Advertisement
Mayo Clinic Proceedings Home

County-Level Contextual Characteristics and Disparities in Life Expectancy

  • Yan Xie
    Affiliations
    Clinical Epidemiology Center, Research and Education Service, VA Saint Louis Health Care System, Saint Louis, MO

    Veterans Research & Education Foundation of Saint Louis, Saint Louis, MO

    Department of Epidemiology and Biostatistics, College for Public Health and Social Justice, Saint Louis University, Saint Louis, MO
    Search for articles by this author
  • Benjamin Bowe
    Affiliations
    Clinical Epidemiology Center, Research and Education Service, VA Saint Louis Health Care System, Saint Louis, MO

    Veterans Research & Education Foundation of Saint Louis, Saint Louis, MO

    Department of Epidemiology and Biostatistics, College for Public Health and Social Justice, Saint Louis University, Saint Louis, MO
    Search for articles by this author
  • Yan Yan
    Affiliations
    Clinical Epidemiology Center, Research and Education Service, VA Saint Louis Health Care System, Saint Louis, MO

    Division of Public Health Sciences, Department of Surgery, Washington University School of Medicine, Saint Louis, MO
    Search for articles by this author
  • Miao Cai
    Affiliations
    Clinical Epidemiology Center, Research and Education Service, VA Saint Louis Health Care System, Saint Louis, MO

    Department of Epidemiology and Biostatistics, College for Public Health and Social Justice, Saint Louis University, Saint Louis, MO
    Search for articles by this author
  • Ziyad Al-Aly
    Correspondence
    Correspondence: Address to Ziyad Al-Aly, MD, Clinical Epidemiology Center, VA Saint Louis Health Care System, 915 N Grand Blvd, 151-JC, Saint Louis, MO 63106.
    Affiliations
    Clinical Epidemiology Center, Research and Education Service, VA Saint Louis Health Care System, Saint Louis, MO

    Veterans Research & Education Foundation of Saint Louis, Saint Louis, MO

    Department of Medicine, Washington University School of Medicine, Saint Louis, MO

    Nephrology Section, Medicine Service, VA Saint Louis Health Care System, Saint Louis, MO

    Institute for Public Health, Washington University in Saint Louis, Saint Louis, MO
    Search for articles by this author

      Abstract

      Objective

      To estimate the contribution of county-level contextual factors to differences in life expectancy in the United States.

      Methods

      We used a counterfactual approach to estimate the years of life expectancy lost associated with 45 potentially modifiable county-level contextual characteristics in the United States in the year 2016. Contextual data and life expectancy data were obtained from the County Health Ranking Project and the U.S. Small-Area Life Expectancy Estimates Project, respectively.

      Results

      Median census-tract–level life expectancy was 78.90 (interquartile range, 76.30-81.00) years, and the range across census tracts spanned 41.20 years. Large variations in life expectancy existed within and between states and within and between counties; the gap between counties was 20.30 years and gaps within counties ranged from 0 to 34.60 years. An array of 45 county-level factors was associated with 4.30 years of life expectancy loss. County-level adult smoking, food insecurity, adult obesity, physical inactivity, college education, and median household income were associated with 1.24-, 0.89-, 0.58-, 0.35-, 0.33-, and 0.14-year losses in life expectancy, respectively; and altogether were associated with a 3.53-year loss in life expectancy. The contribution of contextual factors to years of life expectancy lost varied among states and was more pronounced in states with lower life expectancy and in areas of increased socioeconomic deprivation and increased percentage of Black race.

      Conclusion

      Substantial geographic variation in life expectancy was observed. Six county-level contextual factors were associated with a 3.53-year loss in life expectancy. The findings may inform and help prioritize approaches to reduce inequalities in life expectancy in the United States.

      Abbreviations and Acronyms:

      ADI (area deprivation index), CHR (County Health Rankings), IQR (interquartile range), LOWESS (locally weighted scatterplot smoothing), OR (odds ratio), USALEEP (US Small-Area Life Expectancy Estimates Project)
      Life expectancy has dramatically improved in the United States during the past several centuries, owing largely to significant economic development and scientific advances. However, significant inequities and disparities in life expectancy remain and may be widening, and recent data suggest that improvement in life expectancy may have stalled and even that some gains have been reversed.
      • Arias E.
      • Escobedo L.A.
      • Kennedy J.
      • Fu C.
      • Cisewki J.
      U.S. small-area life expectancy estimates project: methodology and results summary.
      GBD 2017 DALYs and HALE Collaborators
      Global, regional, and national disability-adjusted life-years (DALYs) for 359 diseases and injuries and healthy life expectancy (HALE) for 195 countries and territories, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017.
      • Foreman K.J.
      • Marquez N.
      • Dolgert A.
      • et al.
      Forecasting life expectancy, years of life lost, and all-cause and cause-specific mortality for 250 causes of death: reference and alternative scenarios for 2016-40 for 195 countries and territories.
      • Dowell D.
      • Arias E.
      • Kochanek K.
      • et al.
      Contribution of opioid-involved poisoning to the change in life expectancy in the United States, 2000-2015.
      • Torjesen I.
      Inequalities in life expectancy are widening, data confirm.
      • Singh G.K.
      • Kogan M.D.
      • Slifkin R.T.
      Widening disparities in infant mortality and life expectancy between Appalachia and the rest of the United States, 1990-2013.
      Contextual characteristics, which describe the social, demographic, economic, cultural, political, and physical environment, and attributes related to access and quality of medical care, are associated with life expectancy.
      • Dwyer-Lindgren L.
      • Bertozzi-Villa A.
      • Stubbs R.W.
      • et al.
      Inequalities in life expectancy among US counties, 1980 to 2014: temporal trends and key drivers.
      • Chetty R.
      • Stepner M.
      • Abraham S.
      • et al.
      The association between income and life expectancy in the United States, 2001-2014.
      • Sasson I.
      Trends in life expectancy and lifespan variation by educational attainment: United States, 1990-2010.
      • Bowe B.
      • Xie Y.
      • Xian H.
      • Lian M.
      • Al-Aly Z.
      Geographic variation and US county characteristics associated with rapid kidney function decline.
      However, the contribution of contextual characteristics to differences in life expectancy has not been comparatively evaluated. Undertaking an analysis of the contribution of multiple major contextual characteristics to years of life expectancy lost will facilitate the comparative assessment of their relative importance and may help inform policy priorities aimed at improving and reducing disparities of life expectancy in the United States.
      In this work, we leveraged the availability of county-level contextual data from the County Health Rankings (CHR) data sets, census tract–level life expectancy data from the U.S. Small-Area Life Expectancy Estimates Project (USALEEP), and other data sets to systematically examine the associations between life expectancy and U.S. county contextual characteristics in several domains, including social and economic factors, physical environment, health behaviors, and clinical care. We then developed analyses to estimate the contribution of each contextual factor to losses in life expectancy at the national and state levels.

      Methods

      Life Expectancy Data

      The USALEEP data provided census-tract–level life expectancy data. USALEEP used 2010 to 2015 National Vital Statistics System mortality data, 2010 decennial census population counts, and American Community Survey 5-year estimates (2011-2015) to calculate life expectancy at the census-tract level based on abridged life tables.
      • Arias E.
      • Escobedo L.A.
      • Kennedy J.
      • Fu C.
      • Cisewki J.
      U.S. small-area life expectancy estimates project: methodology and results summary.
      Age intervals (nx) in the life table were defined as less than 1, 1 to 4, and then every 10 years as an interval, ending with age older than 85 years.
      • Arias E.
      • Escobedo L.A.
      • Kennedy J.
      • Fu C.
      • Cisewki J.
      U.S. small-area life expectancy estimates project: methodology and results summary.
      In total, data from 65,662 census tracks from all states except Maine and Wisconsin was available.
      • Arias E.
      • Escobedo L.A.
      • Kennedy J.
      • Fu C.
      • Cisewki J.
      U.S. small-area life expectancy estimates project: methodology and results summary.
      Data for life expectancy at birth, life expectancy at age interval x (ex), total number of person-years lived until age interval x (Tx), person-years lived within the interval (nLx), number dying within the interval (ndx), number surviving before age interval x (lx), and probability of dying within the interval (nqx) for each age interval were used for analyses.

      US County-Level Contextual Factors

      County-level contextual factor data were obtained from the CHR project between 2014 and 2019. The CHR curates county-level measures from a variety of publicly available data sources and is updated every year.
      • Remington P.L.
      • Catlin B.B.
      • Gennuso K.P.
      The County Health Rankings: rationale and methods.
      In this work, we used potentially modifiable contextual factors in several domains, including social and economic factors, physical environment, health behaviors, and clinical care. To enhance the ability to compare across states, measured values of the contextual factors, instead of within-state rankings, were used. Contextual factor values from data sources in 2015 were selected because these are temporally aligned with USALEEP data. If 2015 data were missing, values were carried forward or imputed backward from the year of data source closest to 2015. Factors still missing in more than 10% (302 of 3020) of the counties were removed, and then remaining missing data was imputed based on a fully conditional specification imputation approach. Cubic splines of latitude and longitude of a county’s centroid were used as auxiliary variables in the imputation process. A total of 45 potentially modifiable contextual factors were used in the analyses.
      We then further selected 6 contextual factors for further analyses based on: 1) importance to life expectancy, which was represented by the percentage of census-tract–level life expectancy variance explained by the contextual factor; 2) data quality, evaluated by the portion of measured data vs imputed data; and 3) the factor’s representativeness of the conceptual construct it measures, for which, among factors that measured a similar construct, we selected the one that was most representative. A detailed algorithm of the selection process is presented in Supplemental Figure 1 (available online at http://www.mayoclinicproceedings.org).
      Census-block group–level and census-track–level population size in 2015 were obtained from the American Community Survey 5-year estimates 2011 to 2015. Census-block group–level area deprivation index (ADI) was obtained from the Neighborhood Atlas project by the University of Wisconsin.
      • Kind A.J.
      • Jencks S.
      • Brock J.
      • et al.
      Neighborhood socioeconomic disadvantage and 30-day rehospitalization: a retrospective cohort study.
      ,
      • Kind A.J.H.
      • Buckingham W.R.
      Making neighborhood-disadvantage metrics accessible - the Neighborhood Atlas.
      The national ranking of ADI was aggregated to the census-track level based on the population weighted mean.

      Statistical Methods

      The geographic distribution of census-tract–level life expectancy at birth was presented in the map (Figure 1A). Median, interquartile range (IQR), minimum, and maximum life expectancies by states were also presented. To examine unadjusted associations between county-level contextual factors and census-tract–level life expectancy, coefficients and percentages of variance explained by the contextual factors were reported. The unadjusted associations were plotted through locally weighted scatterplot smoothing.

      Association Between Contextual Factors and Life Expectancy

      To gain a better understanding of the association between contextual characteristics and life expectancy, we modeled age-specific nonlinear associations between contextual factors and death rates and then computed the estimated life expectancy based on a counterfactual scenario. The counterfactual scenario was set to be a hypothetical county that achieved the optimal 5th percentile of all 45 measured contextual factors in our study, representing a theoretically achievable optimal scenario (achieved for each contextual factor by at least 5% of the counties within the United States).
      • Cohen A.J.
      • Brauer M.
      • Burnett R.
      • et al.
      Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study 2015.
      • Burnett R.T.
      • Pope 3rd, C.A.
      • Ezzati M.
      • et al.
      An integrated risk function for estimating the global burden of disease attributable to ambient fine particulate matter exposure.
      • Lim S.S.
      • Vos T.
      • Flaxman A.D.
      • et al.
      A comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor clusters in 21 regions, 1990-2010: a systematic analysis for the Global Burden of Disease Study 2010.
      • Bowe B.
      • Xie Y.
      • Li T.
      • Yan Y.
      • Xian H.
      • Al-Aly Z.
      Particulate matter air pollution and the risk of incident CKD and progression to ESRD.
      • Bowe B.
      • Xie Y.
      • Li T.
      • Yan Y.
      • Xian H.
      • Al-Aly Z.
      The 2016 global and national burden of diabetes mellitus attributable to PM2.5 air pollution.
      • Bowe B.
      • Xie Y.
      • Li T.
      • Yan Y.
      • Xian H.
      • Al-Aly Z.
      Estimates of the 2016 global burden of kidney disease attributable to ambient fine particulate matter air pollution.
      The observed contextual factors better than 5th percentile were kept the same in the counterfactual scenario. The difference between observed and estimated life expectancy at the counterfactual scenario was used to represent the estimated years of life expectancy lost associated with contextual factors.
      Based on abridged life tables from USALEEP, the census-tract–specific fractions of life lived by those who died in each age interval were computed for close-ended age intervals. Fractions (ax) were computed based on ax=Lxnnx(lxdxn)nxdxn. Census-tract– and age-group–specific death rates (nMx) were then computed based on the fractions and probability of death. Death rate for age interval older than 85 years was computed by number of persons surviving at the beginning of the age interval divided by the number of person-years lived in this age interval M85=l85L85.
      For each age group, the association between contextual factor and the death rate was estimated through a population-weighted generalized estimating equation for grouped logistic regression.
      • Danaei G.
      • Rimm E.B.
      • Oza S.
      • Kulkarni S.C.
      • Murray C.J.
      • Ezzati M.
      The promise of prevention: the effects of four preventable risk factors on national life expectancy and life expectancy disparities by race and county in the United States.
      ,
      • Zhao Y.
      • Wright J.
      • Begg S.
      • Guthridge S.
      Decomposing indigenous life expectancy gap by risk factors: a life table analysis.
      Death rates within each age group were modeled independently. Each contextual factor was treated as a restricted cubic spline function in which knots were placed at the 20th, 40th, 60th, and 80th percentiles. Models were constructed to allow for dependence of death rate within each county with an independent correlation structure. The odds ratio (OR) of death when all factors were set to the optimal 5th percentile (counterfactual scenario) compared with observed factor values for each census-tract and age-group was computed from the model. The counterfactual death rate (counterfactual nMx) was computed based on observed death rate (nMx) and OR of death:
      CounterfactualMxn=(Mxn1Mxn)OR1+[(Mxn1Mxn)OR]


      The age- and census-tract–specific counterfactual probability of death was computed based on the estimated counterfactual death rate. The counterfactual probabilities of death in different age groups were used to compute the counterfactual life expectancy for each census tract based on the abridged life table method.
      • Chaing C.L.
      The Life Table and Its Applications.
      Associations were presented as the loss of life expectancy associated with the contextual factors, computed from the difference between observed and estimated life expectancy at the counterfactual scenario.
      In addition to the multicontextual factor analysis for all 45 contextual factors, single contextual factor analyses were conducted to investigate the association between each of the selected 6 contextual factors and life expectancy. We additionally estimated the life expectancy that could be reached if all selected 6 contextual factors performed optimally at the counterfactual scenario.
      To examine the importance of the selected 6 factors in explaining loss of life expectancy, the difference in estimated life expectancy based on 2 scenarios (1, all 45 factors performed optimally and 2, the selected 6 factors performed optimally) was also computed and further compared with an optimal achievable life expectancy, which was defined as optimal 5th percentile of the population-weighted life expectancy. In addition, 3 population-weighted linear regressions were built in which identifiers for the county the census-tract resided in, all 45 contextual factors, and the selected 6 factors were separately as predictors. Census-tract–level life expectancy variances explained by all county-level effects, all 45 contextual factors, and the selected 6 factors were reported from these models.
      We further conducted decomposition analyses of life expectancy by each of the 6 selected factors to understand the contribution of each factor to loss of life expectancy based on methodology developed by Das Gupta.
      • Das Gupta P.
      Standardization and decomposition of rates from cross-classified data.
      ,
      • Das Gupta P.
      Standardization and Decomposition of Rates: A User’s Manual.
      The association of each factor was defined by the increase in life expectancy if the factor were changed from the observed value to their counterfactual value while the other factors were held constant. Decomposition analyses were conducted in the overall data and by states, ADI quantile, and racial composition, which was expressed by percentage of Blacks. To indicate the trend of association between factors and life expectancy loss across census tracts, locally weighted scatterplot smoothing was applied.
      For all analyses, census-tract–level life expectancies were first computed and then county-, state-, and national-level life expectancies were computed as the population-weighted average of the census tracts within them. To examine the robustness of our results, multiple sensitivity analyses were conducted in which we: 1) excluded census tracts with life expectancy purely based on prediction from analyses, 2) additionally controlled for spline terms of the latitude and longitude, and 3) conducted within- and between-state effect analyses to estimate the within-state effect of contextual factors.
      Statistical analyses were conducted in SAS Enterprise Guide 7.2 (SAS Institute Inc). Figures were produced from SAS Enterprise Guide 7.2, Tableau 10.5 (Tableau), and R 3.5.1 (R Core Team) using ggplot2 and tmaptools.

      Results

      Average ± SD life expectancy in the United States was 78.69±3.78 years, median was 78.90 (IQR, 76.30-81.00) years, and range spanned 41.20 years. Life expectancy varied substantially across and within states (Figure 1A and B; Supplemental Table 1, available online at http://www.mayoclinicproceedings.org). Hawaii had the longest life expectancy (median, 81.40; IQR, 79.20-83.80 years) and Alabama had the shortest life expectancy (median, 75.20; IQR, 73.20-77.70 years). The lowest life expectancy in each state (at the census-tract level) varied from 56.30 years in Oklahoma to 72.40 years in Wyoming; the highest life expectancy varied between 97.50 years in North Carolina and 85.50 years in Mississippi. Life expectancy also varied substantially within counties (Figure 1C; Supplemental Figure 1). The within county gaps ranged from 0 to 34.60 years; New York County (New York), Cook County (Illinois), Franklin County (Ohio), the District of Columbia, and Erie County (New York) exhibited the top 5 widest ranges at 34.60, 30.10, 27.60, 27.50, and 27.40 years, respectively. There was also substantial variation between counties in which East Carrol Parish County (Louisiana) had the lowest county mean life expectancy (69.20 years), and Cheyenne County (Colorado) had the highest mean life expectancy (89.50 years), suggesting a gap in life expectancy among counties of 20.30 years (Figure 1C; Supplemental Figure 1).
      Figure thumbnail gr1
      Figure 1A, Geographical distribution of life expectancy. Mapped census-tract–level life expectancy in the United States. Wisconsin and Maine are blank due to lack of data. White indicates a life expectancy close to the mean, red indicates lower than the mean, and blue indicates higher than the mean. B, Distribution of life expectancy across states. Box plots represent the distribution of population-weighted life expectancy in each state. The median was represented by a vertical bar. The filled box indicates interquartile range (IQR) and lines indicate range from minimum (Min) to maximum (Max) values. States were ordered based on increasing mean life expectancy. C, Life expectancy difference within counties. Each dot represents a census tract, where dots on the same line are in the same county. Counties are represented on the Y axis and ordered by highest life expectancy within county from low (on top) to high (on bottom). Life expectancy is on the X axis. Colors represent the absolute difference from the highest life expectancy within the county. Green represents difference less than 5 years; blue, between 5 and 10 years; yellow, between 10 and 20 years, and magenta, greater than 20 years.

      County-Level Contextual Characteristics and Life Expectancy

      We curated a list of 45 potentially modifiable county-level contextual characteristics from the CHR data sets in several domains, including social and economic factors, physical environment, health behaviors, and clinical care. The variance of life expectancy explained by each contextual factor, its ranking (based on variance) within its domain, and its overall ranking are presented in Supplemental Table 2 (available online at http://www.mayoclinicproceedings.org).
      To gain a better understanding of how county-level contextual factors may be associated with differences in life expectancy, we modeled age-specific nonlinear associations between contextual factors and death rates. We then estimated the loss of life expectancy associated with the contextual factors (Table). The results suggest that observed rates of the full array of 45 potentially modifiable U.S. county-level contextual factors was associated with 4.30 years of loss in life expectancy. The effect of unmeasured characteristics, estimated from the difference between the observed top 5th percentile of life expectancy among census tracts and the estimated life expectancy in counterfactual scenario in which all the contextual factors are set at the optimal level of performance, was associated with a 1.71 year loss in life expectancy.
      TableEstimated Years of Life Expectancy Loss in Association With US County-Level Contextual Factors
      Contextual FactorsEstimated Life Expectancy
      Estimated life expectancy in the counterfactual scenario in which contextual factors were set at the optimal 5th percentile level of performance.
      (y)
      Life Expectancy Lost
      Years of life expectancy lost in association with the contextual factors in unicontextual factor analyses.
      (y)
      Decomposed Life Expectancy Lost Associated With Selected 6 Factors
      Years of life expectancy lost in association with the contextual factors in multicontextual factor analysis.
      (y), no. (% contribution)
      An array of 45 contextual factors82.994.30Not Applicable
      Adult smoking81.512.811.24 (35.13)
      Food insecurity80.471.770.89 (25.21)
      Adult obesity80.852.160.58 (16.43)
      Physical inactivity80.862.160.35 (9.92)
      College education80.551.850.33 (9.35)
      Median household income80.892.200.14 (3.97)
      Selected 6 factors
      The 6 factors include adult smoking, food insecurity, adult obesity, physical inactivity, college education, and median household income.
      82.223.533.53 (100)
      a Estimated life expectancy in the counterfactual scenario in which contextual factors were set at the optimal 5th percentile level of performance.
      b Years of life expectancy lost in association with the contextual factors in unicontextual factor analyses.
      c Years of life expectancy lost in association with the contextual factors in multicontextual factor analysis.
      d The 6 factors include adult smoking, food insecurity, adult obesity, physical inactivity, college education, and median household income.
      Six contextual factors meeting criteria for variance explained (percentage of census-tract–level life expectancy variance explained by the contextual factor), on the basis of data quality including missingness and representativeness, were then selected for further analyses (Supplemental Figures 2 and 3A-F; Supplemental Table 3, available online at http://www.mayoclinicproceedings.org). In single contextual factor analyses, increases in county-level adult smoking, food insecurity, adult obesity, and physical inactivity were associated with increased loss of life expectancy, and decreases in college education and county-level median household income were associated with increased loss in life expectancy (Figure 2A-F).
      Figure thumbnail gr2
      Figure 2Bivariate associations between estimated life expectancy lost and county-level contextual characteristics. A, Adult smoking; B, food insecurity; C, adult obesity; D, physical inactivity; E, college education; and F, median household income. Estimated life expectancy lost is calculated from the difference between observed and estimated life expectancy when the related contextual factor is set to the optimal 5th percentile. Locations of circles represent estimated life expectancy lost and the value of related contextual factor for each census-tract. The orange line represents functional form of the association based on a locally weighted scatterplot smoothing regression.
      In multicontextual factor analyses, county-level adult smoking, food insecurity, adult obesity, physical inactivity, college education, and median household income were associated with an estimated 1.24-, 0.89-, 0.58-, 0.35-, 0.33-, and 0.14-year loss in life expectancy, respectively (Table). The average estimated life expectancy loss associated with all 6 factors was 3.53 years (Table; Supplemental Figure 4, available online at http://www.mayoclinicproceedings.org). The distribution of estimated life expectancy in which all 6 contextual factors were set at the optimal 5th percentile showed overall improvement in life expectancy, reduced heterogeneity (reduced range, IQR, and standard deviation), and narrower geographical difference but did not show total elimination of clusters of high and low life expectancy (Supplemental Figure 5A-C, available online at http://www.mayoclinicproceedings.org).

      Variance Analyses

      Analyses of census-tract–level life expectancy variance suggested that whereas 30.49% of the life expectancy variance was explained by county-level effects, 26.53% was explained by the array of 45 county-level contextual factors and 23.32% was explained by the 6 county-level factors selected from the 45 factors for further analyses, suggesting that the array of 45 factors and the selected 6 retained 87.01% and 76.48% of the explanatory power of overall county-level effects on census-tract life expectancy.

      Analyses by State

      Analyses at the state level showed significant variability in the contribution of the contextual factors between states. States with the largest life expectancy loss associated with adult smoking rates included West Virginia (2.53 years), Kentucky (2.29 years), and Tennessee (2.09 years). Food insecurity had the strongest association with years of life expectancy lost in Mississippi (1.79 years), Arkansas (1.52 years), and Alabama (1.45 years) (Figure 3). The relative contribution of obesity increased as observed life expectancy increased, whereas the relative contribution of college education decreased as life expectancy increased. The relative contribution of food insecurity was similar across states (Supplemental Table 4, available online at http://www.mayoclinicproceedings.org).
      Figure thumbnail gr3
      Figure 3Decomposition of the estimated years of life expectancy lost in association with county-level contextual characteristics by states. States were ordered based on increasing mean life expectancy. Colored bars indicate the estimated life expectancy lost associated with each of the 6 county-level contextual factors. A “+” symbol indicates life expectancy based on scenario in which all 45 contextual factors were set to their optimal 5th percentile. An “x” symbol indicates a theoretically achievable optimal life expectancy in each state and was defined based on the 95th percentile of life expectancy within each state. The dashed line was the 95th percentile of life expectancy across all census tracts in the study.
      Whereas variance of observed life expectancy across states was 2.49 years, variance of estimated life expectancy based on the counterfactual scenario of optimal performance of the selected 6 contextual factors was substantially narrower at 0.47 years. In general, states with lower observed life expectancy exhibited larger losses of years of life expectancy associated with contextual factors. However, Utah had the 15th highest observed life expectancy but the 2nd lowest loss of life expectancy. Similarly, District of Columbia had the 41st highest observed life expectancy but the 14th lowest loss (Supplemental Table 4). The additional contribution of all other CHR characteristics and unmeasured characteristics was more evident in states with higher observed life expectancy (Figure 3; Supplemental Table 4).

      Contextual Characteristics and Life Expectancy by ADI and Race

      To better understand the influence of broader contextual determinants of health, including measures of socioeconomic deprivation (expressed by ADI) and measures of racial composition (expressed by percentage of Blacks) on the relationship between contextual characteristics and life expectancy, we decomposed the results of our analyses by ADI quartile and by race. The results suggest that estimated years of life expectancy lost in association with the contextual factors increased with increasing ADI and as percentage of Blacks increased (Figure 4A and B; Supplemental Table 5A and B, available online at http://www.mayoclinicproceedings.org).
      Figure thumbnail gr4
      Figure 4Estimated years of life expectancy lost in association with county-level contextual characteristics by measures of (A) socioeconomic deprivation expressed by Area Deprivation Index and (B) racial composition expressed by percentage of Blacks in the county. Estimated life expectancy lost was calculated from the difference between observed and estimated life expectancy when 6 county-level contextual factors were set to their optimal 5th percentile. Locations of circles represent the estimated life expectancy lost in each census tract and (A) the Area Deprivation Index of each census tract; (B) percentage of Blacks in the corresponding county. The orange line represents functional form of the association based on a locally weighted scatterplot smoothing regression.

      Sensitivity Analyses

      The results remained consistent in analyses in which we: 1) removed census tracts (8559 tracts) in which only model-predicted life expectancy estimates were available from USALEEP, 2) additionally controlled for spline terms of latitude and longitude, and 3) developed analyses to estimate within-state effect (Supplemental Table 6, available online at http://www.mayoclinicproceedings.org).

      Discussion

      In this work, we evaluated the association between county-level contextual characteristics and life expectancy in the United States. Our results suggest substantial variation in life expectancy at the national level (range, 41.20 years), and substantial variation within and between states and within and between counties. We estimated that a broad array of county-level contextual factors was associated with 4.30 years of loss in life expectancy; 6 county-level contextual factors including adult smoking, food insecurity, adult obesity, physical inactivity, college education, and median household income were associated with 3.53 years of loss in life expectancy. The contribution of the contextual factors to years of life expectancy lost varied by state and was more pronounced in states with lower expectancy. The influence of socioeconomic deprivation and race, broader macro level determinants of health, was evident because the number of years of life expectancy lost in association with the contextual factors was more pronounced in areas with high levels of socioeconomic deprivation and higher percentage of Blacks.
      First our analyses document the substantial variation in life expectancy in the United States, a 41.20-year gap between the lowest and highest census-tract–level life expectancy, up to 34.60 years within-county gap, and 20.30-year gap between highest and lowest county mean life expectancy. Although the gap between counties is consistent with a prior report by Dwyer-Lindgren et al,
      • Dwyer-Lindgren L.
      • Bertozzi-Villa A.
      • Stubbs R.W.
      • et al.
      Inequalities in life expectancy among US counties, 1980 to 2014: temporal trends and key drivers.
      the extent of geographic inequalities in life expectancy at the census-tract-level (U.S. range and within-county gap) was much larger and likely a reflection of the higher spatial resolution.
      • Dwyer-Lindgren L.
      • Bertozzi-Villa A.
      • Stubbs R.W.
      • et al.
      Inequalities in life expectancy among US counties, 1980 to 2014: temporal trends and key drivers.
      ,
      • Wang H.
      • Schumacher A.E.
      • Levitz C.E.
      • Mokdad A.H.
      • Murray C.J.
      Left behind: widening disparities for males and females in US county life expectancy, 1985-2010.
      ,
      US Department of Health and Human Services
      Healthy People 2020 Framework.
      The data suggest that higher resolution small-area estimation methodologies offer greater clarity to document and evaluate geographic inequities in life expectancy.
      • Arias E.
      • Escobedo L.A.
      • Kennedy J.
      • Fu C.
      • Cisewki J.
      U.S. small-area life expectancy estimates project: methodology and results summary.
      The observations that the gap in some counties may exceed 3 decades of life and the overall range in the United States exceeds 4 decades of life suggest the need for greater understanding of drivers of these inequities and, most importantly, action to address them.
      • Torjesen I.
      Inequalities in life expectancy are widening, data confirm.
      ,
      • Bennett J.E.
      • Pearson-Stuttard J.
      • Kontis V.
      • Capewell S.
      • Wolfe I.
      • Ezzati M.
      Contributions of diseases and injuries to widening life expectancy inequalities in England from 2001 to 2016: a population-based analysis of vital registration data.
      The drivers of these inequities include a complex interplay of demographic, social, economic, behavioral, environmental, and genomic factors and factors related to access and quality of clinical care; some drivers are captured at the individual level, and some of these drivers are shaped by the contextual reality in which people live.
      • Singh G.K.
      • Kogan M.D.
      • Slifkin R.T.
      Widening disparities in infant mortality and life expectancy between Appalachia and the rest of the United States, 1990-2013.
      ,
      • Wang H.
      • Schumacher A.E.
      • Levitz C.E.
      • Mokdad A.H.
      • Murray C.J.
      Left behind: widening disparities for males and females in US county life expectancy, 1985-2010.
      ,
      • Bowe B.
      • Xie Y.
      • Li T.
      • et al.
      Changes in the US burden of chronic kidney disease from 2002 to 2016: an analysis of the Global Burden of Disease Study.
      In this work, we merged CHR data sets describing a broad and comprehensive array of contextual characteristics with census-tract life expectancy data to gain a better understanding of how contextual factors relate to geographic differences in life expectancy. We built analyses to estimate the years of life expectancy lost in association with the contextual factors in the same computational framework, thus allowing comparative evaluation of their relative importance at both the national and state levels. Suboptimal performance of an array of 45 county-level factors was associated with reduced life expectancy and explained much of the variation (87.01%) in census-tract life expectancy that may be explained by county-level effects.
      Our analyses identified 6 county-level contextual characteristics and ranked them in order of their contributions to loss of life expectancy, including adult smoking, food insecurity, adult obesity, physical inactivity, college education, and median household income, which together contribute 3.53 years of life expectancy loss. These 6 county-level factors retained 76.48% of the explanatory power of overall county-level effects on census-tract life expectancy. The relationship between these contextual characteristics and survival probability reflects the influence of not only that they are aggregate rates of well-characterized individual-level risk factors for early mortality (eg, smoking, obesity, and physical inactivity) but also the broader macro level conditions, including economic prosperity and educational attainment.
      • Chetty R.
      • Stepner M.
      • Abraham S.
      • et al.
      The association between income and life expectancy in the United States, 2001-2014.
      ,
      • Sasson I.
      Trends in life expectancy and lifespan variation by educational attainment: United States, 1990-2010.
      ,
      • Ketenci N.
      • Murthy V.N.R.
      Some determinants of life expectancy in the United States: results from cointegration tests under structural breaks.
      US Burden of Diseases Collaborators
      The state of US health, 1990-2016: burden of diseases, injuries, and risk factors among US states.
      GBD 2017 Risk Factor Collaborators
      Global, regional, and national comparative risk assessment of 84 behavioural, environmental and occupational, and metabolic risks or clusters of risks for 195 countries and territories, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017.
      Our analyses provide estimates of the contribution of contextual characteristics to years of life expectancy lost at the national level and state level. This comparative assessment at the national and state levels illuminates our understanding of the contextual contributors to disparities in life expectancy, and may inform and help prioritize policy levers or potential targets for state and national interventions aimed at reducing inequalities in life expectancy in the United States.
      The decomposition analyses by the ADI and race, broader determinants of health, suggest that the contribution of the county-level contextual characteristics to years of life expectancy lost was more pronounced in counties with a high levels of socioeconomic deprivation and higher percentage of Blacks. Furthermore, our state-level analyses suggest that the contribution of the contextual characteristics to years of life expectancy lost was most pronounced in states with lower life expectancy. Woven together, the results suggest that these 6 contextual factors may be even more important contributors to losses in life expectancy precisely where it matters most, in areas of high socioeconomic deprivation, higher percentage of Blacks, and states with lower life expectancy.
      • Dwyer-Lindgren L.
      • Bertozzi-Villa A.
      • Stubbs R.W.
      • et al.
      Inequalities in life expectancy among US counties, 1980 to 2014: temporal trends and key drivers.
      ,
      • Wang H.
      • Schumacher A.E.
      • Levitz C.E.
      • Mokdad A.H.
      • Murray C.J.
      Left behind: widening disparities for males and females in US county life expectancy, 1985-2010.
      ,
      • Dwyer-Lindgren L.
      • Bertozzi-Villa A.
      • Stubbs R.W.
      • et al.
      US county-level trends in mortality rates for major causes of death, 1980-2014.
      The study has several limitations. Our analyses captured life expectancy at the census-tract level; although this spatial resolution is considered high, differences within a census tract may exist and are not accounted for in our analyses. In addition, because of the small geographic unit (census-tract), not all death rates from different age groups could be observed, and as a result, estimates of life expectancy for some tracts were developed using prediction modeling. The USALEEP data included census-tract–level data for life expectancy in all states except Maine and Wisconsin. Although we comprehensively evaluated all the contextual characteristics from the CHR data sets and selected for further analyses those that met specific criteria for variance and data quality, unmeasured and yet to be characterized contextual factors may also be important and are not accounted for here. However, we conducted analyses to estimate the overall effect of those unmeasured and unknown characteristics on differences in life expectancy. This study did not account for individual-level characteristics, and the associations observed may reflect the joint contribution of both contextual factors, individual-level risks, and other characteristics which were not accounted for in the analyses. The results reflect associations and should not be interpreted as causal in nature. We reported results based on the counterfactual scenario in which contextual factors performed at the optimal 5th percentile and the interpretation of study results should be made with full cognizance of the underlying assumptions; mainly that the counterfactual level for each contextual factor is theoretically achievable.
      • Bowe B.
      • Xie Y.
      • Li T.
      • et al.
      Changes in the US burden of chronic kidney disease from 2002 to 2016: an analysis of the Global Burden of Disease Study.
      The analyses benefited from the recent availability of high spatial resolution life expectancy data (USALEEP) and the merging with the CHR data sets, which enables characterization and understanding of the relationship between a broad array of characteristics in several contextual domains and life expectancy. Furthermore, we leveraged advances in statistical modeling to develop our analyses. Our modeling approach to estimate years of life expectancy lost allowed heterogenous (potentially nonlinear) contextual factor effects on mortality rates across contextual factor levels and age groups. Moreover, in addition to characterizing the relationship between contextual factors and life expectancy, we provided estimates of the number of years of life lost in association with these contextual characteristics using a modeling approach that considers the effect of other contextual characteristics. This comparative assessment at the national and state levels will facilitate greater public understanding of the implications of our study results.

      Conclusion

      In sum, our analyses suggest substantial variation in life expectancy in the United States, within and between counties, and within and between states and identifies several U.S. county-level contextual characteristics associated with life expectancy. Overall, 6 contextual characteristics, including adult smoking, food insecurity, adult obesity, physical inactivity, college education, and median household income, were associated with 3.53 years of life expectancy loss. The contribution of contextual characteristics to years of life expectancy lost was more pronounced in states with lower life expectancy and in areas with high measures of socioeconomic deprivation and higher percentage of Blacks. The results provide a comparative assessment of the contribution of contextual characteristics to disparities in life expectancy that may unlock opportunities for improving life expectancy and help inform policy priorities aimed at reducing the wide inequalities in life expectancy in the United States.

      Acknowledgments

      The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication. The contents do not represent the views of the U.S. Department of Veterans Affairs or the U.S. government.

      Supplemental Online Material

      References

        • Arias E.
        • Escobedo L.A.
        • Kennedy J.
        • Fu C.
        • Cisewki J.
        U.S. small-area life expectancy estimates project: methodology and results summary.
        Vital Health Stat 2. 2018; : 1-40
        • GBD 2017 DALYs and HALE Collaborators
        Global, regional, and national disability-adjusted life-years (DALYs) for 359 diseases and injuries and healthy life expectancy (HALE) for 195 countries and territories, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017.
        Lancet. 2018; 392: 1859-1922
        • Foreman K.J.
        • Marquez N.
        • Dolgert A.
        • et al.
        Forecasting life expectancy, years of life lost, and all-cause and cause-specific mortality for 250 causes of death: reference and alternative scenarios for 2016-40 for 195 countries and territories.
        Lancet. 2018; 392: 2052-2090
        • Dowell D.
        • Arias E.
        • Kochanek K.
        • et al.
        Contribution of opioid-involved poisoning to the change in life expectancy in the United States, 2000-2015.
        JAMA. 2017; 318: 1065-1067
        • Torjesen I.
        Inequalities in life expectancy are widening, data confirm.
        BMJ. 2018; 360: k1017
        • Singh G.K.
        • Kogan M.D.
        • Slifkin R.T.
        Widening disparities in infant mortality and life expectancy between Appalachia and the rest of the United States, 1990-2013.
        Health Aff (Millwood). 2017; 36: 1423-1432
        • Dwyer-Lindgren L.
        • Bertozzi-Villa A.
        • Stubbs R.W.
        • et al.
        Inequalities in life expectancy among US counties, 1980 to 2014: temporal trends and key drivers.
        JAMA Intern Med. 2017; 177: 1003-1011
        • Chetty R.
        • Stepner M.
        • Abraham S.
        • et al.
        The association between income and life expectancy in the United States, 2001-2014.
        JAMA. 2016; 315: 1750-1766
        • Sasson I.
        Trends in life expectancy and lifespan variation by educational attainment: United States, 1990-2010.
        Demography. 2016; 53: 269-293
        • Bowe B.
        • Xie Y.
        • Xian H.
        • Lian M.
        • Al-Aly Z.
        Geographic variation and US county characteristics associated with rapid kidney function decline.
        Kidney Int Rep. 2017; 2: 5-17
        • Remington P.L.
        • Catlin B.B.
        • Gennuso K.P.
        The County Health Rankings: rationale and methods.
        Popul Health Metr. 2015; 13: 11
        • Kind A.J.
        • Jencks S.
        • Brock J.
        • et al.
        Neighborhood socioeconomic disadvantage and 30-day rehospitalization: a retrospective cohort study.
        Ann Intern Med. 2014; 161: 765-774
        • Kind A.J.H.
        • Buckingham W.R.
        Making neighborhood-disadvantage metrics accessible - the Neighborhood Atlas.
        N Engl J Med. 2018; 378: 2456-2458
        • Cohen A.J.
        • Brauer M.
        • Burnett R.
        • et al.
        Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study 2015.
        Lancet. 2017; 389: 1907-1918
        • Burnett R.T.
        • Pope 3rd, C.A.
        • Ezzati M.
        • et al.
        An integrated risk function for estimating the global burden of disease attributable to ambient fine particulate matter exposure.
        Environ Health Perspect. 2014; 122: 397-403
        • Lim S.S.
        • Vos T.
        • Flaxman A.D.
        • et al.
        A comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor clusters in 21 regions, 1990-2010: a systematic analysis for the Global Burden of Disease Study 2010.
        Lancet. 2012; 380: 2224-2260
        • Bowe B.
        • Xie Y.
        • Li T.
        • Yan Y.
        • Xian H.
        • Al-Aly Z.
        Particulate matter air pollution and the risk of incident CKD and progression to ESRD.
        J Am Soc Nephrol. 2018; 29: 218-230
        • Bowe B.
        • Xie Y.
        • Li T.
        • Yan Y.
        • Xian H.
        • Al-Aly Z.
        The 2016 global and national burden of diabetes mellitus attributable to PM2.5 air pollution.
        Lancet Planet Health. 2018; 2: e301-e312
        • Bowe B.
        • Xie Y.
        • Li T.
        • Yan Y.
        • Xian H.
        • Al-Aly Z.
        Estimates of the 2016 global burden of kidney disease attributable to ambient fine particulate matter air pollution.
        BMJ Open. 2019; 9: e022450
        • Danaei G.
        • Rimm E.B.
        • Oza S.
        • Kulkarni S.C.
        • Murray C.J.
        • Ezzati M.
        The promise of prevention: the effects of four preventable risk factors on national life expectancy and life expectancy disparities by race and county in the United States.
        PLoS Med. 2010; 7: e1000248
        • Zhao Y.
        • Wright J.
        • Begg S.
        • Guthridge S.
        Decomposing indigenous life expectancy gap by risk factors: a life table analysis.
        Popul Health Metr. 2013; 11: 1
        • Chaing C.L.
        The Life Table and Its Applications.
        Krieger Pub Co, original edition (December 1, 1983), Malabar, FL1984
        • Das Gupta P.
        Standardization and decomposition of rates from cross-classified data.
        Genus. 1994; 50: 171-196
        • Das Gupta P.
        Standardization and Decomposition of Rates: A User’s Manual.
        US Department of Commerce, Economics and Statistics Administration, Bureau of the Census, Washington, DC1993: 19-36
        • Wang H.
        • Schumacher A.E.
        • Levitz C.E.
        • Mokdad A.H.
        • Murray C.J.
        Left behind: widening disparities for males and females in US county life expectancy, 1985-2010.
        Popul Health Metr. 2013; 11: 8
        • US Department of Health and Human Services
        Healthy People 2020 Framework.
        (Accessed August 11, 2019)
        • Bennett J.E.
        • Pearson-Stuttard J.
        • Kontis V.
        • Capewell S.
        • Wolfe I.
        • Ezzati M.
        Contributions of diseases and injuries to widening life expectancy inequalities in England from 2001 to 2016: a population-based analysis of vital registration data.
        Lancet Public Health. 2018; 3: e586-e597
        • Bowe B.
        • Xie Y.
        • Li T.
        • et al.
        Changes in the US burden of chronic kidney disease from 2002 to 2016: an analysis of the Global Burden of Disease Study.
        JAMA Network Open. 2018; 1: e184412
        • Ketenci N.
        • Murthy V.N.R.
        Some determinants of life expectancy in the United States: results from cointegration tests under structural breaks.
        J Econ Finance. 2018; 42: 508-525
        • US Burden of Diseases Collaborators
        The state of US health, 1990-2016: burden of diseases, injuries, and risk factors among US states.
        JAMA. 2018; 319: 1444-1472
        • GBD 2017 Risk Factor Collaborators
        Global, regional, and national comparative risk assessment of 84 behavioural, environmental and occupational, and metabolic risks or clusters of risks for 195 countries and territories, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017.
        Lancet. 2018; 392: 1923-1994
        • Dwyer-Lindgren L.
        • Bertozzi-Villa A.
        • Stubbs R.W.
        • et al.
        US county-level trends in mortality rates for major causes of death, 1980-2014.
        JAMA. 2016; 316: 2385-2401