Low income, community poverty and risk of end stage renal disease

Study design and population

REGARDS is a population-based cohort of black and white U.S. adults aged 45 years and older [21, 22]. A stratified random sample of eligible individuals was recruited from North Carolina, South Carolina, and Georgia, Alabama, Mississippi, Tennessee, Arkansas and Louisiana (56%), with the remaining 44% of the sample recruited from the other 40 contiguous U.S. states and the District of Columbia. Recruitment was from January 2003 to October 2007. Written consent was obtained from each participant. The institutional review boards (research ethics committees) of the participating institutions approved the study [21]. A full list of participating REGARDS investigators and institutions can be found at http://www.regardsstudy.org.

There were 30,183 REGARDS participants who completed an in-home examination. For this study, we excluded participants with addresses that could not be geocoded (due to their only providing post office boxes rather than street addresses; n = 6,333), those treated with dialysis at study enrollment (n = 116), and those missing follow up data within the study period (n = 420); final N = 23,314.

Data collection and measures

Individual participant data were obtained during a telephone interview and subsequent in-home examination. During the interview, we ascertained participant’s age, sex, race, educational attainment, health insurance status, annual household income, and history of hypertension and/or diabetes. Home addresses used for the in-home visit were geocoded to the U.S. census tract level. Blood pressure was measured twice during the in-home visit with an aneroid sphygmomanometer following three minutes of sitting with both feet on the floor. The average of the two blood pressure measurements was used. Hypertension was defined as self-reported use of antihypertensive medications, a systolic blood pressure ≥140 mmHg, or a diastolic blood pressure ≥90 mmHg. Venous blood was collected for serum creatinine and glucose. Diabetes was defined by either self-report, prescribed oral hypoglycemic medications or insulin, fasting glucose ≥126 mg/dL or non-fasting glucose ≥200 mg/dL. Serum creatinine was measured by colorimetric reflectance spectrophotometry using the Ortho Vitros Clinical Chemistry System 950IRC instrument (Johnson & Johnson Clinical Diagnostics, Rochester, NY). The creatinine assay was calibrated to a creatinine standard determined by isotope dilution mass spectrometry [23]. Estimated glomerular filtration rate (eGFR) was calculated using the available single serum creatinine measurement for each participant, and the 4-variable estimating equation modified for the international calibration standards published by the Chronic Kidney Disease Epidemiology Collaboration [24].

Self-reported household income and the degree of concentrated poverty in the community of residence at the time of the in-home interview were used as individual and community level exposures in our hierarchical models described below. Annual household income for a participant was ascertained by asking “Is your annual household income from all sources less than…?”, and then specifying income levels from U.S. dollars (USD) <5,000 to USD >150,000 [25]. We grouped income levels into five categories: refused to provide, <$20,000; ≥$20,000 and < $35,000; ≥$35,000 and < $75,000; and ≥ $75,000.

The geocoded home address was used to assign a level of geographically concentrated county poverty using data obtained from the 2000 U.S. Census. The degree of poverty in a county was calculated by combining two county-level attributes: 1) a standardized Z score (Z = [county mean poverty level- mean poverty level across all counties]/ [standard deviation of county poverty across all counties]); and 2) the degree of local spatial autocorrelation of the Z score with the score of nearby counties [26]. The Z-score is a dimensionless measure of the deviation of a value from the overall group mean in units of the measure’s standard deviation. For example, a Z-score of 2.0 for any variable means that the value is 2 SD above and a Z-score of –2.0 is 2 SD below the overall mean for that measure and is comparable to a similar Z-score for any other measure. We defined concentrated poverty as counties with a Z-score greater than 2 above the mean poverty rate for U.S. counties and a local spatial autocorrelation (Moran’s score) Z-score greater than 2. In a similar fashion we categorized counties having concentrated affluence and those with neither concentrations of poverty or affluence. After inspection of the resulting distribution of counties on the U.S. map we defined the following categories of concentrated spatial wealth: outlier poverty, extremely high poverty, very high poverty, high poverty, neither, high affluence and outlier affluence [26]. Outlier counties are those that are more impoverished or affluent than would be expected given the level of poverty or affluence of the adjoining counties.

To confirm that geographically concentrated county poverty data from 2000 reflected more recent data, we examined community material disadvantage across county poverty levels using a measure developed and validated by Diez-Roux and colleagues [27]. Briefly, a neighborhood poverty score was calculated as the sum of the Z-scores of six measures of material well-being collected by the 2010 US Census at the census block level. The variables used in the construction of the neighborhood score included log of the median household income; log of the median value of housing units; the percentage of households receiving interest, dividend, or net rental income; the percentage of adults 25 years of age or older who had completed high school; the percentage of adults 25 years of age or older who had completed college; and the percentage of employed persons 16 years of age or older in executive, managerial, or professional specialty occupations. A summary score (neighborhood poverty Z-score) was defined as the sum of the six Z-scores. Additionally, we examined the Gini index, which is a measure of wealth inequality (range of 0-1, with 0 equal to complete equality, and 1 complete inequality) [28]. We also assessed the percentage of households in each county poverty category who were (a) living below poverty threshold, (b) female-headed, (c) with household income < $30,000, (d) without a vehicle, (e) vacant, (f) receiving public assistance, and (g) unemployed. Finally, we examined the rural urban commuting area (RUCA) codes for each participant’s zip code to determine whether they resided in a metropolitan (RUCA score 1-3), micropolitan (score 4-6) or rural (score 7-10) area [29].

Our outcome of interest was incident ESRD identified through linkage of REGARDS study participants with the United States Renal Data System (USRDS). The USRDS ESRD database is a national registry of patients receiving renal replacement therapy [30]. This analysis included incident ESRD cases through August 2009 defined as the first date of dialysis documented on the ESRD Medical Evidence Form of the Centers for Medicare and Medicaid Services (form CMS-2728) and recorded by the USRDS. Person-time was censored at ESRD, death, or date of last follow-up phone contact, whichever occurred first.

Statistical analysis

We used ANOVA and chi-square tests to assess differences across county poverty category. We used multivariable Cox proportional hazards models to examine the independent association between income and county poverty measures and incident ESRD, while accounting for death as a competing risk using the Fine and Gray method [31]. We began with models that included interaction terms between individual income and county poverty. An interaction was assessed based on the statistical significance of these interaction terms. As none were statistically significant, our adjusted Model 1 included age, sex, race, and education. Model 2 also included income. The proportional hazards assumption was tested by examining the log-log survival plots-2 * log likelihood plots. Statistical analyses were performed using SAS version 9.2 (SAS Institute, Cary NC).

Participant characteristics by county poverty category

Nearly half (49.9%) of respondents lived in counties where there was neither concentrated poverty nor affluence and 32.8% lived in counties with varying degrees of concentrated poverty (mean county household income 2 or more standard deviations below the national mean that was correlated with contiguous and nearby counties) (Table 1). Individuals living in counties characterized by greater concentrated poverty were more likely to be female, of black race and with less than a high school education. As was expected from the original description of the concentrated poverty counties by Holt [26] nearly all of the respondents living in counties with concentrated poverty were residents of the U.S. “stroke belt” and these individuals were more likely to live in rural areas than those residing in affluent counties. In contrast, none of the respondents living in outlier poverty counties were “stroke belt” residents and nearly all (98.4%) resided in metropolitan areas.

The association between reported household income and concentrated county poverty was as expected. Incomes of less than $20,000 were more frequently reported by individuals living in the most concentrated poverty. There was no clear pattern for incomes in the $20,000 to $34,999 category, while participants with incomes > $35,000 or > $75,000 comprised a greater proportion of those residing in more affluent counties (Table 1). For example, 10.9% of individuals living in counties in the extremely concentrated poverty category reported personal household incomes of $75,000 or greater compared to 17.2% of individuals in the more affluent category. Participants who refused to report household income comprised a greater proportion of those residing in concentrated poverty counties than those in the affluent counties.

Several clinical factors varied across county poverty categories. Individuals living in concentrated poverty counties had a greater prevalence of diabetes and hypertension than those in affluent counties; and estimated GFR was slightly higher among persons in concentrated poverty counties.

County characteristics by county poverty category

Incident ESRD by county poverty category

Median follow up for our study was 6.0 years, and a total of 2,615 participants died during follow up. There were 158 incident cases of ESRD of follow up. Incident ESRD rates were 178.8 per 100,000 person-years (10⁵ py) (95% confidence interval [CI], 83.0 to 339.5/10⁵ py) among individuals in high poverty outlier counties and were 76.3 /10⁵ py (95% CI 34.9 to 144.9/10⁵ py) in affluent outlier counties, p for trend = 0.06 (Figure 1). In unadjusted competing risk models, persons residing in high poverty outlier counties had higher incidence of ESRD (which was not statistically significant) when compared to those persons residing in counties with neither high poverty nor affluence [hazard ratio (HR) 1.54, 95% Confidence Interval (CI) 0.75-3.20]. This association was markedly attenuated following adjustment for socio-demographic factors (age, sex, race, education, and income); HR 0.96, 95% CI 0.46-2.00. In this multivariable competing risk model, only income was independently associated with risk of ESRD [hazard ratio (HR) 3.75, 95% confidence interval (CI) (1.62-8.64) comparing the < $20,000 group to the > =$75,000 group] (Table 3). There were no statistically significant associations of county poverty with incident ESRD. Furthermore, there was no statistically significant interaction between income and county poverty category. We did not perform further adjustment of our models for other factors noted in Table 1, such as baseline kidney function, diabetes and hypertension status, out of concern for potentially inducing biased estimates [32].