The Milwaukee HRR includes both the city of Milwaukee and a surrounding zone ten times as large in area but roughly equivalent in population. Health care spending in the Milwaukee HRR exceeds the rate in other parts of the upper-Midwest by approximately one third,27 a fact that has concerned Milwaukee’s business community.32
We divided Milwaukee’s adult population into working-age adults (ages 18 to 64) and seniors (ages 65 and over). Our initial studies were carried out among the former. When assessed at the ZIP code level, there were steeply inverse, curvilinear relationship between median household income (MHI) and both the number of hospital days per 1,000 (r
2 = 0.755; Figure 1a) and the per cent of the ZIP code population reporting a disability (r
2 = 0.636). The relationship between MHI and hospital days could also be resolved into two linear components: one at household incomes below $50,000 (39 % of ZIP codes), which had a steep slope and strong coefficient (r
2 = 0.818), and the other at incomes above $50,000, which was relatively flat. Utilization was 3-fold greater in ZIP codes comprising the lowest income decile vs. the highest
These statistical relationships were also evident in geomaps (Figure 2). The spatial distribution of the quintile of ZIP codes with the lowest MHIs (A) was similar to those with the highest percent of disability (B), and these were similar to those with the most hospital days per 1,000 of population (C).
Much of the increase in utilization in low-income ZIP codes was due to admissions for ambulatory care-sensitive conditions. Comparing utilization among adults ages 35–64 in the lowest vs. the highest income quartile of ZIP codes in Milwaukee County, the number of hospital days per 1,000 was greater by 347 % for heart failure, 266 % for diabetes, and 610 % for chronic obstructive pulmonary disease (COPD), increments similar to those observed previously in other urban areas.36–38
Seniors accounted for 17 % of the adult population but utilized 49 % of the total number hospital days. As observed among working-age adults, there were significant associations between MHI and both hospital utilization and disability among seniors, but these were weaker than among working-age adults. Accordingly, when hospital utilization was plotted against MHI (Figure 1b), the data were more scattered (r
2 = 0.304) and the amplitude of differences in utilization between the poorest and richest ZIP codes was less than had been observed for working-age adults. ZIP code maps confirmed these statistical differences (Figure 2).
Two factors appeared to contribute to these differences between seniors and working-age adults. One was a difference in residential distribution. Census tract maps of seniors showed many low-income tracts within higher-income ZIP codes. Similarly, ZIP code maps showed the presence of low-income seniors in areas in which higher income working-age adults resided (Figure 2A, D). This was due, in part, to the distribution of nursing homes and senior housing. Indeed, Milwaukee’s poverty core is devoid of nursing homes. Conversely, census tract maps showed clusters of high-income seniors in predominantly low-income ZIP codes, corresponding to the locations of luxury apartments in the central city. Thus, the high degree of income segregation that exists among working-age adults does not continue beyond age 65, and the patterns of hospital utilization followed accordingly.
A second factor is a difference in income distribution among seniors and working-age adults (Figure 3). During the decades of working life, incomes are skewed to higher incomes, whereas after age 65, incomes are sharply skewed to low income. Some low-income seniors were poor earlier in life and, therefore, may have experienced chronic poverty,39–41 while others became low income in retirement but had the advantages of higher income in earlier decades. This phenomenon decreases the validity of low income as a proxy for poverty as it relates to health care utilization among seniors.
Milwaukee’s Poverty Corridor
Because of Milwaukee’s extreme racial and economic segregation, we were able to define a narrow “poverty corridor” (Figure 4A + B) in which the MHI was 40 % lower than elsewhere in Milwaukee. The corridor included 41 % of the adult population but 85 % of the black and Hispanic populations, and blacks and Hispanics residing there accounted for more than one third of the population, as compared to fewer than 5 % elsewhere. In the core area of extreme poverty (A), blacks and Hispanics comprised more than two thirds of the population and the poverty rate was 72 %. Hospital utilization among working-age adults was 85 % greater in the poverty corridor (A + B) than in the remainder of the Milwaukee HRR (C + D), and it was 145 % greater in the core area of greatest poverty (A).
Compared to other HRRs in Wisconsin, hospital utilization among working-age adults in Milwaukee was 38 % greater. However, when both Milwaukee’s poverty corridor and the poorest ZIP codes of other HRRs were excluded, the difference decreased to 5.4 %. Because low-income seniors were distributed more widely (Figure 2C), excluding the poverty corridor had less effect on their utilization, but even among seniors, removing the corridor from consideration reduced the difference in utilization between Milwaukee and other Wisconsin HRRs from 44 % to 29 %. Thus among working-age adults, the poverty corridor accounted for almost all of the difference in utilization between Milwaukee and other HRRs, and it accounted for almost half among seniors.
The highest-income area of Milwaukee was a rim of ZIP codes that capped the poverty corridor (Figure 4D). Compared to this affluent rim, utilization in the poverty corridor was more than double among working-age adults and one-third greater among seniors. If the utilization of health care throughout Milwaukee had been at the rate of the affluent rim, the number of hospital days per 1,000 would have been 37 % less among working-age adults, 13 % less among seniors, and 25 % less overall (Table 1). Thus, the poorest ZIP codes in Milwaukee were the major contributors to higher hospital utilization in the Milwaukee HRR.
Because Los Angeles is so populous, we were able to study two cohorts of working-age adults, ages 18–44 and 45–64, and third cohort of seniors. Of these, the 45–64-year-old cohort displayed the strongest relationships between hospital utilization both income and disability, and the analyses that follow focus on this cohort.
Like Milwaukee, Los Angeles has both affluent and poor areas, but unlike Milwaukee, where poverty is largely confined to a narrow corridor, poverty exists both in a central core and in scattered clusters elsewhere, many adjacent to affluent neighborhoods. Nonetheless, as in Milwaukee, there was a steeply inverse, curvilinear relationships between hospital days per 1,000 and MHI (r
2 = 0.440; Figure 5a). The magnitude of difference in utilization between ZIP codes containing the poorest and wealthiest deciles of the population was almost 3-fold.
The 18–44-year-old cohort utilized approximately one third as many hospital days per 1,000 as the 45–64-year-old cohort, and accordingly the statistical relationship was weaker (r
2 = 0.259), but the range of difference in utilization between the wealthiest and poorest ZIP codes was the same in both cohorts. Across both age cohorts, there was a strong relationship between the MHI of ZIP codes and the percent disabilities, which best fit a power function (r
2 = 0.613). As in Milwaukee, increases in hospital utilization in low-income ZIP codes could be partially explained by higher rates of admission for ambulatory care sensitive conditions, which across all adult ages were twice as frequent in the area of highest poverty as in the other areas of Los Angeles.42
The relationship between income and hospital utilization was weakest among seniors (r
2 = 0.220), as also observed in Milwaukee, and the magnitude of difference in utilization between the poorest and wealthiest deciles was half as great. When viewed on geomaps, both poverty and high rates of hospital utilization were distributed more broadly among seniors than among working-age adults. Reflecting this wider distribution, the ratio of seniors to 45–64 year olds in the lowest-income quintile was 20 % lower than in the highest. Finally, to an even greater degree than in Milwaukee, there were clusters of seniors in high-income census tracts within predominantly low-income ZIP codes in central Los Angeles. Thus, as in Milwaukee, ZIP codes proved to have less fidelity for assessing the economic characteristics of seniors than of working-age adults.
Mapping Los Angeles
Figure 6 displays ZIP codes that encompass quartiles of the 45–64-year-old population with the highest and the lowest income, the highest and lowest per cent disability, and highest and lowest rates of hospitalization utilization. The average rate of utilization in the quartile with the highest was 3.0-fold that of the lowest (Figure 6a) and MHI in the highest was 3.2-fold the lowest (Figure 6c). There was strong overlap between areas of high-income, low-disability, and low-hospital utilization. The converse was also true, with strong overlap between areas of low income, high percentages of disability, and high rates of hospital utilization. However, the overlap with low income (Figure 6c) was greatest for ZIP codes in which there were more than 10 % blacks (average = 30 %), while a group of low-income ZIP codes with fewer than 10 % blacks fell outside of zone of highest utilization.
Impact of Low-Income ZIP Codes
To assess the contribution of low-income ZIP codes to overall hospital utilization, we calculated the number of hospital days that would have been utilized in Los Angeles if the rates of utilization in all ZIP codes had been at the rate of those with MHIs above $75,000 (mean = $96,600), which included 5 % of the population. Had utilization everywhere been at this rate, it would have been 45 % less among 45–64 year olds, 31 % less among seniors and 37 % less overall (Table 1).
Thus, although separated by 2,000 miles and with populations that differed by a factor of five, the Milwaukee and Los Angeles HRRs proved to be more similar than different. Both had high degrees of income inequality; both had high rates of hospital utilization; and in both, the patterns of utilization followed underlying income differences, with the highest utilization in areas of greatest poverty and disability and the lowest in areas of greatest wealth and health.
San-Framento and California Counties
Because poverty in Los Angeles is not confined to a core area, it was not possible to carve-out a poverty corridor as in Milwaukee. Instead, we compared Los Angeles to a region in northern California (San-Framento) with a similar population but a lower rate of hospital utilization.
San-Framento is a 10-county area stretching from San Francisco to Sacramento. It has 90 % the population of Los Angeles and is principally urban, although its land mass is larger due to farming areas between urban centers. However, the sociodemographic characteristics of these two regions are quite different. San-Framento has fewer Hispanics (20 % vs. 45 % in Los Angeles), more non-Hispanic whites (50 % vs. 31 %) and more Asians (18 % vs. 12 %) but similar percentages of blacks (8 % vs. 9 %). MHI is one-third greater in San-Framento than in Los Angeles, and the poverty rate is one-third lower. Most important in terms of the current study, Medicare enrollees have been reported to use 40 % more hospital days in Los Angeles than in San-Framento,27 which is similar to the 39 % difference that we observed among seniors.
San-Framento vs. Los Ángeles
Despite these differences in sociodemographic characteristics and hospital utilization, the shapes of the curves relating MHI to hospital utilization were virtually identical in Los Angeles and San-Framento (Figure 5a, b), and the goodness of fit in San-Framento was also similar (r
2 = 0.545). How do these similarities reconcile with the overall differences in utilization between the Los Angeles and San-Framento?
The explanation emerges from a comparison of Figure 5a, b, which shows data for the 45–64-year-old cohort. While the arcs that define the regressions in each were virtually identical, there were more low-income, high-utilization ZIP codes in Los Angeles (the shaded area in Figure 5a) and more high-income, low-utilization ZIP codes in San-Framento (the non-shaded area in Figure 5b). Across all ZIP codes, utilization among ages 45–64 in Los Angeles was 27 % greater than in San-Framento. However, when only those ZIP codes with MHIs >$75,000 were compared, it was only 4 % greater. Similarly, among 18–44 year olds, utilization across all ZIP codes was 24 % greater in Los Angeles than in San-Framento but only 3 % greater in high-income ZIP codes, and among seniors, these differences were 39 % and 9 %. Thus, differences in aggregate hospital utilization between Los Angeles and San-Framento appear to be due principally to differences in the relative numbers of low-income ZIP codes.
Even though San-Framento had fewer low-income ZIP codes than Los Angeles, these contributed substantially to overall utilization (Table 1). Had utilization throughout San-Framento been at the rate of its highest-income ZIP codes, the overall rate would have been 18 % less, half the decrement in Los Angeles but substantial.
Figure 7 extends this analysis to the eight counties in California that have both high-income (MHI >$75,000) and low-income (MHI <$50,000) ZIP codes. When all ZIP codes were considered, the range of variation among counties in the 45–64-year-old cohort was 67 % and the coefficient of variation (COV) was 0.161. When only low-income ZIP codes were considered, the range widened to 103 % and the COV to 0.252; whereas, when only high-income ZIP codes were considered, the range of variation decreased to only 18 % and the COV fell to 0.056. Comparable results were obtained at ages 18–44 (COV = 0.127, 0.226, and 0.082, respectively) and at ages 65+ (COV = 0.153, 0.154, and 0.088). Thus, variation in hospital utilization among counties was strongly influenced by the proportion of low-income ZIP codes. Indeed, there was virtually no variation when only the more affluent ZIP codes were considered.
The contribution of low income to utilization was assessed in the seven counties other than Los Angeles included in Figure 7, as was previously done for Milwaukee, Los Angeles, and San-Framento (Table 1). Had utilization in each of these counties been at the rate of its wealthiest ZIP codes, there would have been 31 % fewer hospital days among 45–64 year olds and 16 % fewer among all adults. We extended this analysis to the 18 most populace California counties (from a total of 59), whose combined adult population of 24.8 million represents 85 % of the total adult population of California. Had utilization in each of these been at the rate of its affluent ZIP codes, there would have been 37 % fewer hospital days among 45–64 year olds and 26 % fewer among all adults (Table 1). Thus, the increased utilization in low-income ZIP codes throughout the most populace counties of California proved to be a major contributor to overall hospital utilization and to account for most of the observed variation in utilization among them.