Research A Section 508–conformant HTML version of this article is available at http://dx.doi.org/10.1289/EHP197. Particulate Air Pollution, Exceptional Aging, and Rates of Centenarians: A Nationwide Analysis of the United States, 1980–2010 Andrea A. Baccarelli,1 Nick Hales,2 Richard T. Burnett,3 Michael Jerrett,4 Carter Mix,2 Douglas W. Dockery,1 and C. Arden Pope III 2 1Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA; 2Department of Economics, Brigham Young University, Provo, Utah, USA; 3Environmental Health Directorate, Health Canada, Ottawa, Ontario, Canada; 4Division of Environmental Health Sciences, School of Public Health, University of California Berkeley, Berkeley, California, USA Background: Exceptional aging, defined as reaching age 85 years, shows geographic inequalities that may depend on local environmental conditions. Links between particulate pollution—a wellrecognized environmental risk factor—and exceptional aging have not been investigated. Objectives: We conducted a nationwide analysis of ~28 million adults in 3,034 United States counties to determine whether local PM2.5 levels (particulate matter < 2.5 μm in aerodynamic diameter) affected the probability of becoming 85- to 94-year-olds or centenarians (100- to 104-year-olds) in 2010 for individuals who were 55–64 or 70–74 years old, respectively, in 1980. Methods: We used population-weighted regression models including county-level PM2.5 from hybrid land-use regression and geostatistical interpolation, smoking, obesity, sociodemographic, and age-specific migration variables. Results: On average, 2,295 and 71.4 per 10,000 of the 55- to 64- and 70- to 74-year-olds in 1980, respectively, remained in the 85- to 94- and 100- to 104-year-old population in 2010. An interquartile range (4.19 μg/m3) increase in PM2.5 was associated with 93.7 fewer 85- to 94-yearolds (p < 0.001) and 3.5 fewer centenarians (p < 0.05). These associations were nearly linear, were stable to model specification, and were detectable below the annual PM2.5 national standard. Exceptional aging was strongly associated with smoking, with an interquartile range (4.77%) increase in population who smoked associated with 181.9 fewer 85- to 94-year-olds (p < 0.001) and 6.4 fewer c­ entenarians (p < 0.001). Exceptional aging was also associated with obesity rates and median income. Conclusions: Communities with the most exceptional aging have low ambient air pollution and low rates of smoking, poverty, and obesity. Improvements in these determinants may contribute to increasing exceptional aging. Citation: Baccarelli AA, Hales N, Burnett RT, Jerrett M, Mix C, Dockery DW, Pope CA III. 2016. Particulate air pollution, exceptional aging, and rates of centenarians: a nationwide analysis of the United States, 1980–2010. Environ Health Perspect 124:1744–1750;  http://dx.doi. org/10.1289/EHP197 Introduction Recent declines in mortality have resulted in a worldwide increase in exceptional aging, defined as reaching age 85 years or older [National Institute on Aging (NIA) 2013]. Persons ≥ 85 years of age comprise the fastestgrowing segment of the world’s population. Between 2010 and 2050, the number of indi- viduals ≥ 85 years old will increase worldwide by > 300%, and the number of centenarians is projected to rise as much as 10 times [NIA/ World Health Organization (WHO) 2011]. The distribution of exceptional aging shows substantial geographic inequalities that may depend, at least in part, on local conditions (NIA/WHO 2011). Many cohort studies have shown that long-term exposure to air pollution, particularly to particulate matter < 2.5 μm in is associated waeitrhodcyanrdaimovicasdcuialamr emteor r(bPidMit2y.—5), which affects considerably aging individuals (Brook et al. 2010)—and increased total mortality (Beelen et al. 2014; Cesaroni et al. 2013; Crouse et al. 2012; Dockery et al. 1993; Jerrett et al. 2013; Miller et al. 2007; Pope et al. 2002; Zeger et al. 2008). Although associations of have been well documented, iPtMis 2u.5ncelxepaorshuorews these associations extend to the extremes of the age distribution. Aging and mortality are certainly linked; however, simulation studies have shown key differences in their biological and statistical dynamics, particu- larly at the extreme end of the human life span (Olshansky et al. 2001). To date, no study has investigated whether exposure to air pollu- tion adversely affects the probability of excep- tional aging. For this analysis, we asked the following question: Using U.S. county-level data, is there evidence that PM2.5 air pollution affects population-based measures of excep- tional aging? We hypothesized that counties with less air pollution, as well as those with favorable sociodemographic conditions, have higher probabilities of exceptional aging, even when controlling for other population-level health-influencing factors. We analyzed data on ~28 million individuals ≥ 55 years old across 3,034 counties to examine the associa- ttiiaolndoeftePrmMi2n.5anctosnwceitnhtrtahtieonpsroabnadbiolitthyeor fpaogteinng- to 85–94 years old and separately of becoming a centenarian (100–104 years old) in 2010, given the population in 1980. Methods Demographic and Socioeconomic Data County-level demographic and socioeconomic data were drawn from the 1980, 2000, and 2010 censuses (U.S. Census Bureau 1980, 2000, 2010). Data from Hawaii and Alaska were eexstcilmudaetdes.beTchaeusfeololof winiandgeqcuoautnetPy-Mlev2.e5l exposure data for other demographic and socioeconomic variables were compiled from year 2000 census data: median age, percent of population > 65 years old, percent of black or Hispanic population, population density, percent of population in urban areas, median income, percent of high school graduates, percent below poverty level, and percent unemployed. Because of the importance of adequately addressing migration rates for the elderly, relevant age-specific migra- tion rates were obtained for the decades of the 1980s, 1990s, and 2000s (Winkler et al. 2013). See Table 1 for a description and summary of these variables and their sources. Construction of Population-based Probabilities of Exceptional Aging We constructed two age-specific, populationbased indices of county-level probabilities Address correspondence to A.A. Baccarelli, 665 Huntington Ave., Harvard T.H. Chan School of Public Health, Boston, MA 02115 USA. Telephone: (617) 432-0037. E-mail: abaccare@hsph.harvard.edu Supplemental Material is available online (http:// dx.doi.org/10.1289/EHP197). This work was supported by grants from the National Institutes of Health/National Institute of Environmental Health Sciences (R01ES021733, P30ES000002) and the Mary Lou Fulton Professorship at Brigham Young University (to C.A.P. III). The authors declare they have no actual or potential competing financial interests. Received: 18 August 2015; Revised: 18 November 2015; Accepted: 18 April 2016; Published: 3 May 2016. Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all ­readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your ­accessibility needs within 3 working days. 1744 volume 124 | number 11 | November 2016  •  Environmental Health Perspectives Particulate matter, aging, and centenarians of exceptional aging. The first index was the proportion of 85- to 94-year-old persons per 10,000 in the 2010 census relative to the population of 55- to 64-year-olds in the 1980 census, constructed as follows: P8E5A–94 = 10, 000 # Pop82501–094 Pop 1980 55–64 ,[1] where PE85A–94 is the constructed proportion of exceptionally aged individuals 85–94 years old 85- in to 29041-y0e,aPr-oopl28d05s1–09i4n is the population of 2010, and Pop15958–064 is the population of 55- to 64-year-olds in 1980. The second index was the proportion of 100- to 104-year-old persons per 10,000 in the 2010 census relative to the population of 70- to 74-year-olds in the 1980 census, constructed as follows: P1E0A0–104 = 10, 000 # Pop2100100–104 Pop 1980 70–74 ,[2] where P E10A0–104 is the constructed proportion of exceptionally aged individuals 100–104 years the population old in of 100- 2t0o1100, 4P-yopea21r00-10o0–l1d0s4 is in 2010, and Pop17908–074 is the population of 70to 74-year-olds in 1980. In the absence of migration, the indices above are simply scaled (× 10,000) probability ratios. With control for migration in statistical models, these indices are approximately equivalent to scaled probabilities of survival over the 30-year span for the respective age range. These variables are also summarized in Table 1. Air Pollution, Smoking, and Obesity Data County-level mated using a ehxypborsiduraeps ptrooaPcMh 2th.5atwienrceluedsteid- land-use regression, traffic indicators, and Bayesian maximum entropy interpolation of land-use regression space-time residuals as documented elsewhere (Beckerman et al. 2013). The model is highly predictive of ground-level concentrations with a cross-­ validation R2 of 0.79 and no indication of bias. These estimates were population-weighted averages (using the 2000 census data) from census-tract estimates averaged for all months of 1999–2008. The percentage of adults who smoked daily in 2000 was obtained for each county from the Institute for Health Metrics and Evaluation (2014), and obesity preva- lence data were obtained from the Centers for Disease Control and Prevention (CDC 2013). Statistical Analysis We used population-weighted regression, weighting by the square root of the total population in the year 2000, to estimate associations of probabilities of exceptional aging (PE8A5–94 and PE1A00–104) with PM2.5 levels and with other determinants. Primary results were obtained from a linear regression model including smoking, oPbMes2it.5y,assoacnioeincodnepoemnidce, natndvadrieambloe-; graphic variables; indicator variables for the nine geographic census divisions of the United States; and age-specific migration rate vari- ables. The age-specific migration rates included in the regression models were selected to provide the closest possible temporal align- ment consistent with the initial age groups and with the relevant subsequent age groups for the entire three-decade study period. For PinE8At5h–9e41(9p8e0rsocnens swush)o, were 55–64 years the most relevant old and best temporally matched available migration rate data included the migration rates for ages 55–60 and 60–64 in the 1980s, the migra- tion rates for ages 65–70 and 70–74 in the 1990s, and the migration rate for ages ≥ 75 i7n0–th7e4 2y0ea0r0ss.olFdoirnPtEh1A0e0–1190840(pceernsosunss),wthhoe were most relevant and best temporally matched available migration rate data included the migration rates for ages 70–74 and ≥ 75 in the 1980s and the migration rates for ages ≥ 75 in the 1990s and the 2000s. Sensitivity analyses employing various alternative regression models were conducted to evaluate the robustness of the study findings. The largest modeling concern involved adequately controlling for migra- tion and accounting for outlier counties with extreme migration patterns. Therefore, models were estimated with and without migration Table 1. Description of United States county-level variables with unweighted means (SD), interquartile ranges and sources. Variable Description Mean (SD) IQR Source P E8A5–94 P E1A00–104 PM2.5 Percent smoking Obesity prevalence Median income Percent below poverty Percent black Percent Hispanic Population density Percent urban Percent high school graduate Percent unemployed Population in 2000 Median age Percent < 65 years old Migration 1980, 55- to 60-year-olds Migration 1980, 60- to 64-year-olds Migration 1980, 70- to 74-year-olds Migration 1980, ≥ 75-year-olds Migration 1990, 65- to 70-year-olds Migration 1990, 70- to 74-year-olds Migration 1990, ≥ 75-year-olds Migration 2000, ≥ 75-year-olds (Number of people 85–94 years old in 2010 divided by number of people 55–64 years old in 1980) × 10,000 (Number of people 100–104 years old in 2010 divided by number of people 70–74 years old in 1980) × 10,000 Mean PM2.5 concentrations from 1999–2008 (μg/m3) Percentage of adults in county that smoked daily in 2000 Average, age-adjusted percentage of population that was obese (data from 2004–2010) Median income in 1999 (in thousands of U.S. dollars) Percentage of population below poverty line Percentage of population that was black in 2000 Percentage of population that was Hispanic in 2000 Thousands of people per square mile in 2000 Percentage of population that live in urban areas in 2000 Percentage of population that were high school graduates Percentage of population that was unemployed Number of people living in a county in 2000 Median age for county in 2000 (years) Percentage of population > 65 years old in 2000 Migration rate for 55- to 60-year-olds in 1980sa Migration rate for 60- to 64-year-olds in 1980sa Migration rate for 70- to 74-year-olds in 1980sa Migration rate for ≥ 75-year-olds in 1980sa Migration rate for 65- to 70-year-olds in 1990sa Migration rate for 70- to 74-year-olds in 1990sa Migration rate for ≥ 75-year-olds in 1990sa Migration rate for ≥ 75-year-olds in 2000sa 2,295 (779.5) 71.4 (55.3) 10.4 (2.8) 21.5 (3.7) 28.1 (3.6) 35.3 (8.8) 13.3 (5.6) 8.9 (14.6) 6.2 (12.1) 0.25 (1.67) 40.2 (30.9) 77.3 (8.7) 5.8 (2.7) 89,927 (293,515) 37.4 (4.0) 14.8 (4.1) 4.7 (18.9) 7.4 (22.3) 1.9 (13.5) –0.43 (10.2) 10.1 (20.5) 4.2 (13.3) –0.17 (14.0) –0.81 (17.2) 728.52 56.30 4.19 4.77 3.53 9.60 6.90 10.04 4.19 0.09 53.69 12.70 2.90 50,742 4.60 4.97 12.00 14.00 10.00 10.00 17.00 12.00 13.00 15.00 U.S. Census Bureau 2010 U.S. Census Bureau 1980 U.S. Census Bureau 2010 U.S. Census Bureau 1980 Beckerman et al. 2013 Institute for Health Metrics and Evaluation 2014 CDC 2013 U.S. Census Bureau 2000 U.S. Census Bureau 2000 U.S. Census Bureau 2000 U.S. Census Bureau 2000 U.S. Census Bureau 2000 U.S. Census Bureau 2000 U.S. Census Bureau 2000 U.S. Census Bureau 2000 U.S. Census Bureau 2000 U.S. Census Bureau 2000 U.S. Census Bureau 2000 Winkler et al. 2013 Winkler et al. 2013 Winkler et al. 2013 Winkler et al. 2013 Winkler et al. 2013 Winkler et al. 2013 Winkler et al. 2013 Winkler et al. 2013 Abbreviations: IQR, interquartile range; PM2.5, particulate matter with aerodynamic diameter < 2.5 μm; SD, standard deviation. aAge-specific migration rates calculated as (net migration over the given decade divided by expected population at the end of the decade) times 100, where net migration is observed final population minus expected final population. Environmental Health Perspectives  •  volume 124 | number 11 | November 2016 1745 Baccarelli et al. variables and with and without censusdivision indicator variables. Models were also estimated using data from all counties, excluding observations with model residuals > 3 standard deviations from zero (censoring 37 observations) and excluding the 5% of counties (censoring ~150 observations) with the most extreme migration patterns (based on migration rates for ages ≥ 75 in the 2000s). In addition, because the precision of the exceptional aging variables was dependent upon county population, the regression models were weighted by the square root of the population. All linear models were estimated using the PROC REG procedure in SAS v.9.3 (SAS Institute Inc.). We also fit generalized additive models (GAMs) with a linear fit for (PaMllo2w.5inagndforp≤en 4aldizeegdreeres gorfesfrseioednomsmfooortehaecrhs smooth) for the other covariates (excepting the ­census-­division indicators). Finally, we also fit nonlinear models using a penalized regression smoother for PM2.5 in addition to the other covariates. These GAMs were estimated using the gam function in the R software mgcv package (R Core Team 2015). Results Probabilities of Exceptional Aging for Individuals 55–64 and 70–74 years old in 1980 In 1980, the U.S. population of the 48 contiguous states included 21,595,507 individuals who were 55–64 years old and 6,418,352 individuals who were 70–74 years old. In 2010, the 85- to 94-year-old group included 5,035,379 individuals and was therefore more than four times smaller than the corresponding group of 55- to 64-year-olds in 1980 (mean PE8A5–94 = 2,295 individuals in 2010, scaled to 10,000 individuals in 1980). In 2010, the 100to 104-year-old group included 48,303 individuals and was < 0.8% of the corresponding age group of 70- to 74-year-olds in 1980 s(cmaleeadntoP1E1A000,0–10004 i=n d7iv1i.d4uainlsdiinvi1d9u8a0ls).in 2010, PM2.5 Levels and Probabilities of Exceptional Aging Counties with higher PM2.5 concentrations had reduced probabilities of exceptional aging between 1980 and 2010. Plots of unadjusted county-level probabilities of exceptional aging for the two age groups (P8E5A–94 and P1E0A0–104) showed negative correlations with PM2.5 levels but revealed the presence of several outlier counties (Figure 1A,B). We excluded these outliers and used regression models controlling for possible confounders, which were characterized through relevant variables for smoking, obesity, socioeconomic and demographic characteristics, age-specific migration rates, and regional indicators (Table 2). Plots of the corresponding partial residuals (residuals obtained by adjusting for all covariates except PM2.5) showed a subtle, nearlinear covariate-adjusted association between PM2.5 and probabilities of exceptional aging (Figure 1C,D). The plots illustrate that unexplained variability in exceptional aging remained, particularly for the older group, but that the results were not driven by outliers. As shown in Table 2, the regression models estimated that each interquartile range (4.19 μg/m3) increase in PM2.5 was associated with 93.7 [standard error (SE) = 12.2] fewer remaining individuals in the 85- to 94-yearold group (p < 0.001) and 3.5 (SE = 1.5) sFa1(Di9iezg8)reu0opr)drr ee×ye pn1 s1.rae0eCmn,s0otie0cup0nnda;ttsirPyat-iEtm1lahA0e0leev–tre1ees0lr4qsp,ui 65 years old. For centenarians, there were no associations with median age or percent > 65 years old. For both age groups, there were very strong associations with relevant age-specific migration rates, which were generally consistent with expected structural associations (Table 2). Sensitivity Analysis and Possible Influence of Migration Rates As illustrated in Figure 2, the associations of PstMab2l.e5 with exceptional aging were in sensitivity analyses using robust and alternative regression models. Controlling for migration variables for the 8at5t-entuoat9ed4-tyheeaPr-Mol2d.5 effect somewhat group but not for the 100- to 104-year-old group. Models with and without regional indicator variables and GAMs that allowed for non-linear asso- ciations with all other covariates resulted in remarkably similar estimates. Furthermore, using different strategies to censor for extreme outliers to evaluate the potential influence of outliers provided remarkably consistent statisti- cally significant results in all cases. The largest county in the analysis, Los Angeles, was the most heavily weighted and was one of the most highly polluted counties (Figure 1). Los Angeles, however, tended to have more excep- tional aging than was predicted by the model; therefore, its inclusion slightly mitigated the negative linear associations between PM2.5 and exceptional aging. To further evaluate the possible influ- ence of migration rates on our results, we examined whether migration was differential finodr iPviMdu2a.5lscmonocveendtrfartoiomnsc,otmhmatuins,itwiehs ewthitehr high PM2.5 into communities with low PorerMlenv2ao.5n,tcoomrrrigevrliaactteiioonvnerrbasteaet.sw,WieneecnlufPoduiMnng2d.t5hmeanimndiimgtrhaae-l tion rate for 55- to 60-year-olds in 1980s (r = –0.03), the migration rate for 60- to 64-year-olds in the 1980s (r = –0.07), the migration rate for 70- to 74-year-olds in the 1980s (r = –0.07), the migration rate for ≥ 75-year-olds in the 1980s (r = –0.08), the migration rate for 65- to 70-year-olds in the 1990s (r = –0.07), the migration rate for 70- to 74-year-olds in the 1990s (r = 0.01), the migration rate for ≥ 75-year-olds in 1990s (r = 0.14), and the migration rate for ≥ 75-year-olds in the 2000s (r = 0.21). Linearity of the Association of PM2.5 with Probabilities of Exceptional Aging The GAMs that allowed for nonlinear asso- ciations with PM2.5 and all other covariates (except for the census-division indicators) indicated with the that the probabi laistsyocoiaf tieoxnc eopftPi oMn a2.l5 levels aging for both age groups (P E85A–94 and P E10A0–104) was approximately linear. Figure 3 presents plots of nonparametric smoothed functions (Figure 3A,C). For comparison, plots of the relationship between exceptional aging and percentage of smokers are also included (Figure 3B,D). For PM2.5, the penalized regression smoother was not significantly different from linear; however, for illustrative purposes, Figure 3 presents the estimated smooth with 2 degrees of freedom. In addition, as shown in Figure 2, GAMs with atiolinnseafrorfitthfoeroPthMer2.5cobvuatrisaptleisne(esxmceopotthfofrunthcecensus-division indicators) resulted in linear asasmsoeciaastitohnossefforromPMth2e.5futlhlaltinweaerremnodeaerlsl.y the Discussion In this nationwide analysis of ~28 million individuals ≥ 55 years old across 3,034 counties, higher levels of PM2.5 air pollution were associated with lower population-based probabilities of exceptional aging, even after adjusting for smoking, obesity, demographic and socioeconomic variables, total and agespecific migration rates, and differences across the nine census divisions of the United States. The highest rates of exceptional aging occurred Table 2. Regression coefficients (SE) for measures of exceptional aging regressed on PM2.5 and other covariates using the full linear models with censoring of observations with residuals > 3 standard deviations. Coefficients are scaled per interquartile range difference of each variable. Variable (× IQR) PM2.5 (× 4.19 μg/m3) Percent smoking (× 4.77) Percent obesity (× 3.53) Median income (× 9.60) Percent below poverty (× 6.90) Population density (× 0.09) Percent urban (× 53.69) Percent high school graduate (× 12.70) Percent unemployed (× 2.90) Percent black (× 10.04) Percent Hispanic (× 4.19) Median age (× 4.60) Percent > 65 years old (× 4.97) Migration 1980s, 55- to 60-year-olds (× 12) Migration 1980s, 60- to 64-year-olds (× 14) Migration 1980s, 70- to 74-year-olds (× 10) Migration 1980s, ≥ 75-year-olds (× 10) Migration 1990s, 65- to 70-year-olds (× 17) Migration 1990s, 70- to 74-year-olds (× 12) Migration 1990s, ≥ 75-year-olds (× 13) Migration 2000s, ≥ 75-year-olds (× 15) Regional indicators R2 Number of counties Difference in the rates of 85- to 94-year-oldsa (P E8A5–94) –93.7 (12.2)*** –181.9 (14)*** –83.9 (9.5)*** 62.5 (12.4)*** –60.5 (19.5)** 0.2 (0.1) 86.8 (15.2)*** 13.8 (19.5) 6.1 (10.5) 3.9 (6.4) 0 (3.2) –110.3 (16.9)*** 158.5 (18.9)*** –86.3 (11.6)*** 275.9 (12.8)*** — — –140.9 (13.5)*** 333.7 (11.6)*** — 317.6 (7.1)*** Included 0.89 2,996b Difference in the rates of 100- to 104-year-oldsa (P E1A00–104) –3.5 (1.5)* –6.4 (1.7)*** –3.1 (1.1)** 5.3 (1.5)*** 0.5 (2.4) 0 (0)** –2.5 (1.8) –1.3 (2.4) 3.3 (1.3)** 5.2 (0.8)*** 0.4 (0.4) 1.5 (2.1) –0.3 (2.3) — — –0.6 (0.8) 9.4 (1.2)*** — — 11.6 (0.9)*** 2.9 (0.7)*** Included 0.39 2,996b 7Aa7E00bs––bt77irm44e  vyayieteaeaatsirross(nsosotla:lddnIQidninaR1r,1d9i9n8e80ter0)r )r×o q× ru1) 0a1o,0r0ft,0i0tlh0e0;e0rP;adMPnifgfE21e.Ae05r0,;e–pPn1a0cE8r4Ae5t,i–ci(9nun4,lutahm(tneeburmemaratbeotetsferoprofewfoipniptdehlioevapi1edle0ruo0a8d–l5sy1–n0894a54 m– yy9eiec4aa-rdyrsiesaaomorlldsedtoieinnlrd2<2o00 r211.01050 ddμ0iim–vvi1i;dd0See4Eddy, ebbsaytyarnsnnuudommaldrbbdieenerrr2oor0off1rpp0.,eeooovppelleer 10,000 individuals who were in the corresponding age group (55–64 years old or 70–74 years old) in 1980. The two param- eters represent the probability of exceptional aging in 2010, given the corresponding starting population in 1980, and were labeled listed in PthE8eA5–ta94blaen. dbTPheE1A0m0–o1d04e,lsreesxpcelucdtiveedlya.llRoeustuliletsrss, hdoewfinnewd earseoobbstearivnaetdiofnrsomwirtehgrreessidsiuoanlsmthoadtewlsefritetin>g 3 all the variables standard devia- tions > 0 or < 0. Number of counties included in the analysis varied because of different numbers of outliers excluded from the data for 85- to 94-year-olds and for 100- to 104-year-olds. *, p < 0.05; **, p < 0.01; ***, p < 0.001. Environmental Health Perspectives  •  volume 124 | number 11 | November 2016 1747 Baccarelli et al. in counties with relatively low pollution levels, lower rates of smoking and obesity, and higher median income. Prospective cohort survival studies that follow individuals over time can control for many individual risk factors and have provided some of the most important evidence regarding the health effects of longterm exposures to air pollution (Beelen et al. 2014; Cesaroni et al. 2013; Crouse et al. 2012; Dockery et al. 1993; Jerrett et al. 2013; Miller et al. 2007; Pope et al. 2002; Zeger et al. 2008). The ecological approach used in the present analysis provides alternative evidence that is a simple, direct, and transparent exploration that used U.S. county-level census and related data that are easily accessible and publicly available. Our approach did not require probabilistic data linkage with death records, nor was there a need to specify cause of death. This approach may overcome limitations regarding health assessment of the effects of air pollution, as well as other health determinants, in countries where population data are available, but the quality and/ or availability of mortality data is limited. Furthermore, if regular and reliable population counts are available, repeated analysis could be used to evaluate time trends in risk. The weighted regression techniques used in our analysis are also relatively straightforward and easy to interpret. For example, the results of this analysis demonstrate the relatively large effect of smoking rates. On average, out of 10,000 persons who were 55–65 years old in 1980, ~2,295 persons survived to be 85–94 years old in 2010. The regression coefficients on smoking suggest that, for every 1% increase in smoking rate, there were approximately 38 fewer 85- to 94-year-olds alive in 2010. These results suggest that the current average smoking rate of 18% in the United States (Agaku 2014) is responsible for a reduction of the probability of reaching age 85–94 years by ~30% {[(38 × 18) ÷ 2,295] × 100}. By comparison, regression results suggest that a 10-μg/m3 increase in long-term exposure to PM2.5 is associated with ~225 fewer 85- to 94-year olds or a reduction in the probability of reaching age 85–94 years by ~9.7% [(223 ÷ 2,295) × 100]. Comparisons between the effects of smoking and air pollution are similar for the probability of reaching 100–104 years of age. We also compared the effects of various other factors and estimated reductions in exceptional aging probabilities associated with interquartile-range changes in each ­population-­based variable. Based on effect estimates presented in Table 2, if we could improve any of these factors, the largest increase in exceptional aging would come from reduced smoking. Reductions in obesity, poverty, and air pollution would also provide substantial improvements. Other unmeasured differences resulting in racial inequalities or in residual uncontrolled confounding may contribute to these associations. The most important limitation to our approach concerns population mobility. The constructed indices of exceptional aging would be ideal if there were no migration across counties, and the populations in these counties could therefore be treated like cohorts that were being followed over time. However, because these populations are not strictly cohorts but are populations with uncontrolled migration, we controlled for migration as part of the regression analysis by including in the regression models age-specific migration rates that provided the closest possible temporal alignment consistent with the initial age group and with relevant age groups in subsequent decades, and the migration data available for all three decades. Migration rates did not appear to operate as serious confounders, as indicated by the similar associations of PM2.5 with exceptional aging found in unadjusted data (Figure 1), covariate-adjusted models (Table 2 and Figure 2), and the weak corre- lations between PM2.5 and the age-specific migration rates. Because persons move to and from counties with different levels of pollu- tion, however, even full population–based adjustments could not fully account for resuEltsintigmm­atiesdclapsosilfliuctaitoionnlsevoeflsPMwe2r.5e exposure. limited to the years 1999–2008 in our analysis because PM2.5 data were not collected regularly in the United States until 1999. Previous analyses have shown robust spatial patterns in the ranking order of PM2.5 levels measured at different U.S. locations between 1980 and 1999 (Pope et al. 2002). Therefore, the 1999– 2008 subperiod can be considered a suitable proxy of levels across the entire 1980–2010 period. There is also some temporal mismatch Figure 2. Estimated reduction in an interquartile range increase iinndPicMe2s.5o(f4.e1x9c μegp/tmio3n)afloargvinagri,oPusE8A5m–9o4 d(Ael)s, .aBndlacPkE1A0c0i–r1c04le(sB)r,eapsresosecniattemdowdeitlhs with no censored observations, gray squares represent models excluding observations with residuals > 3 standard deviations from zero, and open triangles represent models excluding observations with the 5% m(onfoupsmetboeepxrlteroef8m5p–ee9om4p liyeger1aa0rts0io–on1l0dp4ai nyttee2ar0rn1ss0obdldaivsiinedde2d0o1nb0ytdhnievuimdmebidgerrbayotifonpnuemroabpteeler fo7of0rp–≥e7 4o7 p5y-leeyae7ra0sr–-o7ol4dld ysienian1r9s280o00l)d 0×.i nP101E8,A950–809040),; ×(Pn 1Eu1A00m0,–0b100e04r,; PM2.5, particulate matter with aerodynamic diameter < 2.5 μm. 1748 volume 124 | number 11 | November 2016  •  Environmental Health Perspectives Particulate matter, aging, and centenarians between the three-decade time period used for aging (1980–2010) and the estimates for obesity prevalence (2004–2010) and smoking and socioeconomic variables (approximately 2000). The association estimates may be not be significantly affected if the spatial contrasts of these variables are reasonably stable over time. The analysis was conducted at a population, not an individual, level; therefore, this approach cannot evaluate risk factors at the individual level or explore sensitive subpopulations. Other unmeasured factors might also influence exceptional aging and result in residual confounding. Finally, information on age was obtained from U.S. census data and was not independently validated. This limitation may be particularly relevant for the 100- to 104-year-old group because birth records were less accurate over a century ago (Sachdev et al. 2012) than in more recent decades. Nevertheless, results were mostly similar for 85- to 94-year-olds and 100- to 104-year-olds. Additionally, age errors are likely to be n­ ondifferential with respect to the exposures and to bias our analysis toward the null rather than generate the observed ­significant associations. The present analysis is population-based, has comprehensive coverage of the United States, and includes a large number of exceptionally aged individuals. In particular, because of the large population of the United States and its relatively high life expectancy, the 48,303 centenarians in our data represent approximately 11% of all centenarians worldwide (United Nations 2010). We excluded potential influences from outliers, including counties with large in- or out-migration rates for elderly individuals. Therefore, our analysis provides statistically robust results that might apply to other countries with similar age structures and possibly to other regions in which life expectancy is increasing. Conclusions Our study supports the association between long-term exposure to air pollution and probability of exceptional aging. This association was Figure 3. Nonparametric smoothed functions illustrating relationships between the indices of exceptional fndaruigevmieinddbgeoed(mrPboE8(yfAE5–npD9ue4Fmosap)bnlaederre7Po0Erf1–Ae0p70p–e4o1 o0yrp4te)elaedarn7sin0do–ePl7daM4c iny2he.51ap9(raA8sn0,Coe) )ll×d. o P1irn0E8p,A510–e909r840c,0;e()Pn n×uE1t Aa0m10g0–be1,0e040or,0fo(;ndfPuapMmiely2bo.5eps,rlmpeoao8frk5tpei–cer9uso4lpa (yBlteee,Da1mr0)s.a0E–totq1eld0ur4iivw nyaietl2eha0nr1ast0edordeolidvgdiryidenneea2dsm0b1oicy0f diameter < 2.5 μm. found in our analysis—at least for part of the PM2.5 distribution—even at PM2.5 concentrations below the annual average limit values set by the U.S. Environmental Protection Agency (2013) (12 μg/m3), as well those set by other countries, such as Japan (15 μg/m3; Ministry of the Environment 2009), the European Union (25 μg/m3; European Commission 2008), and China (15–35 μg/m3; Chinese Ministry of Environmental Protection 2012). Rates of smoking, obesity, and poverty also showed associations with exceptional aging. Although more studies in other nations are needed, particulate matter air pollution is ubiqui­tous, and on the basis of our results, reducing PM2.5—along with improving other sources of inequality—may contribute to increasing the probability of exceptional aging. References Agaku IT, King BA, Dube SR, Centers for Disease Control and Prevention. 2014. Current cigarette smoking among adults — United States, 2005–2012. MMWR Morb Mortal Wkly Rep 63:29–34. Beckerman BS, Jerrett M, Serre M, Martin RV, Lee SJ, van Donkelaar A, et al. 2013. A hybrid approach to estimating national scale spatiotemporal variability of PM2.5 in the contiguous United States. Environ Sci Technol 47:7233–7241. Beelen R, Raaschou-Nielsen O, Stafoggia M, Andersen ZJ, Weinmayr G, Hoffmann B, et al. 2014. Effects of long-term exposure to air pollution on natural-cause mortality: an analysis of 22 European cohorts within the multicentre ESCAPE project. Lancet 383:785–795. Brook RD, Rajagopalan S, Pope CA III, Brook JR, Bhatnagar A, Diez-Roux AV, et al. 2010. Particulate matter air pollution and cardiovascular disease: an update to the scientific statement from the American Heart Association. Circulation 121:2331–2378. CDC (Centers for Disease Control and Prevention). 2013. Diabetes Data and Statistics. County Data. Diagnosed Diabetes Percentage, 2013. http:// www.cdc.gov/diabetes/atlas/countydata/atlas. html [accessed 6 May 2015]. Cesaroni G, Badaloni C, Gariazzo C, Stafoggia M, Sozzi R, Davoli M, et al. 2013. Long-term exposure to urban air pollution and mortality in a cohort of more than a million adults in Rome. Environ Health Perspect 121:324–331, doi: 10.1289/ehp.1205862. Chinese Ministry of Environmental Protection (MEP). 2012. Ambient Air Quality Standards (GB 3095-2012) [in Chinese]. http://kjs.mep.gov.cn/hjbhbz/bzwb/ dqhjbh/dqhjzlbz/201203/W020120410330232398521. pdf. [accessed 3 September 2016]. Crouse DL, Peters PA, van Donkelaar A, Goldberg MS, Villeneuve PJ, Brion O, et al. 2012. Risk of non­accidental and cardiovascular mortality in relation to long-term exposure to low concentrations of fine particulate matter: a Canadian national-level cohort study. Environ Health Perspect 120:708–714, doi: 10.1289/ehp.1104049. Dockery DW, Pope CA III, Xu X, Spengler JD, Ware JH, Fay ME, et al. 1993. An association between air pollution and mortality in six U.S. cities. N Engl J Med 329:1753–1759. European Commission. 2008. Air Quality Standards. http://ec.europa.eu/environment/air/quality/ standards.htm [accessed 7 August 2016]. Institute for Health Metrics and Evaluation. 2014. United Environmental Health Perspectives  •  volume 124 | number 11 | November 2016 1749 Baccarelli et al. States Smoking Prevalence by County 1996–2012. http://ghdx.healthdata.org/record/united-statessmoking-prevalence-county-1996-2012 [accessed 6 May 2015]. Jerrett M, Burnett RT, Beckerman BS, Turner MC, Krewski D, Thurston G, et al. 2013. Spatial analysis of air pollution and mortality in California. Am J Respir Crit Care Med 188:593–599. Masters RK. 2012. Uncrossing the U.S. Black-white mortality crossover: the role of cohort forces in life course mortality risk. Demography 49:773–796. Miller KA, Siscovick DS, Sheppard L, Shepherd K, Sullivan JH, Anderson GL, et  al. 2007. Longterm exposure to air pollution and incidence of cardiovascular events in women. N Engl J Med 356:447–458. Ministry of the Environment. 2009. Japan: Air Quality Standards. http://transportpolicy.net/ index.php?title=Japan:_Air_Quality_Standards [accessed 7 August 2016]. NIA (National Institute on Aging). 2013. Topics of Interest: Exceptional Longevity. https://www.nia. nih.gov/newsroom/topics/exceptional-longevity [accessed 6 May 2015]. NIA/WHO (National Institute on Aging/World Health Organization). 2011. Global Health and Aging. http://www.nia.nih.gov/sites/default/files/global_ health_and_aging.pdf [accessed 6 May 2015]. Olshansky SJ, Carnes BA, Désesquelles A. 2001. Demography. Prospects for human longevity. Science 291:1491–1492. Pope CA III, Burnett RT, Thun MJ, Calle EE, Krewski D, Ito K, et al. 2002. Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. JAMA 287:1132–1141. R Core Team. 2015. R: A Language and Environment for Statistical Computing. Vienna, Austria:R Foundation for Statistical Computing. https://www.R-project. org/ [accessed 12 November 2015]. Sachdev PS, Levitan C, Crawford JD. 2012. Methodological issues in centenarian research: pitfalls and challenges. Asian J Gerontol Geriatr 7:44–48. U.S. Census Bureau. 1980. Population Estimates. 1980s: County Tables. Available: https://www. census.gov/popest/data/historical/1980s/county. html [accessed 6 May 2015]. U.S. Census Bureau. 2000. Census 2000 Gateway. Available: http://www.census.gov/main/www/ cen2000.html [accessed 6 May 2015]. U.S. Census Bureau. 2010. 2010 Census Data. Available: http://www.census.gov/2010census/ data/ [accessed 6 May 2015]. U.S. Environmental Protection Agency (EPA). 2013. 40 CFR Parts 50, 51, 52 et al. National Ambient Air Quality Standards for Particulate Matter; Final Rule. Federal Register 78(10):3085–3287. United Nations. 2010. World Population Ageing 2009. New York:United Nations, Department of Economic and Social Affairs, Population Division. Winkler R, Johnson KM, Cheng C, Beaudoin J, Voss  PR, Curtis KJ. 2013. Age-Specific Net Migration Estimates for U.S. Counties, 1950–2010. Available: http://www.netmigration.wisc.edu/ [accessed 6 May 2015]. Zeger SL, Dominici F, McDermott A, Samet JM. 2008. Mortality in the Medicare population and chronic exposure to fine particulate air pollution in urban centers (2000–2005). Environ Health Perspect 116:1614–1619, doi: 10.1289/ehp.11449. 1750 volume 124 | number 11 | November 2016  •  Environmental Health Perspectives