Access to perinatal care is a well-documented need for pregnant and postpartum women, particularly those living in rural communities. Pregnant patients in rural communities have less access to clinical care in their communities and are more likely to have inadequate transportation options to access obstetric services (1). They are also less likely to have access to obstetric specialists in their communities (18). Obstetric services in rural communities are at increased risk for closure due a variety of factors such as reimbursement difficulties, declining patient volumes, and staffing difficulties (19). Recent closures of obstetric facilities resulted in more than 50% of all rural counties in the U.S. being without obstetric services (20). As closures of maternity care facilities in the U.S. continue to increase, it is essential to extend our knowledge to be able to intervene effectively to improve health outcomes.

Previous studies, largely conducted outside the U.S., have examined the relationship between geographic distance/driving time and adverse outcomes in pregnancy, with mixed results and outcomes due to limited data and a high number of complicating factors [see (21–25)]. In general, travel times >20–30 min are associated with adverse outcomes (22, 23). Further, the rate of out-of-hospital births is more than double for expectant women living 30+ km (18.6 miles) from care, particularly for U3 women (26).

The Netherlands has a national perinatal registry (Netherlands Perinatal Registry or Stichting Perinatale Registratie Nederland) with uniform data collected each year on locations and outcomes in perinatal care. Utilizing this registry, researchers were able to determine that travel times to maternal care of >20 min resulted in increased mortality and adverse outcomes (23). Using this registry and associated research, a national steering committee on perinatal care identified geographic barriers to care as a major contributor to poor maternal child outcomes in the country. They created a nation-wide stakeholder advisory group and conducted listening sessions. They then developed evidence-based interventions and used health-policy solutions to improve perinatal outcomes. Examples of interventions included: auditing perinatal deaths in term babies, instituting prenatal screening for congenital anomalies, and establishing a commission on perinatal care. They also proposed assigning a case manager to every pregnancy, developing birth plans with all expectant women, and instituting mandatory house visits. Maternal and child outcomes improved dramatically nationwide after implementing these interventions (27, 28). This clearly demonstrates the immense value of a national maternal care dataset.

In comparison to the Netherlands, the U.S. has almost 20-fold higher maternal mortality rates and nearly 50-fold higher rates for certain racial/ethnic groups (29). The U.S. is in dire need of evidence-based interventions in maternal care, yet we lack the national datasets necessary for researchers to generate that evidence. A handful of U.S. studies have measured the driving distance to maternal care services, but they do not provide underlying data for other researchers to use, so their ability to inform the study of geographic barriers to care among U3 populations is unclear [see (14, 30, 31)]. A national dataset that incorporates geographic barriers to maternal care with a special focus on U3 women would enable research, policy, and healthcare communities to identify needs and implement interventions to advance maternal health equity.

There are multiple reasons why such a dataset can be critically needed but not yet exist in the U.S. Of note, single-payer or universal healthcare systems like those seen in the Netherlands or the U.K. enable these national datasets by default, whereas the fragmentation of the U.S. healthcare system makes the creation of a dataset much more challenging. First, it is difficult to find and link data on health outcomes, disparities/inequities, and U3 populations. The Federal government, the source of the majority of these data, has been directed where possible to make data publicly available under the OPEN Government Data Act (32). While this would make a wealth of data available and easy to find and use, no funds were allocated to accomplish this work. Second, creating and sharing datasets is of great benefit to other researchers, but limited benefit to those owning/linking/sharing the data, so they are disincentivized to share those data (33). Third, creating and linking these datasets requires technical expertise in data management, big data analysis, medical diagnosis and billing, social determinants of health, geographic information systems and geocoded data, and health equity, which are highly varied skillsets. The technical skills in particular are a substantial barrier to creating and disseminating high-quality, easily accessible data sources (34). This undertaking would require the building and sharing of a highly versatile dataset in easily accessible formats to allow researchers to analyze and compare health equity and geographic barriers to care across U3 populations and racial, social, and economic underlying factors.

Building a comprehensive dataset will require leveraging diverse, publicly available data such as deidentified electronic health record data; geocoded locations of hospitals providing multiple levels of maternal care services; and social determinants and demographic data at the county-level. While individual research teams can collect these data locally, HIPAA, PII, and data sharing restrictions render the creation of a national dataset nearly impossible. Combining the home geographic region, demographic/socioeconomic information of perinatal patients, and maternal and child health outcomes, would likely enable identification of individuals and violate HIPAA. In addition, health data have become a big business. Health data are estimated to be worth $300–$450 billion per year, creating strong incentives against sharing data or making data interoperable for other researchers to use, analyze, or combine (35, 36). This results in small, fragmented, expensive datasets and/or strong restrictions on what data can be used for research and limiting the applicability of potential results. This paradigm extends to maternal health data: multiple researchers have examined maternity care deserts with informative analyses, but none shared the underlying data. In addition, existing reports focus on the availability of facilities (i.e., maternal care deserts) as a measure of health equity rather than the ability to use those facilities (i.e., health outcomes). For example, the March of Dimes Maternity Care Deserts report (37) ranks each county across the U.S. by the number of facilities in that county. However, while researchers and policymakers can reference these reports, they cannot conduct additional research for other county level measures or moderate by population characteristics.

Due to these many restrictions and limitations, it becomes necessary to collect even larger, aggregate datasets and use the resultant increased statistical power to conduct rigorous population-level research rather than individual-level research. Utilizing this approach enables the use of numerous national aggregate data sources. These data sources have standardized reporting, multi-year collection, and rigorous methods that are already understood and accepted by the broader research community. While this approach is challenging, utilizing national datasets (with reporting at the county level) means that for the first time, it is possible in the U.S. There are numerous potential sources of publicly available data that could be utilized for this purpose. For example, the Healthcare Cost and Utilization Project (HCUP) collects discharge data on hospital inpatient visits across the U.S. (38). These data are de-identified and made publicly available through a dashboard. Data can be accessed at the county level (for participating states) and stratified by aggregated, clinically-meaningful categories (Clinical Classifications Software Refined, or CCSR). The publicly available data include 15 CCSR codes related to pregnancy (see Table 1). The CCSR groups in the dataset can easily be binned into routine vs. high-acuity visits, defined as those visits likely to result in SMM. Socio-demographic variables can be accessed via the U.S. Census Bureau (39) and the CDC/ATSDR Social Vulnerability Index (40), and data from the Centers for Medicare and Medicaid Services can be used to identify the geographic locations of hospitals with maternal care services, separated into those able to provide low-acuity and/or high-acuity services (41). It is therefore possible to combine hospital utilization data with demographic, social, and economic data into a maternal healthcare dataset with county-level geographic specificity for the entire country. An added benefit of utilizing these large, public datasets is increased patient privacy and data protection. All data described above are already deidentified and aggregated at levels to prevent re-identification. There can be a potential risk from linking datasets, and protecting patient confidentiality should be the highest priority in any data project, particularly given the need to establish trust among the underrepresented, underserved, and underreported populations these datasets would most benefit. For example, focusing on population-level data (e.g., percent of population or index ratios) would help retain anonymity.