Could Better Predictions Improve End-of-Life Care?
A team of Yale researchers has developed a statistical tool that can improve predictions of whether patients with advanced cancer are likely to die in the near term. Their analysis suggests that better understanding of the end of life could promote patient welfare by transferring more people from aggressive interventions to hospice care.
By Rachel Brody
End-of-life care is among the most pressing issues in modern American health care. “We have a major issue with how we care for people at the end of life, particularly for patients with cancer,” says Kerin Adelson, chief quality officer and deputy chief medical officer for Smilow Cancer Hospital at Yale-New Haven. Patients close to death are entering hospitals in “tremendously high numbers,” she says, receiving aggressive, expensive and futile care, and dying there instead of in hospice, a less costly and more comfortable alternative.
Why? Adelson says it’s due in part to what she calls a “discrepancy in prognostic understanding.” In other words, even patients with terminal illness often don’t understand what’s coming, and thus don’t identify their goals and priorities for end-of-life care. Much of this has to do with overly rosy prognoses from physicians, a phenomenon Adelson says stems from two possible causes. For one, she says, end-of-life conversations “are really hard.” Oncologists in particular, says Adelson, “worry tremendously about causing patients to lose hope and doing harm or causing distress.” Second, end-of-life discussions require time, but financial incentives have long undermined this.
Read the study: “Development of Imminent Mortality Predictor for Advanced Cancer (IMPAC), a Tool to Predict Short-Term Mortality in Hospitalized Patients With Advanced Cancer”
These realities, Adelson and her colleagues argue, often lead doctors to wrongly estimate the amount of time a patient has left to live. “Previous research has shown that physicians, when they predict life expectancy for terminally ill patients, can be off by a factor of five in terms of amount of time,” says Edieal J. Pinker, deputy dean of the Yale School of Management and the BearingPoint Professor of Operations Research. That’s time and resources lost, Pinker says; hospitals could save millions of dollars in readmissions and expensive treatments, and patients could choose to forgo unnecessary and invasive procedures, or choose to leave hospitals sooner for more comfortable hospice care.
So how to predict short-term mortality objectively, accurately and with patients in mind? A new study, authored by Adelson, Pinker, Rogerio Lilenbaum, and Donald K. K. Lee, among others, attempts to answer this.
The study, a collaboration between researchers from Yale’s School of Management (Lee and Pinker) and School of Medicine (Lillenbaum and Adelson), used both data analysis and medical expertise to create a statistical algorithmic tool called IMPAC, or the imminent mortality predictor in advanced cancer, to offer estimates of life expectancy in stage IV cancer patients. Looking at about 700 patients, discharged from Yale’s Smilow Cancer Center, both in New Haven and in associated care centers across the country, the authors hoped to answer the following: What is the objective probability that an individual—based on what physicians know about his or her health data—will survive less than a certain amount of time?
The statistical model they developed used both the Rothman Index, a real-time measure of health that includes vital signs and other assessments, as well as a series of static variables identified by the team. These included both patient-specific characteristics, such as age, and visit characteristics, such as length of stay and whether the patient had another hospitalization in the previous 90 days. The outcome was a constantly updating prediction about the likelihood that a patient would die within 30, 60, 90, and 180 days past the day of admission.
The researchers focused their analysis on the 90-day benchmark, a period in which one may still positively affect end-of-life care, according to Pinker. The study found that patients whom IMPAC found unlikely to die within 90 days survived about 290 days on average, and patients whom IMPAC found likely to die within 90 days only survived around 47 days. The authors then compared these predictions, made retrospectively, with what physicians actually did. “We looked at the patients who [physicians] discharged to hospice,” says Pinker, “and we found that on average they survived around 22 days, which fits with what has been seen in other studies. The patients that we identified correctly as going to die within 90 days, on average, they survived something like 35 days after we would’ve identified them.” In other words, he says, “our method seems to be able to identify patients earlier than the physician would have identified them.”
That’s critical, the authors say; more advance notice for patients means less cost to the patients and to the hospitals and caregivers. Each such patient, they estimate, could save on average $15,413 in health care costs, and the hospital, by moving patients to hospice earlier on, could save roughly $1.9 million per 1,000 hospital visits by advanced cancer patients, according to Lee, who is associate professor of operations at the School of Management as well as an associate professor in Yale’s Department of Statistics and Data Science.
More realistic prognoses can also mean greater comfort for patients close to death. Hospice, Adelson says, focuses “on comfort and quality of life at the end of life,” keeping patients at home with their loved ones instead of, say, in the intensive care unit.
A predictor tool like IMPAC is also in line with patient-centered care, Adelson explains. “If you take a patient with a terminal illness and give them treatment that is not going to extend their life or improve their quality of life, you have actually caused harm,” she says. With IMPAC, “you’re saving money and aligning care better with patients by helping them understand their own priorities by having honest empathic conversations about prognosis.”
The authors see the tool as phase one in a movement toward smarter end-of-life care. The next step, says Pinker, is to integrate the tool into physicians’ “standard clinical workflow” so that they can access it in real-time and bedside. Such an integration, Adelson hopes, will lead to more frequent and earlier “goals of care” conversations with patients; if it does, she asks, “would we start to see a decrease in unwanted or futile health care utilization?” Less frequent hospital admissions? Fewer costly procedures? Fewer patients dying in the ICU?
As doctors and clinicians, the overwhelming inclination is often to do as much as possible and to prolong life, says Lillenbaum, professor of medicine in medical oncology and chief medical officer of Smilow Cancer Hospital. But, in advanced cancer patients, where there is no hope of a cure, “we get to a point at which our focus should no longer be on trying to beat this disease, but on trying to make that process more comfortable, more peaceful,” he says.
There’s real suffering as a result of poor end-of-life care, whether it be frequent hospital admissions for patients near death, or invasive and costly procedures. At this critical stage in a patient’s life, says Adelson, “it is really imperative that we do a better job.”