Portal de Periódicos da CAPES

Bolstering the Value of Information Technologies in Oncology Care

2021; Wolters Kluwer; Volume: 43; Issue: 21 Linguagem: Inglês

10.1097/01.cot.0000800796.99442.e2

1548-4688

Valerie Neff Newitt,

Economic and Financial Impacts of Cancer

information technologies: information technologiesRavi Parikh, MD, believes cancer care delivery can be improved with the help of technologies that are already in place. Using a data-mining process that takes advantage of electronic health records, he is well on his way to proving his point. As Assistant Professor of Medical Ethics and Health Policy and Medicine at the University of Pennsylvania, he practices oncology at the Hospital of the University of Pennsylvania and Corporal Michael J. Crescenz VA Medical Center, both in Philadelphia.Ravi Parikh, MD: Ravi Parikh, MD“I've always been influenced by my parents,” said Parikh, a native of Hernando, FL, of finding his career path. His father is a neurologist (as is his wife, coincidentally), and his mother a psychiatrist. “Throughout my early life, I listened to them talk about patients, shadowed them; I always felt connected to the medical profession.” Another big influence which shaped his future was a volunteer stint as a bereavement counselor for a local hospice. “There were two ways for me to be exposed to patients when I was in high school. I could volunteer in hospitals, which I did, but I was limited to visiting supply rooms and restocking carts,” he recalled. “The other way was to volunteer at a local hospice, which not a lot of people did. That was one of the more transformative periods of my life. “People think of hospice as a gloom-and-doom place where you only are exposed to people who are dying. When I started, I thought the same thing. But these patients were free of the chains of their medical treatments, free of the burdens of pursuing years-long—or decades-long—courses of treating their illness. In hospice, they focused on living their remaining days, rather than on dying.” Parikh said along the way he captured patients' stories on video in a living legacy-style program. The insights he gained from patients focusing on real-time life instead of treatments “...would stay with me in a lot of the research I would eventually do,” he noted. Having earned a master's in public policy and a medical degree at Harvard Medical School before doing his residency at Brigham and Women's Hospital in Boston, Parikh completed a fellowship in hematology at the University of Pennsylvania where he has remained as a faculty member. He admitted he “backed into” oncology, having first set his sights on being a primary care or geriatric physician. “However, one of the things that drew me to oncology was the intensity of the patient relationship. Things I thought would be particularly rewarding in primary care were actually magnified in the oncology clinic,” he explained. “When someone gets the hardest news they'll ever get in their life—a diagnosis of cancer—you are there. You're the person who will help guide them on their path, navigating the trials and tribulations of their initial chemo and possibly all the way to the end of their life. And the integration of palliative care is more intense in oncology than in routine primary care. It all just became a natural fit.” The Power of a Nudge Parikh told Oncology Times his research interest, in its purest form, revolves around investigating “...a lot of the innovations that are going on in oncology care delivery right now, and actually trying to put numbers to how impactful they are.” As an example, he pointed to some of the positive impacts of early palliative care on oncology practice and cost of care. One of the key areas of insight from such work was the realization that it is oftentimes difficult to identify those patients who would benefit most from palliative care. That thought inspired Parikh and colleagues to consider using predictive tools—emerging data-science infrastructure—to try to identify patients most appropriate for palliative care. “A lot of that is centered on life expectancy, which oncologists oftentimes overestimate,” said Parikh. “Through our work came opportunities to be involved in developing algorithms to predict risk of death in cancer care. One of the best sources of information that we have for predicting mortality is the data we're capturing every single day in our electronic health records. We're in this precision medicine era of oncology, but a lot of the best predictors of how long someone might live are contained in our EHR and we don't take advantage of it. “So I spent 3 years in several efforts to use technologies like machine learning to make sense of a lot of the EHR data we were collecting—not just things like patients' ages and where they live, but the hundreds of labs and ways they utilize the health care system,” Parikh noted. “The way patients look in retrospect makes a lot of sense, but sometimes it's hard to integrate all of that information and make those assessments at the time you're seeing a patient. So our hypothesis was [that] using machine learning to make sense of a lot of the data in the EHR could actually help oncologists make better decisions about end-of-life care.” While many others have been interested in using artificial intelligence and machine learning technologies to predict important outcomes in oncology events, Parikh wanted to see if some of these technologies used in clinic could influence how patients receive care. “We designed a multi-step trial that evolved over several years. Only about 10-20 percent of patients at a given time actually have a documented conversation about their goals and wishes. We asked, ‘How can we improve that?’ We thought giving some of this predictive information to doctors could actually help,” he noted. The researchers began by interviewing doctors to see if they would use a tool like this and what the barriers and advantages of using it might be. “Once we got some positive signals, we started developing algorithms on data from several years past, and running them silently so we could actually test whether the algorithms still held truth in a novel era of cancer care,” described Parikh. “We showed predictions to clinicians to try to get a sense of where the model does well and where it does badly. “After that, we ran a randomized trial where the algorithm identified patients who were most appropriate for a conversation. High-risk patients appropriate for such a conversation were flagged for clinicians through text messages. When we ran the trial, we found that a simple text message nudge based on a machine-learning algorithm increased rates of conversations fourfold in just a 6-month period. It is a useful proof-of-concept that we can use tools like machine learning and make them useful to clinicians in real-time to influence things that actually matter to patients.” Feedback from physicians was not all positive, admitted Parikh, but it was instructive, allowing the researchers to improve the effort even more. “We've done a series of qualitative studies where we actually try to get a sense of what's useful and what's not in this intervention. The number one thing that doctors said was there will always be some patients the algorithm flags incorrectly and patients the algorithm might miss. This will not be 100 percent accurate,” said Parikh. “But just having that nudge based on data was useful, because it got doctors thinking about which patients the conversation might be appropriate for and whether this might be the right time to have it. If a doctor is questioning whether it's the right time to have such a conversation, it's probably the right time. Ultimately, the doctors found that these algorithms were confirming their own intuition, and in a more timely way than they would normally do on their own.” The physicians' feedback also suggested that the effort was useful not just for patients who were flagged by the algorithm, but for other patients as well. “In fact, we doubled rates of conversations among patients who were not flagged by the algorithm,” said Parikh. “And what that told me was maybe this is inducing a change in mindset among doctors, with regard to actually thinking about which of these patients might be appropriate for conversations, flagged or not.” Parikh said other improvements in the work included improving the algorithms, as well as delivery of the message flagging a patient. “We originally chose to deliver information by text message because we thought that's where doctors would be checking most frequently, but we learned it was an invasive way of delivering messages. A lot of clinicians said, ‘Why not just do this in the EHR, the system we're using every day?’ So we've taken that feedback back to iterating on this type of program.” Finally, a lot of doctors felt, even with a flag in the EHR, a day in clinic can be so demanding that the end-of-life conversation about goals, wishes, preferences may not be addressed right there and then. “So we took that and said, ‘What if we actually encourage patients to bring this up with their doctor, rather than putting the onus on the doctors themselves?’ A patient bringing up the subject was rated one of the top facilitators of such a conversation,” said Parikh. “So now we're thinking through and designing ways to prospectively use these type of algorithm-based strategies to offer patients the opportunity to record their wishes in advance of their appointment.” Asked how that might realistically be accomplished, Parikh responded, “I think the bottom line is a matter of culture change. A patient can't think, ‘Oh, I'm going to die so I need my wishes recorded.’ That's not the message we're trying to send and not what we're trying to do. Rather, we're saying, ‘No matter where you are in your disease process, having your wishes recorded is really important. And it's probably more important to do it early rather than late. We're offering you an opportunity to record your wishes and discuss this with your doctor.’ Again, this is a nudge-based concept, trying to ease patients into bringing these conversations up with their physicians.” What Works & What Doesn't? Parikh considers himself “a non-traditional clinical trialist,” noting, “I don't run trials on drugs, instead I'm interested in running trials of delivery-system innovations that can better deliver the care patients want and, hopefully at the same time, save the oncology care system money. To the health economics front, a lot of my interest has been in studying existing initiatives, either with regard to care delivery reform or payment reform in oncology, and seeing what works and what doesn't.” Toward that goal, Parikh and colleagues have partnered with payers to evaluate ways they try to save money in the oncology care system and see if they actually work. “We ran a large evaluation of a study, from a large commercial payer, intended to encourage more cost-effective radiation among their population,” Parikh detailed. “We found a simple, utilization-management program or a prior-authorization program saved patients from having long, expensive courses of radiation on breast cancer. Evaluating and then trying to design better real-world initiatives for delivering cancer care is what really interests me.” Another focus of Parikh's work has been examining alternative payment models in oncology, the biggest one being the Oncology Care Model from Medicare, which is a bundled-payment model. “We've been trying to determine whether the Oncology Care Model works, or reasons it might not work, in hopes of engendering better programs,” he said. “Because cancer care is one of the costliest areas of U.S. health care, we need to find ways to reduce spending so that we're not bankrupting the system, and not bankrupting patients.” Parikh said that, while the Oncology Care Model oftentimes encourages use of certain evidence-based care practices, it doesn't end up saving the system money. “In fact, while some of these initiatives give payments to oncologists to do better value-based care, they do not necessarily put oncologists at risk for cost of care. It's hard for these types of things to save money—and lo and behold—they don't. “The biggest reason for rising costs in cancer care is growth in technology and more expensive drugs that are improving patient outcomes, no doubt, but are costing the system enormous sums of money,” said Parikh. “So an alternative payment model changing the way oncologists or health systems are paid won't be a drop in the bucket when faced with the increasing tide of rising drug costs. The lesson learned from this is we need to focus on the highest cost areas of oncology care before we tackle alternative payment models. Drugs are probably number one, two, and three out of all of those.” So what can we do about it? “I think that's the million dollar—or billion dollar—question,” answered Parikh. “In addition to pathway-driven care, so that we're not using drugs that we know don't work but cost the system a great deal, we may need to tie prices for drugs to how effective they are. It doesn't make sense [that] a drug that gives individuals a small amount of time left to live is just as expensive as some of the really revolutionary immunotherapies resulting in years and years of durable survival. So tying some of the cost of those drugs to how effective they are is one key area that we can use to change the system. Current drug spending is costing money, and is shooting payment reforms in the knee.” Improving Survivorship Parikh said too little attention has been directed at survivorship care, and he hopes to correct that oversight as much as he can throughout his career. “How are we caring for the millions of patients with cancer who are in a stable phase of their treatment, but who are now incurring the consequences of those treatments, whether physical, financial, or mental?” he asked. Parikh has been particularly involved in prostate cancer survivorship and has received funding through the VA and the Prostate Cancer Foundation to help study barriers of survivorship care, and strategies to deliver better care for prostate cancer survivors utilizing wearable devices, digital health, and other type of technologies. “In prostate cancer, we give therapies that reduce the amount of testosterone in a man's body, and that can have big effects on cardiovascular and bone health. We've been thinking through ways to use wearable devices to encourage men to have more physical activity. We are exploring other digital health tools to identify men who are at risk for different adverse events, including using artificial intelligence-enabled tools on patients' routine CT scans to better predict whether they might incur a fracture after their prostate cancer therapy. The question is: How do we encourage a 21st century system of survivorship care for our cancer survivors?” Parikh continued, “To identify people at risk of a fracture, in our old system of work, we've used DEXA scans, an X-ray of the hip to assess whether men are at risk for a fracture because of the therapy they received. That poses a couple of problems. First, the DEXA isn't the most accurate measure to get at future fracture risk, and second, we found only about 20 percent of men actually get a DEXA scan. So there's this huge gap. Either doctors aren't getting them for patients who need them or patients don't think they need them. So, we could try to encourage people to get their DEXA scans, or we could think about better ways to determine someone's bone health from the scans they are already getting as part of their prostate cancer care. We've found some interesting preliminary data from outside of prostate cancer that's shown that just a normal CT scan can be a really good predictor of how strong your bones are and whether you're at risk for a future fracture. What that means is that we might be able to identify a whole new crop of men who might be at risk for a fracture—without having them get another scan—and getting them on drugs to help protect their bones earlier, or tailoring some of their treatment so that we're avoiding side effects of therapy.” Parikh said in-roads are just beginning to be made in areas like cardiovascular disease and mental health as a consequence of prostate cancer therapy, as well. It all ties back to that earlier theory of using existing records in the EHR, according to Parikh. “Again, we are trying to use routinely collected information to identify new patterns and new things to read and act upon,” he noted. “The big effort is to use health technology, digital health technology, artificial intelligence, and machine learning to shape the way that we deliver cancer care, because there's so much rich data in what we're already collecting in our routine oncology practices. We just need better ways to measure it and better ways to mine it. “I'm trying not only to develop those ways to mine data, but also ways to deliver it to oncologists and to patients so that it will actually improve care. My hope is this type of work—not in a decade's-long time span, but in a year-long time span—will be able to generalize several types of technologies to improve survival and, most importantly, improve quality of life by preventing adverse events.” Valerie Neff Newitt is a contributing writer.