Artificial Intelligence Is Still a Long Way From Understanding the Subtlety of a Texas Metaphor

Publication
Article
Oncology Business News®September 2017

It is not far-fetched to believe that, one day soon, automated systems will be able to obtain payer approval for therapies without any human involvement.

Debra Patt, MD, MPH, MBA

It is not far-fetched to believe that, one day soon, automated systems will be able to obtain payer approval for therapies without any human involvement, says Debra Patt, MD, a member of the The US Oncology Network breast cancer research committee. “I think that’s coming and will be incredibly useful to patient care.”

It’s something that payers want as well as oncologists. The bottom line is that payers want improved resource management, and it is not their intention to deny claims and preauthorization requests simply for the sake of saving money, Patt says. “Payers aren’t trying to put inappropriate barriers between patients and the care that they need. Those delays are the only way they know how to make sure that utilization is appropriate. If we find a way to make sure that utilization is appropriate by getting payers the data faster, that serves everyone, and so I think there’s a willingness to participate in those kinds of collaborative solutions.”

Patt, who practices in Austin, Texas, recently discussed some of the limitations and advances in electronic health reporting with Oncology Business Management™. Much of the initiative these days is focused on drawing information from electronic health records (EHRs) without human involvement. This relies on the data being in a structured format that computer systems can recognize and make sense of. The advances in developing and implementing these systems have been uneven.

Patt says that The US Oncology Network has been vigilant about using technical teams to design systems that allow for critical clinical data to be carefully structured in the EHR so that it is accessible to retrieval programs. But that isn’t the case everywhere. Many oncology practices don’t have the luxury of large technical support teams to develop this software. Companies like Flatiron that mine oncology information for clues to better treatment must employ human chart reviewers to pore through clinical information that is unstructured, she notes.

“We need to have consistency in the structured data we are using. In my practice, the doctor interacts with the health information systems systematically, but given our robust number of structured data elements, it’s a unique scenario,” Patt said. Even so, one of the problems encountered is that physicians have individual styles of reporting data, which can confuse even the strongest forms of artificial intelligence (AI).

Patt gave an example of how she and another doctor from her own practice might differ in the way they explain things to a hypothetical patient. “I might say to a woman, ‘You have stage II breast cancer, but you’re 80 and you’ve had a heart attack, so although chemotherapy might reduce your risk of recurrence, it might increase your risk of a heart attack and other morbid occurrences, and so I would not recommend chemotherapy.’” Conversely, a practitioner known for his Texas-style metaphorical way of speaking might tell the same patient the same thing, but in a way that no computer could recognize as identical to the opinion voiced by Patt. “He might say, ‘Mrs Smith, your breast cancer is not good, and we could give you chemotherapy, but that dog is not going to hunt.’ There’s no language processing system in the world that would see those 2 reports as the same, but because we oncologists have so much variability today in how we report various elements that influence our treatment and treatment choices, right now that remains a large barrier.” The only solution is to standardize systems and data reporting styles to cause that barrier to diminish, Patt said.

Being able to tap clinical systems for data is vital to the success of using real-world evidence (RWE) for oncology solutions. The potential inherent in RWE is exciting because there are questions about treatment efficacy that can be answered only by examination of clinical feedback on large populations of patients who have been treated. “Ultimately, when we feel comfortable with that data, I think it should be used to support FDA approval, because once we feel really comfortable with the information that we’re getting, we should make the most of that data and learn from all patients and not just the 35 that we may enroll in a particular clinical trial,” Patt said.

Efforts have been under way for a long time to peel away the barriers to putting RWE to good use. Kaiser Permanente and Emcore have been very active in this sphere. “Within The US Oncology Network, too, we’ve been publishing and using our data on RWE for a long time,” Patt said.

Part of the revolution in using RWE involves AI playing more and more of a role in the clinic, now even going into the examination room with the same oncologist would not have been able to research and interpret independently. This type of support is available currently in the form of tools such as Watson for Oncology, which has had hiccups in its development and acceptance, but has gained a following nonetheless among clinicians who see it as able to provide a computerized second opinion. “There’s a role for AI to assimilate all of the external information important to cancer decision making,” Patt says.

But the AI tool is only as good as its access to relevant data, she notes. “You have to make sure that there is good communication between your data systems--that you have a good bridge--and that you have a clinician who understands informatics and information systems, so that those systems are going to be implemented in ways that are clinically useful.” IBM is not the only company working on this problem. Syapse and Orien are among many that are working to funnel clinical data through automated processing systems to inform oncologists and enhance clinical outcomes.

Whenever the physician’s toolkit includes a computer screen, the debate over whether that screen time is going to be useful or burdensome is renewed. Once again, it’s important for computer systems to be well integrated, streamlined, and capable of providing useful guidance. Practices across the country have systems with varying degrees of sophistication. And where less sophistication means less computer capability, humans have to step in and do the work themselves.

The increasing complexity of oncology means that clinical-decision support systems are becoming essential, but most oncologists don’t yet have them, Patt says. The advantage of having these is that they can do the work of eliminating duplication in filling out forms and data fields on-screen for multiple different programs. Although working on-screen can seem like an added chore, once the computer has the data it needs, it can take over time-consuming tasks elsewhere in the process of managing patient care, Patt notes. “Once I’ve filled out that information, it also populates all of our other data systems, which will provide information to the payers that fund care, and getting the information to them allows us to get approval for chemotherapy faster. I don’t have to fill out the same information 7 times, because the systems are integrated.”

In fact, much of what she needs to do to document patient visits and treatment paths can be completed while she is in the examination room with the patient. “One patient I saw recently had stage I breast cancer. I put all of her information into the computer while I was in the room with her. It took just a few minutes. And then I wrote for her therapy while I was in the room because I was using a decision support system. After that, the system nudged me with guideline prompts for her treatment. As a society, we need to have systems that are integrated with our process flow so they don’t diminish our ability to care for patients.”

For this to happen, there has to be cooperation between the people who design the systems and those who use them. “Efficiency can come only when you have people to serve as the Rosetta Stone between these engineers who build these systems and the clinicians who actually use them. Within our US Oncology network, we’ve managed that by having a team of clinicians who advise on our health record endeavor. I spend a lot of time doing that, to try to make sure that these systems are optimized, that they’re efficient and maximize physicians’ time for patient care.”

There’s another payoff, too. All of these efforts help to reduce levels of frustration in the clinic. “There’s no tolerance for putting in the extra time. If you don’t have an efficient system, you have unhappy doctors,” said Patt.

Related Videos