When my Dartmouth colleague Dr. John Wennberg drove through Vermont in the late 1960s collecting data on hospital admissions, surgical procedures, and doctor visits, he found something unexpected--enormous variability in rates of medical care use across seemingly similar Vermont towns. In some school districts, nearly every school child still retained his tonsils, while in other districts, nearly every school child's tonsils had been removed.
Since 1996, the Dartmouth Atlas of Healthcare has used national Medicare claims data to document regional variations in health care patterns, now across 306 "hospital referral regions" in the United States. The most recent data from 2010 shows per capita age-sex-race-price-adjusted Medicare expenditures ranging from $6,176 in LaCrosse, Wisconsin to $13,824 in McAllen, Texas. These variations have captured the attention of a now-expanding group of economists, physicians, and other social scientists. Both NBER and Dartmouth have been centers for this research, much of which has been supported with funding through the National Institute on Aging. My work, with various collaborators, has focused on three general questions:
What are the causes of regional variations in health care utilization?
Health status is one leading candidate for the observed variations; after all, regions in Louisiana or West Virginia should spend more, given the greater burden of disease and poverty in those states. One study drew on rich survey data of elderly Medicare enrollees with information about income, poverty status, health behaviors such as smoking and drinking, and self-reported health, and found that health and income explained one-third of overall regional variations, leaving two-thirds unexplained. (1) A different approach compared treatment patterns for specific diseases such as heart attacks with highly detailed chart-review information; even this "apples-to-apples" approach finds considerable differences in risk-adjusted spending between the highest and lowest quintile of Medicare patients. (2)
Another explanation for the regional variations is patient preferences and demand more generally - people in high-spending regions may demand, and get, more health care. We have tested this hypothesis using a national survey of Medicare enrollees, where enrollees were asked about their preferences for more intensive diagnoses and treatments. (3) However, we did not find that patient preferences in a region could explain more than a small fraction (generally less than 10 percent) of the overall regional variations in Medicare spending across regions.
If not health or demand, then perhaps these variations are the consequence of "supplier-induced demand": physicians prescribing more than medically necessary to augment their income. One problem with this explanation is that Medicare is a federally administered program that pays just about the same for procedures (adjusted for local prices) in LaCrosse, Wisconsin as in McAllen, Texas. If McAllen...