This was a descriptive, cross-sectional observational study designed to evaluate the availability of rehabilitation providers across census tracts using GISs. The primary objective was to estimate and compare population-to-provider ratios at the census tract level and examine how these ratios relate to the sociodemographic characteristics of each tract. This study used a methodology that combined state licensure data (obtained via paid public record requests) with open access demographic datasets. These data sources were integrated using spatial mapping techniques to facilitate small-area workforce analysis. Texas was chosen as the exemplar state because of its demographic diversity; pronounced urban-rural differences; and availability of comprehensive, high-quality licensure data. All processed datasets were stored locally in both CSV and geodatabase (.gdb) formats in ArcGIS Pro (Esri) to support spatial analysis.

When conducting GIS-based analyses of workforce availability, researchers should consider the geographic and demographic context of the region under study. Factors such as geographic land area, population density, and urban-rural distribution shape both the delivery and interpretation of health services and provider access.

Aggregated geographic units such as counties or zip codes, while commonly used, can mask meaningful local variation in provider availability and introduce misclassification bias. Zip codes, for example, often group together heterogeneous populations by combining areas of affluence with areas of poverty, which can distort analyses of access and need [39]. In contrast, census tracts, which typically include populations of 1200 to 8000 people and are designed to be socioeconomically homogeneous, provide a more precise basis for equitable spatial analysis [38].

The use of small-area units such as census tracts requires detailed data on provider location and population characteristics. This study’s methodology is designed to support census tract–level analysis when such data are available. Ultimately, GIS-based access assessments must balance granularity and feasibility to generate meaningful, context-sensitive insights.

This study methodology used secondary, deidentified data from publicly available sources, including licensure databases and census-based demographic datasets. No identifiable confidential information was disclosed, and no direct interaction with human subjects took place. Before implementation, this study was approved by the Texas Woman’s University Institutional Review Board as exempt under title 45 of the Code of Federal Regulations part 46.104(d) as the research involved publicly available or nonidentifiable data. Institutional review board approval was obtained before conducting data analysis. All data were deidentified before data cleaning and analysis. The institutional review approval is available in Multimedia Appendix 1.

This study integrated information from two primary data sources: (1) state-level licensure data for rehabilitation professionals and (2) population and demographic data from the US Census Bureau’s American Community Survey (ACS). For this study, we used Texas as the example state in the analysis.

Licensure data were obtained from the Executive Council of Physical Therapy and Occupational Therapy Examiners through publicly available open record requests in Texas in 2022 [40]. This dataset included rehabilitation provider information, including residential mailing address (including street, city, state, and zip code); license type; issue and expiration dates; and basic demographic characteristics such as gender and ethnicity, where available. Personally identifiable information (eg, name and license number) were removed before analysis. Rehabilitation provider addresses were used for spatial analysis, and all data were stored and processed in a deidentified format. Although the licensure dataset included an optional work or practice address field, this information was often incomplete or inaccurate as many licensees either repeated their residential address or left the field blank. Therefore, residential mailing addresses were used as a consistent and complete proxy for spatial analysis.

Demographic and population data were sourced from the US Census Bureau’s 2020 ACS [41], a publicly available dataset providing census tract–level estimates for population size and characteristics such as race, ethnicity, age, disability status, insurance coverage, and income. Population estimates were obtained from the 2020 ACS dataset to match the licensure data time frame and provide reliable census tract–level denominators for ratio calculations. These data were downloaded as geodatabase files for integration with GIS software (eg, ArcGIS Pro).

These 2 data sources were used to estimate population-to-provider ratios and assess workforce availability at the census tract level.

After obtaining licensure and demographic data, a multistep cleaning and aggregation process was conducted to ensure accurate spatial representation of the rehabilitation workforce.

The population-to-provider ratios across census tracts were investigated using descriptive statistics and choropleth and bivariate maps. A choropleth map is a thematic map in which areas are styled to represent the variation in a single variable. A bivariate map displays 2 variables using variations in colors or symbols, where each variable is assigned a different graded color scheme in a 3 × 3 table that is included in the top right-hand corner of the bivariate map [44,45]. A graduated color choropleth map was created to visually analyze the distribution of the population-to-provider ratios across the state of Texas by quintiles. Population-to-provider ratios were calculated and categorized into quintiles to facilitate relative comparison across areas following the approach used by Brown et al [46], who used quintile-based classifications to highlight spatial variation in the absence of established adequacy benchmarks. Comparing the highest and lowest quintiles facilitates the examination of factors associated with substantial difference in provider availability across areas. Bivariate choropleth maps were then created to visually describe the relationship between population-to-provider ratios and the percentage of disability in each census tract as defined by the ACS. Bivariate choropleth maps allow for the visualization of the specific areas that may be most in need of providers based on specific population demographics.

Descriptive statistics were used to describe the variability in population-to-provider ratio across all census tracts within the state of Texas. To describe differences in provider availability by rurality, census tracts were classified using 2020 Rural-Urban Commuting Area (RUCA) codes [47]. Tracts with a primary RUCA code of <4 were categorized as urban, and those with codes of ≥4 were categorized as rural. Each tract’s population-to-provider ratio was summarized by rurality using measures of central tendency and dispersion (median, IQR, minimum, and maximum). All analyses were conducted in R (version 4.3) using the dplyr and ggplot2 packages. The R code for the RUCA analysis is included in Multimedia Appendix 5.

To evaluate whether the spatial distribution of access to rehabilitation providers exhibited clustering across the study area, a global Moran I statistic was computed for the population-to-provider ratio at the census tract level using ArcGIS Pro (version 3.3). Because Texas census tracts are polygonal units that vary substantially in geographic size, with very large tracts in rural regions and very small tracts in dense urban areas, a contiguity-based conceptualization of spatial relationships (edges only) was selected instead of a fixed distance threshold. This approach defines neighbors as tracts sharing a common boundary segment, ensuring that all tracts, regardless of their physical size, are evaluated based on direct adjacency rather than distance alone. Row standardization was applied to account for variability in the number of neighbors across tracts. The analysis tested whether the distribution of the population-to-provider ratio was spatially random, clustered, or dispersed using 999 random permutations to assess significance. Z scores and P values were used to evaluate departures from spatial randomness.

The following measures of feasibility were assessed throughout the study: ability to obtain licensure data and the percentage of rehabilitation providers included after cleaning, the time and resources needed to clean and geocode the data, the percentage of provider addresses that were successfully geocoded, the ability to merge licensure data with ACS data, the percentage of census tracts that were matched, the feasibility of the geospatial analysis methods, and the relevance of the results to workforce planning and their replication.

Across 6896 tracts, ratios ranged from 4.5 to 11,147 persons per rehabilitation provider (median 1131, IQR 537-2501). Tracts with the lowest ratios clustered in the Texas Medical Center (eg, census tract 313102), whereas the highest ratios were observed in Lubbock (002100) and east of Dallas (census tract 012000). The distribution of population-to-provider ratios across the state of Texas is shown in Figure 2.

When stratified by rurality, urban tracts (5734/6896, 83.1%) had a median ratio of 1141 (IQR 2054), whereas rural tracts (1162/6896, 16.9%) had a median ratio of 1093 (IQR 1690). These values indicate broadly similar distributions of provider availability between urban and rural areas, with slightly greater variability in urban tracts. Box plots (Figure 3) show the distribution of census tract–level population-to-provider ratios for rehabilitation providers classified by RUCA code. Ratios are displayed on a logarithmic scale to account for skewness in the data. Median ratios and IQRs are similar across rural and urban tracts, suggesting comparable variability in provider availability across both settings.

The global Moran I statistic for the population-to-provider ratio across census tracts was 0.305 (Z=40.28; P<.001), indicating significant positive spatial autocorrelation and less than a 1% chance that this clustered pattern was a result of random chance. The ArcGIS Moran I results are included in Multimedia Appendix 6.

Graduated and bivariate choropleth maps were used to visualize provider availability and its intersection with social and demographic indicators. A graduated color choropleth map (Figure 4) displays population-to-provider ratios across all census tracts in Texas, with lighter shades indicating higher provider availability (lower population-to-provider ratio) and darker shades indicating lower provider availability.

To further illustrate equity-related patterns, bivariate choropleth maps overlay these ratios with variables such as percentage of disability, percentage of minority populations, and percentage of the population below the poverty level. For example, disability data were drawn from 6 ACS questions addressing hearing, vision, cognitive ability, ambulatory ability, self-care, and independent living difficulty [48]; respondents answering “yes” to any item were classified as having a disability.

As shown in Figure 5, the overlay of population-to-provider ratios with disability prevalence reveals regional variability across the state, although not a strict urban-rural divide. While areas of west and south Texas include clusters of high population-to-provider ratios, similar patterns also appear in parts of urban and periurban counties. Conversely, some rural tracts demonstrate moderate ratios despite small provider counts, reflecting the influence of low population density on the calculation. Regions in dark blue reflect regions where there is a high population-to-provider ratio coinciding with high prevalence of disability, suggestive of census tracts where there is a mismatch of population need and provider availability.

Feasibility was assessed throughout each phase of the methods, including data acquisition, data processing, descriptive mapping, and statistical analysis. Table 1 illustrates the measures of feasibility assessed in this study.

Rehabilitation provider availability showed substantial geographic variation, with tract-level population-to-provider ratios ranging from 4.5 to 11,147 (median 1131, IQR 2010) and similar central tendencies across rural and urban areas, yet the strong positive spatial autocorrelation (Moran I=0.305; P<.001) indicated clear clustering of high- and low-availability regions.

These patterns align with findings from other GIS-based workforce studies across a range of health professions, which similarly report uneven spatial distribution of health care providers and localized pockets of limited access. National analyses in the United States have identified substantial variation in spatial accessibility to health care providers and dental clinics [25,26], whereas international assessments in Taiwan, Brazil, China, and Ethiopia have documented comparable clustering of high- and low-access areas across rehabilitation and community care services [27-31]. Although these studies differ in setting and methodological approach, they consistently demonstrate that workforce maldistribution emerges at small geographic scales and often reflects underlying sociodemographic gradients. This study extends this body of work to the US rehabilitation workforce by applying comparable geospatial methods to tract-level licensure data, situating our findings within a broader pattern of spatially concentrated workforce inequities.

Statewide patterns in this study demonstrated striking variation in population-to-provider ratios across census tracts, revealing local disparities that county-level summaries fail to capture. Previous statewide reports from the Texas Health Professions Resource Center have suggested that availability challenges are largely a rural issue [17,18]. In 2022, the average population-to-provider ratio for physical therapists was 1593:1 in metropolitan counties and 2870:1 in nonmetropolitan counties, whereas for occupational therapists, it was 2778:1 in metropolitan areas and 5611:1 in nonmetropolitan areas. These data painted a picture of a binary divide between urban and rural access.

When we examined rehabilitation provider availability at the census tract level rather than the more common county designation, our findings diverged from prevailing assumptions. Although state-level reports show markedly higher population-to-provider ratios in nonmetropolitan counties than in metropolitan ones, our tract-level analysis found similar median values for urban and rural tracts, albeit with greater variability in urban areas. This pattern mirrors findings for other health professions: for example, a spatial analysis of primary care access in Philadelphia, Pennsylvania, revealed that geographic access disparities often overlay racial and socioeconomic patterns rather than simply urban versus rural gradients [46]. By shifting to smaller geographic units, our study revealed that within-urban and within-rural heterogeneity matters because access gaps may persist in urban neighborhoods that are often assumed to be well served and, conversely, some rural tracts may fare better than county-level data suggest.

The Moran I statistic confirms that rehabilitation provider availability is spatially clustered rather than randomly distributed. In other words, tracts with low provider availability tend to be near one another, forming regional patterns of limited access. A comparable spatial analysis of rehabilitation human resources in China demonstrated this clustering effect across provinces, underscoring that workforce maldistribution may follow geographic spillovers rather than isolated pockets [49]. However, in our Texas analysis, these clusters did not correspond neatly to urban or rural classifications, indicating that factors beyond simple rurality may drive spatial patterns of provider scarcity. Although further research should investigate these determinants in depth, the observed clustering suggests that improving access will likely require coordinated, regionally targeted interventions rather than isolated, tract-by-tract fixes.

Although this analysis was based on provider residential addresses, the presence of strong spatial clustering indicates that these data still meaningfully represent patterns of service availability. It is unlikely that large numbers of providers are commuting long distances into the clusters of low access identified in this study. Instead, these regions likely represent genuine shortages in the local rehabilitation workforce. Recognizing and addressing these gaps is essential not only from an equity standpoint but also from a broader population health perspective. Limited access to rehabilitation care can lead to preventable disability, loss of function, and reduced participation in work and community life [50,51].

The GIS-based maps generated in this study extend beyond visualization: they function as actionable tools for workforce assessment and planning. Similar spatial approaches have been applied in other areas of public health to identify regions where need and provider availability intersect, informing policy and resource allocation decisions. For example, bivariate choropleth maps have been used in cancer control to guide the placement of screening resources in areas with low service capacity and high disease burden [52] and in studies of lung cancer screening to identify US counties with both high smoking prevalence and low provider density [53]. Such mapping frameworks demonstrate the potential of GIS-based methods to translate complex population and workforce data into actionable insights. Applying this approach to rehabilitation introduces an equivalent capacity to identify where population need and provider scarcity converge. By pinpointing census tracts with high disability prevalence and low provider availability, these maps enable decision-makers to prioritize areas for resource allocation, community partnerships, and program expansion. Beyond workforce planning, they also provide a framework for longitudinal monitoring, allowing policymakers to observe how workforce distribution shifts in response to educational pipelines, policy incentives, or population change over time. In this way, the maps produced in this study serve not only as evidence of current inequities but also as instruments for promoting future workforce equity in rehabilitation.

The approach presented in this paper is replicable; the step-by-step procedures, along with the inclusion of code in the multimedia appendices, are intentional so that the methods can be applied in a variety of geographic contexts or across different health professions. This transparency supports reproducibility and scalability in future research. Additionally, the novel method emphasizes high geographic resolution by conducting analyses at the census tract level. This finer scale allows for a more precise identification of geographic disparities than approaches that rely on broader units such as counties or zip codes [38,39], offering a more detailed understanding of provider availability.

There are several limitations to this study. First, it focused exclusively on availability, one domain within the theory of access by Penchansky and Thomas [7], and did not account for realized access (the actual entry of a patient into the health care system) [54] or outcomes. The use of residential addresses as a proxy for provider practice location introduced potential misclassification, and the population-to-provider ratio did not capture factors such as transportation infrastructure, health insurance coverage, or service capacity.

Additionally, the observational nature of the analysis precludes causal inferences about relationships between availability and health disparities. This study also prioritized comparison of high- and low-availability areas using quintiles, which may not account for regional variations in land mass, population density, or local health systems.

Finally, publicly available licensure and census data are subject to reporting errors and definitional inconsistencies. These data limitations should be acknowledged in future research and addressed, where possible, through improved workforce data infrastructure.

The methods used in this study are adaptable and scalable for use in other professions (eg, speech-language pathology and behavioral health) and other geographic regions. Future studies can build on this framework by incorporating clinic location data, travel time analyses, and patient-level outcomes to provide a more complete picture of access to care. Integration of GISs into national workforce planning tools could support more equitable distribution of health services and guide data-driven decisions at local, state, and federal levels.

Beyond immediate replication, this work lays the foundation for a broader research agenda in rehabilitation workforce equity. The GIS-based approach demonstrated in this paper establishes a platform for developing rehabilitation-specific shortage area indexes, evaluating spatial alignment between workforce supply and population need, and examining equity in workforce diversity. These next steps extend beyond feasibility to inform national policy and education initiatives aimed at improving access for underserved communities. The ability to acquire, clean, and link licensure and census data demonstrates both the practicality and potential of applying this method to larger, longitudinal studies that track changes in provider availability and access over time. In this way, the results directly inform workforce planning by providing an evidence-based foundation for policies and programs aimed at achieving a more equitable distribution of rehabilitation providers.

This study demonstrates the feasibility and value of integrating licensure and demographic data to assess rehabilitation provider availability at the census tract level, which is a novel approach in this field. Quantitatively, tract-level population-to-provider ratios ranged from 4.5 to 11,147 (median 1131, IQR 2010) and demonstrated significant spatial clustering (Moran I=0.305; P<.001). The results reveal that provider availability in Texas is spatially clustered and influenced by factors beyond urban-rural status, highlighting the need for targeted, regionally informed interventions.

By establishing a replicable, GIS-based framework, this work advances both workforce equity research and geospatial health methods, offering a foundation for small-area analyses that move beyond traditional county-level approaches. These findings underscore the potential of spatial mapping to identify underserved communities and guide data-driven workforce planning and policy development that promote equitable access to rehabilitation care.