Identifying Women’s Health Care Deserts

The March of Dimes produces a report and map proposing to identify "Maternity Care Deserts" in the United States at the County Level. They define a desert as a county with zero providers, as well as zero hospitals or birth centers providing obstetric care. Their map displays a simple count of the number of each county's registered OBGYNs, or the sum of the number of OBGYNs in each county, added to the other two metrics. From there, they have color-coded each county from the lowest to highest numbers to display high versus low access. See their full analysis here:  https://www.marchofdimes.org/maternity-care-deserts-report. 

There are a few issues with this analysis. First off, counties are not a standard size, shape, or area in the U.S. In addition, counties often span metro, suburban, and rural areas, plus they can vary greatly in traversable distances, geographies, population numbers, demographics, and more. I propose that creating a summed count of doctors or institutions, then comparing this number across counties, is inadequate to identifying true "deserts" of care at the county level. In theory, people may and do traverse multiple nearby (contiguous) counties to access resources in many areas of the country, especially in metro, suburban, and more populous areas. In short, a county lacking all three of the metrics may indeed border other counties that do.

The Area Health Resource Files’  Health Provider Shortage Area  (HPSA) designation does this type of "contiguous county" analysis, but only for primary care, dental, and mental health, and not for women’s health, maternity care, or other specialties. While March of Dimes uses AHRF data, it fails to include any computation which might take into account these considerations. In fact, March of Dimes cites this limitation with their analysis at the end of their report, specifically highlighting their counting of individual counties' information alone, as islands unto themselves, and their lack of inclusion of the contiguous county context. This project's analysis is meant as a building block and starting point to creating a more adequate, comprehensive metric by which to compare this access to care. It focuses solely on the main OBGYN metric.

This is what the map looks like if we simply compare the counts of OBGYNs per county (below). To me, it appears to somewhat mirror population density in the United States. Almost all major cities appear to have similarly good access (in blue and light blue), while less populous areas do not (in red).

One of the main goals of this analysis was to create and include the Python computation of a new metric which would take into account the contiguous county considerations. I theorized that the original map of “deserts” would change starkly once contiguous counties’ OBGYN providers were factored in, with some of the counties that were originally designated deserts disappearing due to the appearance of contiguous county access. “True” deserts that emerge, then, will be counties that lack access even in neighboring accessible counties. 

Research Questions: What does it mean to have access to “maternity health care”? How do we best define and measure what a maternity health care desert, or oasis, is? What regions (and who) does and doesn’t have access to this care? How does access vary by region, proximity to metro areas (urban versus rural), or by other features?

Data used: Data comes from the same source, the U.S. Health Resources and Services Administration's (HRSA) Area Health Resources Files (AHRF). This aggregates data from multiple sources including Census data, county, state, and national level files from the HRSA, the Bureau of Health Workforce (BHW), and the National Center for Health Workforce Analysis (NCHWA). Luckily, AHRF produces a statistic around contiguous counties for other medical areas, so I was able to use their coding to create a new metric for any counties that touched another county.

Logic: For each county in the row of the dataframe, create a new variable to add together all contiguous county data on top of the original county's OBGYN counts and store them in "contig_obs_sum." Create a for loop that loops through the contiguous county data if there is any for each row. Add results of the contiguous counties to the variable, otherwise keep the same number as the original county.

The for loop iterates through each of the contiguous county columns (available from contgs_cnty_num01 to contgs_cnty_num14). For each row in the DataFrame, it checks whether a contiguous county exists and adds the land area condition of less-than 1,000 square miles in order to focus on counties that are, in theory, navigable in a long commute (2 hours, 120 miles) or less. It then finds the matching FIPS code for that county, retrieves the corresponding OBGYN count value for it (md_nf_obgyn_gen_21), and adds it to the summed OBGYN metric (contig_obs_sum) for the original row.

Data Management: Research AHRF data, download 2021 data files, in Python, clean data to include only variables relevant to maternal care, create geodataframe by joining to Census county and congressional district data, normalize all variables, export and upload to ArcGIS.

Exploratory Data Analysis: Filter out maternal health care variables, focus on 2021 OBGYN and population numbers, review initial descriptive statistics and patterns.

New Variable/Metric Development: Create logic to calculate OBGYN access in neighboring counties based on land area (with acknowledge limitations), then compare to original variable/metric in descriptive statistics and in data visualizations (maps).

Rural vs. Metro Analysis: Using AHRF continuum variable, identify patterns across rural and metro regions.

Mapping: Produce a series of map comparisons in Python and ArcGIS Online.

Beginning with the simplest analysis of OBGYN counts per county, I then compared this to OBGYN counts per county, but with the new metric included. This particularly affected the map on the eastern half of the country. However, the metric still needed normalization to create a more balanced analysis. I normalized the metric to make it the number of OBGYNs per (in an available distance to) 1,000 people in each county, theorizing that this is the upper end of a patient panel load that one single OBGYN might be able to see in one year. This created a wholly different map and the final product that I analyzed for this project.

View each of the four map versions and the differing map analyses in the slideshow below.

Scroll to view the progression:

The final analysis uses the new calculated metric, OBGYNs per 1,000 females in the population of the current county AND its contiguous counties if and only if the county is less than 1,000 square miles.

A closer look at four different states, four different scenarios, in four different geographic regions, using this new metric.

As I thought, the map visually changes quite a lot based on the type of measurement used to designate a "desert." Counties that previously were considered deserts by the analysis that inspired this project are really not entirely in my analysis, solely because of their access to resources in nearby contiguous counties, while some are even worse off than in previous analyses due to their true lack of any nearby OBGYN care.

There are some obvious regional (urban/rural continuum) differences: Metro areas, especially located on the eastern side of the country, have much broader access, both by the numbers and in more geographic directions, than rural areas, especially compared to the West/Mountain West.

From my perspective, less than 1 to even 1-to-5 OBGYNs available to the population in a traversable area still doesn’t seem much like access, let alone an availability of choice. This leaves little consideration of options for the care an individual might need or want from their provider, who in many cases might also function as their primary care doctor who they see once per year.

Next steps: Need to factor in the additional institutions like hospitals and birth centers with obstetrics to the new variable next, as well as find additional ways to address the problem of irregular county size. For policy/resource allocation, there is a need to take a closer look at populations in low to no-access areas (and dive deeper into the demographics of the female population, ages, birth rates, and more). Other precedents to this type of analysis do exist for further context and incorporation, such as the AHRF  Shortage Designation Areas  and this  Maternal and Infant Health Mapping Tool .

My hope is that an analysis like this one would be useful to federal agencies and local/state governments for their own reporting, policy development, campaigns, and resource targeting.

In another project, I created another analysis and  an example dashboard  that is specifically created for policy use. Read more about it here:  https://annaefeldman.com/gis-dashboard-maternal-healthcare .