General practitioners in German metropolitan areas – distribution patterns and their relationship with area level measures of the socioeconomic status

We documented an unbalanced spatial distribution of GPs in terms of GP density in German metropolises on intra-city level. This finding is in line with current literature: A study conducted within Adelaide (Australia) found similar distribution patterns: 16% of residents were reported to live in an area of GP shortage within Adelaide boundaries [26]. An unbalanced distribution of physicians was further reported for Auckland (New Zealand) [27] and Toronto (Canada), where the authors reported a range of 6.18–12.42 GPs per 10^4 residents for the city districts of Toronto [28]. In our study, the number of FTE per 10^4 residents ranged between 1.80–43.98 and therefore variation was higher compared to the Canadian study. However, we included 14 cities whereas the aforementioned studies only reported results from individual cities, which increased the likelihood of data variability. Still, our results can be considered more valid due to the bigger sample size.

Furthermore, the extent of the demonstrated distribution varied between the analysed cities. For example, the high variation in Munich could be due to a greater competition between GPs, which translated into higher GP densities near the city centre and vice versa. However, it has to be noted that the number of districts varied between the cities and especially the number of districts in Munich used in this study was higher compared to all other cities. This could explain the higher variation in Munich.

On inter-city level, the distribution of GP densities was balanced. In other words, there was only little variability in GP density data. We assume that this balanced GP distribution pattern was due to strict German health care regulations, which restrict GP numbers in certain areas such as larger cities. Such areas are defined “oversupplied” as soon as the population-to-GP ratio exceeds 1:1671 by 10% (§ 11 (4) of the ‘directive of capacity planning’). In this case, GPs will not receive permission to establish practices in these areas.

In general, studies have shown that GPs as well as other medical specialties tend to be located in urban areas [5, 6]. The data presented in our study revealed higher GP densities on city level: in our study the overall physician-to-population ratio (PPR) was 1:1550 (SD: 139.3). Compared to the aforementioned threshold value of 1:1671 (the ‘optimum ratio’) the ratio was exceeded by 7.8%. Therefore, applying the definition of oversupply mentioned above, these cities cannot be considered oversupplied. However, it has to be noted that within national capacity planning, adjustments can be made according to demography and/or geography. Regarding the non-physician participation in primary care, they were shown to constitute the smallest share of primary care workforce in urban areas and were more likely to delivery primary care in rural areas [29]. Therefore, lack of primary care services in more rural areas could be compensated by non-physician workforces. This solution to GP workforce shortage has also been reported by recent literature [3, 4].

In addition to the aforementioned variation of district numbers, differing district sizes could conceal further variation. However, city district sizes and numbers are not standardized in Germany, which limits the inter-city comparison. Therefore, the heterogeneity of district sizes and numbers could have concealed inter-city variations.

Using socioeconomic indicators, a correlation with GP densities was present that could explain GP location choices on city district level. In regard to the GDI, a mild negative correlation was present (r = −0.49) indicating a negative influence on GP location choices. Looking at the factors comprising GDI, the most important factor was a spatial factor (‘distance to the city hall’). Therefore, our results suggested spatial factors to be more important than non-spatial factors. However, in a survey [23] among 117 physicians in Germany, 81% of participants stated that neighborhood characteristics concerning the suitability for children and families was the most important factor regarding practice location choice. The high impact of the ‘significant other’ on GP location choices has also been reported by Smith et al. [30]. Furthermore, spatial factors such as ‘proximity to work place of significant other’ or ‘proximity to city center’ were reported to only play a minor role regarding GP location choices [23]. However, a study in Finland [31] further reported ‘location of workplace’ and ‘being near a central hospital’ as the 2nd and 6th most important factor for the GPs workplace choice. In addition, proximity to other health care providers (e.g. dentist, pharmacies, hospitals) was further reported to influence location choices of GPs [31, 32]. In our study, pharmacies, dentists, hospitals and the work place of GP spouses were not measured and therefore not controlled for. In summary, it can be stated that spatial factors as reported in our study are important factors influencing the practice location choice of GPs. However, taking further research into account, non-spatial factors seem to have an even greater impact compared to spatial factors. The non-spatial factors comprising the GDI, namely household size and population density, were also used as high loading factors in the urban index (UX) created by Schulz et al. [24]. In their study, no relevant correlation (r = −0.13) between UX and the demand for GP services was present. Considering both UX and GDI, the reported findings suggested that household size and population density influence the GP practice location choice (i.e. provider supply), but have no influence on the populations’ demand for health care services. Looking at other non-spatial factors, Gosden et al. [33] reported ‘aversion to location in areas of high deprivation’ as the most important factor influencing GP practice location choices. In our study, deprivation was mostly reflected by CEI. However, a weak correlation (r = −0.22) was found, which suggests only a minor influence on practice location choices. Further factors include financial incentives often represented by the percentage of the population with private health insurance, which has been shown to explain 14% of the GP density variation in Germany, while ‘health care needs’ only explained less than 5.2% [34].

It has to be noted that location choices of GPs can be influenced by both external and internal factors. Internal factors such as personal characteristics have been shown to play an important role in the workplace selection process [35]. However, in our article we did not control for internal factors and focused solely on external factors defined by neighborhood statistics.

In regard to mortality, our study found no correlation with the spatial distribution of GPs. This finding is in line with current literature which showed no significant difference of mortality (deaths per 100 beneficiaries) in regard to differing GP densities [36].

Finally, many countries limit the GPs’ practice location choice in order to sustain equal access, regardless of a patients’ living location. The degree of this limitation usually depends on the performance of the health care system in place: The more sophisticated the health care system, the more regulations are in place, which in turn limits the GPs practice location choices. China for example has a rather weak primary care system, whereas the United Kingdom and Germany have a rather strong primary care system [6, 22]. Bearing this in mind, our results only reflect the situation in a developed health care setting.

Although GPs do not fulfill a formal compulsory gatekeeper function in Germany, their availability and accessibility play a crucial role in a populations’ access to health care [6, 37]. Hence, it can be assumed that an unbalanced distribution of GPs could translate into health inequities [38, 39]. A mismatched distribution of GPs constitutes a major challenge for public and community health. Thus, tailored strategies and policies are desperately needed to address these challenges.

Population-to-provider ratios (PPR) have limitations in regard to the measurement of spatial accessibility [16, 40]. As outlined by Guagliardo, limitations of PPR include 1) not accounted border crossing, 2) blindness to variabilities within bordered areas, 3) omission of distance/time and 4) fixation to geographical/administrative boundaries [16]. Especially the omission of border crossing in the presented approach represents a limitation. More sophisticated measurements, which addressed these issues, are based on gravity models such as the Two Step Floating Catchment Area (2SFCA) method [41]. Since in Germany the population-to-provider ratio is still widely used in research and in the political context, we used a PPR despite the known limitations. However, we acknowledge the great value of more sophisticated measures using geographic information systems. Furthermore, using airline distances instead of time distances, represents a distance measurement simplification and therefore a loss of accuracy must be assumed. However, airline distances have also been used in similar research [21].

In addition to the aforementioned limitations, there are four limitations that have to be further addressed: First, there is no best indicator for measuring SES. Residual confounding by unmeasured socioeconomic factors can only be minimized, not excluded. On area level, different single measurements or composite indicators have been established (e.g. English Index of Multiple Deprivation [42], Bavarian Index of Multiple Deprivation [43], Townsend Deprivation Index). However, due to insufficient data availability, none of the described composite indicators could be used for the nationwide inner-city analysis in Germany.

Furthermore, city district data were heterogeneous and therefore the accuracy of the results could be compromised. In regard to GP practice locations, the allocation to city district boundaries has limitations due to possible errors retrieving geo coordinates [44].