The Application of Community-Based Information from the American Community Survey in a Large Integrated Health Care Organization



 

Zhi Liang PhD1; Claudia Nau PhD1; Fagen Xie PhD1; Ralph Vogel PhD2; Wansu Chen PhD1

Perm J 2020;25:20.010 [Full Citation]

https://doi.org/10.7812/TPP/20.010
E-pub: 12/09/2020

ABSTRACT

Background: The American Community Survey (ACS) is the largest household survey conducted by the US Census Bureau. We sought to describe the community-level characteristics derived from the ACS among enrollees of Kaiser Permanente Southern California (KPSC), evaluate the associations between ACS estimates and selective individual-level health outcomes, and explore how using different scales of the census geography and the linearity assumption affect the associations.

Methods: We examined the associations between track-level and block group-level ACS 5-year estimates and 4 individual-level Healthcare Effectiveness Data and Information Set (HEDIS) outcome measures (comprehensive diabetes care, postpartum care, antidepressant medication management, and childhood immunization status) using multilevel generalized linear models. Odds ratios and their 95% confidence intervals were estimated for every 10% increase in ACS measures.

Results: 6,357,841 addresses were successfully geocoded to at least the tract level. The community-level demographic, socioeconomic, residential, and other ACS measures varied among KPSC health plan enrollees. A majority of these ACS measures were associated with the selected HEDIS health outcomes. The directions of the effects were consistent across health outcomes; however, the magnitudes of the effect sizes varied. Within each HEDIS health outcome, the relative size of the effects appeared to remain similar. Differences between the census tract- and block group-level estimates were minor, especially for measures related to race/ethnicity, education, income, and occupation.

Conclusion: These findings support the use of many ACS measures at neighborhood levels to predict health outcomes. The geographic units might have little effect on the results. The linearity assumption should be made with caution.

INTRODUCTION

Mandated and governed by federal laws, the American Community Survey (ACS) is an ongoing community-based survey conducted by the US Census Bureau since 2005 to collect demographic, housing, and socioeconomic information including employment, migration, and disability. The ACS is the largest household survey that the Census Bureau administers, reaching approximately 295,000 addresses monthly (or 3.5 million annually). The individual responses are aggregated and made available at various scales of the census geography and administrative entities such as states, counties, metropolitan statistical areas, tracts, block groups, and census-designated places.1 Two types of ACS estimates are released annually: 1-year estimates for areas with a population of at least 65,000 people and 5-year estimates for areas down to the block group level.2

Information on socioeconomic status (SES), occupation, household type, or certain types of demographics is not routinely collected as part of patient care or upon health plan enrollment. The use of community-level measures as proxies for individual-level SES has gained popularity. In addition, numerous studies have demonstrated that neighborhood SES has an independent effect on individual health outcomes even after individual-level SES and other characteristics are controlled.3,4 These community-level measures include education, income, and occupation and have been shown to be associated with numerous health outcomes, including cardiovascular disease,5-7 obesity,8-10 infant mortality,11 mental disorders,12 certain types of cancer,13,14 chronic conditions,15 self-reported health,16,17 and life expectancy.18 Although the relationships between SES indicators such as poverty level, employment, and education and adverse health outcomes have been commonly examined,19,20 many ACS measures such as household type, travel time to work, and transportation to work have not been widely studied as potential community-level risk factors or confounders.

Among the various levels of census geographic units, researchers tend to select measures at the tract, block group, or zip code level. Most often, researchers only use 1 level of aggregation and do not compare results across multiple levels of aggregation. There is little conceptual guidance on the geographic scale of community effects on health. Many studies have shown that results of block group- and tract-level SES measures including income, education, and occupation fare similarly when used to predict health outcomes.21-33 However, research findings appear to be conflicting.21,22,25,34 Furthermore, most research relies on the linearity assumption to examine the associations between community-level characteristics and health outcomes, assuming that change in the effect of a community-level variable is linear across its continuum.35,36 However, whether this assumption is valid is unknown and the impact on researchers’ interpretations when the linearity assumption is violated is most often unclear. The linearity assumption is rarely, if ever, assessed in the analyses of ACS variables on health outcomes. If the linearity association was routinely violated, results of many publications would likely be biased.

The purpose of this study was 4-fold and aimed to 1) describe the community-level characteristics derived from the ACS among enrollees of each service area of Kaiser Permanente Southern California (KPSC), a large integrated health care organization; 2) evaluate the associations between the community-level characteristics based on ACS and selective individual-level health outcomes in the KPSC population; 3) understand the differences in the associations between the 2 statistical levels (tract level versus block group level) of census and ACS estimates; and 4) explore the impact of the linearity assumption on the modeling approach by using an ACS estimate. Results from this study may inform the choice of ACS predictors when predicting health outcomes, the geographic scale, and the risk of bias in models that do not test for nonlinearity of the predictor of ACS variables.

METHODS

Setting

KPSC is an integrated health care organization that provides comprehensive health care to more than 4.6 million Southern California residents. Members receive medical care in 15 hospitals and more than 232 medical offices in 10 counties of Southern California. Approximately 19% of the general population in the region is enrolled with KPSC. This study included KPSC health plan enrollees who had at least 1 month of active enrollment between 2009 and 2015. This study was approved on March 8, 2017 by the KPSC Institutional Review Board.

Geocoding

A census tract generally comprises a population sized between 600 and 3000 housing units, with an optimum size of 1500 housing units.37 Tracts are divided into census block groups that generally contain between 250 and 500 housing units, with an average size of 400 housing units.38 Block groups are further divided into census blocks, which are the smallest geographic units at which basic demographic data are tabulated and released by the Census Bureau.

Each geographic unit has a unique identifier. Census tracts are identified by 11-digit codes (first 2 digits for state, middle 3 digits for county, and last 6 digits for tract). Block group identifiers add an additional digit to the tract identifier and contain 12 digits (11 digits for tract plus 1 digit for block group). The Census Bureau refers to these geographic identifiers for tract, block group, and block identifiers, along with other geographic identifiers, as GEOIDs.39

A geocoding procedure was invoked to assign geographic coordinates and GEOIDs to an individual address. We first extracted addresses of KPSC enrollees from the KPSC Research Data Warehouse, which contains weekly updated information on enrollees’ demographic and address information. A preprocessing step was performed to remove special characters as part of the addresses that were clearly erroneous and to standardize the addresses based on the US Postal Service street format. If an individual had multiple addresses between 2009 and 2015, the most current address was selected for geocoding. Then MapMarker (version 28)40 was used to map individual addresses to GEOIDs. When an address could not be geocoded to the census block level, it was mapped to the lowest block group, census tract, or zip code level possible. Addresses that could not be mapped to a zip code, the highest level, remained uncoded. For the purposes of this study, we limited the study subjects to those whose addresses were successfully geocoded to the census tract or a more granular level (ie, block group or block).

ACS Measures

The GEOIDs derived from the geocoding process described above and the GEOIDs from the ACS 2011 to 2015 5-year estimates41 were linked at the census tract level. Tract- and block group-level estimates were assigned to each patient based on his or her GEOID.

To study the unadjusted associations between community-level characteristics and individual health outcomes, we chose ACS measures that are commonly used or likely to be of interest to researchers. The following analytical variables were created: bachelor’s degree or higher (sum of bachelor’s degree, graduate or professional degree, and doctorate degree), median household income of less than $50,000 (sum of less than $25,000 and $25,000 to $49,999), median household income of more than $100,000 (sum of $100,000 to $199,999 and $200,000 or greater), poverty level of 150% or greater (sum of between 150% and 199.99% and 200% or greater), and travel time to work of less than 30 minutes (sum of less than 15 minutes and 15 to 29 minutes).

To compare the differences in the association of ACS estimates at the block group and tract levels with selected health outcomes (see the outcomes described below), we also assigned the ACS characteristics at the block group level to the KPSC enrollees whose addresses were successfully geocoded at the block group level. Although the ACS 5-year estimates are available for all geographic areas down to the block group level,42 certain SES 5-year estimates are only available at the census tract level (Supplemental Table E1a). Results are only shown for individuals whose ACS characteristics are available at both the tract and block group levels.

Individual-Level Health Outcomes

Four individual-level Healthcare Effectiveness Data and Information Set (HEDIS)43 measures of KPSC enrollees in measurement year 2015 were selected. The 4 HEDIS measures were as follows: 1) hemoglobin A1C (HbA1C) control as part of comprehensive diabetes care (CDC), defined as a person with diabetes having HbA1C control (<8.0%) among patients aged 18 to 75 years; 2) postpartum care (PC), defined as having a postpartum visit 21 to 56 days after delivery among women who gave live births; 3) antidepressant medication management (AMM), defined as new antidepressant medication usage for at least 12 weeks for adult patients with a diagnosis of major depression (and no prior dispensing in the previous 105 days); and 4) childhood immunization status (CIS), defined as a child having all immunizations (including 4 diphtheria, tetanus, and acellular pertussis; 3 polio; 1 measles, mumps, and rubella; 3 Haemophilus influenzae type B; 3 hepatitis B; 1 chicken pox; 4 pneumococcal conjugate; 1 hepatitis A; 2 or 3 rotavirus; and 2 influenza vaccines) on or before their second birthday.

These 4 HEDIS measures were chosen to address various aspects of healthcare, including pediatric preventive care, maternal well-being, mental health care, and chronic conditions management. Furthermore, the organization’s internal performance monitoring suggested that there could be performance variation on these measures in subgroups of patients and that overall performance in these measures could be improved relative to external benchmarks. For detailed definitions of these measures, please refer to the National Committee for Quality Assurance website.44

Statistical Analysis

First, we examined the distribution of tract-level ACS estimates for all KPSC enrollees and for enrollees of each KPSC medical service area in 2015. The estimates were classified into the following 7 groups: race/ethnicity, education, household income, poverty, housing, employment/occupation, and others. For all ACS measures except for median household income, the average percentages of enrollees having the characteristics (eg, education less than ninth grade) were reported. The estimates were calculated by dividing the sum of the ACS measures for all enrollees (or enrollees in a medical service area) by the total number of enrollees (or enrollees in a medical service area). For example, for poverty level below 50%, we first assigned the tract-level estimate of poverty level below 50% from the ACS to individual enrollees who lived in the tract. Then we summed the estimate for all enrollees (or enrollees in a medical service area) regardless of track information. Finally, we divided the sum by the total number of enrollees (or enrollees in a medical service area). Similarly, median household income was reported as the average measure of all enrollees (or enrollees in a medical service area) at the tract level.

The associations between the selected tract-level characteristics derived from the ACS and the 4 individual-level HEDIS outcomes were examined using multilevel logistic regression analysis.45 Multilevel models allow us to accommodate for the nested structure of our data, with individuals (level 1) being nested within each census geographic unit (block group and tract) (level 2). We estimated 1 unadjusted model for each HEDIS measure and estimated odds ratios (ORs) and their 95% confidence intervals (CIs) for every 10% increase in ACS estimates (except for median household income). For median household income, the OR and 95% CI were estimated for every $10,000 increase in income estimate. Only study subjects who met the HEDIS inclusion and exclusion criteria were included in these analyses. Because diabetes is a common condition that applied to many health plan enrollees, the analyses of CDC were based on a random sample of 5% of enrollees diagnosed with diabetes.

The same multilevel logistic regression analysis was performed to examine the associations between the 4 individual-level HEDIS outcomes and block group-level ACS characteristics. The ORs and their 95% CIs were estimated and compared with those derived from the tract level.

Finally, we demonstrated the impact of the linearity assumption of the regression model by first assuming the relationship between an ACS estimate (12th-grade education or less) and the log odds of a health outcome (CIS) being linear. Then we categorized the study subjects into 3 groups (low = 0% to 29%, medium = 30% to 49%, and high = 50% to 100%) based on the values of the variable 12th-grade education or less, and we ran the regression models individually for each of the 3 groups (intervals) assuming that the ACS estimates and the log odds of the outcome were linear within each interval. The thresholds to divide the groups were selected based on visual examination (Supplemental Figure E1a).

The analyses were conducted using SAS (version 9.3 for Windows; SAS Institute, Cary, NC). The multilevel analysis was conducted using the GLIMMIX procedure with a random intercept.46,47

RESULTS

Geocoding

Of the 6,514,463 addresses we submitted for geocoding, 6,357,841 (97.6%), 6,219,724 (95.5%), and 6,199,120 (95.2%) were successfully geocoded to tract, block group, and block levels, respectively. An additional 154,803 addresses (2.4%) were geocoded to the zip code level. Only 1819 of the 6,514,463 addresses (0.03%) were not geocoded.

Tract-Level Characteristics of KPSC Enrollees

The distributions of ACS tract-level characteristics of KPSC enrollees in each of the 15 KPSC medical service areas (referred to as “service area” below) are presented in Table 1. These community-level estimates varied significantly from one service area to another. For example, 71.4% of enrollees in the Downey service area were Hispanic compared to 27.2% in the Woodland Hills service area. The percentage of non-Hispanic white enrollees ranged from 11.2% in the Downey service area to 51.8% in the Irvine service area. The percentage of Black or African American enrollees was highest in the West Los Angeles service area (28.2%) and lowest in the Irvine service area (1.3%). Having a bachelor’s degree or higher was most common in the Irvine service area (44.4%) and least common in the Downey service area (14.9%). The results showed that 41.6% of enrollees living in the Irvine service area had a median household income of $100,000 or greater, while only 20.1% of enrollees living in the Downey service area met or passed the income threshold. With regard to commute times, 25.6% of enrollees living in the Antelope Valley service area traveled 60 minutes or more to work compared with 6.4% in the San Diego service area.

Table 1. Distributions of American Community Survey tract-level estimates in Kaiser Permanente Southern California overall and in each of its 15 medical service areas

Demographic and socioeconomic status Antelope Valley Baldwin Park Downey Fontana Kern Los Angeles Anaheim Irvine Ontario Panorama City Riverside San Diego South Bay West Los Angeles Woodland Hills Overall
Race/ethnicity
      Hispanic 44.1 54.8 71.4 53.9 47.7 46.1 43.0 24.5 49.4 50.4 42.4 34.2 39.4 37.5 27.2 44.6
      Non-Hispanic white 34.0 16.1 11.3 29.4 39.8 28.4 32.7 51.8 29.4 33.6 40.6 46.0 26.2 24.1 54.0 33.5
 Non-Hispanic Black 14.4 2.2 7.2 9.1 4.6 4.1 1.9 1.3 5.8 3.5 5.3 4.8 13.9 28.2 3.0 6.8
 Non-Hispanic Asian 4.0 24.8 8.2 4.6 5.1 18.9 19.5 18.6 12.0 9.4 8.0 10.8 15.5 6.8 12.1 11.8
 Non-Hispanic other 3.5 2.1 1.9 3.0 2.8 2.5 2.9 3.8 3.4 3.1 3.7 4.2 5.0 3.4 3.7 3.3
Education
      12th grade or less 20.6 23.5 32.6 23.7 23.8 24.7 20.4 11.5 17.9 22.9 17.4 15.0 19.0 21.5 11.5 20.4
 High school graduate 27.9 25.2 25.3 27.3 26.9 18.4 21.1 15.2 22.7 22.4 25.4 20.1 20.7 18.7 19.1 22.2
 Some college, no degree 27.3 19.7 20.4 25.2 24.5 16.9 22.0 20.5 24.7 21.0 26.5 23.2 22.4 20.7 21.7 22.4
 Associate’s degree 8.6 7.8 6.7 8.0 7.6 6.5 7.4 8.4 8.1 7.3 8.2 9.1 8.1 6.2 8.1 7.8
 Bachelor’s degree 10.2 16.9 10.6 10.1 11.4 22.1 19.7 28.3 17.4 18.8 14.5 20.7 19.8 20.3 25.2 17.8
 Graduate degree or higher 5.4 6.9 4.4 5.7 5.8 11.4 9.4 16.1 9.2 7.6 8.0 11.9 10.0 12.6 14.4 9.4
Household income ($)
      Median 56.1 64.1 55.4 56.7 57.9 56.4 72.0 87.8 71.6 66.4 70.1 68.5 65.7 55.7 84.8 65.8
 <25,000 23.1 18.0 21.7 22.3 23.0 27.0 16.7 13.6 16.2 19.2 16.7 18.7 20.7 28.0 14.7 19.8
 25,000-49,999 23.7 22.8 25.8 24.3 23.7 23.1 20.3 16.4 19.8 21.5 20.7 21.1 21.7 23.0 17.2 21.8
 50,000-99,999 32.2 32.8 32.4 32.9 31.3 26.6 31.3 28.3 33.2 30.9 33.0 30.9 29.8 26.1 28.5 30.9
 100,000-199,999 18.1 22.1 17.5 17.9 18.4 17.1 24.6 29.3 25.1 22.7 24.7 23.1 21.5 16.4 28.0 21.7
 ≥200,000 2.9 4.3 2.6 2.6 3.6 6.2 7.1 12.4 5.7 5.7 4.9 6.2 6.3 6.5 11.6 5.8
Poverty
      Family below FPL income 20.0 13.1 18.0 19.6 19.9 20.2 14.0 10.4 12.6 15.4 13.8 14.2 15.4 20.5 10.6 15.7
      Household below FPL income 16.7 10.5 15.6 16.5 17.2 16.8 11.2 7.5 9.9 12.5 10.9 11.0 12.5 17.0 7.8 12.7
 FPL (%)                                
  <50 8.5 3.5 5.7 6.6 6.5 5.8 4.0 2.9 3.7 4.4 4.1 4.8 4.9 6.7 3.0 4.9
  50-99.99 8.1 7.1 9.9 10.0 10.8 11.0 7.2 4.7 6.2 8.1 6.8 6.2 7.7 10.4 4.8 7.9
  100-199.99 22.3 21.2 25.0 22.4 21.1 22.6 17.3 12.3 17.3 19.5 16.4 16.0 18.2 20.6 12.4 18.9
  >200 61.1 68.2 59.4 61.0 61.6 60.6 71.5 80.1 72.8 68.0 72.7 73.0 69.2 62.3 79.8 68.3
  Household receiving public assistance 5.3 3.7 6.1 6.2 6.2 4.1 3.0 2.1 3.5 3.6 3.6 3.0 4.2 5.4 2.4 4.1
  Household with no car 5.6 5.5 7.3 5.2 5.9 13.8 5.3 4.1 4.4 6.4 4.0 5.9 7.3 12.4 5.0 6.5
Housing
      Households with ≥1 person per room 5.7 11.5 18.7 10.2 8.6 14.4 13.6 7.4 7.6 11.8 7.1 6.9 11.6 10.6 6.1 10.3
Employment/occupation
      Unemployed 6.9 6.1 6.6 8.1 7.2 6.6 5.5 4.7 6.5 6.9 7.7 5.7 6.4 7.3 5.3 6.5
 Management occupation                                
  Women 1.8 2.4 1.7 1.7 1.7 3.8 2.8 4.0 2.1 3.0 1.8 3.7 3.1 4.2 4.2 2.9
  Men 2.3 3.3 2.2 2.2 2.3 5.3 4.3 7.9 3.4 3.6 3.0 6.1 4.2 5.2 7.1 4.3
Other
      Born in the US 81.0 61.5 65.3 77.9 80.3 57.4 66.0 71.4 73.8 63.8 78.2 74.8 70.5 70.4 69.3 70.7
 Population aged >65 y 9.6 13.4 10.0 10.1 10.1 12.7 12.0 13.5 10.1 10.7 10.8 12.7 12.7 12.2 14.2 11.8
 English speaking only 70.0 36.9 32.5 57.9 59.8 36.0 47.4 61.6 54.4 43.1 63.9 61.8 54.0 54.3 59.5 52.5
 Spanish speaking only 25.7 39.8 59.5 36.2 34.7 39.7 33.5 18.5 33.0 42.0 27.4 25.9 31.1 34.5 20.6 34.0
 Marriage 48.9 48.9 45.8 48.3 51.1 43.5 51.3 53.5 50.2 48.6 52.5 49.2 45.4 38.4 51.7 48.5
 Medicare 11.9 13.9 10.8 11.6 11.8 13.4 12.4 13.9 11.1 11.3 11.8 13.8 13.3 12.9 14.7 12.7
 Medicaid 26.6 22.6 27.5 28.1 27.6 24.7 20.0 13.0 18.2 22.9 19.6 16.8 21.1 23.8 13.9 21.5
 Household move in last 12 mo 11.8 10.1 10.1 14.3 16.4 12.6 13.2 14.6 14.1 11.7 14.6 14.6 13.2 14.1 13.0 13.4
 Female householder with own children aged <18 y 9.7 6.8 10.8 9.8 10.0 7.1 7.2 5.2 7.5 7.4 6.8 6.8 8.2 9.0 5.5 7.8
 Family household                                
  Married-couple family 53.1 55.4 51.4 53.3 53.4 41.6 56.2 55.0 55.2 51.7 58.8 50.8 45.3 33.9 54.1 51.3
  Male only 7.0 8.1 8.9 7.8 7.8 7.1 6.5 5.1 6.6 7.5 6.3 5.6 6.8 7.0 5.5 6.8
  Female only 17.5 17.0 21.0 17.9 16.1 15.6 14.7 10.9 14.8 15.2 12.8 13.3 17.1 18.5 12.2 15.5
 Nonfamily household                                
  Living alone 18.1 15.4 14.9 16.5 17.8 27.4 17.0 22.0 18.7 19.4 17.0 22.5 24.9 32.3 21.6 20.4
  Not living alone 4.3 4.1 3.8 4.5 4.9 8.3 5.6 7.0 4.7 6.2 5.1 7.8 5.9 8.3 6.6 6.0
  Travel time to work (min)                                
  <15 22.0 17.4 17.1 22.5 29.4 18.4 20.5 22.7 20.5 17.8 19.2 21.1 20.3 16.3 23.2 20.5
  15-29 32.6 30.6 34.3 35.9 42.9 33.0 36.2 39.4 32.8 32.0 29.1 43.0 37.0 34.6 31.0 35.5
  30-59 19.8 36.8 37.4 26.6 20.2 37.1 33.5 29.7 30.1 35.1 30.3 29.5 32.1 38.7 31.9 31.7
  ≥60 25.6 15.2 11.2 15.0 7.5 11.5 9.8 8.2 16.6 15.1 21.4 6.4 10.6 10.4 13.9 12.3
 Transportation to work                                
  Car 92.7 89.5 88.5 91.7 91.9 77.4 89.4 87.9 91.3 86.2 90.5 87.5 86.5 79.7 86.0 87.7
  Public transportation 1.5 3.4 5.1 1.6 1.1 10.8 3.2 1.8 2.1 5.4 1.4 3.0 4.9 9.2 3.0 3.8
  Other 1.8 3.0 3.6 2.4 4.0 6.3 4.0 3.7 2.7 3.6 2.8 3.8 4.5 5.3 4.3 3.7
  Work at home 4.0 4.1 2.8 4.3 3.0 5.5 3.4 6.6 3.9 4.8 5.3 5.7 4.1 5.8 6.7 4.8

FPL = federal poverty level.

Associations Between Tract-Level Characteristics and Individual Health Outcomes

The sample size and the frequency for each of the 4 outcome measures are presented in Table 2. For example, of 255,187 enrollees who met the eligibility criteria for CDC, 174,775 (68.5%) received CDC. The associations between the ACS tract-level characteristics and CDC are presented in Figure 1. A 10% increase in the percentages of non-Hispanic white and Asian populations was associated with an increase in the odds of CDC by 7% (OR = 1.07; 95% CI = 1.05 to 1.09) and 5% (OR = 1.05; 95% CI = 1.02 to 1.08), respectively, while a 10% increase in the percentage of the Hispanic population was associated with a decrease in odds of CDC by 7% (OR = 0.93; 95% CI = 0.92 to 0.95). A 10% increase in the percentage of enrollees having a bachelor’s degree or higher in the community was associated with a 15% increase in the odds of CDC (OR = 1.15; 95% CI = 1.13 to 1.16), and an increase of $10,000 in median household income in the community was linked to increased odds of CDC by 6% (OR = 1.06; 95% CI = 1.04 to 1.07). Additionally, a 10% increase in the percentages of residents (households) who were Medicare enrollees, older than age 65 years, spoke English only, were married, were born in the United States, worked at home, lived in an area with a poverty level of 150% or greater, or held a management occupation (for women and men) was associated with an increase in the odds of CDC. In addition, a 10% increase in the percentages of residents (households) who were Medicaid enrollees, receiving public assistance, had no car, lived in an area with a poverty level below 50%, were unemployed, or use of public transportation to work was associated with a decrease in the odds of CDC (Figure 1).

Table 2. Four Healthcare Effectiveness Data and Information Set outcome measures of Kaiser Permanente Southern California enrollees in 2015

HEDIS measure Denominator Numerator Percentage
Comprehensive diabetes care 255,187 174,775 68.5
Postpartum care 38,029 33,867 89.1
Antidepressant medication management 58,727 40,388 68.8
Childhood immunization status 32,890 19,028 57.9

HEDIS = Healthcare Effectiveness Data and Information Set.

Figure 1

Figure 1. Odds ratios of comprehensive diabetes care and postpartum care for every 10% increase in American Community Survey estimates at the tract level. Odds ratios and their 95% confidence intervals were estimated for every 10% increase in American Community Survey estimates (except for median household income). For median household income, the odds ratio and the 95% confidence interval were estimated for every $10,000 increase in income estimate. ACS = American Community Survey; CDC = comprehensive diabetes care; CI = confidence interval; FPL = federal poverty level; HH = household; NH = non-Hispanic; OR = odds ratio; PC = postpartum care; TT = travel time to work; TW = transportation to work.

The relationships between the ACS measures and the other 3 health outcomes (PC, AMM, and CIS) in terms of the significance, direction, and relative strength of effects are illustrated in Figures 1 and 2. The effects of tract-level ACS estimates for all 4 health outcomes are summarized in Table 3. A 10% increase in non-Hispanic whites or non-Hispanic Asians increased the odds for all 4 health outcomes (p < 0.01), while the same level of increase in Hispanics or non-Hispanic Black residents decreased the odds (p < 0.01), except for the association between non-Hispanic Black residents and CDC (Table 3). Low level of education (12th grade or less), low household income (less than $50,000), all unfavorable poverty measures, households with 1 or more person per room, unemployment, Spanish speaking only, Medicaid, and male- or female-only households were associated with unfavorable results (p < 0.001). In contrast, high household income ($100,000 or greater), federal poverty level of 150% or greater, management occupation, Medicare, and married-couple family led to positive results (at least p < 0.01). Although some of the ACS estimates studied were not significantly related to one or more of the outcomes, none had conflicting results (ie, related to one outcome positively and to another outcome negatively) except for measures related to travel time and transportation to work (Table 3).

Figure 2

Figure 2. Odds ratios of antidepressant medication management and childhood immunization status for every 10% increase in American Community Survey estimates at the track level. Odds ratios and their 95% confidence intervals were estimated for every 10% increase in American Community Survey estimates (except for median household income). For median household income, the odds ratio and the 95% confidence interval were estimated for every $10,000 increase in income estimate. ACS = American Community Survey; AMM = antidepressant medication management; CI = confidence interval; CIS = childhood immunization status; FPL = federal poverty level; HH = household; NH = non-Hispanic; OR = odds ratio; PC = postpartum care; TT = travel time to work; TW = transportation to work.

Table 3. Summarized effects of tract-level American Community Survey estimates on 4 Healthcare Effectiveness Data and Information Set measures

ACS estimate CDCa PC AMM CIS
Race/ethnicity
      Hispanic − − − − − − − − − − − −
 Non-Hispanic white + + + + + + + + + +
 Non-Hispanic Black O − − − − − − −
 Non-Hispanic Asian + + + + + + + + + + +
Education
 12th grade or less − − − − − − − − − − − −
 Bachelor’s degree or higher + + + + + + + + + + + +
Household income ($)
 <25,000 − − − − − − − − − − − −
 <50,000 − − − − − − − − − − − −
 ≥100,000 + + + + + + + + + + + +
 ≥200,000 + + + + + + + + + + + +
 Median + + + + + + + + + + + +
Poverty
 Family below FPL income − − − − − − − − − − − −
 Household below FPL income − − − − − − − − − − − −
 FPL <50% − − − − − − − − − − −
 FPL ≥150% + + + + + + + + + + + +
 Household receiving public assistance − − − − − − − − − − − −
 Household with no car − − − − − − − −
Housing
      Household with ≥1 person per room − − − − − − − − − − −
Employment/occupation
      Unemployed − − − − − − − − − − − −
 Management occupation        
  Women + + + + + + + + + + + +
  Men + + + + + + + + + + + +
Other
      Born in the US + + + O + + + + + +
 Population aged >65 y + + + + + + + + + + +
 English speaking only + + + O + + + O
 Spanish speaking only − − − − − − − − − − − −
 Marriage + + + + + + + + + + +
 Medicare + + + O + + + + + +
 Medicaid − − − − − − − − − − − −
 Household move in the last 12 mo O − − − O − − −
 Female householder with own children aged <18 y − − − − − − − − − − − −
 Family household        
  Married-couple family + + + + + + + + + +
  Male only − − − − − − − − − − − −
  Female household only − − − − − − − − − − − −
 Nonfamily household        
  Living alone + + + + + + +
  Not living alone O + + + + +
 Travel time to work (min)        
  <15 O − − + + + O
  <30 O − − − + + O
  ≥60 O + + − − − − − −
 Transportation to work        
  Car O O O − − −
  Public transportation − − O − − − +
  Work at home + + + O + + + + +

aORs and their 95% CIs were estimated for every 10% increase in ACS estimates (except for median household income). For median household income, the OR and the 95% CI were estimated for every $10,000 increase in income estimate. O indicates no association at the 95% level. When OR > 1, + indicates p < 0.05, + + indicates p < 0.01, and + + + indicates p < 0.001. When OR < 1, − indicates p < 0.05, − − indicates p < 0.01, and − − − indicates p < 0.001.

ACS = American Community Survey; AMM = antidepressant medication management; CDC = comprehensive diabetes care; CI = confidence interval; CIS = childhood immunization status; FPL = federal poverty level; HEDIS = Healthcare Effectiveness Data and Information Set; OR = odds ratio; PC = postpartum care.

Effects of Tract-Level versus Block Group-Level Estimates

The comparisons between census tract-level and census block group-level estimates are presented in Table 4. For CDC and PC, the effects of ACS estimates at the tract level and those at the block group level were comparable. For AMM, the ORs derived from the 2 census levels were comparable except for the following ACS estimates: households receiving public assistance, worked at home, population older than 65 years, family households with a male householder without a wife, and family households with a female householder without a husband. Finally, the associations between the ACS estimates and CIS were also similar except for households receiving public assistance and family households with a male householder without a wife.

Table 4. Odds ratios (95% confidence intervals) for each Healthcare Effectiveness Data and Information Set measure for every 10% increase in American Community Survey estimates at the census tract and block group levels

ACS estimate CDC PC AMM CIS
Tract Block group Tract Block group Tract Block group Tract Block group
Race/ethnicity
 Hispanic 0.93 (0.92-0.95) 0.94 (0.93-0.95) 0.96 (0.94-0.97) 0.96 (0.95-0.97) 0.91 (0.90-0.91) 0.91 (0.91-0.92) 0.96 (0.95-0.97) 0.96 (0.96-0.97)
 Non-Hispanic white 1.07 (1.05-1.09) 1.06 (1.05-1.08) 1.03 (1.01-1.04) 1.02 (1.01-1.04) 1.11 (1.10-1.12) 1.11 (1.10-1.12) 1.02 (1.01-1.03) 1.02 (1.01-1.03)
  Non-Hispanic Black 1.00 (0.97-1.03) 1.00 (0.97-1.02) 0.96 (0.92-0.99) 0.96 (0.94-0.99) 0.91 (0.90-0.93) 0.93 (0.91-0.94) 0.96 (0.94-0.98) 0.97 (0.95-0.99)
  Non-Hispanic Asian 1.05 (1.02-1.08) 1.04 (1.01-1.07) 1.15 (1.11-1.19) 1.13 (1.10-1.16) 1.06 (1.05-1.08) 1.05 (1.04-1.07) 1.14 (1.12-1.16) 1.12 (1.10-1.14)
Education
 12th grade or less 0.90 (0.88-0.92) 0.91 (0.89-0.93) 0.92 (0.90-0.94) 0.92 (0.91-0.94) 0.85 (0.84-0.86) 0.86 (0.85-0.87) 0.95 (0.93-0.96) 0.95 (0.94-0.97)
 Bachelor’s degree or higher 1.11 (1.08-1.13) 1.10 (1.07-1.12) 1.12 (1.10-1.15) 1.11 (1.09-1.13) 1.15 (1.13-1.16) 1.14 (1.13-1.15) 1.12 (1.10-1.13) 1.11 (1.09-1.12)
Household income ($)
 <25,000 0.92 (0.89-0.95) 0.94 (0.91-0.96) 0.86 (0.83-0.88) 0.88 (0.86-0.90) 0.87 (0.86-0.89) 0.91 (0.89-0.92) 0.91 (0.89-0.93) 0.93 (0.92-0.95)
 <50,000 0.93 (0.91-0.95) 0.94 (0.92-0.96) 0.90 (0.88-0.92) 0.91 (0.89-0.93) 0.90 (0.89-0.91) 0.92 (0.91-0.93) 0.93 (0.92-0.94) 0.94 (0.93-0.95)
 ≥100,000 1.10 (1.07-1.13) 1.09 (1.06-1.11) 1.12 (1.10-1.15) 1.10 (1.08-1.12) 1.13 (1.12-1.15) 1.11 (1.10-1.12) 1.10 (1.08-1.12) 1.09 (1.08-1.10)
 ≥200,000 1.26 (1.19-1.35) 1.19 (1.13- 1.26) 1.26 (1.18-1.34) 1.19 (1.13-1.25) 1.31 (1.27-1.35) 1.24 (1.22-1.28) 1.25 (1.20-1.29) 1.22 (1.18-1.26)
 Median 1.06 (1.04-1.07) 1.05 (1.04-1.07) 1.07 (1.06-1.09) 1.06 (1.05-1.08) 1.08 (1.07-1.08) 1.06 (1.05-1.07) 1.06 (1.05-1.07) 1.05 (1.04-1.06)
Poverty
 Household receiving public assistance 0.72 (0.65-0.79) 0.79 (0.74-0.85) 0.70 (0.64-0.761) 0.81 (0.756-0.86) 0.60 (0.57-0.63) 0.74 (0.71-0.77) 0.69 (0.64-0.73) 0.80 (0.77-0.84)
Housing
Household with ≥1 person/room 0.87 (0.83-0.90) 0.89 (0.86-0.92) 0.91 (0.88-0.94) 0.93 (0.90-0.95) 0.79 (0.78-0.81) 0.83 (0.82-0.84) 0.96 (0.93-0.98) 0.97 (0.95-0.99)
Other
Population aged ≥65 y 1.29 (1.21-1.38) 1.18 (1.13-1.24) 1.10 (1.02-1.18) 1.10 (1.05-1.16) 1.25 (1.22-1.29) 1.18 (1.16-1.21) 1.17 (1.12- 1.23) 1.10 (1.06-1.14)
Marriage 1.13 (1.09-1.18) 1.09 (1.05-1.13) 1.06 (1.02-1.10) 1.04 (1.01-1.07) 1.14 (1.116-1.16) 1.10 (1.08-1.119) 1.08 (1.05-1.11) 1.06 (1.04-1.09)
Family household
 Married-couple family 1.04 (1.01-1.07) 1.03 (1.00-1.05) 1.05 (1.02-1.08) 1.04 (1.02-1.06) 1.04 (1.03-1.06) 1.03 (1.02-1.04) 1.03 (1.01-1.05) 1.02 (1.01-1.04)
 Male only 0.75 (0.68-0.82) 0.86 (0.81-0.92) 0.79 (0.72-0.87) 0.91 (0.86-0.97) 0.63 (0.60-0.66) 0.80 (0.78-0.83) 0.83 (0.78-0.88) 0.93 (0.90-0.97)
 Female only 0.83 (0.79-0.87) 0.90 (0.86-0.93) 0.82 (0.79-0.87) 0.88 (0.85-0.91) 0.75 (0.73-0.77) 0.83 (0.82-0.85) 0.89 (0.86-0.93) 0.92 (0.90-0.95)
Nonfamily household
 Living alone 1.07 (1.03-1.11) 1.05 (1.02-1.08) 1.04 (1.00- 1.07) 1.02 (0.99-1.05) 1.09 (1.07-1.11) 1.07 (1.05-1.08) 1.03 (1.01-1.06) 1.02 (1.00-1.04)
 Not living alone 1.03 (0.94-1.13) 1.02 (0.93-1.06) 1.10 (1.01-1.20) 1.06 (0.99-1.12) 1.21 (1.159-1.26) 1.13 (1.09-1.162) 1.06 (1.00-1.13) 1.01 (0.97-1.06)
Travel time to work (min)
 <15 1.01 (0.97-1.06) 1.01 (0.98-1.05) 0.92 (0.88-0.97) 0.96 (0.93- 0.99) 1.06 (1.03-1.08) 1.04 (1.02-1.05) 0.97 (0.94-1.00) 0.99 (0.97-1.02)
 <30 0.99 (0.96-1.03) 1.01 (0.98-1.04) 0.90 (0.87-0.93) 0.94 (0.92-0.96) 1.03 (1.01-1.05) 1.02 (1.01-1.03) 1.02 (1.00-1.04) 1.02 (1.00-1.04)
 ≥60 0.99 (0.94-1.05) 1.00 (0.96-1.05) 1.10 (1.05-1.16) 1.03 (0.99-1.08) 0.93 (0.91-0.96) 0.96 (0.94-0.98) 0.84 (0.81-0.869) 0.89 (0.867-0.92)
 Transportation to work
  Car 1.03 (0.98-1.08) 1.02 (0.98-1.07) 0.99 (0.94-1.04) 1.01 (0.97- 1.05) 1.02 (0.99-1.05) 1.02 (1.00-1.04) 0.93 (0.90-0.96) 0.94 (0.92-0.97)
  Public transportation 0.88 (0.82-0.94) 0.88 (0.83-0.93) 0.96 (0.90-1.03) 0.96 (0.90-1.01) 0.80 (0.77-0.83) 0.83 (0.81-0.86) 1.06 (1.00-1.11) 1.05 (1.00-1.09)
  Work at home 1.33 (1.18-1.51) 1.22 (1.12-1.33) 1.10 (0.98-1.24) 1.02 (0.94-1.10) 1.59 (1.50-1.68) 1.28 (1.23-1.33) 1.7 (1.08-1.27) 1.09 (1.03-1.15)

ACS = American Community Survey; AMM = antidepressant medication management; CDC = comprehensive diabetes care; CIS = childhood immunization status; FPL = federal poverty level; HEDIS = Healthcare Effectiveness Data and Information Set; PC = postpartum care.

For all ACS estimates and all 4 outcomes, the ORs and 95% CIs derived from both levels were very similar most of the time. However, when the point estimates from the 2 levels were not the same, the one from the block level was always closer to null (1.0) compared to the corresponding estimate derived from the tract level.

Impact of Linearity Assumption of the Modeling Approach

When we assumed that the relationship between the ACS estimate of 12th-grade education or less and the log odds of CIS is linear, for every 10% increase in the percentage of 12th-grade education or less, the odds of CIS decreased by 5% (OR = 0.95; CI = 0.93 to 0.96; n = 32,890). When we categorized the study subjects into 3 groups based on the value of 12th-grade education or less (low = 0% to 29%, n = 24,370; medium = 30% to 49%, n = 6732; and high = 50% to 100%, n = 1788) and ran the regression models individually for each of the 3 groups (low, medium, and high) assuming that the relationship between 12th-grade education or less and the log odds of CIS is linear only within each one of the 3 intervals, the ORs and the 95% CIs for the 3 groups were 0.86 (0.83 to 0.89), 1.06 (0.96 to 1.17), and 1.25 (1.02 to 1.53), respectively.

DISCUSSION

We observed stark variations in socioeconomic community characteristics among KPSC medical service areas when considering ACS census tract-level characteristics of KPSC enrollees. For example, health plan enrollees who lived in the Irvine medical service area were 3.0 times more likely to have a bachelor’s degree or higher, and they were 2.1 times more likely to have a median household income of $100,000 or greater than enrollees who lived in the Downey service area. Our results also revealed that many ACS characteristics were consistently associated with individual-level HEDIS health outcomes in the KPSC enrollee population. More specifically, living in neighborhoods with a higher education level, higher income level, and lower poverty level was favorably associated with selected HEDIS health outcomes, while living in neighborhoods with a lower education level, lower income level, and higher poverty level was negatively associated with these health outcomes. For example, a 10% increase in the percentages of residents who were unemployed decreased the odds of CDC by 27% (Figure 1). Conversely, a 10% increase in the percentages of female or male residents with management occupations increased the odds of CDC by 62% and 54%, respectively (Figure 1).

Prior studies have investigated the relationships of neighborhood-level SES with HbA1C or medication. For example, after classifying block group-level SES to quintiles, Cozier et al48 reported that Black women living in the poorest neighborhoods had the least favorable HbA1C (p for trend = 0.07) when the study subjects were ranked according to quintiles of neighborhood SES. Using the data of 63,053 patients with diabetes, Gabert et al49 found strong correlations between zip code-level income, education, and insurance coverage and targeted measures based on HbA1C, blood pressure, low-density lipoprotein cholesterol, and tobacco cessation and suggested using this approach to identify areas that can be targeted for intervention. Using HEDIS measures for the quality of outpatient depression care, Akincigil et al50 found a higher level of adherence to antidepressant medications in study subjects with a higher median household income at the zip code level. These findings are consistent with ours.

On the contrary, our finding on the relationship between community-level race estimates and childhood immunization was not in harmony with a previous study led by Lieu et al,19 in which a higher percentage of Asians and Hispanics in a block group was shown to be associated with higher income and a higher rate of immunization. Each unit increase in the percentage of households with incomes below the poverty line and in the percentage of persons with graduate degrees increased the odds of underimmunization by 2.19 (95% CI = 1.44 to 3.34) and 2.63 (95% CI = 1.72 to 4.02), respectively.19 In the KPSC population, we found that neighborhoods with higher percentages of non-Hispanic white and Asian residents were associated with improved health outcomes, including childhood immunization (Figures 1 and 2; Table 3).

Many studies have shown that results of block group and tract-level SES measures including income, education, and occupation produced similar outcomes.21-33 However, the conclusions from the literature are inconsistent. Krieger et al23 reported tighter CIs generated by block group-level estimates compared to those derived from the tract level. Soobader et al21 suggested that researchers consider measures from smaller geographic units, as “measures from smaller geographic units may produce results for aggregate SES measures that are slightly less biased,” although both “the tract and block group measures were as effective as the individual measures in controlling for SES confounding in the interpretation of the effect of race on health outcomes.” Nevertheless, there was evidence that tract-level estimates most consistently revealed associations between community socioeconomic characteristics and residents’ health.34 Interestingly, Lee et al51 concluded that community-level effects have been shown to vary by the geographic scale at which they operate and by the aggregating community characteristics. The current study found no difference or very minor differences in the point estimates between the tract- and block group-level ACS measures related to race/ethnicity, education, and income as well as their 95% CIs. However, our findings revealed that for certain ACS estimates (eg, households receiving public assistance, population older than 65 years, family household/male householder only, family household/female householder only, and transportation to work) and some health outcomes, the ORs derived at the block group level were all closer to null (1.0) compared with those derived at the tract level. As a result, an association at the tract level could be statistically significant, while the corresponding one at the block group is not. For example, the association between a travel time of 60 minutes or more and PC was 1.10 (1.05 to 1.16; statistically significant at the 95% level) at the tract level, while the one derived at the block group level was only 1.03 (0.99 to 1.08; not statistically significant at the 95% level). The larger effects observed at the tract level could be partially due to the smaller margin of error (MOE) of these ACS estimates compared to those at the block group level. The MOE is a measure of variation of an ACS estimate from the population value; the MOE is released for each ACS estimate as part of the ACS data and can be used to calculate CIs.52 Even for 5-year ACS estimates, MOEs can be large for small geographic or little populated areas (eg, block group or even tracts in rural areas). It is unclear whether the larger effects for tract-level estimates observed in the current study, compared with some of the block group-level estimates, were closer to the true effects of the measures at KPSC. We recommend using tract-level estimates for race/ethnicity, education, and income, and reporting the effects at both levels for other community-level estimates, if they are the interest of the study and the measures at both levels are available.

The linearity assumption is commonly applied to neighborhood-level measures in research studies.35,36,52 However, the actual effects of these neighborhood-level measures on a health outcome may not be linear. The current study explored the impact of the linearity assumption by using 2 examples. In one example, the linearity assumption did not seem to be violated. However, in the other example, a 10% increase in 12th-grade education or less was negatively associated with CIS in enrollees whose values were in the lowest group (0% to 29%) but was positively associated with CIS in those whose values were in the highest group (50% to 100%). Using vital records and hospital data, Shmool et al53 modeled nonlinear relationships and effect modification between nitrogen dioxide and area-level depreciation on term birth weight and compared the results with those derived from linear models. Users need to be cautious about the application of the linearity assumption and consider quadratic or nonlinear effects in multilevel models when necessary.

There is a large body of literature that has linked neighborhood and community characteristics to health.54 Prior research has traced the history of racial and social segregation that has shaped an urban landscape in which health-furthering resources and health risks such as access to healthy foods, opportunity for exercise, neighborhood safety, traffic, and pollution tend to cluster in neighborhoods where the majority of residents have low income or are people of color.55-61 Our findings provide further evidence of the systematic effects of community on health and provide guidance to researchers for the analysis of these pervasive effects. The consistency of the observed effects across the 4 selective outcomes was very high in our analyses. These data suggest that there may be opportunities for Kaiser Permanente and other organizations for outreach to help address care gaps in disadvantaged communities.

Several limitations should be acknowledged. First, the ACS community-level characteristics were shown to be associated with the 4 individual-level HEDIS health outcomes selected. However, it is unclear whether the observed associations also exist for other health outcomes. Second, the measures of associations we present in this study are unadjusted and thus may reflect compositional effects by age, gender, or race/ethnicity and need to be interpreted with caution. Some of the reported associations could also be driven at least partly by confounding variables such as local infrastructure resources that are more numerous and better maintained in higher-income neighborhoods. Third, many of the community-level ACS estimates can be used simultaneously or jointly as a single index in health research. Readers are encouraged to apply the wide range of community-level estimates to study the joint effects of these characteristics.

CONCLUSION

In summary, the community-level demographic, socioeconomic, residential, and other ACS measures vary among KPSC health plan enrollees. A majority of these ACS measures were associated with the selected HEDIS health outcomes. These ACS measures may be considered as potential risk factors in studies of health-related outcomes in future research. The differences between the census tract- and block group-level estimates were minor, especially for measures related to race/ethnicity, education, income, and occupation. Finally, the linear assumption should be examined before applying ACS measures either as potential risk factors or as confounders.

Supplemental Material

aSupplemental Material is available at: www.thepermanentejournal.org/files/2020/20.010suppl.pdf.

Disclosure Statement

The author(s) have no conflicts of interest to disclose.

Acknowledgments

The authors thank Ms. Melanie Balasanian for creating the reference database. The project described was supported, in part, by the Department of Research and Evaluation within the Southern California Permanente Medical Group with funds from the Medical Group and the Kaiser Permanente Community Benefit Fund. The contents and conclusions are solely the responsibility of the author and do not necessarily represent the official views of the Department of Research and Evaluation and the Southern California Permanente Medical Group.

Author Affiliations

1Department of Research and Evaluation, Kaiser Permanente Southern California, Pasadena, CA

2Department of Clinical Analysis, Kaiser Permanente Southern California, Pasadena, CA

Corresponding Author

Wansu Chen, PhD (wansu.chen@kp.org)

Author Contributions

Zhi Liang, PhD, and Wansu Chen, PhD, conceived and led the design of the study and drafted the manuscript. Zhi Liang extracted the data and conducted all of the analyses. Claudia Nau, PhD, provided critical suggestions to improve the design as well as the presentation of the results. All authors participated in the design of the study, interpreted the results, critically reviewed the manuscript, and have given final approval to the manuscript.

How to Cite this Article

Liang Z, Nau C, Xie F, Vogel R, Chen W. The application of community-based information from the American Community Survey in a large integrated health care organization. Perm J 2020;25:20.010. DOI: 10.7812/TPP/20.010

References

1. About the American Community Survey. Accessed January 9, 2020. https://www.census.gov/programs-surveys/acs/about.html

2. Understanding and using ACS single-year and multiyear estimates. Accessed January 9, 2020. https://www.census.gov/content/dam/Census/library/publications/2018/acs/acs_general_handbook_2018_ch03.pdf

3. Ellen G, Mijanovich T, Dillman KN. Neighborhood effects on health: Exploring the links and assessing the evidence. J Urban Aff 2001 Sep;23(3-4):391-408. DOI: https://doi.org/10.1111/0735-2166.00096

4. Pickett KE, Pearl M. Multilevel analyses of neighbourhood socioeconomic context and health outcomes: A critical review. J Epidemiol Community Health 2001 Feb;55(2):111-22. DOI: https://doi.org/10.1136/jech.55.2.111, PMID:11154250

5. Kaplan GA, Keil JE. Socioeconomic factors and cardiovascular disease: A review of the literature. Circulation 1993 Oct;88(4 Pt 1):1973-98. DOI: https://doi.org/10.1161/01.cir.88.4.1973, PMID:8403348

6. Ramsay SE, Morris RW, Whincup PH, et al. The influence of neighbourhood-level socioeconomic deprivation on cardiovascular disease mortality in older age: Longitudinal multilevel analyses from a cohort of older British men. J Epidemiol Community Health 2015 Dec;69(12):1224-31. DOI: https://doi.org/10.1136/jech-2015-205542

7. Pollack CE, Slaughter ME, Griffin BA, Dubowitz T, Bird CE. Neighborhood socioeconomic status and coronary heart disease risk prediction in a nationally representative sample. Public Health 2012 Oct;126(10):827-35. DOI: https://doi.org/10.1016/j.puhe.2012.05.028, PMID:23083844

8. Ogden CL, Lamb MM, Carroll MD, Flegal KM. Obesity and socioeconomic status in children and adolescents: United States, 2005-2008. 2010 Dec;(51):1-8. PMID:21211166

9. Roswall J, Almqvist-Tangen G, Holmén A, et al. Overweight at four years of age in a Swedish birth cohort: Influence of neighbourhood-level purchasing power. BMC Public Health 2016 Jul;16(1):546. DOI: https://doi.org/10.1186/s12889-016-3252-1, PMID:27400741

10. Powell-Wiley TM, Ayers C, Agyemang P, et al. Neighborhood-level socioeconomic deprivation predicts weight gain in a multi-ethnic population: Longitudinal data from the Dallas Heart Study. Prev Med 2014 Sep;66:22-7. DOI: https://doi.org/10.1016/j.ypmed.2014.05.011, PMID:24875231

11. Donabedian A, Rosenfeld LS, Southern EM. Infant mortality and socioeconomic status in a metropolitan community. Public Health Rep 1965 Dec;80(12):1083. DOI: https://doi.org/10.2307/4592617, PMID:4954380

12. Hudson CG. Socioeconomic status and mental illness: Tests of the social causation and selection hypotheses. Am J Orthopsychiatry 2005 Jan;75(1):3-18. DOI: https://doi.org/10.1037/0002-9432.75.1.3, PMID:15709846

13. Yost K, Perkins C, Cohen R, Morris C, Wright W. Socioeconomic status and breast cancer incidence in California for different race/ethnic groups. Cancer Causes Control 2001;12(8):703-11. DOI: https://doi.org/10.1023/a:1011240019516

14. Ward E, Jemal A, Cokkinides V, et al. Cancer disparities by race/ethnicity and socioeconomic status. CA Cancer J Clin 2004 Mar;54(2):78-93. DOI: https://doi.org/10.3322/canjclin.54.2.78

15. Connolly V, Unwin N, Sherriff P, Bilous R, Kelly W. Diabetes prevalence and socioeconomic status: A population based study showing increased prevalence of type 2 diabetes mellitus in deprived areas. J Epidemiol Community Health 2000 Mar;54(3):173-7. D DOI: https://doi.org/0.1136/jech.54.3.173

16. Foraker RE, Rose KM, Chang PP, et al. Socioeconomic status and the trajectory of self-rated health. Age Ageing 2011 Nov;40(6):706-11. DOI: https://doi.org/10.1093/ageing/afr069, PMID:21737460

17. Wada K, Higuchi Y, Smith DR. Socioeconomic status and self-reported health among middle-aged Japanese men: Results from a nationwide longitudinal study. BMJ Open 2015 Jun;5(6):e008178. DOI: https://doi.org/10.1136/bmjopen-2015-008178, PMID:26109119

18. Signorello LB, Cohen SS, Williams DR, Munro HM, Hargreaves MK, Blot WJ. Socioeconomic status, race, and mortality: A prospective cohort study. Am J Public Health 2014 Dec;104(12):e98-107. DOI: https://doi.org/10.2105/AJPH.2014.302156, PMID:25322291

19. Lieu TA, Ray GT, Klein NP, Chung C, Kulldorff, M. Geographic clusters in underimmunization and vaccine refusal. Pediatrics 2015 Feb;135(2):280-9. DOI: https://doi.org/10.1542/peds.2014-2715, PMID:25601971

20. Schmittdiel JA, Dyer WT, Marshall CJ, Bivins R. Using neighborhood-level census data to predict diabetes progression in patients with laboratory-defined prediabetes. Perm J 2018;22:18.096. DOI: https://doi.org/10.7812/TPP/18-096, PMID:30296398

21. Soobader M, LeClere FB, Hadden W, Maury B. Using aggregate geographic data to proxy individual socioeconomic status: Does size matter? Am J Public Health 2001 Apr;91(4):632. DOI: https://doi.org/10.2105/ajph.91.4.632, PMID:11291379

22. Roux AVD, Kiefe CI, Jacobs DRJr, et al. Area characteristics and individual-level socioeconomic position indicators in three population-based epidemiologic studies. Ann Epidemiol 2001 Aug;11(6):395-405. DOI: https://doi.org/10.1016/s1047-2797(01)00221-6

23. Krieger N. Overcoming the absence of socioeconomic data in medical records: Validation and application of a census-based methodology. Am J Public Health 1992 May;82(5):703-10. DOI: https://doi.org/10.2105/ajph.82.5.703, PMID:1566949

24. Krieger N. Women and social class: A methodological study comparing individual, household, and census measures as predictors of black/white differences in reproductive history. J Epidemiol Community Health 1991 Mar;45(1):35-42. DOI: https://doi.org/10.1136/jech.45.1.35, PMID:2045742

25. Geronimus AT, Bound J. Use of census-based aggregate variables to proxy for socioeconomic group: Evidence from national samples. Am J Epidemiol 1998 Sep;148(5):475-86. DOI: https://doi.org/10.1093/oxfordjournals.aje.a009673, PMID:9737560

26. Geronimus AT, Bound J, Neidert LJ. On the validity of using census geocode characteristics to proxy individual socioeconomic characteristics. J Am Stat Assoc 1996 Jun;91(434):529-37. DOI: https://doi.org/10.1080/01621459.1996.10476918

27. Cherkin DC, Grothaus L, Wagner EH. Is magnitude of co-payment effect related to income? Using census data for health services research. Soc Sci Med 1992 Jan;34(1):33-41. DOI: https://doi.org/10.1016/0277-9536(92)90064-w, PMID:1738854

28. Greenwald HP, Polissar NL, Borgatta EF, McCorkle R. Detecting survival effects of socioeconomic status: Problems in the use of aggregate measures. J Clin Epidemiol 1994 Aug;47(8):903-9. DOI: https://doi.org/10.1016/0895-4356(94)90194-5

29. Carr-Hill R, Rice N. Is enumeration district level an improvement on ward level analysis in studies of deprivation and health? J Epidemiol Community Health 1995 Dec;49(Suppl 2):S28-9. DOI: https://doi.org/10.1136/jech.49.suppl_2.s28, PMID:8594129

30. Mustard CA, Derksen S, Berthelot JM, Wolfson M. Assessing ecologic proxies for household income: A comparison of household and neighbourhood level income measures in the study of population health status. Health Place 1999 Jun;5(2):157-71. DOI: https://doi.org/10.1016/s1353-8292(99)00008-8, PMID:10670997

31. Reijneveld SA, Verheij RA, de Bakker DH. The impact of area deprivation on differences in health: Does the choice of the geographical classification matter? J Epidemiol Community Health 2000 Apr;54(4):306-13. DOI: https://doi.org/10.1136/jech.54.4.306, PMID:10827914

32. Hyndman JC, Holman CD, Hockey RL, Donovan RJ, Corti B, Rivera J. Misclassification of social disadvantage based on geographical areas: Comparison of postcode and collector’s district analyses. Int J Epidemiol 1995 Feb;24(1):165-76. DOI: https://doi.org/10.1093/ije/24.1.165, PMID:7797339

33. Krieger N, Chen JT, Waterman PD, Soobader MJ, Subramanian SV, Carson R. Geocoding and monitoring of US socioeconomic inequalities in mortality and cancer incidence: Does the choice of area-based measure and geographic level matter?: The Public Health Disparities Geocoding Project. Am J Epidemiol 2002 Sep;156(5):471-82. DOI: https://doi.org/10.1093/aje/kwf068, PMID:12196317

34. Krieger N, Chen JT, Waterman PD, Rehkopf DH, Subramanian SV. Race/ethnicity, gender, and monitoring socioeconomic gradients in health: A comparison of area-based socioeconomic measures--the Public Health Disparities Geocoding Project. Am J Public Health 2003 Oct;93(10): 1655-71. DOI: https://doi.org/10.2105/ajph.93.10.1655, PMID:14534218

35. Laraia BA, Karter AJ, Warton EM, Schillinger D, Moffet HH, Adler N. Place matters: Neighborhood deprivation and cardiometabolic risk factors in the diabetes Study of Northern California (DISTANCE). Soc Sci Med 2012 Apr;74(7):1082-90. DOI: https://doi.org/10.1016/j.socscimed.2011.11.036

36. Jokela M. Does neighbourhood deprivation cause poor health? Within-individual analysis of movers in a prospective cohort study. J Epidemiol Community Health 2015 Sep;69(9):899-904. DOI: https://doi.org/10.1136/jech-2014-204513, PMID:25878354

37. Census tracts and block numbering areas. Accessed January 9, 2020. https://www2.census.gov/geo/pdfs/reference/GARM/Ch10GARM.pdf

38. Census blocks and block groups. Accessed January 9, 2020. https://www2.census.gov/geo/pdfs/reference/GARM/Ch11GARM.pdf

39. Understanding Geographic Identifiers (GEOIDs). Accessed January 9, 2020. https://www.census.gov/programs-surveys/geography/guidance/geo-identifiers.html

40. MapMarker US Accessed January 9, 2020. https://www.pitneybowes.com/us/location-intelligence/geographic-information-systems/mapmarker.html

41. American Community Survey 2011-2015 files. Accessed January 9, 2020. https://www2.census.gov/programs-surveys/acs/summary_file/2015/data/

42. American Community Survey 5-year data (2009-2018). Accessed January 9, 2020. https://www.census.gov/data/developers/data-sets/acs-5year.html

43. HEDIS and performance measurement. Accessed December 2, 2020. http://www.ncqa.org/HEDISQualityMeasurement/WhatisHEDIS.aspx

44. National Committee for Quality Assurance. HEDIS measures and technical resources. https://www.ncqa.org/hedis/measures/

45. Greenland S. Principles of multilevel modelling. Int J Epidemiol 2000 Feb;29(1):158-67. DOI: https://doi.org/10.1093/ije/29.1.158, PMID:10750618

46. SAS/STAT 9.3 user’s guide: The GLIMMIX procedure. Accessed January 9, 2020. https://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_glimmix_a0000001398.htm

47. Li J, Alterman T, Deddens JA. Analysis of large hierarchical data with multilevel logistic modeling using PROC GLIMMIX. SUGI 2006 Mar;31:26-9.

48. Cozier YC, Albert MA, Castro-Webb N, et al. Neighborhood socioeconomic status in relation to serum biomarkers in the Black Women’s Health Study. J Urban Health 2016 Apr;93(2):279-91. DOI: https://doi.org/10.1007/s11524-016-0034-0, PMID:27000125

49. GabertR, ThomsonB, GakidouE, RothG. Identifying high-risk neighborhoods using electronic medical records: A population-based approach for targeting diabetes prevention and treatment interventions. PLoS One 2016 Jul;11(7):e0159227. DOI: https://doi.org/10.1371/journal.pone.0159227

50. Akincigil A, Bowblis JR, Levin C, Walkup JT, Jan S, Crystal S. Adherence to antidepressant treatment among privately insured patients diagnosed with depression. Med Care 2007 Apr;45(4):363-9. DOI: https://doi.org/10.1097/01.mlr.0000254574.23418.f6, PMID:17496721

51. Lee BA, Reardon SF, Firebaugh G, Farrell CR, Matthews SA, O'Sullivan D. Beyond the census tract: Patterns and determinants of racial segregation at multiple geographic scales. Am Sociol Rev 2008 Oct;73(5):766-91. DOI: https://doi.org/10.1177/000312240807300504, PMID:25324575

52. Using American community survey (ACS) estimates and margins of error. Accessed January 9, 2020. https://www.census.gov/programs-surveys/acs/guidance/training-presentations/acs-moe.html

53. Shmool JL, Bobb JF, Ito K, et al. Area-level socioeconomic deprivation, nitrogen dioxide exposure, and term birth weight in New York City. Environ Res 2015 Oct;142:624-32. DOI: https://doi.org/10.1016/j.envres.2015.08.019, PMID:26318257

54. Sood MM, Tangri N, Hiebert B, et al. Geographic and facility-level variation in the use of peritoneal dialysis in Canada: A cohort study. CMAJ Open 2014 Jan;2(1):E36. DOI: https://doi.org/10.9778/cmajo.20130050, PMID:25077124

55. Diez-Roux AV, Nieto FJ, Muntaner C, et al. Neighborhood environments and coronary heart disease: A multilevel analysis. Am J Epidemiol 1997 Jul;146(1):48-63. DOI: https://doi.org/10.1093/oxfordjournals.aje.a009191, PMID:9215223

56. Massey DS, Gross AB, Shibuya K. Migration, segregation, and the geographic concentration of poverty. Am Sociol Rev 1994 Jun;59(3):425-45. DOI: https://doi.org/10.2307/2095942

57. Duncan DT, Kawachi I, Kum S, et al. Trees, population health and community wellbeing: Benefits of trees a spatially explicit approach to the study of socio-demographic inequality in the spatial distribution of trees across Boston neighborhoods. Spat Demogr 2005;2(1):1-29. DOI: https://doi.org/10.1007/BF03354902

58. Wilhelm M, Qian L, Ritz B. Outdoor air pollution, family and neighborhood environment, and asthma in LA FANS children. Health Place 2009 Mar;15(1):25-36. DOI: https://doi.org/10.1016/j.healthplace.2008.02.002, PMID:18373944

59. Auchincloss AH, Diez Roux AV, Mujahid MS, Shen M, Bertoni AG, Carnethon MR. Neighborhood resources for physical activity and healthy foods and incidence of type 2 diabetes mellitus: The Multi-Ethnic Study of Atherosclerosis. Arch Intern Med 2009 Oct;169(18):1698-704. DOI: https://doi.org/10.1001/archinternmed.2009.302, PMID:19822827

60. Walker RE, Keane CR, Burke JG. Disparities and access to healthy food in the United States: A review of food deserts literature. Health Place 2010 Sep;16(5):876-84. DOI: https://doi.org/10.1016/j.healthplace.2010.04.013, PMID:20462784

61. Nau C, Ellis H, Huang H, et al. Exploring the forest instead of the trees: An innovative method for defining obesogenic and obesoprotective environments. Health Place 2015 Sep;35:136-46. DOI: https://doi.org/10.1016/j.healthplace.2015.08.002, PMID:26398219

Keywords: American Community Survey, community-level estimates, geocode, health outcome, HEDIS, linearity, multilevel model, neighborhood-level estimates

ETOC

Click here to join the eTOC list or text ETOC to 22828. You will receive an email notice with the Table of Contents of The Permanente Journal.

CIRCULATION

2 million page views of TPJ articles in PubMed from a broad international readership.

Indexing

Indexed in MEDLINE, PubMed Central, EMBASE, EBSCO Academic Search Complete, and CrossRef.


                                             

 

 

ISSN 1552-5775 Copyright © 2021 thepermanentejournal.org

All Rights Reserved