This is an interactive tool to help you visually explore the geographic
associations between disparities in socio-economic opportunity and perinatal
outcomes, such as preterm birth.
How this tool can help
This tool can help inform you about geographic patterns of maternal and
infant risk and resilience, racial and ethnic disparities, and the
geographic patterns of community-based socio-economic factors. The tool
is designed to offer a new way of seeing perinatal risk.
We hope it will spark conversation, communicate new insight, lead to new
questions and catalyze new opportunities for local collaboration and
action to improve the health of women and infants.
Features of the tool
The Birth Equity Locator has two types of data indicators for mapping:
Perinatal health indicators – These measures of the perinatal health of local populations
are available for every county in the U.S., and are derived from a statistical modeling summary
of vital statistics designed to produce reliable estimates even for sparsely populated counties.
These indicators of perinatal health include live birth prevalence (‘risk’) of very preterm
(<32 weeks), early preterm (<34 weeks), late preterm (34-36 weeks), preterm (<37 weeks) and
early term (38-39 weeks) birth. In addition, there are several indicators of ‘racial disparity’
in each of these outcomes, because sometimes the geographic patterns of high perinatal risk are
not the same as the geographic patterns of large racial inequities.
Community contextual variables – These measures of the health care, social and
economic context of each U.S. county are provided to show how places are different
from one another. Indicators in this category are wide reaching, ranging from the
concentration of obstetricians and location of Maternity Care Deserts, to the rate
of poverty, violent crime and housing instability
Get started now
Follow our brief tutorial that explains how to use and interpret the data as well as how
to use the tool itself. Learn more about the perinatal indicators and how they were
estimated by reading the technical documentation.
Decades of research has pointed to the critical importance of the places where we live,
work and play as contextual regulators for both the opportunities that help make us
healthier (e.g. access to health services, neighborhoods free from violence, and economic
opportunity), and many of the exposures that make us less healthy (e.g. violent crime,
neighborhood deprivation, low social support and poor housing). While many other factors
such as medical risk, individual behaviors and genetic predisposition are also important
factors in why one woman has a healthy pregnancy and another woman does not, it is also
clear that individual behaviors and genetics do not explain why some groups of women have
higher average risk for poor outcomes than other groups of women.
Examining the geographic variation in perinatal risk can shed new light on complex problems, and
bring new partners to the table in discussions about improving perinatal outcomes. This is true for
at least two reasons. First, places such as counties or county-equivalents represent an environment
to which women and families are exposed, and which may influence health and perinatal risk. Second,
counties units are one potential target for local action to improve health and health equity.
We can see that ‘place matters’ when we look at how risk for poor outcomes varies
dramatically between counties (hyperlink to separate pages describing the modeling procedure
for perinatal outcomes). In the map below, it is clear that the risk for preterm birth varies more than
3-fold across U.S. counties, with clusters of higher risk counties in the South and clusters of
lower risk counties in the Northeast, Plains and Western states. Even in states with “lower”
than average risk, there are still differences between counties. Why do these differences exist?
In some instances it may be because of demographic characteristics of counties, and in others it may
be about the social, economic, and health service characteristics of counties.
Figure 1: Preterm birth estimates in U.S. counties, 2017
What do the colors on the map mean? All of the maps in the
Birth Equity Locator show the average risk (e.g. estimated percent of
births born preterm) for
perinatal outcomes in each county, with map colors
ranging from light blues representing lower risk, to deep purple representing
higher risk. To assign a map color to each county, the counties are ranked from
lowest to highest, and then categorized into equal-sized groups, each with an
equal number of counties (‘quantiles’). This makes it easy to see the relative
ordering of counties, but it does not always tell us how big the absolute
differences are between color categories. To see the absolute risk associated
with each color, look at the map legend.
Place also matters for community-level risk factors (e.g. access to health
services or poverty rate) for poor health outcomes. In this map, counties without
access to maternity health care services (‘Maternity Care Deserts’) are evident
in nearly every state:
Figure 2: Access to maternity care in U.S. counties, 2016
The Birth Equity Locator includes dozens of community-level indicators of
risk and resilience that are social, economic, and health service related. Each
of them were selected because research suggests they may be related to between-county
differences in risk for poor perinatal outcomes, and because they vary geographically.
While the Birth Equity Locator can only show correlation, but does not prove
causation, the variables selected are nonetheless useful and important measures
for asking better questions and engaging local partners in better understanding
the local drivers of health and equity.
Why outcomes matter
Preterm birth is not just one thing, but is really the outcome from a mix of many
different disease processes (it is a ‘heterogeneous outcome’)! What this means is that
the causes (and solutions) are not the same for every baby born preterm. It is difficult
to account for all of these differences using only birth certificate data. But one way
to begin to see differences is by looking at the severity of prematurity. For instance
babies born at <32 weeks gestation (sometimes called ‘very preterm birth’) have much
higher risk for morbidity and mortality than babies born just one week before 37 weeks
of gestation (more generally called ‘preterm birth’). Not only are the consequences
different, but in some instances the causes or reasons for prematurity may be different.
For instance, it is very unlikely that any baby would be intentionally delivered (e.g.
labor induction or cesarean section) before 32 weeks except in the most serious of
situations. But it is much more common for obstetric decision making about labor
induction or cesarean section to play a role in late preterm births. Therefore the
underlying drivers of each may be different (e.g. maternal health, inflammation or chronic stress
may play a relatively larger role in the case of very preterm birth; obstetric management
and health care practices playing a larger role in preterm birth).
In the Birth Equity Locator we provide indicators for the live birth prevalence
(‘risk’) of overall preterm birth (<37 weeks gestation) but also several other
subsets including the proportion of babies born <32 weeks (very preterm birth); <34 weeks (early preterm birth);
34-36 weeks (late preterm birth); and, recognizing recent evidence of differences among
term babies, even 37-39 weeks (early term).
Are different perinatal outcomes more or less concentrated in different geographic areas?
In the plot below we show how the risk for preterm birth (<37 weeks) correlates with
the risk for very preterm birth (<32 weeks) in the very same counties. The
absolute risk is different between these two outcomes, but we do see there is a correlation.
A correlation means that as the risk for preterm birth goes up in a county, we tend to see
the risk of the very preterm birth go up as well. The steeper the line in the plot, the
stronger the correlation. But one thing that is notable is that the two outcomes are not
perfectly correlated. In other words, some counties have relatively higher risk
for preterm birth but relatively lower risk for very preterm birth. This pattern is
especially evident for Black mothers.
Figure 3: Association of perinatal outcomes within U.S. counties, 2017
These differences can also be seen on maps. In the two maps below of the risk
for very preterm birth and preterm birth among non-Hispanic
Black women, we can see that counties in Tennessee, Kentucky, and Ohio have
relatively lower Preterm Birth, but relatively higher very
preterm birth. The opposite pattern is evident in southeastern Texas counties.
These differences may point to different risk factors in these counties.
How can we compare the colors in two maps side-by-side? As discussed
above, the Birth Equity Locator uses a mapping rule that ranks all
counties in a single map from the lowest to the highest risk, and then groups
them into categories with equal numbers of counties in each category (‘quantiles’).
The colors are assigned
to these categories from the lowest risk (lighter blue) to highest risk
(deeper purple). If two maps are side by side, the colors in one map do not
necessarily reflect the same absolute risk as that color in another map.
It is always important to look at the map legend to understand what the colors
Figure 4: Comparing geographic patterns of different perinatal outcomes, Black women, 2017
Why race matters
The notion that ‘place matters’ when it comes to health suggests that some places are more or less
healthy for everyone in the county. Many aspects of community context likely do affect
everyone living in that community. For example, counties with no maternity health care
services means lower access for all. However, because of the history of slavery, Jim Crow
laws, economic inequality, racial residential segregation, and other processes of
structural racism and inequality, there can be differences in where Black and White
families live, but also differences in how Black and White families experience even the
same places in which they live.
For example, there can be place-based racial disparities in educational and economic
opportunity, racial differences in access to and quality of health care, and differences
in experience of chronic stressors known to be risk factors for preterm birth. These
differences can produce racial differences in the geographic patterns of relatively
higher or relatively lower risk for perinatal outcomes.
There may be distinct geographic patterns of perinatal outcomes for the total population,
and separate patterns for groups defined by race/ethnicity, or even when viewing measures of
health equity between race groups.
For example, below are maps of preterm birth for Black and white women. In general the
absolute risk of preterm birth is higher for Black women than it is for white women.
However, the regions where white women have relatively higher or lower risk are not
always the same as the regions where Black women have relatively higher or lower risk.
Figure 5: Comparing geographic patterns in preterm birth by maternal race, 2017
Why measures matter
Numbers don’t lie, but a single number doesn’t always tell the whole
truth. This is the case for measures of health disparity. Sometimes
differences in health between groups exist due to innate biological
differences, such as the case of sex differences in breast cancer.
However, when differences in health among population groups are not
due to innate biology, we consider the difference not only unequal
but also inequitable (e.g. unjust).
In the United States, on average Black women and infants experience
excess burden of poor perinatal outcomes as compared to non-Hispanic
white women. As discussed above, there is little evidence that genetics
can explain these differences. Instead, many researchers posit that the
reason for the persistence of racial inequity in perinatal outcomes is
due to historical inequality by race and socioeconomic opportunity,
factors that not only affect pregnancy, but can influence girls and
women’s health across their life and even across generations.
There are two common calculations that can be used to describe the
severity or magnitude of health disparities: the risk ratio and
the risk difference.
What is a Risk Ratio? When the risk in one group is divided
by the risk in a comparison group, we call the measure a risk ratio.
If the risk is the same in both groups then the risk ratio will be 1.0.
However, when the risk in the first group is larger than the risk in the
comparison group, the risk ratio will be greater than 1.0 and indicates a
health disparity. For example if the risk in one group is 15% and the
risk in the comparison group is 10% then the risk ratio is 15/10 = 1.5.
This means that the first group has 50% greater relative risk than the
comparison group. A risk ratio describes relative or proportionate
differences between groups.
What is a Risk Difference? Instead of dividing the risk in one
group by the other, a risk difference subtracts the risk in one
group from another (e.g. takes the absolute arithmetic difference). If
the two groups have the same risk, then the risk difference will be zero.
On the other hand, if one group had a risk of 15% and the comparison
group had a risk of 10%, then the risk difference would be 5%. This
means that for every 100 births to the each group, the first group
would experience 5 extra negative perinatal outcomes than the
comparison group. A risk difference describes the absolute difference
In the figure below, we plot the Black-white risk ratio for U.S.
counties against the Black-white risk difference for the very same
counties. Clearly the two values are generally correlated (go up together).
But looking at counties along the purple line (representing a relative three-fold higher
risk of very preterm birth among Black women as compared to white women), we can
see that there are a range of Black-white risk differences. The colors
in this plot symbolize how the risk for very preterm birth in each county
among the lower risk group (e.g. non-Hispanic white women) varies. In counties where
White women have low risk (colored blue), and a Black-white risk ratio
of 3.0, the Black-white risk difference is low (between 11 and 13 per
1,000 births). But if we look at counties where white women have higher risk
(colored red), and the very same Black-white risk ratio of 3.0 results in
a Black-white risk difference of 15-18.
Figure 6: Comparing racial disparities measured as 'risk ratios' and 'risk differences' for preterm birth, 2017
Why does this happen? One way to understand this is to imagine two different
counties. In County A, non-Hispanic White women have a 1% risk for very preterm birth
(<32 weeks), while non-Hispanic Black women have a 3% risk. In County B, non-Hispanic
white women have a risk of 2% and non-Hispanic Black women have a 4% risk. It is clear
that in each county Black women have a risk for very preterm birth that is 2% higher
in absolute terms. In other words, the Black-white disparity measured with an absolute
risk difference is 2 excess very preterm births per 100 births. However, if we measured
the disparity with the relative risk ratio, in County A, Black women have three-times
(risk ratio 3% / 1% = 3.0) the risk for very preterm birth. In County B, Black
women have two-times the risk (4% / 2% = 2.0).
Once again, these differences in how big the racial disparity is can be seen on maps.
Figure 7: Comparing geographic patterns of Black-white racial disparities in very preterm birth by choice of disparity measure, 2017
In this map it appears that the Black-white racial disparity in some deep south counties
is relatively larger when using the absolute risk difference measure, but
in those same counties the Black-White racial disparity is relatively smaller
when using the relative risk ratio measure. This is because the risk for very preterm
birth is higher for both Black and white women in deep southern counties as compared
to many counties in the Northeast.
What do these measures tell us? Both measures are mathematically correct. However,
they tell slightly different stories about where disparities in perinatal outcomes are larger
or smaller. Making maps of health equity can be a useful complement to maps of only risk in
the whole population or for individual racial or ethnic groups. However, it is important to
consider whether a large or small disparity is because both groups are doing well or both
are doing poorly.
Details about the data
Restricted access data from NCHS Natality files from 2007 through 2017
were collected under a Data Use Agreement.
Gestational age categories were determined from the best
obstetric/clinical estimate when available.
Data preparation proceeded through several steps:
Exclude births with birthweight that is implausible for a live birth
(e.g. less than 500 grams). This excluded between 0.14% and 0.16% of
records per year
Exclude births with gestational age estimates that were missing or
implausible for a live birth (e.g. <21 weeks). This excluded between
0.14% and 0.23% of records per year
Include only singleton pregnancies, because multiple gestations is
an independent risk factor for preterm birth. This excluded 3.3% to
3.5% of records per year.
Include only births to mothers self-reporting as non-Hispanic white,
non-Hispanic Black, or Hispanic. This excluded between 7.3% and 8.7%
of records per year. There are other race/ethnic groups in the U.S.,
but these three groups represent the largest and most geographically
diverse. Inclusion of these groups draws attention to possible
geographic patterns of risk within race/ethnicity but fails to highlight
important disparities for other groups including American Indian/Alaska
The count of live births, and the count of births in each of the
preterm categories were aggregated separately for every maternal
race/ethnic group, year of birth, and U.S. county.
Perinatal Variables: Data Sources and Estimation
What's behind the modeled data.
Indicators of perinatal health include live birth prevalence (‘risk’) of:
Why are the perinatal indicators in the Birth Equity Locator only estimates rather than the actual risk?
It is true that birth certificate information on nearly every live birth in
the United States is collected by the National Center for Health Statistics (NCHS).
These records include the gestational age and weight of the infant at birth,
information about the mother and the county in which she lived at the time of
birth. So wouldn’t it be better to just map the “actual” risk
rather than an estimate? We believe the answer is ‘no’ for two reasons.
The first is that the NCHS does not publicly distribute the county of residence
for vital records to protect the privacy of individual women and small demographic
sub-groups. These data are available under a special Data Use Agreement with NCHS,
but only for specific, targeted use by individuals who are part of the Agreement.
The second reason is that, even if we could share the actual count of preterm
births for every county, the statistics would be considered unreliable
or unstable for many counties where there are not very many births
to the specific race/ethnic group being mapped. Imagine there are two counties, each
measured for three separate years. The number of live births stayed the same in each
county for all three years. However each year, perhaps due to random chance, there
was one additional baby born preterm than the previous year. In County B, the risk
of preterm birth is very stable around 10 percent. In other words the
relatively random fluctuation of one or two more or fewer preterm babies did not
change the estimate of risk very much. In contrast, in County A the risk ranged from
0 percent to 20 percent, both very extreme values for preterm birth. This is because there are not
many births at risk in County A. When there are a very small number of births or
very few poor outcomes (e.g. sparse data), the calculated risk can be statistically
unreliable or unstable, and can easily take on unreasonably extreme values.
What does it mean to estimate risk and how was it done here?
Risk estimates are an effort to produce the most reliable prediction of the
underlying ‘true’ risk of poor perinatal outcomes experienced by a group of women
(e.g. in a specific race/ethnic group in a specific county). To the extent that
the methods and approach used to calculate the estimate are valid and robust, the
resulting estimate is more useful for comparing across counties or across years.
In other words, the estimation process aims to describe what is really different
about risk without too many extreme swings that are due only to sparse data and random
The approach to perinatal risk estimation in the Birth Equity Locator
uses sophisticated spatio-temporal statistical models called Bayesian disease mapping.
The more technical details of the modeling procedure are described below. However in
broad terms, it is well known that, on average, the risk in one year is more correlated
with the risk in the previous year than it is with risk much longer ago. Similarly, on
average, the risk in one county is more similar to the risk in nearby counties than it
is to a randomly selected, distant county. Of course there are exceptions to both
situations. However the Bayesian disease mapping approach is a way to stabilize risk
estimates from sparse populations (counties or race/ethnic groups) by ‘borrowing
statistical information’ about risk from the previous year(s) and from the nearby
counties. This approach has been widely used by the Centers for Disease Control
and Prevention, NCHS and the World Health Organization to develop statistically
reliable maps of geographic patterns of population health.
Why are there ‘insufficient data’ for many counties?
The Bayesian statistical
model makes an estimate of risk for every racial/ethnic group and every county
by ‘borrowing statistical information’ from nearby counties and over time.
However, because of the geographic settlement patterns of different
racial/ethnic groups across the U.S., there are some counties with very
few or even zero women of color. Therefore we chose to suppress the Bayesian
statistical estimates in those counties. Specifically, we suppressed data for
all counties where there were fewer than 150 births to women of a given
racial/ethnic group summed across the years 2005 to 2018.
Contextual Variables: Definitions and Sources
What's behind the contextual variables data.
Contextual variables were selected to meet each of several criteria:
Indicators represent a range of factors that have been hypothesized in
peer-reviewed published literature
to be predictors of maternal and child health and perinatal health
Indicators must have quantitative measures available for all U.S.
counties during the period 2007 to 2017.
Indicators are publicly available for use.
There are important social and contextual determinants of health
which are not included because they do not meet the above criteria.
For example differences in implicit bias experienced by women in health
care settings may be important, but are not easily quantified at the
county level across all U.S. counties.
We aim to represent indicators measured for each year from 2007-2017.
However, several data sources are only available for a subset of the
years. When maps represent contextual indicators that are from a
different year than the perinatal indicator, the appropriate dates
are reported in the figure legend.
Contextual indicators come from multiple data sources. Several are
derived from the American Community Survey of the U.S. Census Bureau
5-year moving window summaries. For example the 2005-2009 ACS estimates
were used to represent the social and economic context for 2007; the
2006-2010 ACS estimates were used to represent the context for 2008.
This is illustrated in the figure below.
Figure 1: Mapping of ACS 5-year summaries to specific years in Birth Equity Locator
The Perinatal Birth Equity Locator project was created and developed by:
Michael R. Kramer (Principal Investigator), Rollins School of Public
Health at Emory University in Atlanta
Erin Stearns (Spatial epidemiologist and Tool developer) of EpiMap
Kevin Weiss (Modeler), Rollins School of Public Health at Emory University
The project was supported by a grant from the March of Dimes Center
for Social Science Research and EMD Serono, Inc., the biopharmaceutical
business of Merck KGaA, Darmstadt, Germany in the U.S. and Canada.
Find patterns. Make data-based connections. Inform action.
Disclaimer: Correlation is not causation. All perinatal indicators are statistical estimates of local population risk.