Over the past few weeks a lot of media buzz was generated by a Papanicolas et al. JAMA report that compared the US healthcare system with that of 10 other developed nations to identify the key drivers of higher healthcare spending in the US. The authors (and media reports) largely came to the same conclusion as a 2003 Health Affairs article by Andersen et al., “It’s the prices, stupid!”
What received significantly less attention was an editorial in the same issue of JAMA by Zeke Emanuel, which drew somewhat different conclusions from the same data set used by Papanicolas et al. Specifically, the editorial concludes that, “…4 areas – pharmaceuticals; high-volume, high-margin procedures; CT and MRI imaging; and administration – account for just under two-thirds of the difference in health care costs between the United States and other developed countries.”
We tend to agree with the conclusions of this editorial, that BOTH prices and utilization play a role, but rather than focusing on international comparisons (useful, but also fraught with methodological issues), below we draw on data and research that document and examine the variability in cost of care within the US.
One way to begin to disaggregate the price vs. use rate contributions to healthcare spending is by focusing on the data that comes from the Medicare program. Since Medicare sets provider prices directly (through a complex set of calculations, rather than provider negotiations), any regional variation in Medicare spending (once adjusted for geographic cost of doing business and beneficiary health status) is likely driven by the availability and use rate of services.
Although regional variation in Medicare utilization and spending has been documented by the Dartmouth Atlas of Health Care for over 20 years, the notion that use rate may be important received wide-spread attention with the publication of “The Cost Conundrum,” a 2009 New Yorker article by Atul Gawande. The article reported that Medicare spending for the elderly in McAllen, TX was about double that of El Paso, TX despite essentially the same demographic profiles, availability of medical services/technology, and patient outcomes. Since prices are set top down by Medicare, the use rate of services is the likely culprit of the observed cost differences between the two locations. And in fact, Gawande found that, “Compared with patients in El Paso and nationwide, patients in McAllen got more of pretty much everything—more diagnostic testing, more hospital treatment, more surgery, more home care.”
Several other studies have concluded that the use rate of services is a key driver of the variation in Medicare cost per capita around the country:
- Wennberg et al. (2008), “Most of this variation was not due to differences in the price of care in different parts of the country, but rather to differences in the volume….”
- Gottlieb et al. (2010), “… utilization – not local price differences – drives Medicare regional payment variation….”
- MedPac report to Congress (2011), “…there is nearly a twofold difference between the MSA with the greatest service use (the Miami, FL, MSA) and the MSA with the least service use (the La Crosse, WI, MSA) [after adjusting for regional prices, added payments for Graduate Medical Education, demographics, beneficiary health status, etc.].”
Moreover, a 2008 Congressional Budget Office report —“Geographic Variation in Health Care Spending” — summarized previous research on the topic as follows:
“Researchers have reported that, after controlling for local practice costs, health status, and demographics, between one-half and three-fourths of total variation in spending remains unaccounted for….researchers generally attribute much of the remaining geographic variation to differences in the way medicine is practiced.”
Some of this variation in use rate may be attributed to medical uncertainty/lack of clear guidelines for appropriate care, e.g., using data from the Dartmouth Atlas, Peter Orszag (then Director of the CBO) showed that there is little difference between high spending areas and low spending areas in services were clear clinical practice consensus exists (e.g., mammograms are done at the same rate per 1,000 population). However, high spending areas use about twice as many ICU days per 1,000 population, with ICU use being an example of service where individual physician judgment and practice style play a key role. The ICU example is perhaps not surprising considering that the hospital is the most expensive place to get care, and the ICU is the most expensive place to get care within the hospital setting.
Papanicolas et al. do not examine prices directly, rather they conclude that prices are the major culprit in high levels of US healthcare spending because broad categories of utilization of healthcare services in the US do not differ significantly from other countries. As highlighted in Emanuel’s editorial, this observation does not hold true for all of the provided indicators (e.g., C-sections, knee replacements, CABG). Take an expensive procedure like CABG. The US (highest use rate country in the Papanicolas et al. article) performed 3 times as many CABG’s per 100,000 population as the UK (lowest use rate). Or take expensive diagnostic tests like CT and MRI where the US performs 3 times and 2.9 times as many tests per population than the lowest use rate country. These seem like significant utilization rate differences to us. With the authors just using broad utilization statistics like hospital discharges and physician visits per population, they completely overlook the issue of intensity of services within those broad categories.
Moreover, as pointed out by Baicker and Chandra in another JAMA editorial, the original research report does not account for the variation in the intensity of services provided in the context of a given procedure (e.g., angioplasty, where patients could “receive drug-eluting stents instead of bare-metal stents”) or hospital discharge (as we have highlighted with the variability in ICU use above).
And of course, this is just the cost side of the value equation. One should also be looking at patient outcomes – not life expectancy, which is a poor measure of the healthcare delivery system given the limited role the delivery system plays in the overall health of the population. It is of interest to note that, the clinical indicators reported by Papanicolas et al. (e.g., 30-day AMI and stroke mortality, where a lower rate is better) actually place the US at or near the top in terms of quality. In contrast, the low-cost UK system (which the Commonwealth Fund recently ranked as the best healthcare system) fares worse, consistent with its poor performance on a broader set of patient outcome indicators examined in the same Commonwealth Fund report.
It is of interest to note that regional variability in healthcare spending extends beyond the Medicare population, with the lowest cost state (Utah in 2014) spending per capita of $5,982, on par with that of comparison nations (reported range is $3,377-$6,808). Perhaps, as alluded to by Baicker and Chandra, we may be better served by within-US, rather than international comparisons, and strive to learn from those providers who are consistently able to deliver high-value care within the context and constraints of the current US healthcare system.