Approximately 20 percent of advanced cancer patients receive treatment that weakens their immune systems and increases the risk of life-threatening infections—in particular, a form of infection known as febrile neutropenia. A good proportion of these complications can be prevented using FDA-approved medications. The problem? There has been no clear standard of care with regard to the prescribing of those medications, with countless patients remaining at undetermined levels of risk. It is in that context that a collaboration emerged among the University of Minnesota Health and Fairview Health Services—both based in the Minneapolis-St. Paul metropolitan area and which since September 2018 have been operating under a broad joint operating agreement under the brand name M Health Fairview—and Amgen, the Thousand Oaks, Calif.-based international pharmaceutical manufacturer.
Leaders from all three organizations created a multidisciplinary research team to investigate this issue, extracting data from over 50,000 patients to identify over 8,000 eligible patients at risk for febrile neutropenia, and provide granular details into its rates of occurrence. The initiative identified the patients most at risk, and helped to generate a multi-variable risk model which is planned to be deployed free-of-charge for use by any oncologist around the country. It offers innovation in a key clinical area that until now has not been extensively addressed. And ultimately, their innovative risk model has been able to identify three times as many chemotherapy patients at risk than the model that has long been used by a national cancer care collaborative.
Arpit Rao, M.D., a practicing oncologist and the chair of oncology quality and safety at the University of Minnesota Health, explains the problem and, how and why he and his colleagues got involved. “One of the largest service lines at the University of Minnesota Health is oncology,” Rao explains, “and I am the quality and safety chair for the entire service line. We have 11 inpatient and outpatient care sites across the state, treating tens of thousands of patients Febrile neutropenia is a very common complication, meaning that the body becomes weak because of chemotherapy. In fact,” he says, “your risk of life-threatening complications goes up dramatically for three to five days after each chemo cycle. And these patients are exposed” during the time when they’re undergoing chemotherapy; “even a small viral infection or urinary tract infection can become life-threatening.”
And here’s the specific clinical challenge: there are approved medications available to reduce the risk of febrile neutropenia during chemotherapy. The problem has to do with processes around when to prescribe those medications. As Rao explains it, “About 2 to 5 percent of chemo patients develop febrile neutropenia, which involves a low-performing immune system with fever, and is obviously a red flag for infection. A drug given as an injection takes five to 10 minutes to administer, and can cut the risk of neutropenia by 80 to 85 percent. But we didn’t know whether the people who were supposed to get that shot were [the ones] getting it. There’s nothing in the literature that can guide clinicians on prescribing it,” based on risk profile.
As a result, with support from Amgen, which manufactures products that decrease neutropenia risk, a two-year project ensued, which involved the creation of a joint steering committee from all three organizations, and which encompassed the participation of data scientists, analysts, statisticians, project managers, and others. The members of that team worked with IT leaders at Fairview to retrieve detailed, anonymized, patient-level data. From the Amgen side, says Rao, there were a number of senior leaders involved, including their vice president for value-based partnerships, and three executive directors from different divisions and four directors from different divisions, but mainly value-based partnerships and health economics.
And from the Fairview side, notes Rao, “There was me, our service line leadership, and Fairview Pharmacy leadership—including the two vice-presidents, one for specialty pharmacy, and the other for value-based partnerships. We also had a project manager from Fairview Pharmacy. And that was the real key; that project manager was the common thread in the year-and-a-half of work in establishing deadlines and timelines, and holding us to it.”
Adam Rhodes, health outcomes manager at Fairview Pharmacy Services, recounts the data challenge that he and other team members took on. “One of the difficulties with febrile neutropenia is that there isn’t a straightforward ICD-10 code involved,” he explains. “There’s a neutropenia code, but that’s not the same as febrile neutropenia. It just indicates a patient’s white blood cell level. Febrile neutropenia sets a lower-blood-count threshold and also involves a fever. So given that there’s an ICD-10 for neutropenia, but not for febrile neutropenia, we created a data set. If we were going to implement this and have a decision-making point, we needed a practical point. So I created a data set based on treatment plans. Every patient gets a treatment plan; sometimes, patients will change treatment plans. So within six months of the start of their latest treatment plan, we ask whether the patient has been documented with febrile neutropenia. We’re feeding the data set historical information, using a standard logistical regression-type of algorithm.”
And, Rhodes says, as a result, he and his colleagues are now able to accurately predict which patients are at risk for future febrile neutropenia. “Healthcare outcomes are so difficult to predict, because there are always a zillion variables, and there are genetic predispositions and lifestyle elements, and we really don’t capture data on everything,” he notes.
“There are three things that we’ve planned for and are in the process of doing,” Rao explains. “One was to get an estimate of febrile neutropenia rate in our system, and that was done. Our beginning estimate was 2 to 5 percent; ultimately, we found through analysis, a febrile neutropenia rate of 5.9 percent system-wide. So we created a benchmark with this study, for quality improvement projects. That was the first piece of our results. The other thing that we wanted to do was to dive down and look at over 100 variables for patients—age, gender, type of cancer, type of chemo, number of doses of chemo, risk factors like diabetes and heart disease. We looked at over 100 variables for each patient, and collected data for over 10,000 patients, and we created a predictive model where you can plug in information about a patient you’re starting on chemo, and getting an estimate of the percentage of chance of the patient getting febrile neutropenia.”
Doing so is important, Rao says, because the existing model of reimbursement for care is “very simplistic. It says if you give chemo to Patient X, the risk is 25 percent, so you should give that drug; if you give chemo to Patient Y, the risk is 15 percent, so you shouldn’t give that drug. That’s the National Comprehensive Cancer Network (NCCN) model. That model just looks at clinical trials and what happened to the patients in clinical trials with that chemotherapy; and the only risk factor included is the type of chemotherapy. In other words, it’s a crude measure. And sometimes, patients get caught up in insurance issues,” he adds. “Let’s say the risk of Patient X coming down with febrile neutropenia is 21 percent and they get the drug, and their risk drops to 2 percent. The insurance will only reimburse above 20 percent, so a patient might not get the medication, and ends up with neutropenia.” The key with this initiative? “We found our model was able to identify three times more people at risk—70 percent versus 23 percent, in the original NCCN model.”
Importantly, Rao says, “We can now use this information to create quality improvement projects in our system. Innovation doesn’t happen without a good understanding of what’s wrong, and without benchmarks. So this benchmark is now helping to propel additional quality projects.”
Rhodes believes that interdisciplinary collaboration and teamwork have been key. “I think, for one, the importance of collaboration between different disciplines,” he says. “This is specialized stuff. There are doctors out there who are into statistics and data, and they do some of this themselves. I can jump in, and I’ve learned about healthcare over time, but I’m not a clinician. So the collaboration between clinicians and data scientists is huge, and there’s so much room for innovation. I’m always struck by the tremendous amount of data available in healthcare.”
