back-end office functions such as billing, supply-chain management and identifying patients for insurance enrollment.
Morgan of Carilion Clinic said that his system has already begun exploring robotic processes in revenue-cycle management as well as researching predictive algorithms for patient outreach for insurance programs.
“When we’re looking at this, we're trying to apply it across the board because there are so many more opportunities, certainly from my perspective, than the clinical ones,” Morgan said.
Forum participants said they expect AI to support clinicians to make care decisions but not to replace human caregivers altogether.
Baptist Health South Florida is conducting a pilot program in which three emergency medical services (EMS) teams do follow-up home visits with patients. These teams have a high level of trust with patients in the community, making them a natural resource for high-quality, follow-up care, Wenke said. Baptist Health has discussed how to apply AI to these visits for enhanced data collection and monitoring, he said. Today, there is no reimbursement mechanism for the visits, so Baptist Health is funding the EMS team visits.
Similarly, Atlantic Health System is exploring using AI for patient education, said Kloos. The idea is that avatars or other AI-enabled applications that are specifically designed to engage patients can be used to help establish patient goals and aid their understanding of treatment options.
Who will pay for the implementation and delivery of AI-enabled patient care is an issue that the executives said has not been resolved. It’s a sticking point that potentially could slow adoption of AI applications, they said.
“I think the outlier is the payer,” said Scogna. “We know the pool of money is not going to change. What are the government and the Blue Crosses of the world going to do? Will they encourage this disruption as a positive thing, or will they consider it a threat?”
Streamlining claims could spur adoption, suggested Kloos. “Truth be told, it takes six to nine months to bring in one claims file from one payer,” he explained. “Performance is 10 months behind strategy. … So, I think where AI has a real opportunity is the ability to take the data and not worry if it is missing a decimal point and just normalize the data for more rapid process.”
Choosing the right partners to implement AI-enabled programs requires some thought into the tolerance for risk, budget and how the technology will help meet strategic goals, the executives said.
“We’re trying to figure out who to partner with and how to partner,” said Morgan. “Health care is still local. Our patients still view it as local, and the solutions we’re looking at we want to apply locally.”
Thinking about the local needs of each community gives pause when considering whether to implement these big ideas in AI with multinational, powerful partners, he said.
“I’m not convinced yet that my patients would respond to Google health care, for instance,” Morgan said. “I think they still look to a local health system to provide that.”
Health care delivery has largely been driven by nonprofit systems, but the sector is now a target for disruption by large for-profit companies including Amazon, Apple, Google and retailers like Walgreens and Walmart.
“We’re sort of edging our way there, from a business perspective, and we’re not thinking about the implications,” Scogna said. “What happens to caring for communities that can’t afford it? As you start to pick your partners, understand they will have skin in the game, and that’s going to lead to greater and greater pressure on the concept of nonprofit care.”
Still, that concern won’t deter hospitals from seeking opportunities to leverage AI to improve efficiencies, streamline processes and ultimately create a better, more personalized care experience, the executives said.
“For me, at least, we're talking about it and the AHA [American Hospital Association] is talking about it and getting people to think about it,” said Morgan.
- 510(k) pending at U.S. FDA. Not available for sale in the United States.