1Unequal allocation of healthcare resources poses significant challenges for maintaining equity within healthcare systems across many countries. “Medical deserts”, often defined as medically underserved areas in the European literature, are described with issues in territorial planning, decline of general practitioners (GPs), and geographical imbalances in healthcare human resources (Brînzac et al., 2023). To tackle health inequalities in the European Union (EU), the European Commission has prioritized addressing medical deserts in their funding programmes. Some authors have proposed defining medical deserts through consensus-building approaches (Brînzac et al., 2023), while others have provided overviews of their definitions, characteristics, contributing factors, and mitigation strategies across Europe (Flinterman et al., 2023). At the national level, in Spain, Dubas-Jakobczyk et al. 2024 (Dubas-Jakóbczyk et al., 2024) applied a mixed methods approach combining a scoping literature review and a case study analysis of medical deserts. In France, Bonal et al. have proposed a typology capturing the multiple dimensions of the “medical desert” concept (Bonal et al., 2024). Launay et al. proposed a SCALE index to measure healthcare accessibility at a fine spatial scale, incorporating both the proximity and availability of healthcare professionals (Launay et al., 2024).
2The concept of medical deserts remains complex, as no two such areas are identical. They may refer to rural, geographically isolated areas that are difficult to reach by ambulance or where residents must travel long distances to access a general practitioner. Similarly, urban areas where healthcare services do not adequately meet the needs of the local population can also fall into this category. Such situations can worsen health inequalities, particularly for vulnerable groups such as the elderly. However, this concept appears too vague to fully capture patients’ diverse challenges in accessing care or to provide clear guidance for developing appropriate interventions. A more accurate approach is to address this issue by focusing on disparities in access to care and by including the multiple dimensions that define the medical deserts.
3Defining the scale(s) at which the health issue will be observed is crucial. Changes in scale not only alter the spatial dimensions under consideration, but also affect the possible appreciation of the social, political, or economic decision. In France, decisions are made at the national level, while policy development at the regional level. The health project implementation has capacity development at the Inter-municipal Cooperation (EPCI) level. The Regional Health Agencies (ARS) are responsible for implementing national health policy at the regional level, considering local specificities. The definition of relevant territories for action can draw on fine administrative scales, useful for studies considering populations within their living environments. Methodologies developed to understand medical deserts rely on such administrative scales, thus contributing to the construction of decision-making territories. This presupposes that these action territories are built by aggregating finer administrative territories for which statistical information are available and can be mobilized.
4To better understand this context, place-based tools describe how accessibility is associated with a place or spatial unit. These spatial planning tools that allow authorities to identify areas with a less balanced distribution of accessibility and to support a vision of reducing inequalities (Apparicio et al., 2008). Different approaches are traditionally used to describe healthcare accessibility. Classical measures of density and distance are still used in France (Launay et al., 2024), and in other European countries (Dubas-Jakóbczyk et al., 2024). More complex and efficient indicators such as local potential accessibility (Barlet et al., 2012) allow the French administration to define medically underserved areas to deploy measures to attract and retain GPs (Chevillard et al., 2018; Mangeney, 2023). The advantage of the xSFCA method is to consider both the supply of care and demand within the geographical unit under consideration and the surroundings. Most studies focused on one healthcare provider as general practitioners (Lucas-Gabrielli & Mangeney, 2019), nurses, physiotherapists (Legendre, 2021), or hospitals (Gao et al., 2021).
5In 2022, “medical deserts” of General Practitioners (GPs) based on xSFCA indicators concerned around 20 million people in France, which is equivalent to 30% of the population (“Arrêté du 1er octobre 2021 – Légifrance”). The factors responsible for this shortage are diverse, ranging from an aging population requiring more care, to the delayed increase in the numerus clausus for medical students, and the mass retirement of general practitioners.
6In this context, primary healthcare promoted by the World Health Organization as “the first level of contact of individuals with the national health system”(Rifkin, 2018) appears relevant. Involving several primary care workers alongside GPs (nurses, clinical pharmacists, physiotherapists, and other physicians…) in the diagnosis and treatment of patients contributes to promoting a better patient-centered and integrated approach to care (de Bont et al., 2016; Garattini et al., 2023).
7This study aims to describe the implications and consequences of multidimensional definitions of healthcare accessibility across three scales (municipality and two supra-municipal), taking full advantage of the classification approach of healthcare accessibility. This research is multidimensional as it explores different aspects (e.g. healthcare accessibility, healthcare needs, and the supply dynamics). It also adopts a 'multi-scale' approach, studying inequalities in primary care accessibility at different levels (e.g. population-living environment and policy and decision-making levels).
8Our study investigates France’s mainland and its overseas departments using data for 2019-2020. To analyze the multi-scale approach, the study area includes 34,990 municipalities as a population-living level and two supra-municipal scales as policy and decision-making levels corresponding to 2,805 “territoires de vie sante” and 1,250 EPCIs (Table 1).
9The municipality scale is a fine administrative unit for which socio-demographic and health data are available in France with the advantage of being the reference level of aggregation in many databases (INSEE population data and medico-administrative databases). The EPCI and the ‘territoire de vie santé’ scales, are relevant and more appropriate than municipalities- the smallest scale- for capturing the dynamics of mobility and healthcare provision. However, both scales also have limitations. The EPCI scale is a recognized political and administrative level that facilitates the operational translation of findings into local actions. However, it is not based on healthcare usage patterns. The ‘territoire de vie santé’ scales is less familiar to local elected officials and stakeholders and does not correspond to the institutional boundaries of local policies. However, it is a zoning system specifically designed for healthcare organizations and is used by the regional health agencies to define medically underserved areas (Chevillard et al., 2018). The “territoire de vie santé” is defined as the smallest group of municipalities within inhabitants can access the most common facilities and services. (Réseau français des villes santé de l’OMS, 2021).
Table 1: Characteristics of the three spatial boundaries of France: Municipality, EPCI, and “Territoire de vie santé” scales
|
Spatial boundaries
|
Number of spatial units
|
Average of population (habitants)
|
Population min, max
|
|
Municipality
|
N=34,990
|
1,917
|
[0 - 2,165,423]
|
|
“Territoire de vie santé”
|
N= 2,805
|
23,630
|
[96 - 486,828]
|
|
EPCI
|
N= 1,250
|
53,548
|
[3 909 -7,094,649]
|
10The data for this study come from multiple sources (Table 2). Healthcare accessibility, supply dynamics and healthcare needs variables at the supra-municipal level were constructed by aggregating finer administrative units (municipalities) for which statistical information is available and can be mobilized.
Table 2: Contextual data, data sources, and scales according to the dimensions of healthcare accessibility
*LPA: local potential accessibility
11The construction of geographic variables is necessary to characterize healthcare accessibility levels, supply dynamics, and healthcare needs. Spatial healthcare accessibility refers to both the distance to the nearest providers or services and to local potential accessibility indicators available for certain providers. First, the distance to the nearest providers (pharmacies and laboratories) or services (radiology and emergency) is used as a proximity indicator (measured in meters and minutes). It was calculated from the centroid of each geographical unit to the closest healthcare provider or service. These calculations account for road network characteristics, topography, and operational factors, using the distance matrix provided by IRDES (Lucas-Gabrielli et al., 2022). Second, Local Potential Accessibility (LPA) is a floating density indicator, relative to the age-standardized population, used to assess spatial accessibility to GPs, nurses and physiotherapists. Routinely calculated by the Direction for Research, Studies, Evaluation and Statistics (DREES) (Vergier, Chaput, 2017), this indicator represents the full-time equivalents of healthcare workers or consultations available(Barlet et al., 2012; Lucas-Gabrielli et al., 2016).
12The evolution of supply will highlights increasing inequalities in access to care, due to a decline in the number of general practitioners. This dimension is described using two indicators: the annual average rate of change in GPs’ LPA between 2015 and 2019, and the proportion of GPs over 60 years old.
13The healthcare needs dimension characterizes areas where populations have socio-sanitary disadvantaged. It is described by the median income and standardized global and premature mortality rates per 100,000 inhabitants. A sensitivity analysis showed that income alone was the best discriminator of the area, serving as a strong proxy for the social dimension of municipalities. An additional file presents the univariate analysis of all variables across the three scales, providing a comprehensive description of the French territory.
14Three classifications were created using standardized data and a consistent three-step method: first, computing scores for each dimension, second, applying Principal Component Analysis (PCA), and finally, performing Hierarchical Clustering (HAC). These steps allow each territory to be characterized according to the multidimensional definition of healthcare accessibility presented above. This approach enhances the statistical stability and robustness of the classification process, minimizing the risk of territorial misclassification (Ben-Hur, Guyon, 2003).
Figure 1: Methodological framework used to create and compare multidimensional healthcare accessibility classifications
Credits: Author
15Firstly, accessibility to general practitioners and emergencies services was introduced directly into the next two steps (PCA and HAC) due to their distinction from the other variables, considering that GPs are the cornerstone of healthcare accessibility and that emergencies concern the hospital setting (Vergier, Chaput, 2017). Four scores were calculated and included in the PCA: primary care providers, primary care services, supply dynamics, and healthcare needs (Figure 1). Creating these scores allows greater emphasis onthe healthcare accessibility dimensions, which involve two variables and two scores, while reducing the influence of supply dynamics and population needs, which initially involved many variables. These scores were developed using principal component analysis (PCA) by taking the coordinates of the first axis to summarize groups of variables within the same dimension. This approach used previously (Fayet et al., 2020) enhances robustness by simplifying the information.
16Then, two variables and four scores were included in each of the three PCAs. PCA analysis is one of the most widely used dimensionality reduction methods for continuous variables and helps to avoid redundant information. The variances explained were 57.8%, 60.9%, and 59.2% for the municipal, EPCI, and “territoire de vie santé” scales, respectively (Table 3). The internal validity of the PCAs was assessed using Kaiser normalization (eigenvalue threshold = 1), confirmed by the visualization of the elbow of the scree plot graph. Across allthe three scales, the PCA revealed the same dimensions: variables related to healthcare accessibility contributed most strongly to the first component (higher accessibility to GPs and providers corresponds to shorter travel times to emergenciess and services), supply dynamics on the second component and healthcare needs on the third component (Additional file).
Table 3: Indicators of quality for each classification
|
|
Number of class
|
The number of PCA components that explain 80% of the variation in the aggregated data
|
% of Inertia explained
1 and 2 components
|
Number of municipalities
mean [min, max]
|
|
Municipality
|
7
|
4
|
38%
|
19.8%
|
|
|
“Territoire de vie santé”
|
5
|
4
|
37.9%
|
21.3%
|
12 [1 - 102]
|
|
EPCI
|
5
|
4
|
39.1%
|
21.8%
|
28 [3 - 158]
|
17Finally, a HAC was performed to classify municipalities according to all dimensions, diversified by nature (healthcare accessibility, supply dynamics, and healthcare needs). HAC analysis consists of grouping spatial units based on multiple variables, considering similarities (within a cluster) and differences (between clusters). Based on the analysis of the inertia gain graph and the dendrogram, the clustering identified 7 classes at the municipal scale, while the number of classes decreased to 5 at the EPCI and “territoire de vie santé” scales. For each scale, the number of classes was chosen to optimally fit our objective of analyzing the impact of scale to describe the multidimensional approach to healthcare accessibility. Statistical analyses were performed using the Factominer package with R version 4.2.1.
18This study compares three scales to examine how the spatial context of healthcare accessibility varies. These scales include municipal and two supra-municipal levels, constructed by aggregating finer administrative units: “territoire de vie santé” and Inter-municipal Cooperation (EPCI).
19However, when spatial data are aggregated into arbitrary units, a problem known as the “modifiable area unit problem” (MAUP) arises (Openshaw, 1984). As Grasland et Madelin (2006) and Sanders (2011) noted, this issue stems from the fact that spatial units — such as municipalities, census tracts, or administrative regions — are not natural entities, but social and political constructs (Grasland, Madelin, 2006; Sanders, 2007, 2011). Consequently, the results of statistical analyses may vary depending on the scale (size of spatial units) and zoning system (how boundaries are drawn), leading to differences in the characterization of healthcare accessibility.
20In this sense, two scenarios are possible:
-
Municipalities geographically close have similar characteristics in terms of healthcare accessibility; the risk of differences in classification is minimized when aggregating geographical units and creating a new classification at a supra-spatial level.
-
Municipalities geographically close but do not have similar characteristics in terms of healthcare accessibility; the risk of differences remains.
21First, we computed a joincount global autocorrelation indicator to test the geographic proximity betweenclasses of municipalities according to their accessibility to primary care. The null hypothesis of a random distribution of municipal classes across the territory (i.e., absence of spatial autocorrelation) was tested. For a pair of municipal classes, a statistically significant Z score means that some classes of municipalities are geographically grouped. Possible scenarios include: one class of municipalities with high accessibility is located near another class of municipalities with high accessibility, or conversely, a class with low accessibility is located near another low-accessibility class. A p-value for each two-class distribution allows us to identify which classes are significantly autocorrelated and thus geographically grouped rather than randomly distributed (function joincount.test, package R spdep) (Zhukov, 2010).
22Then, we calculated the Theil index to analyze disparities in the distribution of municipal classes according to their multidimensional accessibility to primary care within and among “territoires de vie santé” and EPCI territories in France (Theil, 1967). The within-level Theil index (TW) reflects the distribution (homogeneity or heterogeneity) of municipal classes within each “territoire de vie santé” or EPCI, indicating internal disparities in healthcare accessibility . In contrast, the between-level Theil index (TB) reflects the distribution of municipal classes between “territoires de vie santé” or EPCI territories, highlighting broader disparities in healthcare accessibility. The T values range from 0 (homogeneous) to 1 (heterogeneous). The Theil index is a hierarchical measure of entropy, allowing simultaneous comparison of areas across different levels of spatial organization. The total Theil index combines the within- and between-level components as follows:
23Finally, we quantified how municipal classifications of healthcare accessibility changed using Sankey diagrams. The Sankey diagram allows us to analyze the effects of changing spatial scale by visualizing the flow between classifications using arrows: i) from EPCI to municipalities and ii) from “territoires de vie santé” to municipalities. The lines represent the relationships between classes, and their width is proportional to the frequency and intensity of these relationships.
24Figure 2 presents the spatial distribution of healthcare accessibility classes at the municipal level (A), “territoires de vie santé” level (B), and EPCI level (C). The three maps reveal substantial regional disparities, with greater variability at the municipality level, whereas at the supra-municipal scales, is the results appear more smoothed.
25Across all scales, the maps reveal contrasts between more and less attractive areas. More attractive areas shown in dark and light blue (classes 6 and 7 at the municipal level) are located along the Atlantic and Mediterranean coasts as well as in suburban areas. These socio-sanitary advantaged areas also accumulate favorable healthcare accessibility dimensions, maintaining a good supply of GPs and access to a full range of services. Less attractive areas, shown in red, pink and purple (classes 1, 2 and 5 at the municipal level) are found along the diagonal and in northern France; and include isolated rural areas or small urban centers. Classes 1 & 2 combine low attractiveness with disadvantaged healthcare accessibility dimensions , describing areas with low healthcare accessibility to all providers and services, considering all healthcare workers combined with low supply dynamics, and high socio-sanitary disadvantage. In contrast, the mitigated class 5 (municipal level) represents geographically less attractive areas with socio-sanitary disadvantage but relatively good levels of primary care providers and services.
Figure 2: Maps of the classifications of multidimensional healthcare accessibility municipalities (A), “territoire de vie santé” (B), and EPCI scales (C). A) Municipality scale
Mapping tool: QGis 3.44.4; Source: IGN, Natural Earth, 2022
B) “Territoire de vie santé” scale
Credits: Author
C) EPCI scale
Mapping tool: QGis 3.44.4; Source: IGN, Natural Earth, 2022
26To analyze the significant geographic proximity of certain municipal classes , a joincount global autocorrelation test was performed. The results revealed spatial dependence with two pairs of municipal classes showing significant autocorrelation. The positive Z score and associated p-value (p<0,05) indicate that classes 6 and 7 (light and dark blue) are geographically grouped. These classes are similar in terms of healthcare accessibility, characterized by areas with high accessibility to care services and providers. The combination of similarity in healthcare accessibility, and geographic proximity, suggests that, for these classes, differentiation in classification will be minimized when aggregating geographical units to create a supra-spatial classification. In contrast, although classes 1 and 4 are also geographically grouped, their healthcare accessibility profiles differ. Aggregating these contiguous areas will result in heterogeneous groupings at the supra-spatial level.
27Table 4 presents the Theil index to analyze disparities in the distribution of municipal classes according to their healthcare accessibility within and between “territoires de vie santé” and EPCI areas. Several key findings emerge from the analysis. First, the within-level Theil index shows that municipal classes within each EPCI are more heterogeneous in healthcare accessibility than those within each “territoire de vie santé”. The percentage of inequality in the within index differs by 5 to 8 points. Second, EPCIs and “territoires de vie santé” belonging to low accessibility or mixed healthcare accessibility classes exhibit greater heterogeneity within these areas than those in high accessibility classes. Finally, the between-level Theil index indicates that municipalities in high accessibility classes display many disparities, regardless the scale, suggesting that scale choice is less discriminating for high accessibility municipalities than for those with lower healthcare accessibility.
Table 4: Index value of inter and intra inequalities and proportion of inequalities within classification of municipalities according to their level of healthcare accessibility
|
“Territoire de vie santé” scale
|
ECPI scale
|
|
Between “territoire de vie santé”
|
Within “territoire de vie santé”
|
% inequalities within “territoire de vie santé”
|
between EPCI
|
within EPCI
|
% inequalities within EPCI
|
|
Lowest accessibility
|
0.141
|
0.253
|
64.4%
|
0.119
|
0.275
|
69.7%
|
|
Middle or mitigated accessibility
|
0.170
|
0.437
|
71.9%
|
0.131
|
0.476
|
78.4%
|
|
Highest accessibility
|
0.215
|
0.226
|
51.1%
|
0.188
|
0.254
|
57.5%
|
28To visually analyze and quantify disparities between municipal healthcare accessibility classifications, we used Sankey diagrams (Figure 3). The lines represent relationships between classes from the municipal level to supra-municipal classes. Middle accessibility classes show the greatest diversity of line colors. Among municipalities in class 3 at the “territoire de vie santé” level (middle accessibility class): 2.9% and 3.6% belong to low accessibility municipalities (classes 1 & 2), 35.3% and 13.5% belong to high accessibility municipalities (classes 6 & 7), and 20,8%, 12,6%, 11,2% belongs to mitigated accessibility municipal classes (classes 3, 4 & 5). Furthermore, 67.5% of municipalities in the highest accessibility municipal class (class 6 and 7 in blue, geographically close) also belong to the best accessibility class (class 5) at “territoire de vie santé”level. The same holds for ECPIs with 66.1% of the highest healthcare accessibility municipalities remaining in the top class. The proportion decreases for municipalities in the lowest healthcare accessibility classes which are not located geographically close: only 36.4% and 41.4% (class 1 and 2 in red and pink) remain in the lowest accessibility classes (class 1 and 2) at the “territoire de vie santé” level.
Figure 3 A&B: Sankey diagrams to diagnostic changes in classification of municipalities after aggregation. Colors and flows from right “territoire de vie santé” (A) and EPCI (B) scales to left municipality scale in France. A) Municipality classes to “territoire de vie santé” classes
Credits: Author
B) Municipality classes to EPCI classes
Credits: Author
29This research examines the consequences of scale choice in the assessment of area-based healthcare accessibility. It adopts a multidimensional perspective, as different aspects of healthcare accessibility (e.g., healthcare accessibility, healthcare needs, and supply dynamics) are summarized into seven classes at the municipal level and five classes at the EPCI or “territoire de vie santé” level. It also follows a multiscale approach, studying healthcare accessibility inequalities at different levels (e.g., population-living level and policy and decision-making levels). The measurement and understanding of contextual accessibility to healthcare providers, as well as its spatial inequality, are highly dependent on the chosen spatial scale.
30Several studies have examined multidimensional and multiscale healthcare accessibility. One previous study in Ecuador, which considered two indicators separately (a deprivation index and a healthcare accessibility index [LPA]), found no scale effect between census blocks and census tract levels (Cabrera-Barona et al., 2018). While previous studies have focused on scale effects to deprivation using composite indices (Cabrera-Barona et al., 2016; Cebrecos et al., 2018; Petrović et al., 2022), Brousmiche et al., used two indices in a region in northern France to investigate the implications of spatial unit choice in assessing environmental and social health inequalities (Brousmiche et al., 2023). Their results show that the infra-communal level highlights substantial disparities, while higher resolution analyses provide a clearer description of the health determinants, particularly in the most densely populated municipalities. Other studies have explored territorial inequalities in access to healthcare services in France through the prism of distance and spatial levels of access to healthcare services (Chevillard et al., 2018; Lucas-Gabrielli, Mangeney, 2019; Coldefy et al., 2011).
31The findings of our study are consistent with previous literature, which shows that larger spatial units generally exhibit less variability due to the averaging or smoothing of the underlying variability (Clerval, Delage, 2014; Marzi et al., 2018; Buzzelli, 2020; Brousmiche et al., 2023). The spatial distribution of classification reveals greater variability at the municipal level than tat higher scales. Mitigated classes (classes 3, 4 and 5 at the municipal level) which are useful for identifying areas with mixed accessibility (areas with long distances to healthcare providers but a strong supply of GPs, and other providers, or vice versa) disappear at the EPCI level in favor of a less informative middle class (areas with average accessibility across all providers and services). The “territoire de vie santé” level retains the advantage of maintaining an mitigated class with different configurations across healthcare providers. As highlighted by Coldefy et al. (Coldefy et al., 2011), the notions of distance and proximity vary depending on the type of healthcare professional. GPs, nurses and physiotherapists are more locally distributed and tend to reduce spatial contrasts, unlike radiologists and emergency services, which are more distant. The EPCI scale reflects this hierarchy of primary care services, whereas the “territoire de vie santé” scale captures a more differentiated structure.
32Our results show that municipalities with the highest accessibility (classes 6 and 7, in blue) exhibit significant overall autocorrelation (joincount) and are better represented at supra-municipal scales than other classes. The Sankey diagram indicates that the proportion of municipalities changing class —from high to medium or low accessibility —is lower than for other groups. Furthermore, the Theil’s index reveals that inequalities within EPCIs or “territoires de vie santé” are less pronounced for these classes than for others. These findings reflect the well-established organization of healthcare services in France based on centrality and hierarchy, as theorized by Christaller’s central place theory (Férérol, 2013; Humain-Lamoure, Laporte, 2022). We show that high healthcare accessibility classes typically include small coastal towns with a high concentration of primary care providers, and GPs tend to cluster in larger cities with strong geographic proximity, particularly where services such as laboratories and emergency services are available. When similar areas are spatially dependent, the influence of scale is reduced.
33The results show that scale strongly influences the multidimensional definition of healthcare accessibility and the comparison between places. As demonstrated by Pellé and Martin (Pellé, Martin, 2012), observation and measurement units (such as territories in geography) are not fixed, but result from social, political, and scientific choices. The MAUP clearly illustrates this issue, showing that statistical results depend on the spatial divisions used. However, when areas with a similar healthcare accessibility level exhibit spatial dependence, the influence of scale is reduced. Consequently, it is essential to assess whether the selected reporting units or spatial representations of healthcare accessibility are meaningful for end users, such as decision-makers and practitioners.
34This study provides an original contribution in the French context, offering new and relevant insights into healthcare accessibility across different scales. The multidimensional synthesis of healthcare accessibility constitutes a valuable tool for obtaining a comprehensive and reliable perspective. Our study fully benefits from this multidimensional classification, highlighting a diversity of configurations in accessibility to primary care. The classifications reveal areas combining low accessibility and low supply dynamics, as well as others with the opposite profile. Some classes present contrasting profiles, with accessibility varying according to the healthcare professions considered. These findings provide insights that previous studies based solely on indices were unable to capture (Bonal et al., 2024, Launay et al., 2024).
35Our study is not without limitations. First, individual determinants of healthcare access were not included. Waiting delays, as well as financial and spatial accessibility, should be documented, alongside more personal factors such as satisfaction with the healthcare received. To best describe the French context, the first point of contact with the healthcare system was used to select primary care providers, services, emergency departments and general practitioners. This methodological choice is consistent with our objective. However, other healthcare professionals were not considered, such as dentists, midwives, gynecologists, and speech therapists, even though they are also affected by shortages due to declining availability in France. Finally, the three spatial scales were carefully selected to best capture the territorial organization in the French context. Nevertheless, even at the municipal level —the finest scale considered —disparities within metropolitan areas are not described with sufficient precision. An additional scale would be necessary to better capture these highly urban areas. Fourth, our study area includes Corsica and the overseas departments and regions (DROMs), which have more specific characteristics than mainland areas and cannot be fully accounted for in this analysis.. A specific analysis of these territories using tailored indicators, would be relevant for future research. Fifth, the combination of PCA and HAC can be a valuable tool for exploring the multiple dimensions of medical deserts, but it also has limitations when applied to spatial data (Demšar et al., 2013). In particular, PCA assumes that observations are independent. Finally, our results show that certain classes are spatially autocorrelated, which may lead to an overestimation of the variance explained by the first components at the municipal level.
36We demonstrate that changing the geographic unit used to define access to proximity healthcare alters the diagnosis, with a loss of precision when larger spatial units are used. The spatial distribution of the classification shows greater variability at the municipality level than at higher scales. The geographic contiguity of municipalities with similar healthcare accessibility profiles helps reduce information loss. We show that it is essential to situate healthcare accessibility within its specific spatial context, as variations exist both between and within areas. This distinction is particularly relevant at a time when local actions are implemented to modify the distribution of healthcare provision, most often at the “territoire de vie santé” scale and, to a lesser extent, at the EPCI level. Based on these results and the French territorial and healthcare organization, no single optimal scale can be identified. We recommend conducting descriptive analysis at the finest geographical unit available, alongside analyses at a scale more appropriate for interventions and decision-making. Challenges arising at different scales may require distinct policy responses.
37All authors conceptualized the study. All authors interpreted the results. MB and GC acquired the data and performed the data analysis. CP drafted the manuscript. VLG supervised the project. All authors have read and approved the final version of the manuscript for publication.
38None declared.
39We would like to thank the FNORS for providing their mortality data. We received funding from the European Commission as part of the promoting evidence-based reforms on medical deserts (OASES) Program (Program 2020 of the 3rd Health Program).
-
Additional file 1 (PDF): Table of variables that are the most contributive to the third components at the municipality level; territoire de vie santé and EPCI level
-
Additional file 2 (PDF): Univariate analysis of all dimensions of the classifications