You Don't Get What You Pay For

The enormous interest in getting good “value” for every dollar spent on health care, whether by individuals, insurers, government, or anyone else neglects certain basic realities—for example, that medical care isn’t a consumer good like toasters: it’s a very sophisticated service provided by highly trained professionals;  and that health insurance by its very nature makes the operation of a free market impossible. There’s still another basic reality that is even more often neglected, and that is the widespread belief that “you get what you pay for.” Or, if you pay less for one treatment than another, the cheaper one is necessarily inferior. Any claims that the two are of equal quality are suspect. And claims that the cheaper one is higher quality are, on their face, deemed outlandish.

Translated into practice, this means that patients and doctors alike tend to assume that more is better. More x-rays (or, as plain radiographs, CT scans, MRIs, and PET scans are collectively known, “imaging studies”), more medications, more doctors is superior care and must result in better outcomes. As a result, I’m not at all surprised that changing physician behavior and patient expectations has proved difficult, even when professional guidelines assert that less is more. And unfortunately (unfortunate since, from a geriatric perspective, less often is more), a new study that purports to show that greater spending per hospitalized patient fails to improve outcomes is hardly convincing.

Previous retrospective studies, especially those comprising the Dartmouth Atlas of Health Care, have shown that expenditures on apparently similar patients differ by geographic region, by hospital, and within regions—without any measurable difference in outcomes. But the Dartmouth Atlas has been criticized for working backwards from death even though death could not have been predicted in advance, it has been criticized for failing to adequately consider differences between the patient population in different locales, and it has been critiqued for not acknowledging that patient preference might account for some of the observed differences in health care utilization and, as a result, in cost. The new study asks whether physicians working in the same hospital nonetheless exhibit differences in their pattern of test- and treatment-ordering and whether that variation results in different outcomes for their patients. Looking at over 1.3 million hospitalizations occurring at over 3000 hospitals and involving 72,000 physicians, they found large variability in expenditures and no difference in outcomes—just like the Dartmouth Atlas findings.

The authors were careful to look at Medicare Part B spending because this is involves services that are at the discretion of physicians (Part A spending is determined largely by the DRG, the reason for admission, and is set by Medicare) and is a “proxy” for the intensity of resource use by physicians. They were careful to confine their analysis to Medicare fee-for-service beneficiaries who were age 65 or older and hospitalized for an acute medical condition. And they examined separately the behavior of general internists and hospitalists. They made some adjustments to account for differences among patients, including age (in 5-year increments), sex, race/ethnicity, median income, and existing comorbidities, and other adjustments to account for differences among physicians, including age (also in 5-year increments), sex, and site of medical school education. They found that the variation in spending across physicians within a hospital was greater than across hospitals. Among hospitalists, adjusted spending was more than 40 percent higher among doctors in the highest spending quartile compared with the lowest quartile. And higher expenditures had no effect on either the 30-day readmission rate or mortality, the two measures of quality used to examine outcomes.

Regrettably, this study has a number of glaring weaknesses. First, there are the odd omissions: the authors report on the gap between the highest and lowest quartiles of hospitalists but not the corresponding figure for general internists, even though nearly twice as many patients were cared for by internists than by hospitalists. Next, it’s not clear that the two outcomes examined—mortality and readmission rate—are good indicators of quality. Or rather, even if the two groups were indistinguishable based on these two measures, perhaps one group fared far better than the other on some other measure that wasn’t looked at, say quality of life. Finally, the study wasn’t randomized and it wasn’t prospective, allowing for the possibility that there were important differences between the patients on whom much money was spent and those on whom less was spent. In fact, maybe the patients on whom more resources were expended were sicker. If they were sicker but had the same mortality rate and readmission rate as those on whom fewer resources were spent, then arguably they fared better than their counterparts!

So where do we go from here? Contrary to the prevailing wisdom, the answer may not lie with “big data.” Too many things are going on at once with these patients to be able to reliably conclude that ceteris paribus, all things being equal, overall expenditure on tests and treatments had no bearing on outcomes. I think it would make sense to look at a small number of detailed case examples—20 or 30 patients of the same age with the same admitting diagnosis, matched for severity of illness, co-morbidities, race, ethnicity, and socioeconomic class, some of whom are cared for by prolific test-orderers and some of whom are not—following them prospectively over time to see what happens to them. And the study would try to ascertain why various choices were made, perhaps by interviewing the patients and/or their doctors, perhaps by gleaning the answer from free text in medical records, and what their outcomes turned out to be.

All Things Dartmouth

For the last 20 years, the Dartmouth Institute for Health Policy and Clinical Practice—it’s added a few words to its name over time—has produced the “Dartmouth Atlas,” a compendium of information about the American health care system, principally culled from Medicare data. Some people are highly critical of the Atlas because it analyzes health care resource utilization (including hospital, ICU, SNF, physician visits, medications, procedures) in the last six months or the last two years of life—which requires starting with the time of death and looking backwards. The problem with this approach is that we don’t know in advance who’s going to die when, at least not with any degree of certainty. It’s easy to use the retrospectoscope and say that a patient shouldn’t have spent so much time in the ICU or gotten a feeding tube, given that he was going to die soon. But the intent of sending him to the ICU or putting in the tube was to enable him to live longer, or at least better, and it wasn’t so clear up front that the endeavor would fail.

The critics have a legitimate gripe, but I confess that for all its faults, I love the Dartmouth Atlas. I’m fascinated by its principal finding, which is that whenever you look at “preference-sensitive care,” at areas of medical treatment where there is no unequivocally right answer for what to do, what happens to patients varies enormously across the country. And while some kinds of care track together—in Portland, Oregon, patients spend less than the national average number of days in the hospital and in the ICU and more days enrolled in hospice, whereas in New York City, they spend a higher than average number of days in the hospital and the ICU and fewer days enrolled in hospice—the data are full of surprises. Given my general level of enthusiasm for all the reports that come out of the Dartmouth Atlas project, I was very happy to discover, by chance, I might add, that a new report was released just last week, “Our Parents, Ourselves: Health Carefor an Aging Population.”

Much of what’s in this report covers familiar ground. It summarizes the data on the average number of days that older patients spend in the hospital in the last six months of life, it talks about where people die, and reviews how much, or rather how little, time they spend enrolled in hospice. But some of what’s here is new, or at least presented in a different way, and it’s enlightening.

Because the focus of the new report is on the older patient’s experience of the health care system—which incidentally is what the book I’m writing is about—the authors have come up with a few ways of quantifying that experience. They define a new measure, the average number of “contact days,” by which they mean the total amount of time a person interacts in some way with American medical care, the number of days they have a doctor’s appointment, the number of days they go for a test or a procedure, and the number of days they are inpatients in a hospital or a skilled nursing facility. 

I’m not entirely satisfied with a measure that counts a day on which you have a doctor’s appointment or a test as a full day rather than attempting to add up the number of hours you actually spend going to the office, waiting, being seen, and coming home and translating that number into a fraction of a day, but it’s a good first step at measuring the intensity of exposure to the health care system. For what it’s worth, the average number of “contact days” in 2012 was 17, but in certain areas, such as Long Island, New York, it was 25.

Other useful sections of the report address “areas needing improvement” and “areas where we have made progress.” The list of areas needing improvement starts with screening for prostate cancer in men over 75 (we shouldn’t be ordering PSA tests for this group but 20% of older men get them overall and 30% of men in Miami get them). Similarly, there’s screening for breast cancer in women over 75. That’s an area that’s more controversial since its utility depends on life expectancy and there are plenty of 75 year old women who will live another 10 or 15 years and thus may benefit from screening. 

Then there’s late referral to hospice: as many as 27.3% of the residents of Fort Lauderdale who die in hospice were enrolled in the last 3 days of life, way too late to reap much benefit. Next comes feeding tubes in patients with advanced dementia, a subject near and dear to my heart since I’ve written quite a bit about it. It’s still the case, or it was in 2012, that 6% of people who die with dementia got a feeding tube in the preceding 6 months—though this happens less than 2% of the time in Portland, Oregon and as much as 12.8% of the time in Los Angeles. 

Finally, there’s the old favorite, days spent in an ICU in the last 6 months of life. Interestingly, the situation here has gotten worse, if you assume that the time in spent in an ICU before death is very uncomfortable and not at all beneficial. How much time people spend in an ICU is also tremendously variable, ranging from 1 day to 9 days, with La Crosse, Wisconsin at the low end of the spectrum and Miami and LA at the upper end.

And then there are the areas where we’ve made progress, according to the report. There is less use of high risk medications, there is more widespread monitoring of diabetics, there has been a decline in preventable hospital admissions, and there’s been a decrease, albeit slight, in readmissions to the hospital within 30 days of discharge. Finally, the opportunities for coordination of care and for advance care planning, which for some reason are lumped together, have increased.

What’s missing from the report is a compelling explanation of why the things that have improved have gotten better and why those that haven’t improved are still problematic—and why there’s so much variability in what happens where. The authors argue that there are different factors at work: there’s less use of high risk medications because one of the bad actors (propoxyphene, or Darvon) was taken off the market; and there are fewer readmissions because there’s better discharge planning. But Darvon was taken off the market in all 50 states and the rate of use of so-called high risk medications varies between 10% and 30%, depending on geography. Why? The national average for readmissions fell (though not terribly dramatically) between 2008 and 2012, but why did it fall for some hospitals and not others? If we know how to prevent readmissions, at least some of them, why doesn’t everybody do it?

I guess I still need to write that book.