OBR hosted a webinar that featured a discussion about how oncology practices can leverage artificial intelligence (AI) to improve cancer care.
Yesterday, OBR hosted a webinar that featured a discussion about how oncology practices can leverage artificial intelligence (AI) to improve cancer care.
Presenters included Amy Valley, PharmD, BCOP, vice president of clinical strategy and technology solutions at Cardinal Health; John Frownfelter, MD, FACP, chief medical officer at JVION; and Sibel Blau, MD, president and chief executive officer of the Quality Cancer Care Alliance Network. The discussion sought to answer questions such as: What are the types of AI oncology solutions within the market? How is AI being applied across stages of illness and care settings? How is AI driving smarter spending and improved quality of care?
Frownfelter began the discussion by explaining the potential importance AI holds in healthcare. “The maturity model of analytics is extremely important… 5 to 10 years ago, if we were really lucky, we’d have a retrospective report on a patient that explains what was happening with them last month instead of just last year. But that’s still not great because that doesn’t account for where patients are at today. As analytics get more sophisticated—a higher level of maturity—we start to get predictive modeling, and that can be a form of AI,” said Frownfelter.
However, right now, healthcare is decidedly behind in the AI revolution.
Frownfelter said this can be attributed to multiple causes, such as a significant resistance to change—and for good reason—as lives are at stake. There is also a barrier in the form of an unwillingness to share data, as well as the legal barrier of sharing data that the Health Insurance Portability and Accountability (HIPAA) represents. This being said, there remains great potential for AI in healthcare.
“Right now, there have been many federal reimbursement changes in oncology. We’re moving from a traditional fee-for-service system where volume equals profit, to a value-based system where you have to be frugal and thoughtful for who and when you initiate programs. This raises the opportunity for AI to support those requirements to be thoughtful and smart. Over the next 5 years, there will be a huge investment in AI,” said Frownfelter.
The ideal use for AI in healthcare, as Frownfelter sees it, is to leverage it to identify individual patient risks, inform the clinical team what those risk factors are, and then rank the interventions that are most effective for that individual patient.
Valley explained that one of the most important parts about AI is the ability to drive quick results that have a return on investment.
“Using the same Eigen platform for each unique patient population ensures the same performance that has delivered proven results across the hundreds of providers. The advantage of prescriptive analytics is the speed. There’s no need to create a predictive model from scratch. We strive to take the established machine and point it to pertinent questions to an oncology population. That speed is important because as it continues to learn, every time you want to ask it a new question you don’t have to start from scratch, and it provides actionable insight,” she said.
Valley’s employer, Cardinal Health, worked with JVION to create 7 questions, also called vectors, to enable their machine learning to identify at-risk patients. Together, the organizations trained the algorithm to identify patients at risk for:
“In the JVION machine, there’s over 10 years of data. The machine has a wealth of data from a variety of sources, and when we superimpose practices’ [electronic health records] EHRs onto the machine, it only takes a few weeks to gain insights,” said Valley.
She explained that critical data such as social and behavioral determinants of health that may sometimes be invisible to healthcare providers are listed and married with the clinical data. Importantly, these insights can be accessed through a portal, or integrated into the EHR workflow.
The machine will display a list of patients who are at risk for each indication, and why they’re classified as such. It also provides a list of actionable interventions in an order ranked by what is most likely to be impactful to the patient, however, it does not yet tell a healthcare provider what drug treatments to give a patient. These interventions can be customized for a practices’ specific clinical pathways.
Blau provided an overview of her experience with the machine learning during a pilot program across her hospital system. The pilot began in May 2018, she explained, and healthcare providers started to see results in June 2018, with positive results still coming in this past month.
“We implemented this workflow by having a patient coordinator screen the patient-centric dashboard to identify high risk patients. This information was forwarded to case managers and providers who then followed closely by case managers and care coordinators. For example, patients identified on the 30-day mortality vector are set up for an advanced care planning visit with a provider. Their oncologist is notified of the patient’s risk status, and the patient is tracked by a patient navigator until advanced planning is complete,” said Blau.
Overall, Blau reported that her health system found that 6-month depression detection increased by 22%, and 30-day pain management was reduced by a 33% reduction in moderate and severe pain.
“I’m most impressed with the 30-day mortality findings… in our pilot study we found that 45% of patients passed away within 30 days of being identified at high risk. Due to this finding, we’ve increased our hospice referrals,” said Blau.