What influences national and foreign physicians’ geographic distribution? An analysis of medical doctors’ residence location in Portugal

The debate over physicians’ practice location and geographical distribution has attracted the attention of both the economic and public health literature over the last forty years [1]. The issue of uneven distribution of health professionals worldwide has long been recognised [2], and much discussion exists on the most effective strategies to attract and retain physicians in rural and underserved areas [3]. In Portugal universal health coverage is offered through the private and public service free of charge at the point of use [4]. The system has undergone considerable development in the last two decades, with the objective to improve both its accountability and efficiency [5]. Although its physician to population ratio is within the European average [6], the perceived lack of medical doctors across Portugal has been recurrently discussed in the national press [7].

The present study analyses the characteristics of the physician population in Portugal and their home residence, with the objective to identify specific factors influencing their geographical distribution, looking specifically at the possible motivations behind foreign physicians’ choices. First, a brief review of the literature is offered on the issue of physicians’ geographical location worldwide and in Portugal, followed by a description of the methods for data collection and analysis. The results section is organised in two parts, presenting first our findings on physician characteristics and distribution across the country, and then on those municipality and demographic characteristics associated with residence location. The paper closes with a discussion on strengths and weaknesses of the present work, as well as on its policy implications and areas of further research for governments aiming at addressing physician geographical distribution unbalances worldwide.

A cross-sectional observational study of physicians registered in Portugal was conducted with the objective to: (a) describe and analyse the physician population, and (b) explore the association of their geographical residence with relevant municipality and personal characteristics. Although this kind of studies is unable to prove the cause-effect relationship between variables, they are widely deemed valid to give a snapshot of the current situation, and explore the association between variables [24]. Existing guidelines on observational studies were followed to report the study findings [25].

The study focuses on the supply side of medical services, and its objective is to understand what attracts physicians to specific municipalities, and specifically foreign ones, with the ultimate aim of creating an evidence base for policy-makers aiming at attracting physicians in underserved areas. The underlying assumption of the research is to check the validity for the Portuguese physicians case of economic theories on physician location and of foreign physicians as a ‘safety net’ for distribution gaps.

A database was obtained from the PMC, providing information on the totality of the physicians registered in Portugal in 2008. The database included only aggregated data accordingly to variables such as physicians’ address, nationality, specialty, age, and country of training. Municipality of residence was considered a proxy for physician practice location. Besides having been used extensively in the literature on International Medical Graduates in the US [10, 26], in the Portuguese case, data from the National Medical Council are considered a reliable and complete source, and often feed Government statistics on physician activities (e.g. data from the Ministry of Health or from the Ministry of Labour).

Data on municipality characteristics were obtained from the Portuguese National Statistics Institute (INE), providing information on the 2008 population, number of nurses, number of hospital beds, proportion of foreigners applying for residency in the municipality, and on 2007 purchasing power per capita. Number of beds and nurses were included as proxies of health infrastructures in the municipalities. The Municipality Development Index (MDI) was used to evaluate quality of life in specific municipalities in 2008. The MDI is a weighted welfare index developed by Municípia SA©, covering areas such as municipality’s population growth, public services, environment, economic activities and infrastructures [27].

In order to analyse physician geographical distribution, municipalities were grouped according to the INE Nomenclature of Territorial Units for Statistics of greater metropolitan areas, including 18 municipalities for the Lisbon area, and 16 for the Oporto region (see Additional file 1: Table  S1). To investigate the impact of teaching hospitals (TH) on physicians’ residence, municipalities were also divided between those in the proximity of a TH (i.e. with a TH in its territory or in a bordering municipality), and those further away. In order to analyse the impact of Medical Schools’ location on physicians’ geographical distribution, municipalities were divided into those located in proximity (around) the 10 Medical School operating in Portugal at the time of the study (42 in total) and those further away (267). Specific nationality groups were created to allow the analysis of major migrant ^a and linguistic groups ^b.

Basic correlation analysis was performed to check for possible problems of multicollinearity. As the correlations were not very close to 1 or −1 (maximum correlation value was 0.647) and no increase in coefficient estimates’ standard errors was detected, it was decided to maintain in the model all the variables available from our database.

Taking into consideration the limitation on how the original dataset was presented, we were able to construct two databases at the municipality and the aggregate individual levels in order to answer the two main objectives in this study, namely to describe and analyse the physician population, and to explore the association of their geographical residence with relevant municipality and personal characteristics. The two separate analyses are described as follows.

For the first objective, we perform a negative binomial regression where the response variable, Y, is the total number of physicians (Portuguese, foreigners or both) in each municipality and the explanatory variables, X_i’s, are Pop, Beds_PC, PP_PC, Nurses_PC, Foreign_PC, MDI, Metro and TH defined in Table 1.

The model can be written as

with ε the error term, and de variable Y following the distribution of

with E ( Y ) = μ , θ the shape parameter, Γ ( . ) the gamma function, V a r ( Y ) = μ + μ 2 θ , and the link function

with β i ' s as unknown parameters.

For the second objective, the focus was on whether the variables Sex, Age_Group, Nat_Group, and Lang_1, defined in Table 1, were significant in the location of physicians. Thus, a binary logistic regression model was performed with the variables as explanatory variables and with the reference class, Y = 1 , whether a doctor resides outside of a metropolitan area (Y is the variable MT in Table 1). Due to the aggregation level of the database made available by the PMC, this analysis was performed in SPSS by weighting each combination of the set of variables (MT, Sex, Age_Group, Nat_Group, and Lang_1) by the number of doctors in each class. The model can then be expressed by:

The selection of variables in the model was based on the forward stepwise likelihood ratio method. In the following section we present the results of these approaches to the database made available by the PMC.

The present study provides a thorough analysis of physician population in Portugal, and showed that the physicians registered in the country are almost equally male and female, mostly Portuguese and between 41 and 55 years of age. Spanish, Brazilian, PALOP and other European nationalities are also a significant presence among foreign physicians. A considerable unbalance was detected in the physician-to-population ratio between metropolitan areas and the rest of the country, favouring particularly those municipalities in the proximity of a teaching hospital. Among the variables considered, population, nurses and purchasing power per capita were the most important municipality characteristics associated with physician distribution in Portugal. For national doctors, the Municipality Development Index was significant too, and so was for international doctors the proportion of foreigners applying for residence. Foreign physicians resulted, in proportion, more likely than the Portuguese to be found outside metropolitan areas. However, such effect appeared to be mostly due to the Spanish cohort, as other nationalities – especially the African PALOP - displayed location preferences similar to those of the Portuguese. Among foreigners, males and over-50s were more likely to reside outside the Lisbon and Oporto areas. Being able to speak Portuguese or Spanish as mother tongue did not seem to carry further influence on location decisions beyond nationality.

A number of limitations need to be considered when interpreting the results of our study. First of all, our ‘physician residence’ endpoint does not necessarily correspond to physicians' geographical area of practice, although it may arguably provide a good indication of it. Due to privacy concerns, the PMC database employed for our analysis did not disclose information on individual physicians, and this limited considerably the statistical analysis that could be performed. As many physicians appeared to be registered under more than one key characteristics such as specialty or years practice, this made it impossible to apply our statistical methods to explore potential explanatory variables of physician location.

A further point is that the cross-sectional nature of our data limited inference on the cause-result relationships between physicians’ residence and municipality characteristics. However, as our dataset used the most reliable and comprehensive source of information in Portugal, and as similar sort of data have been used in the past in similar studies, the authors still consider the results obtained a fair representation of physician population in Portugal. There is recognition of a need for more complete municipality data to be included in our analysis as proxies for need, such as mortality, morbidity and epidemiological data. As the impact of these may well not be fully captured by the population variable. To this respect, a request to relevant sources has been already made, and a plan exists to make use of a more complete set of variables for future work. Finally, the logistic regression model on physician characteristics to investigate the odds of residing outside metropolitan areas could only be employed for the foreign physicians population, as it displayed a poor fit for the larger Portuguese physicians cohort.

Despite its limitations, a few considerations can be safely drawn from our study. The study confirms the conventional wisdom from the literature that physicians distributions is related to the distribution of infrastructures - here represented by nurses and beds - and by population affluence –represented by MDI and PPP per capita -, possibly exacerbating the imbalance between poor, underserved areas and rich and well-provided ones. However, the study findings also suggest that physicians reside disproportionally in metropolitan areas, and that factors influencing their geographical residence may be in some respect different for national and international doctors. This appears to be consistent with the literature on international medical graduates [11, 13, 16]. For national doctors, the relevance of those municipality characteristics linked to supply-side variables– such as the MDI and PP_PC – as well as their concentration in the Lisbon and Oporto areas, may be signalling a higher discretionary power enjoyed in selecting residence location [28]. For foreign physicians, the importance of more demand-related variables - such as hospital beds and presence of foreign population – as well as their stronger presence outside metropolitan areas, may be interpreted as a more pronounced dependency of foreign physicians on public sector employment opportunities, as well as a lesser access to premium locations. If this were the case, incentives to locate in under-served rural areas may be effective policy options to redistribute Portuguese physicians, either through better public sector salaries or linked to the possibility of developing their own private practice. To foreign doctors, incentives linked to possibility to practice among, or in the proximity of, their ethnic community of origin may represent sensible solutions to attract this type of professionals in the long run.

A closer investigation of foreign physician distribution also showed that the claim that these can be used indiscriminately to fill geographical gaps is somewhat inaccurate, as rural residence patterns appeared to vary considerably among nationality groups, with PALOP physicians concentrating in metropolitan areas even more than their national peers. Such finding resonates with the existing literature on foreign physicians in the United States [16], and its implication would be that, relying on international recruitment might not be the most adequate policy to address physician shortages in rural areas. To this respect, policy-makers may consider alternative policy options such as:

The present study investigated physicians’ population and residence location across Portugal, with the objective of gaining a better understanding of the factors influencing their distribution across the country. Through descriptive and regression analysis, the study identified some of the municipality and physician characteristics displaying association with medical doctors’ distribution and their decision to locate outside metropolitan areas.

The study showed the value of collecting and analysing data on physicians and municipality characteristics, in order to design effective policies to address distribution imbalances across within a country. The study also highlighted the need for better and more detailed data on physician distribution to inform policies addressing health personnel geographical disparities countrywide, possibly through longitudinal studies.

Evaluating the impact of such policies would be an area for further research, as limited evidence seems to exist on the subject, especially with reference to the Portuguese case [33]. Qualitative investigation of physicians residence choices would also help shed light on which of these policies may be more effective than others, and under what conditions [34]. Discrete choice experiments may too provide an insight into those strategies to attract and retain health professionals not yet tested in Portugal or in a similar policy context [35].

^a Portuguese, Spanish, Brazilian, Other European, African Portuguese-Speaking Countries – PALOP.

^b Portuguese-speaking countries, Spanish-speaking ones, European, and Others.

^c University of Açores, University of Beira Interior, University of Coimbra, University of Lisboa, New University of Lisbon, University of Madeira, University of Minho, University of Oporto (2 medical schools), and University of Algarve. The newly inaugurated Aveiro Medical School was not operational at the time of the study and thus not considered to construct the teaching hospital variable.