the first decades of the twenty‐first century have been a challenging period for American
mortality. Life expectancy in the United States ranked 30th in the world in 2010 and
is much lower than in other high‐income countries (World Health Organization 2017).
Between 2010 and 2016, US life expectancy fell further behind other developed countries,
increasing by only 0.08 years, the smallest 5‐year increase since 1970 (Ho and Hendi
2018). These relatively slow mortality declines occurred against a background in which
US mortality in the 1990s and 2000s was already high by the standards of other OECD
countries (Ho and Preston 2010; Crimmins et al. 2011; Ho 2013; Institute of Medicine
and National Research Council 2013; Palloni and Yonker 2016). At the same time, there
have been large and growing geographic and socioeconomic inequalities in health and
mortality within the United States (Fenelon 2013; Wang et al. 2013; Hendi 2015, 2017;
Chetty et al. 2016; Montez, Sasson, and Hayward 2016a).
Several recent studies of the national‐level mortality stagnation have documented
adverse mortality trends among middle‐aged non‐Hispanic whites (Kochanek, Arias, and
Bastian 2016a; Squires and Blumenthal 2016; Case and Deaton 2017), particularly among
women (Astone, Martin, and Aron 2015; Gelman and Auerbach 2016; Kochanek et al. 2016b)
and those with lower levels of education (Hendi 2017) and income (Chetty et al. 2016).
Case and Deaton (2015, 2017) drew attention to the role that “deaths of despair”—consisting
of accidental poisoning (linked to the epidemic of prescription opioids and heroin),
suicide, and chronic liver disease—play in mortality increases among non‐Hispanic
whites. Elevated mortality from these causes of death is especially concentrated among
individuals with low levels of education (see also Ho 2017; Kochanek et al. 2016b).
However, these “deaths of despair” cannot fully explain the slowdown in mortality
declines, since the adverse trends persist even after eliminating mortality from these
causes (Squires and Blumenthal 2016; Monnat 2018; Rigg, Monnat, and Chavez 2018).
Other causes of death are also hypothesized to be important contributors to stagnating
mortality declines. Case and Deaton (2017) show that between 1999 and 2015, declines
in cardiovascular disease and cancer mortality at ages 50–54 were relatively slow
among non‐Hispanic whites in the United States compared to other OECD countries. One
study estimates that rising obesity has reduced the annual rate of decline in US death
rates at ages 40–84 by 0.5–0.6 percentage points between 1986 and 2011 (Preston, Vierboom,
and Stokes 2018).
A parallel literature has highlighted widening geographic inequalities in mortality
in the period leading up to the recent mortality stagnation. Mortality improvements
in Appalachia and the South, particularly the East South Central Division, have lagged
behind other regions (Fenelon 2013; Wang et al. 2013), a pattern that has been partly
linked to behavioral risk factors such as smoking and obesity (Fenelon 2013; Singh
and Siahpush 2014; Dwyer‐Lindgren et al. 2016; Dwyer‐Lindgren et al. 2017; Mokdad
et al. 2017; Roth et al. 2017). Trends in all‐cause and cause‐specific mortality rates
have also varied considerably among US states (Chetty et al. 2016; Montez, Sasson,
and Hayward 2016a) and counties (Ezzati et al. 2008; Dwyer‐Lindgren et al. 2016; Roth
et al. 2017). These studies typically find that although some counties, states, and
regions have experienced improvements in life expectancy, others have experienced
more moderate gains or even declines over the past few decades (Murray et al. 2006;
Cullen, Cummins, and Fuchs 2012; Wang et al. 2013; Chetty et al. 2016).
Other studies have documented widening mortality differences between metropolitan
and nonmetropolitan areas since the 1980s (Cosby et al. 2008; Cossman et al. 2010;
James 2014; James and Cossman 2017; Moy et al. 2017). Singh and Siahpush (2014) examined
data through 2009 and found that rural, nonmetropolitan areas made slower progress
in life expectancy in the preceding decades than urban, metropolitan areas. More recently,
Stein et al. (2017) focused on mortality trends at ages 25–64 between 1999 and 2015
by age, race/ethnicity, cause of death, and level of urbanization. They documented
increasing death rates for non‐Hispanic whites, mainly outside large urban areas,
with suicide, poisoning, and liver disease contributing to these adverse trends. Death
rates were the highest in rural areas for all racial/ethnic groups. Similarly, James
and Cossman (2017) documented growing urban‐rural disparities in age‐adjusted mortality
rates for both whites and Blacks since the mid‐1980s. Percent poor, region, and emergency
room visits and medical doctors per 1,000 were significant predictors of age‐adjusted
death rates among whites in most metropolitan‐nonmetropolitan subcategories in 2012,
but not among Blacks, suggesting that factors that predict the metropolitan‐nonmetropolitan
mortality disparities vary by race/ethnicity.
In this article, we build on this prior research by providing a comprehensive examination
of trends in non‐Hispanic white mortality between 1990 and 2016 by metropolitan‐nonmetropolitan
status and region. Prior studies have often focused on finer levels of geographic
variation, for example, counties or states, whereas our study examines 40 different
geographic areas consisting of 10 broad geographic regions cross‐classified by four
metropolitan‐nonmetropolitan categories. This cross‐classification allows us to incorporate
the substantial amount of variation within states (e.g., across metro‐nonmetro categories)
while identifying shared factors at the regional level that may be driving mortality
trends. We focus on non‐Hispanic whites because their mortality trends were particularly
adverse in the last decade and differ from those of non‐Hispanic Blacks and Hispanics.
In contrast to non‐Hispanic whites, mortality has continued to decline among Hispanics
and non‐Hispanic Blacks, although death rates remain substantially higher for Blacks
than for whites (Harper, MacLehose, and Kaufman 2014; Murphy et al. 2017; Stein et al.
2017). The underlying mechanisms driving mortality trends for non‐Hispanic whites
are fundamentally different from those driving trends for other racial/ethnic groups,
as suggested by James and Cossman (2017). Thus, we limit the scope of this article
to non‐Hispanic whites.
This article is an effort to unite the literature on adverse mortality trends among
non‐Hispanic whites at the national level with the literature on growing geographic
inequalities in mortality. Our goals are to carefully document life expectancy trends
in each of the 40 geographic areas described above, building upon previous research
in several ways. First, we extend analyses to 2016 and include all age groups. Second,
we estimate the contributions of four key age groups to changes in life expectancy
at birth between 1990 and 2016 by metropolitan‐nonmetropolitan status and region.
Third, we examine the contribution of 14 broad, cause of death categories to these
changes, highlighting categories that are strongly linked to behavioral factors and
access to health care. In the Discussion section, we consider several potential explanations
of the patterns we describe.
Data and methods
We use the 1990–2016 Multiple Cause of Death data files (provided by the National
Center for Health Statistics under a data user agreement) to tabulate deaths by age,
sex, race/ethnicity, cause of death, county, and year. To estimate person‐years of
exposure, we use the public‐use Census bridged‐race population estimates by age, sex,
race/ethnicity, county, and year. These data are combined to estimate age‐specific
death rates for all causes combined and for 14 specific, cause of death categories
described below. Death rates are estimated for non‐Hispanic white men and women by
age, year, metropolitan‐nonmetropolitan category, and geographic region. For parsimony,
we use “white men” and “white women” to refer to non‐Hispanic white men and non‐Hispanic
white women, respectively, from this point forward.
We focus on 14 mutually exclusive and exhaustive cause of death categories (described
in Table A‐1). Several of these categories are closely linked to behavioral factors:
alcohol‐attributable deaths and deaths from drug overdose, HIV/AIDS, homicide, suicide,
lung cancer, respiratory diseases, and diabetes. Two categories, screenable cancers
and influenza/pneumonia, were chosen as indicators of access to and quality of health
services. Cardiovascular disease, which is the leading cause of death and influenced
by both health behaviors and health care system variables, constitutes another category.
We also consider the role of a composite category, “deaths of despair” (the aggregation
of alcohol‐attributable, drug overdose,1 and suicide mortality), which has been hypothesized
to play a key role in adverse mortality trends among whites in recent years (Case
and Deaton 2015, 2017; Ho 2017). In addition, we separate out mental and nervous system
disorders, a category that includes Alzheimer's disease and is of emerging importance
(Xu et al. 2016; Ho and Hendi 2018). The remaining categories are the ill‐defined
causes category, which is used for nonspecific causes of death and accounts for a
relatively small proportion of overall deaths, and the residual category.
To classify counties by metro‐nonmetro status, we use the codes developed by the United
States Department of Agriculture (USDA) Economic Research Service (ERS), which were
modified and made available by the National Center for Health Statistics2 (https://www.cdc.gov/nchs/data_access/urban_rural.htm).
We use four categories: large central metros, their suburbs (“large metro suburbs”),
small/medium metros, and nonmetro areas (definitions of these categories are provided
in Table 1). Results from preliminary investigations separating nonmetro counties
by whether they were adjacent to metro areas were very similar for the two groups,
so they were combined in the final analyses. To maintain consistency over time, we
use the counties’ metropolitan category as of 2013 (preliminary analyses showed only
minor differences if we used earlier classification schemes). Our 10 broad geographic
regions are based on nine Census divisions and Appalachia, as defined by the Appalachian
Regional Commission. Appalachia includes all of West Virginia and counties from 12
other states. The Appalachian counties are excluded from their overlapping Census
divisions (definitions of these regions are provided in Table 1). When counties are
cross‐classified by region and metropolitan‐nonmetropolitan category, we identify
40 distinct geographic units.
Table 1
Region and metropolitan‐nonmetropolitan category classifications
Region (State)
Metropolitan‐nonmetropolitan categoriesb
1. New England (CT, ME, MA, NH, RI, VT)
1. Large central metro: Counties in MSAs of more than 1 million population, including
counties that contain all or a part of the area's inner cities
2. Middle Atlantic (NJ, NY, PA)
2. Large metro suburb: Surrounding counties of the large central metros
3. East North Central (IL, IN, MI, OH, WI)
3. Small/medium metro: Counties in MSAs of 50,000–999,999 population
4. West North Central (IA, KS, MN, MO, NE, ND, SD)
4. Nonmetropolitan area (nonmetro): All other counties
5. South Atlantic (DE, DC, FL, GA, MD, NC, SC, VA)
6. East South Central (AL, KY, MS, TN)
7. West South Central (AR, LA, OK, TX)
8. Mountain (AZ, CO, ID, MT, NV, NM, UT, WY)
9. Pacific (AK, CA, HI, OR, WA)
10. Appalachiaa
MSA = metropolitan statistical area.
a
As defined by the Appalachian Regional Commission and includes all of WV and selected
counties in AL, GA, KY, MD, MS, NY, NC, OH, PA, SC, TN, and VA. These counties are
excluded from the remaining regions, which are based on the nine Census divisions.
b
Based on USDA ERS and NCHS urban‐rural classification scheme (Ingram and Franco 2014).
John Wiley & Sons, Ltd.
Our main measure of mortality is life expectancy at birth. We start by examining trends
in life expectancy at birth and age‐group contributions to changes in life expectancy
across metro–nonmetro categories at the national level. Next, we investigate whether
the metro‐nonmetro mortality patterns observed at the national level also hold within
regions. We focus on three periods (1990–1992, 2009–2011, and 2014–2016) and the change
between 1990–1992 and 2014–2016 and between 2009–2011 and 2014–2016. Data are pooled
across three‐year periods to create more stable estimates. In estimating the life
tables, nax values are produced using graduation (Preston, Heuveline, and Guillot
2001). Mortality estimates for ages 85 and older are corrected using a variant of
the procedure outlined in Horiuchi and Coale (1982) to account for differences across
geographic areas in the age distributions of people aged 85 and older. The variant
involves the use of a parametric smoothing procedure (as opposed to direct computation)
to estimate growth rates and mortality above age 85.
Next, we estimate broad age group contributions (0–24, 25–44, 45–64, and 65+) to changes
in life expectancy at birth using Arriaga's (1984) decomposition. We focus on these
four age groups because they are socially and economically meaningful: people aged
0–24 capture the young and school‐aged population; people aged 25–44 have typically
left school and entered the workforce, and are not yet at the ages where chronic diseases
dominate; people aged 45–64 are often labeled middle aged and, while still in the
workforce, are more subject to chronic diseases; and people aged 65 and older are
usually out of the workforce. In addition, the 45–64 age group overlaps with the age
groups that were the focus of several prior studies of midlife mortality.
Third, we investigate the contribution of the specific cause of death categories described
above to trends in life expectancy using Arriaga's (1984) decomposition. We estimate
these contributions for the four metro‐nonmetro categories at the national level,
and then for each of the 40 geographic areas (metro‐nonmetro categories cross‐classified
with region). All analyses are performed separately by sex, and all decompositions
sum to 100 percent of the change in life expectancy over time for a specific geographic
area.
In the Discussion section, we introduce five variables that are potentially linked
to the trends we observe and examine the correlation between trends in these variables
and trends in life expectancy between 1990–1992 and 2014–2016 by the 40 region/metro‐nonmetro
categories. The five variables measure are as follows: educational attainment (percentage
of college graduates from the 1990 US Census Summary files and 5‐year 2011–2015 American
Community Survey), physician availability (active, nonfederal physicians per 1,000
population from the Area Health Resource Files [US Health Resources & Services Administration
2018]), obesity (percentage of population aged 20 and older with body mass index greater
than 30 kg/m2 from the University of Wisconsin [2018]), transfer dependency (transfers
as a percentage share of personal income from the Bureau of Economic Analysis 2018),
and net in‐migration rate for the working‐age population (cumulative net in‐migration
rate for people aged 22.5–62.5, estimated using data from the NCHS and the Census
Bureau and indirect methods detailed in the Appendix).3
Results
Trends by metro‐nonmetro status
Table 2 presents changes in life expectancy between 1990 and 2016 by metro‐nonmetro
status. Life expectancy levels themselves are presented in Table A‐2 in the Appendix.
It is clear from Table 2 that the United States has experienced growing geographic
inequality in life expectancy gains for white men and women over this period. This
divergence has been driven by more rapid increases in life expectancy in large central
metros and slower improvements elsewhere. White male life expectancy increased 5.09
years in large central metros, compared to 3.45 years in large metro suburbs, 2.81
years in small/medium metros, and 2.25 years in nonmetro areas. Large central metros
are now among the areas with the highest white male life expectancy in the country.
This ascendance is particularly noteworthy because large central metros had the lowest
life expectancy levels in 1990–1992. The gain in large central metros was 2.3 times
greater (5.09/2.25) than that in nonmetro areas, which had the second lowest life
expectancy levels in 1990–1992.
Table 2
Age group contributions (years) to changes in non‐Hispanic white life expectancy at
birth between 1990–1992 and 2014–2016 and between 2009–2011 and 2014–2016 by metropolitan‐nonmetropolitan
category and sex
(1990–1992) to (2014–2016)
Δe0
0–24 years
25–44 years
45–64 years
65+ years
Males
Large central metro
5.09
0.57
0.83
1.23
2.46
Large metro suburb
3.45
0.41
−0.18
0.85
2.37
Small/medium metro
2.81
0.46
−0.20
0.42
2.14
Nonmetro
2.25
0.53
−0.31
0.26
1.77
Females
Large central metro
2.98
0.37
0.03
0.68
1.91
Large metro suburb
2.23
0.26
−0.21
0.52
1.65
Small/medium metro
1.24
0.25
−0.31
0.08
1.22
Nonmetro
0.20
0.24
−0.46
−0.21
0.63
(2009–2011) to (2014–2016)
Δe0
0–24 years
25–44 years
45–64 years
65+ years
Males
Large central metro
0.31
0.07
−0.13
0.08
0.29
Large metro suburb
0.05
0.03
−0.29
−0.01
0.32
Small/medium metro
−0.11
0.02
−0.20
−0.17
0.24
Nonmetro
−0.18
0.06
−0.16
−0.22
0.14
Females
Large central metro
0.36
0.04
−0.05
0.03
0.35
Large metro suburb
0.20
0.02
−0.14
−0.01
0.32
Small/medium metro
−0.06
0.01
−0.13
−0.18
0.23
Nonmetro
−0.29
0.03
−0.14
−0.25
0.08
NOTE: Numbers may not add up due to rounding.
SOURCE: Vital Statistics and Census Data. Calculations by the authors.
John Wiley & Sons, Ltd.
Among white women, life expectancy differences by metro‐nonmetro status were relatively
small in 1990–1992. However, by 2014–2016, substantial variation emerged, with the
largest gains recorded in large central metros (2.98 years), followed by large metro
suburbs (2.23 years), small/medium metros (1.24 years), and nonmetro areas (0.20 years)
(Table 2). These differences are particularly striking for large central metros compared
to nonmetros, with gains in white female life expectancy that were 14.9 times greater
(2.98/0.20) in large central metros than in nonmetro areas.
The most recent period stands out for its stark metro‐nonmetro differences (Table 2,
bottom panel). Between 2009–2011 and 2014–2016, large central metros and large metro
suburbs continued to experience gains in life expectancy, but small/medium metros
and nonmetros experienced life expectancy declines. Thus, while the metro‐nonmetro
gradient widened between 1990–1992 and 2009–2011 due to quicker gains among large
central metros, the gradient has since widened due to a combination of gains in large
metros and their suburbs and declines in small/medium metros and nonmetro areas.
The pattern of life expectancy gains was similar for men and women, but women experienced
smaller gains than men, leading to a narrowing of sex differences in life expectancy.
By 2014–2016, the sex difference ranged from 4.65 years in large metro suburbs to
4.96 in nonmetro areas, down from 5.87 years in large metro suburbs and 7.01 years
in nonmetro areas in 1990–1992 (Table A‐2).
There are important variations in the contribution of different age groups to life
expectancy trends across the metro‐nonmetro categories. In large central metros, all
age groups contributed to gains in life expectancy between 1990 and 2016 (Table 2,
top panel). In the other three areas, however, mortality increased at ages 25–44 for
both white men and women and additionally at ages 45–64 for white women in nonmetro
areas. Between 2009–2011 and 2014–2016, mortality increased in the 25–44 age group
for white men and women in all four categories, including large central metros. Furthermore,
in all areas except large central metros, mortality increased at ages 45–64 (Table 2,
bottom panel).
Trends by metro‐nonmetro status and region
The national trends in life expectancy by metro‐nonmetro status are also observed
within most regions of the country, with important regional variation. Figure 1 and
Table 3 present trends in life expectancy between 1990–1992 and 2014–2016 for the
40 areas, representing the four metro‐nonmetro categories and the 10 geographic regions.
Life expectancy levels by metro‐nometro areas and region are shown in Table A‐3 in
the Appendix. The spatial pattern that was evident for the nation as a whole in Table 2—greatest
life expectancy gains in large central metros and smallest life expectancy gains in
nonmetros—is, with minor exceptions, maintained within each of the 10 regions. At
the same time, regional variations are also evident: the Middle Atlantic and Pacific
regions stand out as having particularly rapid life expectancy gains for white men
in large central metros, on the order of 7.13 and 6.11 years, respectively. White
men in nonmetro areas of the Appalachian, East South Central, and West South Central
regions experienced the smallest gains, amounting to 1.42–1.80 years.
Figure 1
Change in non‐Hispanic white life expectancy at birth (in years) by metropolitan‐nonmetropolitan
category and region, 1990–1992 and 2014–2016
NOTE: APP = Appalachia, ENC = East North Central, ESC = East South Central, MA = Middle
Atlantic, MTN = Mountain, NE = New England, PAC = Pacific, SA = South Atlantic, WNC
= West North Central, WSC = West South Central.
Table 3
Age group contributions (in years) to changes in non‐Hispanic white life expectancy
at birth between 1990–1992 and 2014–2016 by metropolitan‐nonmetropolitan category,
region, and sex
(1990–1992) to (2014–2016)
Large central metro
Large metro suburb
Small/medium metro
Nonmetro
Δ e0
0–24
25–44
45–64
65+
Δ e0
0–24
25–44
45–64
65+
Δ e0
0–24
25–44
45–64
65+
Δ e0
0–24
25–44
45–64
65+
Males
APP
2.54
0.24
−0.43
0.51
2.22
2.61
0.42
−0.65
0.76
2.08
2.10
0.40
−0.48
0.26
1.92
1.80
0.55
−0.51
0.10
1.65
ENC
3.86
0.50
0.13
0.92
2.31
3.15
0.43
−0.34
0.74
2.32
2.49
0.44
−0.37
0.37
2.06
2.46
0.52
−0.36
0.44
1.85
ESC
2.43
0.33
−0.16
0.22
2.03
2.86
0.54
−0.33
0.59
2.06
2.25
0.38
−0.23
0.25
1.85
1.42
0.49
−0.32
−0.01
1.26
MA
7.13
0.76
1.65
1.98
2.74
3.97
0.35
−0.16
1.08
2.70
3.12
0.24
−0.22
0.78
2.33
3.77
0.52
−0.27
0.95
2.57
MTN
4.03
0.64
0.52
0.79
2.08
3.82
0.50
−0.02
0.57
2.77
2.98
0.49
−0.06
0.33
2.22
2.92
0.63
−0.26
0.27
2.27
NE
4.23
0.45
0.45
1.01
2.31
3.41
0.31
−0.47
1.09
2.48
3.31
0.36
−0.43
0.95
2.43
3.10
0.38
−0.52
0.85
2.38
PAC
6.11
0.67
1.33
1.47
2.64
4.32
0.73
0.34
0.91
2.34
3.36
0.56
0.22
0.45
2.12
2.97
0.60
0.06
0.25
2.06
SA
5.34
0.61
0.92
1.42
2.39
3.19
0.37
−0.18
0.75
2.25
2.98
0.53
−0.17
0.37
2.24
2.63
0.55
−0.05
0.27
1.87
WNC
4.79
0.57
0.56
1.07
2.60
3.62
0.37
−0.10
0.92
2.44
3.11
0.61
−0.06
0.49
2.07
2.33
0.56
−0.23
0.24
1.77
WSC
4.58
0.49
0.81
1.02
2.26
3.92
0.60
0.14
0.98
2.21
2.22
0.50
−0.14
0.15
1.71
1.55
0.49
−0.20
−0.13
1.39
Females
APP
1.55
0.09
−0.39
0.28
1.57
1.02
0.28
−0.52
0.13
1.12
0.24
0.21
−0.51
−0.24
0.79
−0.39
0.23
−0.63
−0.47
0.48
ENC
2.20
0.29
−0.15
0.55
1.51
2.06
0.25
−0.23
0.53
1.50
1.04
0.24
−0.40
0.15
1.04
0.72
0.25
−0.37
0.08
0.76
ESC
0.84
0.20
−0.48
0.04
1.09
0.77
0.25
−0.35
0.11
0.76
0.31
0.25
−0.29
−0.34
0.70
−0.98
0.20
−0.55
−0.67
0.04
MA
4.66
0.58
0.46
1.15
2.48
3.15
0.27
−0.10
0.79
2.20
2.03
0.18
−0.24
0.50
1.60
2.01
0.30
−0.25
0.50
1.47
MTN
1.99
0.32
−0.15
0.30
1.53
2.73
0.35
−0.12
0.61
1.90
1.44
0.29
−0.26
0.06
1.35
1.30
0.28
−0.31
0.06
1.26
NE
2.32
0.19
−0.03
0.54
1.62
2.50
0.26
−0.21
0.78
1.67
2.21
0.20
−0.23
0.71
1.53
1.49
0.13
−0.37
0.41
1.32
PAC
4.00
0.48
0.20
0.94
2.38
2.91
0.35
0.01
0.66
1.89
1.96
0.31
−0.11
0.33
1.42
1.82
0.30
−0.25
0.10
1.67
SA
2.93
0.43
0.02
0.64
1.85
2.04
0.28
−0.27
0.36
1.67
1.47
0.29
−0.33
−0.09
1.60
0.12
0.18
−0.55
−0.42
0.90
WNC
2.32
0.31
−0.02
0.59
1.44
1.88
0.23
−0.21
0.53
1.34
1.24
0.27
−0.18
0.22
0.92
0.35
0.29
−0.30
−0.07
0.42
WSC
2.20
0.26
−0.06
0.44
1.55
1.67
0.25
−0.22
0.34
1.29
0.38
0.26
−0.34
−0.34
0.80
−0.99
0.19
−0.60
−0.72
0.14
APP = Appalachia, ENC = East North Central, ESC = East South Central, MA = Middle
Atlantic, MTN = Mountain, NE = New England, PAC = Pacific, SA = South Atlantic, WNC
= West North Central, WSC = West South Central.
NOTE: Numbers may not add up due to rounding.
SOURCE: Vital Statistics and Census Data. Calculations by the authors.
John Wiley & Sons, Ltd.
A similar pattern is observed for white women, with the largest gains occurring in
large central metros in the Middle Atlantic (4.66 years) and Pacific (4.00 years)
regions. Nonmetro areas experienced the smallest gains in female life expectancy,
especially in the East North Central, West North Central, and South Atlantic regions,
with actual declines in life expectancy observed in nonmetro areas of the Appalachian,
East South Central, and West South Central regions.
In all 40 areas, white men's life expectancy gains outpaced white women's life expectancy
gains. If we compare the best‐performing region/metro category/sex combination to
the worst‐performing combination, men in large central metros in the Middle Atlantic
gained 7.13 years of life expectancy during this period whereas women in nonmetros
in the West South Central and East South Central regions lost nearly a year in life
expectancy between 1990–1992 and 2014–2016.
Despite the strong performance of large central metros and the relatively poor performance
of nonmetros within each region, striking patterns emerge when comparing across both
region and metro category. For example, large central metros outperformed nonmetros
within the Appalachian and East South Central regions. However, life expectancy gains
in large central metros in these two regions were significantly smaller than gains
in nonmetros of the Middle Atlantic. The life expectancy gains in nonmetros of the
Middle Atlantic exceeded those in large central metros of the Appalachian and East
South Central regions by 1.2–1.3 years for men and 0.5–1.2 years for women.
Age‐specific contributions to gains in life expectancy between 1990–1992 and 2014–2016
by metro‐nonmetro status and region are shown in Figures 2A and 2B and Table 3. Again,
the patterns observed for the nation as a whole are repeated in many of the 40 areas.
Mortality increases among white men and women aged 25–44 were widespread in large
metro suburbs, small/medium metros, and nonmetro areas, with the most noticeable increases
typically occurring in Appalachia and New England. Mortality improvements at ages
65 and above made the largest contributions to gains in life expectancy in all 40
geographic areas.
Figure 2A
Age‐group contributions (in years) to changes in non‐Hispanic white male life expectancy
at birth by metropolitan‐nonmetropolitan category and region, 1990–1992 to 2014–2016
NOTE: APP = Appalachia, ENC = East North Central, ESC = East South Central, MA = Middle
Atlantic, MTN = Mountain, NE = New England, PAC = Pacific, SA = South Atlantic, WNC
= West North Central, WSC = West South Central.
Figure 2B
Age‐group contributions (in years) to changes in non‐Hispanic white female life expectancy
at birth by metropolitan‐nonmetropolitan category and region, 1990–1992 to 2014–2016
NOTE: APP = Appalachia, ENC = East North Central, ESC = East South Central, MA = Middle
Atlantic, MTN = Mountain, NE = New England, PAC = Pacific, SA = South Atlantic, WNC
= West North Central, WSC = West South Central.
When we consider life expectancy trends between 2009–2011 and 2014–2016 (Table 4 and
Figures 3A and 3B), the adverse mortality trends in the 25–44 age group are even more
striking. In all four metro categories and across almost every region, mortality at
ages 25–44 contributed negatively to life expectancy trends in this most recent period.
These contributions were particularly large in the New England and Appalachian regions
for both men and women. Ages 45–64 also contributed negatively to life expectancy
trends in many areas, and more so among women than among men. We observed a metro‐nonmetro
gradient with respect to these age contributions: mortality at ages 25–44 mattered
relatively more in large central metros and their suburbs, while mortality at ages
45–64 contributed relatively more to negative trends in small/medium metros and nonmetros.
Table 4
Age group contributions (in years) to changes in non‐Hispanic white life expectancy
at birth between 2009–2011 and 2014–2016 by metropolitan‐nonmetropolitan category,
region, and sex
(2009–2011) to (2014–2016)
Large central metro
Large metro suburb
Small/medium metro
Nonmetro
Δ e0
0–24
25–44
45–64
65+
Δ e0
0–24
25–44
45–64
65+
Δ e0
0–24
25–44
45–64
65+
Δ e0
0–24
25–44
45–64
65+
Males
APP
−0.17
0.01
−0.45
−0.04
0.30
−0.22
−0.04
−0.47
−0.05
0.34
−0.23
0.02
−0.26
−0.20
0.20
−0.18
0.04
−0.14
−0.27
0.19
ENC
0.00
0.05
−0.25
−0.01
0.21
−0.04
0.07
−0.31
−0.01
0.21
−0.35
−0.01
−0.34
−0.15
0.15
−0.27
0.03
−0.27
−0.15
0.12
ESC
−0.50
−0.02
−0.31
−0.24
0.07
−0.10
0.05
−0.25
−0.16
0.27
−0.32
−0.05
−0.20
−0.26
0.19
−0.42
0.02
−0.18
−0.29
0.03
MA
0.36
0.01
−0.19
0.22
0.31
0.04
0.04
−0.38
0.04
0.35
−0.22
0.01
−0.30
−0.11
0.18
0.23
0.25
−0.23
−0.16
0.37
MTN
0.29
0.14
−0.09
0.01
0.23
0.23
−0.10
−0.14
0.03
0.45
−0.03
0.02
−0.14
−0.19
0.29
0.00
0.13
−0.24
−0.13
0.24
NE
0.00
0.05
−0.40
0.11
0.24
−0.50
−0.04
−0.67
−0.08
0.28
−0.61
−0.06
−0.57
−0.14
0.16
−0.41
0.06
−0.52
−0.22
0.26
PAC
0.53
0.08
−0.03
0.16
0.33
0.31
0.09
−0.07
0.03
0.25
0.08
0.06
−0.03
−0.20
0.25
−0.28
−0.15
−0.10
−0.20
0.17
SA
0.57
0.05
−0.07
0.13
0.46
0.08
0.04
−0.28
−0.06
0.37
0.10
0.06
−0.18
−0.13
0.35
−0.17
0.02
−0.14
−0.25
0.20
WNC
0.69
0.13
0.00
0.24
0.33
0.36
0.05
−0.15
0.08
0.38
0.04
0.05
−0.09
−0.13
0.21
−0.11
0.09
−0.08
−0.22
0.10
WSC
0.30
0.13
−0.07
0.01
0.23
0.41
0.04
−0.09
0.05
0.41
−0.01
0.07
−0.02
−0.20
0.14
−0.10
0.14
0.03
−0.28
0.02
Females
APP
−0.05
0.00
−0.22
−0.12
0.28
−0.08
0.02
−0.26
−0.12
0.28
−0.30
0.05
−0.18
−0.27
0.10
−0.29
0.07
−0.16
−0.31
0.11
ENC
0.05
0.08
−0.15
−0.02
0.14
0.20
0.05
−0.14
0.00
0.28
−0.32
−0.01
−0.22
−0.18
0.09
−0.27
0.07
−0.15
−0.16
−0.03
ESC
−0.41
−0.02
−0.35
−0.18
0.14
−0.20
−0.02
−0.14
−0.15
0.11
−0.03
0.07
0.00
−0.27
0.18
−0.46
0.07
−0.15
−0.37
0.00
MA
0.41
0.02
−0.03
0.13
0.28
0.26
0.01
−0.16
0.07
0.33
0.05
0.03
−0.19
−0.09
0.30
0.23
0.10
−0.08
−0.09
0.29
MTN
0.33
0.02
−0.06
−0.05
0.43
0.69
0.14
−0.05
0.18
0.42
−0.01
0.00
−0.08
−0.20
0.26
0.12
0.02
−0.05
−0.16
0.31
NE
0.08
−0.02
−0.02
−0.05
0.16
0.04
0.04
−0.25
−0.04
0.29
−0.16
−0.03
−0.25
0.00
0.12
−0.32
−0.09
−0.25
−0.14
0.15
PAC
0.63
0.06
0.01
0.11
0.45
0.39
0.02
−0.04
0.03
0.38
0.24
0.00
0.02
−0.07
0.30
−0.02
−0.07
−0.13
−0.31
0.49
SA
0.65
0.03
−0.03
0.06
0.59
0.20
0.00
−0.15
−0.03
0.38
0.03
0.03
−0.13
−0.25
0.38
−0.38
−0.08
−0.17
−0.30
0.17
WNC
0.47
0.03
0.01
0.08
0.34
0.23
0.01
−0.12
0.05
0.30
0.08
−0.03
−0.03
−0.09
0.22
−0.38
0.04
−0.09
−0.23
−0.11
WSC
0.44
0.07
0.02
−0.03
0.38
0.26
0.04
−0.06
−0.11
0.39
−0.11
0.05
−0.11
−0.24
0.18
−0.45
0.00
−0.14
−0.34
0.03
APP = Appalachia, ENC = East North Central, ESC = East South Central, MA = Middle
Atlantic, MTN = Mountain, NE = New England, PAC = Pacific, SA = South Atlantic, WNC
= West North Central, WSC = West South Central.
NOTE: Numbers may not add up due to rounding.
SOURCE: Vital Statistics and Census Data. Calculations by the authors.
John Wiley & Sons, Ltd.
Figure 3A
Age‐group contributions (in years) to changes in non‐Hispanic white male life expectancy
at birth by metropolitan‐nonmetropolitan category and region, 2009–2011 to 2014–2016
NOTE: APP = Appalachia, ENC = East North Central, ESC = East South Central, MA = Middle
Atlantic, MTN = Mountain, NE = New England, PAC = Pacific, SA = South Atlantic, WNC
= West North Central, WSC = West South Central.
Figure 3B
Age‐group contributions (in years) to changes in non‐Hispanic white female life expectancy
at birth by metropolitan‐nonmetropolitan category and region, 2009–2011 to 2014–2016
NOTE: APP = Appalachia, ENC = East North Central, ESC = East South Central, MA = Middle
Atlantic, MTN = Mountain, NE = New England, PAC = Pacific, SA = South Atlantic, WNC
= West North Central, WSC = West South Central.
These adverse mortality trends among working‐aged whites led to declines in life expectancy
at birth between 2009–2011 and 2014–2016 for both sexes in all nonmetro areas except
in the Middle Atlantic and Mountain regions. Among men, life expectancy also declined
in small/medium metros in seven out of 10 regions, in large metro suburbs in four
out of 10 regions, and in large central metros in two out of 10 regions. Women experienced
life expectancy declines in small/medium metros in six regions, in large metro suburbs
in two regions, and in large central metros in two regions. The setbacks suffered
by working‐aged whites during this most recent period were clearly widespread by sex,
region, and metropolitan category, but the most severe problems were encountered in
nonmetropolitan areas.
Cause‐specific contributions to trends by metro‐nonmetro status
Table 5 and Figure 4 present cause of death contributions to the change in life expectancy
by metro‐nonmetro status between 1990 and 2016. Causes of death with positive values
in Table 5 contribute to life expectancy improvements, while those with negative values
contribute to life expectancy reductions. In all four areas, reductions in cardiovascular
disease mortality made the largest contributions to improvements in life expectancy.
Mortality from causes of death amenable to health care, such as screenable cancers,
influenza and pneumonia, and HIV/AIDS, also contributed to life expectancy gains in
all four metro‐nonmetro categories, but these contributions were the smallest in nonmetro
areas. Reductions in mortality from HIV/AIDS (likely related to the introduction of
highly active antiretroviral therapy) were more important for men than for women,
and their impact was particularly large for men in large central metros, where it
made the second largest contribution to life expectancy gains, after cardiovascular
disease (Chiasson et al. 1999; Messeri et al. 2003).
Table 5
Cause‐specific contributions (in years) to changes in non‐Hispanic white life expectancy
at birth between 1990–1992 and 2014–2016 by metropolitan‐nonmetropolitan category
and sex
Cause of death
Large central metro
Large metro suburb
Small/medium metro
Nonmetro
Males
HIV/AIDS
0.86
0.27
0.19
0.09
Homicide
0.24
0.06
0.03
0.04
Alcohol‐related causes
−0.01
−0.06
−0.10
−0.07
Drug overdose
−0.46
−0.71
−0.57
−0.47
Suicide
0.01
−0.09
−0.12
−0.14
Screenable cancersa
0.32
0.36
0.30
0.24
Lung cancer
0.64
0.61
0.58
0.48
Respiratory disease
0.14
0.06
0.04
‐0.05
Circulatory disease
2.59
2.53
2.31
2.09
Mental and nervous system disordersb
−0.35
−0.41
−0.41
−0.37
Diabetes
−0.04
−0.02
−0.07
−0.11
Influenza/pneumonia
0.25
0.21
0.21
0.19
Symptoms and ill‐defined
0.10
0.06
0.07
0.05
All other
0.82
0.59
0.35
0.30
All causes combined
5.09
3.45
2.81
2.25
Females
HIV/AIDS
0.07
0.03
0.01
0.00
Homicide
0.06
0.02
0.02
0.01
Alcohol‐related causes
−0.04
−0.05
−0.08
−0.07
Drug overdose
−0.26
−0.36
−0.36
−0.35
Suicide
−0.02
−0.06
−0.08
−0.09
Screenable cancersa
0.47
0.49
0.39
0.32
Lung cancer
0.26
0.18
0.09
−0.05
Respiratory disease
−0.09
−0.20
−0.28
−0.47
Circulatory disease
2.58
2.37
2.14
1.84
Mental and nervous system disordersb
−0.88
−0.91
−0.97
−0.94
Diabetes
0.03
0.07
0.03
0.00
Influenza/pneumonia
0.28
0.24
0.22
0.18
Symptoms and ill‐defined
0.02
0.00
0.01
−0.02
All other causes
0.52
0.40
0.09
−0.15
All causes combined
2.98
2.22
1.24
0.20
a
Breast, prostate, colorectal, and cervical cancer.
b
Includes Alzheimer's disease.
NOTE: Negative values indicate contributions to reductions in life expectancy and
positive values indicate contributions to increases in life expectancy between 1990–1992
and 2014–2016.
SOURCE: Vital Statistics and Census Data. Calculations by the authors.
John Wiley & Sons, Ltd.
Figure 4
Cause‐specific contributions to changes in life expectancy at birth (in years) by
metropolitan‐nonmetropolitan category, non‐Hispanic whites, 1990–1992 to 2014–2016
Among causes of death closely tied to health behaviors, declining lung cancer mortality
made important contributions to life expectancy increases among men, but much smaller
contributions among women. For women in nonmetro areas, lung cancer mortality increased
over time. Mortality from respiratory diseases, which are also related to smoking,
increased among women, with the largest increases recorded in nonmetro areas.
Mortality from drug overdose, suicide, and alcohol‐related causes of death, which
largely comprise “deaths of despair” category, increased and contributed to life expectancy
reductions across the metro‐nonmetro categories. Drug overdose was the most important
contributor among these causes, and it made a larger contribution to life expectancy
declines among men than women. Among men, drug overdose made the largest impact in
large metro suburbs, followed by small/medium metros, nonmetros, and large central
metros. Among women, the impacts were greater in large metro suburbs, small/medium
metros, and nonmetros than in large central metros. Alcohol‐related causes were a
minor factor in all metro categories. One of the most important contributors to life
expectancy reductions were mental and nervous system disorders, including Alzheimer's
disease. This category was particularly important for women, making large negative
contributions to life expectancy trends in all metro‐nonmetro areas.
Cause‐specific contributions to trends by metro‐nonmetro status and region
The basic pattern of cause‐specific contributions across the metro‐nonmetro continuum
at the national level is preserved across regions (Figures 5A and 5B), with a few
key variations. In all 40 areas, reductions in cardiovascular disease mortality made
the largest contribution to increases in life expectancy at birth. This was observed
for both men and women. Declines in lung cancer mortality were evident in all regions
and all metro‐nonmetro categories among men, but were small or absent among women,
especially in nonmetro areas. One exception is the Pacific region, where women did
experience reductions in lung cancer mortality in all metro‐nonmetro categories. As
was the case nationally, increases in respiratory disease mortality contributed to
life expectancy reductions among women in all regions, with the largest impact generally
observed in nonmetro areas. Increases in respiratory disease mortality were particularly
large among women in nonmetro areas of the Appalachian, East South Central, West South
Central, West North Central, and South Atlantic regions, where they contributed to
a roughly 0.5‐year reduction in life expectancy at birth.
Figure 5A
Cause‐specific contributions to changes in life expectancy at birth (in years) by
metropolitan‐nonmetropolitan category and region, non‐Hispanic white men, 1990–1992
to 2014–2016
NOTE: 1 = Large central metro, 2 = Large metro suburb, 3 = Small/medium metro, 4 =
Nonmetro.
Figure 5B
Cause‐specific contributions to changes in life expectancy at birth (in years) by
metropolitan‐nonmetropolitan category and region, non‐Hispanic white women, 1990–1992
to 2014–2016
NOTE: 1 = Large central metro, 2 = Large metro suburb, 3 = Small/medium metro, 4 =
Nonmetro.
Across both region and metro‐nonmetro categories, the contributions of causes of death
related to medical care (e.g., screenable cancers, HIV/AIDS, and influenza/pneumonia)
were similar to those documented for the country as a whole. In all regions, these
causes contributed to an increase in life expectancy. The contribution of HIV/AIDS
was particularly large in large central metros in the Middle Atlantic (1.45 years),
South Atlantic (0.94 years), and Pacific (1.12 years) regions among men, whereas among
women, HIV/AIDS made a sizeable contribution only in large central metros of the Middle
Atlantic (0.30 years). Screenable cancers and influenza and pneumonia generally made
larger contributions to life expectancy improvements among women than among men.
Furthermore, the cause of death categories contributing to adverse life expectancy
trends in Table 5 and Figure 1 are also implicated in adverse life expectancy trends
in all regions across metro‐nonmetro categories. There is some evidence, especially
among men, that suicide tends to make larger contributions to life expectancy reductions
in nonmetros and small/medium metros than in large central metros and their suburbs.
Drug overdose is a key contributor to life expectancy reductions for men in all 40
areas, making the largest contributions in Appalachia and New England and the smallest
contribution in the Pacific. In most regions, there is no clear gradient in the contribution
of drug overdose to life expectancy reductions by metro‐nonmetro category. However,
there are two exceptions: in Appalachia, there is a positive relationship between
this contribution and the level of urbanicity (i.e., drug overdose makes the largest
contribution to life expectancy reductions in large central metros and the smallest
contribution in nonmetros), and in the Pacific, a negative relationship holds. Among
women, drug overdose made important contributions in the Appalachian and East South
Central regions. Thus, the patterns of cause of death contributions are surprisingly
similar across the country with some variation in the magnitudes of their impact.
Summary and discussion
Over the last quarter century, we have witnessed growing geographic inequalities in
mortality in the United States. Two notable features of the last 25 years are the
sizable increase in life expectancy in large central metros and the slow improvement
or decline, especially among women, in nonmetro areas. For the United States as a
whole, white male life expectancy at birth in large central metros increased by 5.09
years between 1990–1992 and 2014–2016; the comparable figure for white women was 2.98
years. In contrast, nonmetro areas experienced the smallest life expectancy gains:
2.25 years among white men and only 0.20 years among white women. This pattern of
larger increases in life expectancy in large central metros and small or negligible
increases or even declines in nonmetro areas was pervasive in all 10 regions of the
country examined in these analyses.
Importance of mortality trends at the working ages
The extant literature has largely focused on adverse trends among middle‐aged whites,
especially at ages 45–54 (Case and Deaton 2015, 2017). Our study demonstrates that
adverse mortality trends in a younger age group, ages 25–44, also slowed life expectancy
improvements in large metro suburbs, small/medium metros, and nonmetro areas between
1990–1992 and 2014–2016. Since 2010, mortality increased at ages 25–44 in all metro‐nonmetro
categories and at ages 45–64. The mortality increases at younger ages are particularly
troubling. The causes of death that predominate at these ages, like drug overdose,
are largely preventable and may be related to worsening social and economic conditions.
Young adults today have experienced difficulties coming of age during the Great Recession,
that is, delayed transition to adulthood, declines in marriage, and increased rates
of coresidence with parents. An indicator of the poor economic conditions among these
cohorts is that 41 percent of men aged 25–34 earned less than $30,000 per year in
2016; this figure was only 25 percent in 1975 (Vespa 2017). A recent study documented
large increases in mortality from alcoholic liver cirrhosis among adults aged 25–34
(Tapper and Parikh 2018). Furthermore, the adverse mortality conditions and the underlying
factors driving these trends may have life course implications. For example, adults
in this age group have increased rates of drug and alcohol abuse and may experience
increased morbidity and mortality related to these behaviors in future decades (Corrao
et al. 2004; Ronan and Herzig 2016; Ho 2019).
Another surprising new finding from our study is that adverse mortality changes in
the recent period were greater at ages 25–44 than at ages 45–64 in large metros and
their suburbs, whereas adverse mortality changes at ages 45–64 were generally greater
than those at ages 25–44 in small/medium metros and nonmetros. These results suggest
that there are distinct underlying mechanisms driving negative trends in large metro
areas, where younger adults appear to be more vulnerable, and in small/medium metros
and nonmetros, where middle‐aged adults appear to be more vulnerable. In addition,
we find that mortality increases in the recent period at the younger working ages
(25–44) are greater for males than females in all residential categories, whereas
there is little sex difference in trends at ages 45–64. This pattern is potentially
related to the younger age profile of drug overdose mortality for men relative to
women (Ho 2019).
Deaths of despair and neurological disorders
Prior studies have documented adverse trends in mortality related to poisonings, liver
cirrhosis, and suicide within particular age groups (Case and Deaton 2015, 2017; Stein
et al. 2017). We find that among these causes, drug overdose makes the largest contribution
to trends in life expectancy between 1990–1992 and 2014–2016 in all metro categories
and all regions. Its contribution was not limited to only nonmetros or to Appalachia.
The impact of drug overdose has been greatest in large metro suburbs for men and additionally
in small/medium metros and nonmetros for women. We also find that regions have been
differentially impacted: drug overdose had the largest impact on trends in the Appalachian,
East South Central, and New England regions and the smallest impact in the Pacific
region.
In addition to drug overdoses, mental and nervous system disorders—including Alzheimer's
disease—have made negative contributions to trends in life expectancy between 1990–1992
and 2014–2016, particularly among women. Included among mental and nervous system
disorders are those associated with substance abuse (e.g., opioids, alcohol, cocaine,
and other substances), as well as psychological disorders such as depression and anxiety
disorders. The overall geographic pattern of the contribution of mental disorders
is similar to the pattern for drug overdose mortality, and many of these deaths may
also be associated with the recent opioid epidemic. For example, both mental disorders
and drug overdose made large contributions to reductions in life expectancy in large
metro suburbs in New England between 1990–1992 and 2014–2016. In the region/metro‐nonmetro
combinations where the contribution of mental disorders was high, the contribution
of Alzheimer's disease tended to be smaller. These patterns may, in part, reflect
regional differences in coding practices over time and substitution between cause
of death categories (Taylor et al. 2017). The increase in death rates from mental
and neurological disorders, including Alzheimer's disease, is not unique to the United
States, and has also been observed in several European countries (Mackenbach, Karanikolos,
and Looman 2014; Ho and Hendi 2018).
Smoking
If drug overdoses and mental and neurological disorders do not explain the widening
mortality gap between metro‐nonmetro categories across all regions, what other causes
or conditions might help explain these growing disparities? Studies estimating the
impact of smoking on state‐level and regional differences in mortality indicate that
health behaviors likely play an important role. Fenelon and Preston (2012) estimated
that in 2004, the fraction of deaths attributable to smoking at ages 50 and older
in US states ranged from 11 to 30 percent among men and from 7 to 23 percent among
women. Smoking has also been identified as an important contributor to the low US
life expectancy ranking among developed countries (Preston, Glei, and Wilmoth 2010)
and at least some of the poor US performance internationally is likely related to
the geographic variation in smoking‐related mortality within the United States. Between
1965 and 2004, smoking explained a large fraction (up to 70 percent) of the growing
gap in male mortality between the worst‐performing Census division, East South Central,
and other Census divisions, and around 50 percent of the growing gap in female mortality
(Fenelon 2013). Given the higher prevalence of smoking in nonmetro areas, we anticipated
that smoking would account for a sizable fraction of the growing metro‐nonmetro gap
in life expectancy (Fenelon and Preston 2012; Fenelon 2013; Roberts et al. 2016).
The two causes of death most closely tied to smoking are lung cancer and respiratory
diseases. Together, these causes accounted for 0.78 year of the 5.09‐year gain in
life expectancy among white men in large central metros compared to 0.43 years of
the 2.25‐year gain in nonmetro areas, resulting in a difference of 0.35 year between
these two areas (Table 5). This same pattern was the most evident in the midwestern
states, the South, and Appalachia. Among white women, these two smoking‐related causes
made a smaller contribution to the increase in life expectancy in large central metros
(0.17 year) and contributed to life expectancy reductions in large metro suburbs (–0.02
year), small/medium metros (–0.19 year), and nonmetro areas (–0.52 year). These patterns
were evident among women in most regions. Thus, smoking‐related causes likely played
a key role in the widening of metro‐nonmetro inequalities in life expectancy at birth,
especially among women (Preston and Wang 2006).
Obesity
A number of studies have documented that sedentary lifestyle, poor diet, and obesity
are more common in nonmetro than in metro areas (Eberhardt and Pamuk 2004; Michimi
and Wimberly 2010; Befort, Nazir, and Perri, 2012; Meit et al. 2014; Wen et al. 2018).
Diabetes is the cause of death most closely tied to obesity. Trends in diabetes mortality
make a small impact on the trends that we observe, but this may be a result of underreporting
of diabetes on death certificates (Saydah et al. 2004). In contrast, cardiovascular
disease, which is tied to both health behaviors such as obesity and smoking as well
as to medical care (Ford et al. 2007), was the cause of death responsible for the
largest improvements in life expectancy in all areas and for most of the variation
in life expectancy trends across the 40 geographic areas between 1990–1992 and 2014–2016.
In Figures 6A and 6B, the change in life expectancy between 1990–1992 and 2014–2016
is graphed against the change in percent obese between 2004 and 2013 in our 40 areas.
The correlation between increases in life expectancy and change in percent obese is
–0.74 for both men and women. These figures suggest that obesity, a risk factor for
cardiovascular and other chronic diseases, quite possibly plays some role in the trends
we observe. Changes in obesity prevalence have greater correlation with geographic
changes in life expectancy than any other variable that we examine.
Figure 6A
Change in non‐Hispanic white male life expectancy at birth between 1990–1992 and 2014–2016
and change in select area‐level characteristics between 1990 and 2015, (% obeses 2004–2013)
Figure 6B
Change in non‐Hispanic white female life expectancy at birth between 1990–1992 and
2014–2016 and change in select area‐level characteristics between 1990 and 2015, (%
obeses 2004–2013)
Health care
Other causes of death influenced by access to and quality of health care include screenable
cancers, influenza and pneumonia, and HIV/AIDS. These causes made larger contributions
to life expectancy increases in large central metros than in nonmetro areas. This
was especially true for HIV/AIDS among white men, which is partly attributable to
the fact that this cause of death was much more prevalent in large central metros
than in nonmetros. HIV/AIDS made the largest contribution to life expectancy improvements
among men in large central metros in the Middle Atlantic, South Atlantic, and Pacific
regions.
Historically, nonmetro areas have suffered from shortages of both health care personnel
and facilities (Rosenblatt and Hart 2000). These shortages stem from rural areas’
difficulty in recruiting and retaining high‐quality medical personnel due to lower
wages, remoteness, greater population dispersion, poverty, and economic instability
(Rosenblatt 2004; Rosenblatt et al. 2006; Burrows, Ryung, and Hamann 2012). Rural
and urban areas also differ in terms of specialty mix: generalists provide much of
the care in rural areas, whereas specialty care dominates in urban areas (Rosenblatt
2004; Reschosky and Staiti 2005). Shortages of mental health care facilities, substance
abuse treatment, and pharmaceutical services are particularly acute in rural areas,
and may be one factor contributing to the trends in mortality from drug overdose and
mental disorders discussed above (Reschovsky and Staiti 2005; Burrows, Ryung, and
Hamann 2012). Finally, rural residents face greater transportation barriers and may
have to travel great distances or have longer wait times to access health care (Burrows,
Ryung, and Hamann 2012; Meit et al. 2014). Together, this constellation of factors
may result in higher levels of unmet need for health care, lower quality health care,
later detection and poorer management of chronic disease, and lack of access to care
for acute conditions in nonmetro versus metro areas. Our cause of death analyses suggest
that these factors are likely to play some role. In Figures 6A and 6B, we have graphed
the change in life expectancy against the change in physician availability for the
40 areas. The increase in physician availability was greater in large central metros
than in nonmetro areas, although increases in life expectancy were only moderately
positively correlated with changes in physician availability (0.28 for white men and
0.42 for white women). Nonmetro areas had the fewest physicians per 1,000 residents
at the beginning and end of the period.
Poverty
It is also possible that differences in life expectancy trends across metro‐nonmetro
categories and regions are related to the changing characteristics of these areas
between 1990 and 2016. For example, it is possible that life expectancy trends are
related to variation in levels of or trends in poverty across the 40 geographic areas.
The rural mortality disadvantage is sometimes attributed to higher levels of poverty
in nonmetro areas (e.g., Stein et al. 2017). However, the data do not support the
common narrative of rural economic disadvantage when measured by the level of poverty.
According to a recent analysis by the US Census Bureau, poverty rates during 2011–2015
were consistently lower for those living in rural areas than for those living in urban
areas in all Census regions, with the largest differences observed in the Midwest
and Northeast. In 32 states, median household income was higher for rural households
than for urban households (Bishaw and Posey 2016). The poverty advantage for rural
areas is a product of adjustments for geographic differences in housing costs, the
receipt of noncash benefits, and taxes paid; without these sensible adjustments, rural
areas have slightly higher levels of poverty (United States Department of Agriculture
Economic Research Service 2017).
Nolan, Waldfogel, and Wimer (2017) extend the analysis of rural/urban differences
in poverty back to 1967 and adjust for real differences in cost of living and for
taxes paid and transfers received. They find that in 1967, more than 30 percent of
rural residents were living in poverty compared to 20 percent of urban residents.
However, the rural poverty level fell much more quickly and by about 1992 was equal
to that of urban residents. From 1992 onwards, rural poverty continued to fall more
rapidly and was 3 percentage points below urban poverty levels by 2015. Both the cost
of living adjustment and the introduction of taxes and transfers were instrumental
in driving the relative improvements in rural poverty. A higher proportion of rural
residents received assistance from social security, food stamps, and heating subsidies.4
The implications for interpreting the results of our analysis are clear: the deterioration
of nonmetropolitan mortality relative to large central metros cannot be attributed
to worsening absolute or relative poverty trends in rural areas. On the contrary,
trends in poverty since 1990 should have worked to reduce metro‐nonmetro differences
in life expectancy.
That rural areas have benefited from government transfers and receive a higher proportion
of public assistance can also be an indicator of deteriorating labor market conditions
and increased dependence on government subsidies. In Figures 6A and 6B, we have graphed
the changes in life expectancy in the 40 areas against the change in the percentage
of total personal income that is composed of government transfers (i.e., retirement
and disability insurance benefits, medical benefits, income maintenance benefits,
unemployment compensation, veterans’ benefits, and education and training assistance)
and receipts of benefits from nonprofit institutions (Bureau of Economic Analysis
2018). The two variables are highly negatively correlated (–0.57 for men and –0.60
for women). Increases in life expectancy were smaller in areas where dependence on
government transfers grew the most between 1990 and 2015.
Educational attainment
Another compositional change that is related to both mortality and changing characteristics
of the population is educational attainment—one of the strongest predictors of mortality
at the individual level (Elo 2009; Hendi 2015, 2017; Montez, Zajacova, and Hayward
2016b; Sheehan, Montez, and Sasson 2018). Both metro and nonmetro areas witnessed
increases in the educational attainments of their residents between 1990 and 2015.
However, these changes have been more favorable in metro than nonmetro areas. Given
the steep educational gradient in mortality and its steepening, especially for white
women, over the last decades (Hendi 2017), the increase in the proportion of college
graduates in an area should translate into a decline in area‐level mortality over
time. The change in the proportion college educated was highly correlated with the
change in life expectancy in our 40 areas (0.52 for men and 0.58 for women, Figures 6A
and 6B).
Migration
Finally, we examined the relationship between life expectancy and migration of the
working‐age population. The migration rate of the working‐age population is a useful
proxy for the economic conditions and desirability of living in an area. If working‐aged
people are leaving an area, it may be in response to a lack of job opportunities.
If working‐aged people are moving to an area, it may indicate that that place provides
greater opportunity (Castles, de Haas, and Miller 2014). We use variable‐r methods
to compute net in‐migration rates by sex for people aged 22.5–62.5 years in each of
our 40 areas in 1990–1992 and 2014–2016 (see Appendix for further detail). We find
that in‐migration of the working‐age population is strongly and positively correlated
with life expectancy gains (0.44 for women and 0.58 for men). Large central metros
and suburbs have seen an increase in the rate of in‐migration for working age people
and have also seen the largest gains in life expectancy. Small and medium metros and
nonmetros, on the other hand, have mostly experienced declines in net in‐migration
of the working‐age population and have seen much smaller life expectancy gains.
Conclusions
Strong and steady improvements in life expectancy have been observed in industrialized
countries for the better part of a century (Oeppen and Vaupel 2002). This salutary
pattern has begun to fade in the United States in recent decades (Crimmins et al.
2011; Institute of Medicine and National Research Council 2013). In this article,
we identify the geographic areas in the United States where the trends in mortality
since 1990, and especially since 2010, are most problematic. We find that improvements
in life expectancy have been slowest in nonmetropolitan areas and that actual declines
have been observed there in recent years. People living in large central metros, on
the other hand, have experienced rapid improvements in life expectancy, producing
a striking metro‐nonmetro divergence.
We have documented this divergence for white men and women by sex, age, cause of death,
and geographic region. Our analysis provides some clues about the principal causal
processes at work, but it is hardly definitive. Screenable cancers, HIV/AIDS, and
influenza/pneumonia helped to widen the gap and suggest the possibility of a role
for the quality of medical services. On the other hand, our study shows warning signs
about rising mortality resulting from factors less clearly tied to the health care
system. In particular, younger adults aged 25–44 have experienced rapid increases
in mortality and are now contributing to declines in life expectancy across all metro‐nonmetro
categories. This rise in young adult mortality is largely attributable to drug overdose.
We have presented suggestive evidence that changes in smoking and obesity are also
implicated in the divergence. Metropolitan areas have benefited from a more rapid
increase in the population with college degrees, a group with below‐average mortality.
Poverty trends are more advantageous in nonmetropolitan areas, but part of the reason
is that these areas are receiving larger government benefits associated with labor
force inactivity. There are many layers to the mosaic of metropolitan‐nonmetropolitan
mortality patterns. Continued efforts to identify the main factors at work are clearly
justified by the massive divergence in the length of life that we have documented.
Supporting information
APPENDIX
Click here for additional data file.