Around 25% of Medicare spending in the US occurs in the last year of people’s lives. This is sometimes discussed as a questionable use of resources but a Harvard/Massachusetts Institute of Technology (MIT) study found that very little of this spending is on patients whose death within the year is highly likely. For example, the researchers discovered, less than 5% of Medicare spending is applied to the single highest-risk percentile of all individuals – and their predicted one-year mortality rate is still just 46%.
“What we discovered is, very little money is spent on people who we know with high probability are going to die in a short amount of time,” says Amy Finkelstein, a professor in MIT’s department of economics and co-author of the paper. To the extent that such cases exist, she adds, “they’re just not the drivers of spending” in bulk.
The study also illuminates the general circumstances of late-in-life mortality. Fewer than 10% of people who die in a given year have a predicted one-year mortality rate over 50%. As the researchers found, even when people are admitted to a hospital in what turns out to be their last year of life, fewer than 4% of those patients have a predicted one-year mortality rate of 80% or higher at the time of admission.
In a sense, the study shows, the apparent concentration of spending on last-year-in-life patients is a by-product of the fact that even relatively low-mortality health scenarios for the elderly will include a certain number of deaths – not that the individual treatment decisions represent longshot cases.
“I do hope we stop pointing to end-of-life spending as an obvious problem,” Finkelstein says. “That’s not to say there aren’t problems in the US health care system, but this is not a symptom of them.”
The paper’s authors are Liran Einav, a professor of economics at Stanford University; Finkelstein, the John and Jennie S MacDonald professor at MIT; Sendhil Mullainathan, a professor in the economics department at Harvard University; and Ziad Obermeyer, an assistant professor at Harvard Medical School.
To conduct the study, the research team examined a random sample of almost 6m Medicare enrolees who were in the programme as of 1 January, 2008. For survivors, the study examines health care spending for all of 2008; for people who died in 2008, it examines spending over the year prior to death. The analysis produces mortality predictions as of 1 January, 2008, using data on demographics, health care use, and more.
The analysis also deployed a standard form of machine learning to evaluate the impact of a wide range of variables on health care trajectories, and to produce a probability of death within one year for every enrolee in the study.
The overarching result, as the authors write in the paper, is simple: “Death is highly unpredictable.”
Take, for instance, that top percentile of high-risk Medicare enrolees, those whose one-year predicted mortality rate is 46%: Of those patients, 44% survived for at least one year after the start of the study. Similarly, the predicted one-year mortality rate at the 95th percentile of people in the study is just 25%.
Moreover, the study finds, the basic fact that we spend more money on people who are sick – in the study, both those who recovered and those who died – accounts for 30% to 50% of the concentration of spending on people in their last 12 months of life.
“I think the typical narrative is: ‘Wow, the US spends so much on health care and a quarter of that is in the last 12 months of life. That money is obviously a waste; we spent all this money and they died,'” says Finkelstein. “But that’s not the right way to look at it. We don’t know in advance who’s going to die this year, and some of the people we spend money on survive.”
The paper’s authors suggest that productive new avenues for research on efficiency and medical spending will look concretely at more specific parts of the picture. Or, as they write in the paper, “a focus on end-of-life spending is not, by itself, a useful way to identify wasteful spending.”
Instead, Finkelstein contends, it would be more productive to zoom in on particular kinds of treatments and procedures, among other things, to assess their effectiveness, rather than generalising about efficiency based on massive aggregate numbers such as the last-year-of-life Medicare figure.
“What we need to do is engage in the challenging task of figuring out which kinds of spending are yielding health benefits, and what types aren’t,” Finkelstein says. “Let’s focus on the real problem and not get distracted by misleading statistics. There’s a lot of hard work ahead of us, not easy answers.”
Finkelstein adds: “The policy upshot is: It’s important we understand the things we’re talking about.”
The study was supported, in part, by the National Institute on Aging, the National Institutes of Health, and the National Institute for Health Care Management.
That one-quarter of Medicare spending in the United States occurs in the last year of life is commonly interpreted as waste. But this interpretation presumes knowledge of who will die and when. Here we analyze how spending is distributed by predicted mortality, based on a machine-learning model of annual mortality risk built using Medicare claims. Death is highly unpredictable. Less than 5% of spending is accounted for by individuals with predicted mortality above 50%. The simple fact that we spend more on the sick—both on those who recover and those who die—accounts for 30 to 50% of the concentration of spending on the dead. Our results suggest that spending on the ex post dead does not necessarily mean that we spend on the ex ante “hopeless.”
Liran Einav, Amy Finkelstein, Sendhil Mullainathan, Ziad Obermeyer