General discussion
Whether or not financial assistance should be awarded to physicians strongly depends on the selected method and spatial unit of analysis. Policy makers often define shortage areas by calculating PPR per physician zone, for the simple reason that it is an easy calculation and offers a readily understandable measure of accessibility. The advantage of this method is that it considers both the number of physicians and the population within the zone. However, it only offers a very crude representation of accessibility to primary health care because physician zones cover too large geographic areas [22, 23, 34]. Therefore, it cannot detect local variations in accessibility.
When calculating PPR per municipality, we observe slightly more underserved census tracts. This means that when using physician zones, some municipalities are not identified as shortage areas, while in fact they should be. There are however also some municipalities that are considered underserved, while they should not be. There can nevertheless be variations at an even smaller scale (e.g. census tracts), which cannot be detected using this method. Another disadvantage of this method is that interaction across borders is not sufficiently taken into account [10, 14].
Other simple GIS methods (Dist1, Dist3, Cum5, and Cum10) are solely based on the supply (physicians), while the demand (population) is not accounted for. The results show that when using the Dist3 method, only few census tracts maintain their status as shortage area. The Cum10 method provides a result that coincides more with the official method, because both are based on the number of physicians.
FCA-based methods have the advantage of the small geographical scale of analysis at the level of census tracts, and taking interaction between population and physicians into account. From the FCA-based methods, the E2SFCA method is preferred because it accounts for distance decay by using a weight function [3, 6]. The use of this method results in more shortage census tracts compared to the official Impulseo I method. However, only 51.6% of these census tracts were originally indicated as shortage areas. This means that 48.4% of all census tracts should be seen as shortage areas, while now they are not. When geographically comparing the results of the official Impulseo I method (PPR per physician zone) with the results of the E2SFCA method, the ascription of financial assistance is very different. Despite high population densities, urban areas are mostly not identified as shortage areas because of a dense concentration of physicians. Rural and suburban areas are often considered as shortage areas because physician accessibility is low. When using the official Impulseo I method, this pattern is less pronounced, because extreme values are filtered out. This aligns with the findings of Apparicio [16] and McGrail [21], who found that most accessibility problems occur in suburban areas, with low population density and mostly non-residential land use. Interestingly, however, the defined shortage areas follow the distribution of physicians much better when using the E2SFCA method.
The total number of census tracts where financial assistance should be awarded when a physician settles there is slightly higher with the E2SFCA method, so more money would be needed to invest in helping underserved areas. However, approximately the same population (33.1%) and a much bigger area (60.2%) is reached. Therefore, we would advice policy makers to use this method in future evaluations of accessibility to primary health care, because it aligns better with the actual distribution of physicians. In this way, and according the spatial analysis, the current policy in Belgium could be adjusted towards a more area-oriented approach.
Additionally, we want to propose a different way of awarding financial assistance to physicians settling in shortage areas. Now, shortage areas are defined based on a sharp threshold (PPR <90 physicians/100,000 inhabitants). Alternatively, one could vary the financial award in function of the magnitude of shortage (see Figure 6 for an example). The higher the shortage, the higher the award a physician receives when settling there. Doing so, unequal accessibility to primary health care would possibly be conquered even more effectively, since more underserved areas would have a higher attraction to physicians.
Study strengths and limitations
This study has several strengths. First, most previous studies using FCA-based methods use the centroid of the municipality where physicians live as physician location [2, 3, 14, 27], whereas we use the exact location of physicians, leading to more accurate estimations of accessibility and reducing the influence of the MAUP.
Second, distance in this study has been considered following the street network, instead of following a straight line. In many studies (e.g. [2, 16]) the lack of using street network data is considered a major limitation.
Third, the study area (Belgium) is larger and more populated relative to other applications of FCA-based methods in the context of accessibility to primary care. Our study area measures 30,528 km2 and has 10.8 million inhabitants, whereas in other studies the spatial coverage was limited to 19,774 km2 and 3.8 million inhabitants (nine counties in central Texas, USA; [25]), 14,331 km2 and 1.6 million (9 counties surrounding DeKalb in northern Illinois, USA; [2, 3]), 4,258 km2 and 3.4 million (Montreal census metropolitan area, Canada; [16]), 499 km2 and 1.9 million (island of Montreal, Canada; [30]), and 177 km2 and 601,000 (Washington DC, USA; [14]). Two studies have bigger study areas, but a lower population: 230,000 km2 and 1.5 million inhabitants (rural Victoria, Australia; [34]), and 227,000 km2 and 5.5 million inhabitants (Victoria, Australia; [21]).
Fourth, the proposed study adds to the spatial coverage of evidence by spatially complementing existing studies that have been carried out primarily in North America (e.g. [2, 3, 14, 16, 25, 30]) and Australia (e.g. [21, 34]) with evidence from Europe.
Fifth, previous studies (ibid.) are all regional, while ours is nation-wide. A disadvantage of a regional study is that there can occur edge effects, because people can also go to a physician in a neighbouring region [2]. Our nation-wide study limits this, because it is less likely that inhabitants of Belgium will go to a doctor in a neighbouring country. Small edge effects can still occur within Belgium however. Belgium is separated in two regions with different languages, which implies that people prefer to go to a physician that speaks their native language. It was however difficult to control for this, because the language of physician and aggregated population was not known and there is a lot of bilingualism along the borders between the two regions. Also, with our nation-wide study, we can link our results with the conducted policy of the entire country to check whether the policy decisions correspond with the scientific results.
However, this study also has some limitations, most of which constitute interesting avenues for future work. First, accessibility is considered from the home location. However, people can also access primary health care from their working location, which can influence accessibility [35–37]. Nevertheless, in Belgium people shall probably be inclined to go to a physician in their residential neighbourhood whom they are familiar with, rather than searching for a physician near their work location.
Second, according to some studies, the size of the catchment should vary depending on whether it is urban or rural [3, 29, 34]. Despite the small differences between urban and rural populations in Belgium, adding a varying catchment size function (larger catchment sizes for rural populations) could improve the results.
Third, the population per census tract is now centered at its centroid. This is more accurate than looking at a scale level of a municipality or physician zone, but still is an approximation of reality. To improve this, one could consider each home location as a population location, from where accessibility is calculated. However, such data is often not available because of privacy issues and the calculation would be very computationally intensive.
Fourth, various socio-economic factors can also influence accessibility to primary health care [27]. Several studies have considered such factors as financial barriers, car-ownership, and educational level [10, 38–42]. Also, data about the actual use of health services could provide information about revealed accessibility, instead of potential accessibility what is studied now. However, collecting this data is expensive [2], definitely at the scale of our study. Socio-economic attributes of physicians (e.g. ethnicity, gender, age) could also provide interesting information. This could however be incorporated in future research. Gender could be accounted for since the sex of a physician is known to be a barrier for certain population groups (e.g., young women [43]). Age could be dealt with because it will enable to identify and anticipate future shortage areas (i.e. areas that are likely to become underserved because of ageing physicians). Some other factors could also be incorporated in future research concerning this topic: e.g. the fact that physicians can also visit patients, visiting hours of physicians, average visit length which can vary per physicians, and congestion problems along the road network.