Rural–urban disparities in pregestational and gestational diabetes in pregnancy: Serial, cross‐sectional analysis of over 12 million pregnancies

To compare trends in pregestational (DM) and gestational diabetes (GDM) in pregnancy in rural and urban areas in the USA, because pregnant women living in rural areas face unique challenges that contribute to rural–urban disparities in adverse pregnancy outcomes.


| I N TRODUC TION
The rate of diabetes in pregnancy, including both pregestational (types 1 and 2) diabetes mellitus (DM) and gestational diabetes (GDM), has continued to increase in the USA. 1,2In 2020, DM and GDM occurred in 1% and 8% of all pregnant women, 2 respectively.For GDM, this frequency has increased 30% since 2016. 3,4Reasons for the increasing frequency of diabetes in pregnancy have been attributed to multiple factors, including increases in older maternal age, 5 obesity, 6 and adverse lifestyle factors, such as worse diet and less physical activity. 7,8Diabetes during pregnancy increases the risk of severe maternal morbidity and mortality, 9 and adverse maternal and infant health outcomes. 10There are persistent disparities in these outcomes by maternal race and ethnicity (due to adverse social determinants of health), 11 age, 12 and US region. 13regnant women living in rural areas in the USA face unique challenges that contribute to rural-urban disparities in adverse pregnancy outcomes. 14Access to prenatal care, which includes regular check-ups, generally with a physician or midwife provider, to treat and prevent health problems throughout pregnancy, in rural communities is rapidly declining, and 80% of maternity care deserts are currently located in rural counties. 15,16Pregnant women in rural communities are also more likely to lack adequate health insurance coverage or be publicly insured by Medicaid and experience a higher burden of adverse social determinants of health. 17Increasing population-level data in the USA demonstrate that rising rates of severe maternal morbidity and mortality, 18 hypertensive disorders of pregnancy, 19 and chronic hypertension, 20 have been of greater magnitude in rural compared with urban communities.Similarly, outside the USA, rural-urban disparities in diabetes care and adverse outcomes in pregnancy exist in other high-income countries, including Canada and Australia, where nearly one-fifth of individuals reside in rural communities. 21,22here remains limited information about rural-urban disparities in diabetes in pregnancy over time using population-level US data.Whether rural residence may intersect with other disparities, inclusive of racial and ethnic and geographic identities, to reinforce and even increase persistent disparities in diabetes in pregnancy remains to be investigated. 11he objective of this analysis was to examine trends in the frequency of DM and GDM among pregnant women in rural and urban areas of the USA from 2011 to 2019.
Because rural-urban differences in maternal metabolic health and risk factors for diabetes exist, 23,24 we hypothesised that the rate of DM and GDM would differ between rural and urban communities, and that this disparity would change over time and be modified by maternal age, reported racial and ethnic identity, and US region of residence.

| Study design and data collection
We performed a serial, cross-sectional descriptive analysis of all nulliparous women aged 15-44 years who delivered singleton live births in the USA from 2011 to 2019.We excluded those who were parous, or who had non-singleton deliveries (e.g.twins) to avoid duplication of maternal data (because records could not be linked to the same individual in the de-identified data set), and who were missing information on their diabetes status.For the final study population, we modelled the frequency of DM and GDM separately.For DM, we compared those with versus those without this diagnosis; and for GDM, we compared those with versus those without this diagnosis among those without DM (Figure 1).
We used data from the Centers for Disease Control and Prevention National Center for Health Statistics (NCHS) Natality Files, which included maternal demographics, medical history and obstetric complications.Birth certificate data were collected using standardised maternal and facility worksheets. 25NCHS provided guidance to birth facilities with regards to data abstraction and documentation. 25Per NCHS guidance, an individual's diabetes status was ascertained hierarchically per the following criteria: prenatal care record, nursing notes from the delivery admission, delivery admission history and physical examination and the labour and delivery summary.This study was deemed exempt by the Northwestern University Feinberg School of Medicine Institutional Review Board Review given that data were deidentified and publicly available.Patients were not involved in the development of the research.

| Exposure
Rural/urban designations were based on the 2013 NCHS Urban-Rural Classification Scheme for Counties based on the pregnant woman's residence recorded in the birth ethnicity and in women who lived in the South.These findings have implications for delivering equitable diabetes care in pregnancy in rural US communities.

K E Y W O R D S
diabetes mellitus, disparities, gestational diabetes mellitus, pregnancy, rural health, urban health certificate. 26The levels of the NCHS scheme were chosen for their utility in studying health differences across the urbanrural continuum.A rural area was defined as a county in a metropolitan statistical area with a population less than 50 000 individuals, and an urban area as a county in a metropolitan statistical area with a population of more than 50 000 individuals.We secondarily assessed the six-level NCHS urban-rural classification scheme for US counties, which includes two subgroups for rural areas (micropolitan and non-core) and four subgroups for urban areas (large metropolitan, large fringe/suburban, medium metropolitan, small metropolitan).

| Outcomes
The primary outcomes were diagnoses of DM and GDM, assessed separately. 1 GDM, per the NCHS protocol, was defined as 'diabetes that was diagnosed during pregnancy'.In this data set, DM and GDM were mutually exclusive, so that individuals could not be categorised as having both conditions.Those completing the birth certificate were advised to select either DM or GDM, but not both. 25The validity of diagnoses of DM (sensitivity: 46%-82%; specificity: >98%) and GDM (sensitivity: 83%; specificity: 99%) in the NCHS Natality Files have been reported elsewhere. 27,28Diagnostic criteria including testing methods and diagnostic thresholds were not available from US birth certificate data.This study did not include a core outcome set because the primary outcome was a diagnosis of DM and GDM, and not adverse outcomes due to diabetes in pregnancy. 29

| Covariates
Additional assessed covariates included demographic characteristics, such as age (years, categorised to 15-19, 20-24, 25-29, 30-34, 35-39 and 40-44 years), race and ethnicity categories (Hispanic, non-Hispanic Asian-Pacific Islander, non-Hispanic American Indian, non-Hispanic Black and non-Hispanic White), timing of prenatal care initiation (trimester) and US census region (Midwest, Northeast, West and South).Race and ethnicity were abstracted from the maternal worksheet based on patient self-identification, and categorised to be consistent with recent analyses from this data set evaluating diabetes in pregnancy. 1,11,30We used the terms 'race' and 'ethnicity' recognising these terms as a social rather than biological construct.We used bridged-race categories due to differences in state-level categorisation to permit estimation and comparison of race-specific statistics over time, and to facilitate comparison with previous analyses from this data set. 31US census region was defined per maternal residence on the birth certificate.

| Statistical analysis
We calculated the frequency of both DM and GDM per 1000 live births by urbanisation status (rural or urban), and then secondarily, within each rural and urban subgroup.To analyse trends in the rate of DM and GDM from 2011 to 2019, we used Joinpoint statistical software to model consecutive linear segments on a log scale to identify points in time (joinpoints) at which there were statistically significant changes in trends over time.We calculated the mean or average APC stratified by urbanisation status.The mean APC is a rate measured in units of 100 personyears and provides a relative percentage change per year. 32hese trends were standardised to the age distribution of all deliveries in 2011 (first year of the study sample) to allow for direct comparisons across years.To compare the frequency of DM and GDM between rural and urban areas, we calculated unadjusted and age-adjusted rate ratios (RR, aRR) and their corresponding 95% confidence intervals (CI).To assess whether the associations between rural versus urban residence and DM and GDM varied by subgroups (effect modification), including year of delivery as a continuous variable from 2011 to 2019, race and ethnicity, and US census region, we calculated the interaction term (rural versus urban × subgroup category versus reference) as modelled in the multivariable regression models above.Missing data for covariates were represented in the model with a categorical-variable term given the low frequency of missing values, rather than as imputed values.All statistical tests were two-sided and statistical significance was assessed based on a 95% CI excluding the null value.Analyses were conducted with Stata version 17 (Stata Corp, Cary, NC, USA)and Joinpoint Regression version 4.9 available from the National Cancer Institute (Surveillance, Epidemiology, and End Results Online Age Period Cohort Analysis Tool). 33

| Study population
Among 35 043 649 live births to individuals aged 15-44 years in the USA between 2011 to 2020, we excluded parous individuals and those with nonsingleton births, resulting in 13 258 039 live births (Figure 1).For the DM analysis, we excluded those missing data on a DM diagnosis, resulting in a final analytical sample of 12 493 134 live births; and for the GDM analysis, we excluded those missing data on a DM (n = 91 246) or GDM (n = 764 905) diagnosis, resulting in a final analytical sample of 12 401 888 live births.

| Overall trends in the frequency of DM and GDM
The overall age-standardised frequency of DM increased from 6.3 to 8.6 per 1000 live births between 2011 and 2019, and GDM increased from 40.8 to 60.9 per 1000 live births.The frequency of both DM and GDM rose for individuals living in both rural and urban areas.Specifically, the agestandardised frequency of DM increased from 7.6 to 10.4 per 1000 live births for rural residents and from 6.1 to 8.4 per 1000 live births for urban residents; the age-standardised frequency of GDM increased from 41.4 to 58.7 per 1000 live births for rural residents and from 40.8 to 61.2 per 1000 live births for urban residents (Figure 2).

| Race and ethnicity-and US region-specific trends in the frequency of DM and GDM
The average APC of the frequency of DM from 2011 to 2019 increased in rural areas for Hispanic, White and American Indian individuals; and in urban areas for Hispanic, White, Black and Asian/Pacific Islander individuals (Table 2; Figure S2a).For GDM, the average APC increased for all groups in rural and urban areas, except for Black individuals in rural areas; the highest increase was for American Indian individuals in rural areas by 7.7% (95% CI 5.3%-10.0%)and urban areas by 9.7% (95% CI 7.2%-12.2%)with an inflection point in 2014 (Figure S2b).
By US region, the average APC of the frequency of DM in rural and urban areas increased in all four regions except rural areas in the Northeast; and the highest increase was in the West, including rural areas by 4.3% (95% CI 1.6%-7.1%)and urban areas by 5.7% (95% CI 4.3%-7.1%;Figure S3a).For GDM, the average APC increased in rural and urban areas across all four regions, and the highest increase was similarly in the West, including rural areas by 4.6% (95% CI 3.0%-6.1%)with an inflection point in 2016 and urban areas by 5.3% (95% CI 4.4%-6.1%;Figure S3b).

| Overall rural-urban disparities in the risk of DM and GDM
The overall risk of DM was higher for individuals living in rural versus urban areas (aRR 1.48; 95% CI 1.45%-1.51%)and this difference did not change over time (interaction p-value of year as a continuous variable × rural/urban = 0.8; Table 3).The overall risk of GDM also was also higher for individuals living in rural versus urban areas (aRR 1.17; 95% CI 1.16%-1.18%),and this difference increased over time (interaction p-value of year as a continuous variable × rural/urban < 0.01).Abbreviations: DM, pregestational; GDM, gestational diabetes.

| Main findings
The rate of both DM and GDM increased in rural and urban areas of the USA from 2011 to 2019 among nulliparous pregnant women.Moreover, there was a notable geographic disparity, with risks of DM and GDM significantly higher among pregnant women living in rural compared with urban areas.The magnitude of this rural-urban disparity remained similar over time for DM, but increased for GDM.The disparity also was greater for those identified as Hispanic, and those living in particular regions of the country, most notably the South (the one region which had wider rural-urban disparities for both DM and GDM).

| Strengths and limitations
Strengths of this study include, to our knowledge, that this is the first analysis to assess rural-urban disparities in the incidence of diabetes in pregnancy.This analysis included a contemporary and generalisable cohort of all nulliparous live births in the USA through 2019.There have been no major changes in US guidelines for diabetes screening in pregnancy during the study period. 34his study has several limitations.This analysis was restricted to individuals with a singleton live birth because multiparity and history of GDM are risk factors for both DM and recurrent GDM.De-identified NCHS data also do not allow for accounting for multiple deliveries to the same individual during the study period.It is possible the frequency of DM and GDM may have been underestimated, but this would probably have occurred in both rural and urban areas (i.e.nondifferential misclassification).The birth certificate includes both type 1 and type 2 diabetes as a single diagnosis of pregestational diabetes, so the relative contribution of each could not be assessed in this analysis.Miscoding of diabetes diagnoses is possible, which may be related to screening and prenatal care.However, birth certificate data are completed by a professional birth attendant and do not use administrative coding.Previous data suggest a high level of agreement between the birth certificate and the medical record. 35It is possible that the frequency of DM and GDM may have been underestimated in birth certificate data, 28 but the systematic reasons for this would likely have occurred in both rural and urban areas (i.e.nondifferential misclassification).Additional factors that can impact diabetes risk, such as diet and physical activity, 36 were not available in this national data set.

|
The persistent rural-urban disparity in the rate of DM and increasing disparity in the rate of GDM among pregnant women is consistent with emerging data with regards to rural-urban disparities in adverse pregnancy outcomes, severe maternal morbidity and mortality, as well as access to high-quality clinical and obstetric care services during pregnancy in the USA, 18,37 as well as other high-income countries, such as Canada and Australia. 21,22Whether the risk of adverse outcomes is higher among pregnant women with diabetes in rural versus urban areas requires further study.][40][41] Pregnant women living in rural communities are more likely to live in maternity care deserts, which are counties without a hospital or birth centre offering obstetric care and without any obstetric providers.Over 80% of these maternity care deserts are rural counties. 16Differences in GDM screening in rural communities in the USA are probably lower than in urban areas and require further study, data from other high-income countries including Australia suggest that GDM screening may be as low as 50% in rural communities. 42If there was a lower frequency of screening in rural areas, the actual disparity for GDM may be even higher than that documented in this analysis.
Taken together these results highlight the need for prenatal and diabetes care delivery interventions that address this rural-urban divide in access to care and adverse outcomes.
In the current study, the rural-urban disparity in diabetes in pregnancy was even wider for some subpopulations.The rural-urban disparity was wider for Hispanic individuals compared with White individuals for both GDM DM, albeit lower for Asian/Pacific Islander individuals for DM and American Indian, Asian/Pacific Islander and Black individuals for GDM.Also, the rural-urban disparity was higher in the Northeast, South and West compared with the Midwest for GDM, and for GDM, higher in the South, but lower in the West.Outside pregnancy the rural-urban disparity in DM has been shown to vary in some US regions, such as the South. 43The current study emphasises the importance of considering intersectional identities in order to determine within rural areas which pregnant women and which areas are at the greatest risk due to possibly converging social determinants of health. 44he current study highlights the potentially unique challenges faced by rural residents, including individual risk factors, adverse social determinants of health and inadequate health delivery system factors, 45 that impact diabetes risk in pregnancy.Pregnant women in rural areas are more likely have higher body mass index and more chronic comorbid conditions, including hypertension, compared with those in urban areas. 20Those in rural areas are also more likely to have lower educational attainment and inadequate insurance coverage. 15,17The provision of subspecialty diabetes focused care may not be available in many rural medical centres, which generally have a lower birth volume and fewer clinicians and resources for specialised obstetric care than urban hospitals. 46The uptake of evidence-based clinical care, technology and care bundles in pregnancy is also frequently slower in rural than in urban communities. 47Hence, rural pregnant women with diabetes may be more likely to require care from multiple providers or in different locations. 37hese findings emphasise the need for public health efforts that emphasise interventions aimed at preventing diabetes, and in particular GDM, as well as the provision of diabetes care in pregnancy in rural communities.Interventions that address both individual and social adverse determinants of health, including food insecurity and poor dietary quality, 48 lack of physical activity, 8,36 and access to quality health care, in rural communities could have an impact on the incidence of diabetes in pregnancy and its severity, 49 including glycemic control. 50For example, rural residents may require assistance not only in accessing clinical services, but also with transportation and housing around the time of birth (especially if they need to travel for care).The clinical management of diabetes in pregnancy frequently requires collaborative care, including access to prenatal ultrasound, intensive pharmacotherapy, diabetes education and nutrition support, and subspecialist consultation with maternal fetal medicine specialists and endocrinologists. 34,51Practice models and policies that include logistical support and resources and training for providers for further collaboration across rural communities and higher acuity facilities in urban settings should be evaluated.Telehealth may be an option to bridge this divide to decrease rural-urban variation in care delivery, allow for remote patient monitoring and improve patient engagement with their prenatal and diabetes care. 52

| CONCLUSIONS
In conclusion, the rate of both DM and GDM increased over the study period and was persistently higher for pregnant women in rural compared with urban areas of the USA from 2011 to 2019.These rural-urban differences varied by race and ethnicity and US region.Attention to the particular challenges faced by rural pregnant women living with diabetes is an important public health priority for delivering equitable diabetes care in pregnancy.

AU T HOR C ON T R I BU T ION S
KKV, XH, LCP, WAG and SSK wrote the manuscript and analysed the data.MBL, JJ and NAC contributed to the discussion and reviewed/edited the manuscript.The authors have no conflicts of interest to disclose.

T A B L E 1
Characteristics of nulliparous pregnant women in the USA from 2011 to 2019 by rural and urban status (N = 12 401 888).

F
I G U R E 2 (A) Trends in the age-adjusted frequency of pregestational diabetes per 1000 live births among nulliparous pregnant individuals in the USA from 2011 to 2019 by rural versus urban residence.(B) Trends in the age-adjusted frequency of gestational diabetes per 1000 live births among nulliparous pregnant individuals in the USA from 2011 to 2019 by rural versus urban residence.

T A B L E 2
Average annual percent change (AAPC; 95% CI) in the rate of pregestational and gestational diabetes among nulliparous pregnant women in the USA from 2011 to 2019, by rural and urban status.

| 33 RURAL
-URBAN DISPARITIES IN DIABETES IN

F
U N DI NG I N FOR M AT ION Drs Venkatesh and Joseph were supported by the Care Innovation and Community Improvement Program at The Ohio State University.Dr Khan was supported by NHLBI grant #HL161514.C ON F L IC T OF I N T E R E S T S TAT E M E N T None declared.DATA AVA I L A BI L I T Y S TAT E M E N TThese data are publicly available from the US National Center for Health Statistics (NCHS) National Vital Statistics System via a data use agreement.

E
T H IC S A PPROVA L This was deemed exempt by the Northwestern University Feinberg School of Medicine Institutional Review Board Review given that data were de-identified and publicly available.ORC I D Kartik K. Venkatesh https://orcid.org/0000-0002-8043-556XR E F E R E NC E S Six-level NCHS urban-rural classification scheme for US counties, which includes two subgroups for rural areas (micropolitan and non-core) and four subgroups for urban areas (large metropolitan, large fringe/suburban, medium metropolitan, small metropolitan) were assessed.Rate ratios of pregestational and gestational diabetes in rural versus urban areas among nulliparous pregnant women in the USA from 2011 to 2019 overall and by age, race and ethnicity, and US region.Values of p are based on the interaction term in the regression models.
a Joinpoint regression introduced an inflection point in [year].b T A B L E 3 b