Introduction
The test of our progress is not whether we add more to the abundance of those who have much. It is whether we provide enough for those who have too little.
Franklin D. Roosevelt, 1937
Geographic variation in health care has played a prominent role in shaping health care reform.
1‐
5 It also has been a subject of interest to social epidemiologists.
6‐
9 However, there are important differences. First, the focus of interest among planners has been on variation in
health care spending, while epidemiologists have focused principally on
health. In addition, while planners have attributed some of the variation to differences in patients’ burden of disease, they have attributed little to income, and much remains “unexplained.”
10‐
15 In contrast, a broad body of epidemiological literature links low income to poor health and shorter life-spans.
6‐
8,
16‐
23 From the perspective of social epidemiologists, poverty has a crushing effect on health.
7,
23
Why has poverty been so prominent in epidemiological studies and so out of view in studies of health care spending? One reason is that epidemiologists generally examine data at the level of individuals or within units more reflective of neighborhoods, such as census tracts or postal codes.
24‐
26 In contrast, health planners have generally studied much larger units, such as counties, hospital referral regions (HRRs), or states.
10,
12,
15,
27 Aggregating populations in units as large and diverse as these has tended to blur the effects of social factors that are so readily apparent in units of smaller size.
28‐
31 As Krieger has warned, “Blot poverty from view and not only will we contribute to making suffering invisible but our understanding of disease etiology will be marred.”
7
We have attempted to gain insight into the basis for geographic variation in health care among larger units by disaggregating them into their constituent ZIP codes and census tracts and assessing hospital utilization, household income, and the prevalence of disability, both statistically and spatially. Our studies centered on two urban HRRs, Milwaukee and Los Angeles. The Milwaukee HRR is not only the most populace in Wisconsin but also the most racially and economically segregated, and it utilizes more health care per capita than other HRRs in the upper Midwest.
27,
32,
33 The Los Angeles HRR is the most populous in the nation, and its rate of health care utilization is among the nation’s highest.
34 We compared Milwaukee to other Wisconsin HRRs and Los Angeles both to a region of comparable size, stretching from San Francisco to Sacramento (termed “San-Framento”), and other populace California counties. These studies revealed the profound contribution of the poorest ZIP codes of each region to geographic variation in health care utilization among regions.
Methods
ZIP code-level hospital data for Wisconsin were obtained from the Person-Level Data and Analysis Section of the State of Wisconsin Bureau of Health Information and were averaged for the years 1999 through 2002. The Milwaukee HRR was compared with seven others in Wisconsin: Appleton, Green Bay, La Crosse, Madison, Marshfield, Neenah, and Wausau. ZIP code level hospital data for California were obtained from the Patient Discharge Data File of the Office of Statewide Health Care Planning of the State of California for the year 2008. The Los Angeles region consisted of Los Angeles County, which overlaps the Los Angeles HRR. The San-Framento region encompassed San Francisco, Marin, San Mateo, Santa Cruz, Alameda, Contra Costa, Santa Clara, San Joaquin, Solano, and Sacramento counties. In both states, measurements of inpatient hospital days were limited to adults in acute care hospitals, exclusive of admissions related to pregnancy and child-birth. Admissions to skilled nursing, intermediate care, psychiatric, chemical dependency, and physical rehabilitation facilities were excluded. Only the Wisconsin and California portions of HRRs that extended into adjacent states were analyzed.
Population and income data at the ZIP code level were from the Census Bureau, either directly or through other sources.
35 For studies in Wisconsin, census data for 2000 were also extracted from GeoLytics Professional (GeoLytics, Inc., East Brunswick, NJ). For studies in California, estimates for 2008 were obtained from Claritas PopFacts (Tetrad Computer Applications, Inc., Ferndale, WA). Data on poverty and disability from all causes by age were from the 2000 census, as complied by GeoLytics.
ZIP codes were excluded where the principal populations were university students, military personnel, or institutionalized populations or where the total adult population was less than 1,500. The final analyses included 107 ZIP codes in Milwaukee, 266 in Los Angeles, and 287 in San-Framento. The total adult population included was 1.44 million in Milwaukee, 7.48 million in Los Angeles, and 6.94 million in San-Framento. Data were mapped at the ZIP code level using Mapland Professional (Software Illustrated, Tracy CA) and at the census tract level using GeoLytics Long Form. Goodness of fit was calculated using the power trend function of Microsoft PowerPoint. Pearson correlation coefficients were calculated using the statistical tool of Microsoft Excel.
Discussion
Four principal conclusions emerge from these studies. First, understanding geographic variation among large regions, such as counties and HRRs, requires disaggregation into their constituent ZIP codes and census tracts. Second, residents of low-income ZIP codes have greatly increased rates of disability and hospital utilization. Third, assessments of the relationship between income and hospital utilization are more valid among working-age adults than among seniors. And finally, poverty varies geographically and its variation explains a great deal about geographic variation in health care utilization. A series of observations contributed to these conclusions:
-
In Milwaukee, Los Angeles, and San-Framento, per capita rates of both hospital utilization and disability were steeply increased in ZIP codes with lower MHIs.
-
The much higher rates of hospital utilization in Milwaukee as compared with other Wisconsin HRRs were largely explained by the very high rates in Milwaukee’s dense poverty corridor.
-
Similarly, the much higher rates of utilization in Los Angeles as compared with San-Framento could be explained by a greater proportion of low-income ZIP codes in Los Angeles and a greater proportion of high-income ZIP codes in San-Framento, while the underlying statistical relationships between income and utilization were the same in both.
-
Among eight populace California counties that had both high-income and low-income ZIP codes, the wide variation in utilization that was observed overall was further exaggerated when only low-income ZIP codes were compared but was virtually absent when only high-income ZIP codes were considered.
-
In each region, the results of ZIP code analyses were statistically stronger and the impact of low income was quantitatively greater among working-age adults than among seniors.
-
These weaker results for seniors appeared to be due both to the wider residential distribution of seniors with respect to income and to weaker associations between low income and chronic poverty among seniors than among working-age adults.
-
If hospital utilization within the various regions and counties studied had been at the rate of the high-income ZIP codes in each, it would have been approximately 35 % less among working-age adults, 20 % less among seniors, and 30 % less overall.
Taken together, these studies demonstrate the profound association between poverty and health care utilization.
Because Milwaukee is so highly segregated, most low-income ZIP codes were clustered in a narrow poverty corridor, which also proved to be the zone of highest hospital utilization. While hospital utilization among working-age adults was one third higher in the Milwaukee HRR than in other HRRs in Wisconsin, utilization in the portion of the Milwaukee HRR outside of the corridor was within 5 % of other Wisconsin HRRs.
Los Angeles presented a greater challenge, but the conclusions were the same. Hospital utilization in Los Angeles was greater than in San-Framento, but this was simply because Los Angeles had a higher proportion of low-income, high-utilization ZIP codes while San-Framento had proportionately fewer, while utilization at comparable levels of income was the same. However, because the regression arcs in Figure
5 transcend differences in utilization of 3-fold and more, many-fold greater than the 25–35 % differences in aggregate utilization between these two regions, small shifts in the proportion of low-income ZIP codes were sufficient to account for the aggregate differences observed. Similarly, variation in hospital utilization among California counties virtually disappeared when only their high-income ZIP codes were considered.
One reason for this discordance, which was apparent on geomaps, was an out-migration of low-income seniors from the poorest ZIP codes into surrounding areas of higher income and, to a lesser extent, an in-migration of wealthy seniors into high-income enclaves within low-income ZIP codes. The former was also inferred from the lower ratio of seniors to 45–64 year olds in low-income ZIP codes than in high-income ones and is accounted for, at least in part, by the location of senior housing and nursing homes. The latter is related to the location luxury apartments within inner-city ZIP codes. These phenomena, which were most apparent at the census tract level, resulted in greater economic heterogeneity at the ZIP code level for seniors than for working-age adults. Previous studies indicate that such movements are not random with respect to health but, rather, that seniors migrating from lower to higher-income areas have higher medical expenditures, while wealthier seniors migrating into lower-income ZIP codes have lower expenditures.
53 Thus, while ZIP code income appears to provide a valid representation of the economic status of working-age adults, it is a much poorer proxy among seniors.
A second reason for discordance relates to the increase proportion of seniors with low-income as compared with working-age adults. While some of these seniors were poor earlier in life and experienced durable and often multigenerational poverty,
21‐
23,
39‐
41 others newly acquired low income after a lifetime of higher income and better health. Assessing poverty has been a challenge at all ages, but it is a particular problem at older ages.
54‐
56 Indeed, some have suggested that wealth or education may be better indices.
53,
54,
57,
58
This presents an enigma. While there were strong associations between income and hospital utilization in the 45–64-year-old cohort, these associations were much weaker and of lesser magnitude in the next decade. It seems implausible that such income-related differences would suddenly diminish after age 65. Rather, it is likely that aggregation of dissimilar income groups within ZIP codes and uncertainty over the meaning of low income over age 65 created ambiguities. The aggregation error becomes compounded when ZIP codes are further aggregated into counties or HRRs, further masking income-related differences.
11‐
15 Yet, it is the Medicare population that has been the principal object of study in defining geographic variation in health care, and it is from such studies that the notion of “unexplained” variation was derived.
10‐
15,
27,
59 Our research suggests that much of this previously “unexplained” variation simply reflects the inability to adequately measure the contribution of low income to health care utilization in the Medicare population, even at the ZIP code level and especially at the level of HRRs.
Table
1 lists the differences in hospital utilization that would have occurred within various regions if utilization rates in each ZIP code had been at the level of the region’s wealthiest ZIP codes. Taken together, these differences account for approximately 35 % of the total number of hospital days among working-age adults, 20 % among seniors, and 30 % among all adults. The 20 % increment that we observed among seniors is similar to increments in aggregate spending above the expenditure level of high-income Medicare enrollees reported elsewhere.
14,
63,
64 Similarly, the 30 % increment across the entire adult population is similar to increments in hospital admissions, preventable hospitalizations, and expenditures attributable to lower income in previous studies.
65‐
67 It also is similar to Marmot’s estimate that one third of spending in the British National Health Service (NHS) results from income inequality.
72
While low-income patients consume more services today, that was not always the case. Forty years ago, they consumed less, both through Medicare and the NHS.
73‐
75 It was not until the early 1980s that parity was reached, and health care spending among low-income patients has risen disproportionately ever since.
14,
48,
53,
63,
64,
76 Yet, this added spending for the poor is still not viewed as commensurate with their burden of illness,
65,
77 and despite it, the gap in life expectancy between rich and poor continues to widen.
70,
78
Third is the issue of homogeneity. Census tracts encompass relatively homogeneous populations, but ZIP codes were created for postal routes. While they generally provide valid measures,
24,
25,
31 that ability depends on their homogeneity with respect to the characteristics studied. This proved to be greatest in Milwaukee, one of the most segregated cities in the nation, but wealth and poverty were more comingled in Los Angeles, San-Framento, and elsewhere. In addition, although our studies focused on areas of higher population density, ZIP codes with fewer than 5,000 adults comprised 5 % of all ZIP codes studied in Los Angeles, 9 % in San-Framento, 17 % in other California counties, 25 % in the Milwaukee HRR, and 75 % elsewhere in Wisconsin. The resulting errors were magnified among 18–44 year olds, whose hospital admission rates were low, and among seniors, who account for fewer than 20 % of adults and who are more dispersed relative to income. Thus, although disaggregation of counties and HRRs into ZIP codes resolved many of the errors of aggregation that existed in larger units, the problem persisted even in units as small as ZIP codes.
It follows that, since poverty is distributed geographically, geographic differences in health care utilization are largely the result of geographic differences in poverty. That proved to be the case in our studies. Indeed, when only ZIP codes with higher degrees of wealth were considered, there was very little variation at all, which serves to emphasize the need to disaggregate large units of analysis, such as HRRs, if differences in health care utilization among them are to be understood.
Finally, our studies demonstrate that the relationship between poverty and health care utilization, which is so evident among working-age adults, is partially obscured among retirees. This suggests caution in interpreting studies of geographic variation in health care among the Medicare population, which have played such a prominent role in shaping policy.
As the USA seeks to slow the growth of health care spending, it will be important not to conflate the greater amounts of health care utilized by low-income patients with inefficiencies in clinical practice. Even with continued efforts to increase clinical efficiency, it seems unlikely that the inexorable growth in health care spending can abate as long as income inequality continues to widen. The real “inefficiency” is the existence of a population that has not been adequately nurtured in childhood nor given the tools to be healthy adults.
39‐
41 Poverty is not only an unsustainable failure of social justice. It creates an unsustainable financial burden for our health care system. Accepting this reality is a necessary first step. Confronting it should be our Nation’s highest priority.