Background: Differences in health care utilization across geographical areas are well documented within several countries. If the variation across areas cannot be explained by differences in medical need, it can be a sign of inefficiency or misallocation of public health care resources. Methods: In this observational, longitudinal panel study we use regional level data covering the 21 Swedish regions (county councils) over 13 years and a random effects model to assess to what degree regional variation in outpatient physician visits is explained by observed demand factors such as health, demography and socio- economic factors. Results: The results show that regional mortality, as a proxy for population health, and demography do not explain regional variation in visits to primary care physicians, but explain about 50% of regional variation in visits to outpatient specialists. Adjusting for socio-economic and basic supply-side factors explains 33% of the regional variation in primary physician visits, but adds nothing to explaining the variation in specialist visits. Conclusion: 50–67% of regional variation remains unexplained by a large number of observable regional characteristics, indicating that omitted and possibly unobserved factors contribute substantially to the regional variation. We conclude that variations in health care utilization across regions is not very well explained by underlying medical need and demand, measured by mortality, demographic and socio-economic factors. Keywords: Regional variation, Health care utilization, Demand, Panel data, Random effects Background care drives these regional variations due to the depend- Levels of health care utilization and expenditure differ, ence between determinants [4, 5]. not only among individuals but also across geographical Factors known to determine health care utilization on areas within the same country [1, 2]. From a health pol- an individual level include demand factors such as health icy perspective and concerning equity in care, it is fun- status, demographic and socio-economic background damental to establish if the variations in health care . Factors on the supply side that may drive regional utilization and expenditure across geographical entities variations in health care include access to care, medical are justified based on underlying demand factors, or if practice and provider characteristics [1, 7]. If the popula- they are a sign of inefficiency, misallocation of public tion of an area has worse health status and a higher health care resources, or over- and underuse . How- medical need, we may expect a higher level of health ever, it is well documented that it is difficult to distin- care utilization. On the other hand, if differences in guish to what degree “demand” and “supply” of health medical practice and access to health care are the main drivers of regional variation, the variations are unwar- ranted and the health care system is not fulfilling the legislators’ intentions [3, 8, 9]. * Correspondence: email@example.com Despite a large number of studies demonstrating geo- Health Metrics, Sahlgrenska Academy, University of Gothenburg, PO Box graphical variations in medical practice and health care, 463, SE-405 30 Gothenburg, Sweden Full list of author information is available at the end of the article © The Author(s). 2018 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Johansson et al. BMC Health Services Research (2018) 18:403 Page 2 of 9 assessing what factors drive the variations are less stud- it is purposeful to evaluate different types of health ied [2, 3]. Cutler and Sheiner  used aggregated care separately as determinant factors are likely to cross-sectional data of US Medicare expenditure and differ [15, 19]. In the setting of a public, universal concluded that health status and demography accounted health care system, and using regional level data from for 45% of the variation (measured using the standard national population and health care registers, we analyze deviation of residuals) in spending across hospital refer- regional variation in the general population. From an ral regions. Several of the more recent studies conclude international perspective, the Swedish system is homoge- that supply-side factors, such as physician preferences, neous with low levels of private insurance, few providers style of practice and incentives, have more importance outside the public reimbursement system, and low patient than the demand side in explaining geographical vari- cost sharing, which limits the effect of some potential con- ation in health care [1, 4, 5, 10]. But, there are also other founders (price, diverse insurance terms, and access). The studies arguing that demand factors play a small but sig- main reason to assess geographical variations on the ag- nificant part in explaining regional variation , or gregation level of regions, as opposed to for example mu- even explain a large share of the variation [5, 12]. There nicipal level or hospital level, is the decentralized structure are studies that also find that most , if not all , of of the Swedish health care system, described below, where regional variation in health care can be explained by ob- the executive power lies with the regional governments. servable characteristics of demographic, employment, In the universal health care system of Sweden, the cen- health status and infrastructural factors. tral government sets principles and guidelines, whereas In sum, we see somewhat conflicting empirical evi- the responsibility to finance and provide health care lies dence on whether the demand side or the supply side is with each of the 21 regions (county councils). Propor- the driving force for regional variation in health care. tional taxes are set and collected within each region to This relates back to whether geographical differences fund the health care services. Providers are either public can be justified based on underlying demand factors or private, but private providers are typically reimbursed such as medical need or if they can be seen as a misallo- through the public funds. Upon visits, patients are cation of common resources on the supply side. The charged a (relatively small) copayment. Primary care purpose of this study is to examine the variations in out- physicians (general practitioners) are usually employed patient health care utilization across the 21 Swedish re- at a primary health care center, which is the first port of gions. We explore to what extent regional variation of call for (non-acute) health related problems. Patients are outpatient physician visits can be explained by common assigned to a specific primary care center based on geo- observable demand factors such as health status, demog- graphic proximity, but are free to register at another raphy and socio-economic structure. Similar to previous center of their choice. Specialists are usually employed authors (e.g. [7, 13, 15]), we add covariates successively at hospitals and may be involved in both out- and in- and estimate how much each set of covariates addition- patient services. Supply is rationed by gate keeping and ally explains the regional variation in physician visits, waiting times, and to allocate the patients’ priority the considering the estimated standard deviation of official rule is to follow the principle of need. Gate keep- region-specific random intercepts. Furthermore, we ing is applied by the use of a phone triage system to pri- evaluate how much of the regional variation remains un- mary care, and by the use of referrals to specialist care. explained after having adjusted for demand, and some In international comparisons, Swedish health care fares basic supply-side factors. well on most health outcomes such as low levels of in- Our study contributes to the literature by focusing on fant mortality and high life expectancy, but accessibility the demand side, in contrast to a large part of the previ- and long waiting times, which differ slightly across re- ous literature where there has been a focus on gions, are often-debated issues where the health care supply-side factors (e.g. [4, 16–18]), probably in part due system comes out poorly. to difficulties to observe and measure factors on the demand-side . Additionally, we contribute by asses- Methods sing regional variation in health care in the level of Variables and data sources utilization, as opposed to expenditure as in many previ- The panel data set of the 21 Swedish regions (corre- ous studies have done (e.g. [4, 10–13]). This gives the sponding to Eurostat NUTS level 3 ) covers the advantage of not having to adjust for price and structural period from 2001 to 2014. Regional borders have conditions associated with expenditure. Using visits to remained the same throughout this period, except a outpatient physicians as the outcome of interest have transfer in 2007 of one small municipality to the neigh- the advantage that it is a well-defined internationally boring region, which is unlikely to affect our analysis. comparable measure and that the vast majority of health The data is gathered from online national registers of care service takes place in outpatient care. Furthermore, the Swedish municipality and region database, Statistics Johansson et al. BMC Health Services Research (2018) 18:403 Page 3 of 9 Sweden, the National Board of Health and Welfare and, individual health behavior and efficiency in health the Swedish Association of Local Authorities and Re- self-production, and thereby the motivation and capacity gions. The data included are aggregated at the regional to seek health care. Covariates of supply include number level from population registers (except unemployment, of primary care centers (per 100,000 inhabitants), pro- which is based on a sample survey) or health care regis- portion of non-public providers and physicians (per ters, thus without any individual research subjects in- 1000 inhabitants). Variables on the supply side are volved ethical review by the ethics committee was not expected to affect the level of utilization as higher acces- necessary. sibility facilitates utilization, the possible presence of The focus of the analysis is on level of utilization of supplier induced demand and the interdependency be- health care services (see e.g. [5, 15, 19]), rather than tween supply and other included explanatory covariates using expenditure as the outcome variable, as is the case . Also, for supply covariates, the issue of endogeneity in many other studies (see e.g. [4, 10, 12, 13]). Given the is present as higher (lower) levels of utilization (due to level of medical need and health care services used, re- need and/or demand) may increase (decrease) the gional variation in expenditure may arise due to regional number of providers, and we need to be careful not to labor market differences causing higher (lower) costs in interpret the relationship between supply and utilization some regions, therefore it makes sense to directly assess as causal. utilization. We assess two outcome variables: per capita The patient out-of-pocket price is theoretically an im- visits to physicians in primary care (general practi- portant determinant of demand. In some recent Swedish tioners) and per capita visits to specialists in outpatient studies it has been shown that out-of-pocket price has care. The outcome variables are measured as the total little effect on health care utilization, at least on an ag- number of physician visits occurred in the region, di- gregated level [21, 22]. The out-of-pocket price of visits vided by the regional population. The purpose for separ- to primary physicians and to specialists are included in ating visits to primary physicians and visits to specialists sensitivity checks. is because (i) previous authors have found determinants Table 1 shows descriptive statistics for included vari- of variation and the degree of explanation to differ for ables. Due to restrictions of years covered for a few vari- different levels of health care [15, 19], and (ii) the devel- ables, the regressions of visits to primary physicians opment over time and variation across regions differ cover the period 2002–2014 (total 273 region-year ob- (see graphs in Additional files 1 and 2). Numbers include servations) and the regressions of visits to specialists planned and non-planned visits, somatic and psychiatric cover the period 2001–2013 (total 273 region-year ob- care. servations). See Additional file 3 for a table specification Determinants of health care use are grouped into four of variable units, data sources and years covered. categories: mortality, demography, socio-economic structure, and supply. We use the regional mortality rate Description of health care utilization in Sweden (standardized by age and gender of the average popula- In terms of physician visits per capita, health care tion) as a proxy for general level of health. The demo- utilization in Sweden is low in international comparison. graphic covariates include proportion of women, In 2014, the per capita number of physician’s consulta- proportion of seniors 65–79 years, proportion of seniors tions was e.g. 4.3 in Norway, 6.3 in France, 7.6 in Canada 80 years or older and, proportion of foreign born. To ad- (2013), whereas it was only 2.9 in Sweden . Separat- just for health (medical need), previous authors have ing visits into primary and specialist care, the national used self- or physician assessed morbidity in combin- per capita has fluctuated around 1.4 throughout the ation with standardized mortality rate or demographic 2000s for both levels of physician’s care (see graph in measures (see e.g. [5, 13, 15]). Mortality rate alone is a Additional file 1). Across the 21 regions of Sweden, the rather rough measure of health, but it is an objective, number of visits to physicians in primary care ranged comparable measure based on high quality data of the from 1.2 to 1.6 per capita in 2000, and from 1.1 to 2.0 whole population. Using health as a determinant of per capita in 2015. As shown in the left panel of Fig. 1, health care utilization assumes that health is exogenous, there are only a few observations (actually two regions, a simplifying assumption. Obviously, no claims of causal the capital region of Stockholm and a small southern re- inference can be made but the aim of the analysis is to gion, Halland) that have considerably higher values dur- estimate the extent to which regional variation can be ing the second part of the period. Median, inner quartile explained by included covariates. range and min–max range of the box plots vary over the The covariate set of socio-economic structure includes years, but not in an increasing or decreasing manner. educational level, gross regional product (GRP) per See graph in Additional file 2 for a display of the devel- capita, financial assistance and unemployment. These opment over time for each region separately, where no factors, are included because they are likely to affect obvious geographical patterns over time are seen. Johansson et al. BMC Health Services Research (2018) 18:403 Page 4 of 9 Table 1 Descriptive statistics of included variables, from 2001 to 2014 Mean Min Max St.dev. Overall St.dev. Between St.dev. Within Health care utilization Visits to primary physician (per capita) 1.37 0.99 2.02 0.17 0.16 0.08 Visits to specialist (per capita) 1.21 0.86 2.28 0.23 0.21 0.09 Mortality Mortality rate (per 100,000) 963.34 767.34 1155.01 77.94 48.59 61.80 Demography Women (%) 50.12 49.02 51.03 0.37 0.35 0.16 65–79 years (%) 13.74 9.58 17.93 1.73 1.31 1.16 80 years or older (%) 5.65 3.88 6.51 0.57 0.57 0.15 Foreign born (%) 10.86 3.88 23.43 4.12 3.89 1.59 Socio-economic structure Education lower (% 23.76 15.33 32.40 3.54 2.37 2.68 with only primary educ.) Education intermed. (% with 46.96 37.71 52.66 3.04 3.06 0.55 secondary educ. at highest) Education higher (% with some 27.75 19.60 43.48 4.81 4.33 2.29 type of tertiary educ.) GRP/capita (SEK 1000) 328.92 241.79 571.00 52.76 48.59 22.96 Financial assistance (SEK 1000) 1.02 0.45 1.82 0.27 0.24 0.13 Unemployment (%) 6.89 2.47 11.00 1.89 0.89 1.68 Supply Primary care centers (per 100,000) 13.14 7.81 22.76 3.05 2.92 1.06 Non-public primary care (%) 25.31 0.00 67.31 17.03 14.65 9.21 Density of physicians (per 1000) 2.95 2.21 4.27 0.41 0.33 0.24 Copayments Copayment primary (SEK) 147.01 99.77 209.69 26.03 18.90 18.33 Copayment specialist (SEK) 284.15 199.64 350.00 36.41 25.62 26.42 Prices in 2014 price level, 1 SEK ≈ €0.10 For visits to outpatient specialists, the region averages (lower panels in Fig. 2), the relative difference across the ranged from 0.9 to 2.2 in 2000 and from 0.9 to 1.6 in regions in 2000 ranged from 35% below to 50% above 2015. The right panel of Fig. 1 show that the variation the national mean, but in 2015 the range was ±30% of between regions has decreased over the period; both the the national average. In general, low-utilizing regions re- inner quartile range and the min–max range seem to di- main at the bottom and high-utilizing regions remain at minish over the years. There are more outliers among the top, although there are some regions that have chan- specialist visits than among visits to primary physicians; ged considerably over this time period. again, the capital region of Stockholm has the highest values almost every year, but two large urban regions Empirical approach with major university hospitals (Uppsala and Skåne) also We apply a random effects model, estimated by GLS, stand out. with model specification Displaying only two years from the panel separately gives the advantage of marking out each region, enabling y ¼ α þ βX þ δ þ ε ð1Þ i it it it a comparison of regional ranking over the years. Figure 2 2 2 (top panels), shows how the variation between regions in where δ Nð0; σ Þ; ε Nð0; σ Þ. i it δ ε visits to primary physicians is greater in 2015 than in y is the outcome variable, per capita visits to it 2000. In 2000, the relative difference across the regions physician, in region i in year t. X is a vector of explana- it ranged from around ±15% of the national mean, whereas tory variables grouped in the four covariate sets as shown in 2015 the range had increased to about ±30% of the in Table 1 (mortality, demography, socio-economics, national average. For specialist visits in outpatient care and supply). δ is a random effect of unobserved, i Johansson et al. BMC Health Services Research (2018) 18:403 Page 5 of 9 Primary physician visits Specialist visits 2.2 2.2 2 2 1.8 1.8 1.6 1.6 1.4 1.4 1.2 1.2 1 1 .8 .8 Fig. 1 Spread of region per capita number of visits to physician in primary care and to specialist. Note. Outliers defined as observations more than 1.5*(inner quartile range) away from the 25th resp. 75th percentile time-invariant disturbance, allowing for region-specific socio-economics, 11% of regional variation is explained, intercepts, and ε is an error term . The random and including supply the added covariates together ex- it effects, δ , is assumed to be normally distributed with plain 33% of variation on regional level, leaving two thirds of regional variation unexplained. mean zero and variance σ . The variance is esti- In Model 2, of visits to specialists, the estimated re- mated by σ , yielding a measure of the (unexplained) gional variation is larger than that of visits to primary regional variation. Due to a limited data set of 21 re- physicians (Table 2). The mortality rate reduce σ^ by gions, we use a stepwise backward elimination of co- 18%, and adding demographic covariates explain another variates with p-values > 0.2 . We estimate two 31% of the regional variation. Including socio-economic models, Model 1 with visits to physicians in primary and supply covariates marginally adds to the degree of care as the dependent variable; and Model 2 with explanation, and in total included covariates account for visits to specialists in outpatient care as the 50% of the regional variation. In both Model 1 and 2, dependent. The two models are estimates five times about 70% of the total variation after adjusting for covar- each, adding covariate sets successively. By looking at iates is found at the regional level, the remaining 30% how the standard deviation of the random effects, σ^ , being variation over time (Additional files 4 and 5). is reduced in successive regressions, we can assess In robustness checks, the results do not change from the amount of regional variation explained. The stat- including neither year-dummies nor cluster robust istical analysis is performed using Stata software 14.1. standard errors. Excluding Stockholm County Council from the analysis, the unadjusted variation in smaller Results and less of regional variation is explained by included Assessing the stepwise backward elimination of covari- covariates (19% and 43% in Model 1 and 2 respectively). ates, we include eight covariates in Model 1 and nine co- Including patient out-of-pocket price introduces more variates in Model 2. Mortality rate, as a proxy for the regional variation in Model 1 but explains additionally general level of health, is kept in both models despite be- 5% of regional variation in Model 2. (Results from ro- ing non-significant. In Model 1, of visits to physician in bustness checks are available on request). primary care, the mortality rate reduce σ^ only by 1%, and adding demographic covariates actually introduces Discussion more variation (Table 2, for full regression results see ta- Regional differences in health care utilization in Sweden bles in Additional files 4 and 5). Including covariates of are, as in many other countries, of relatively large Per capita physician visits Johansson et al. BMC Health Services Research (2018) 18:403 Page 6 of 9 Primary physician visits: 2000 Primary physician visits: 2015 Halland Stockholm Gävleborg Halland Kalmar Gävleborg Stockholm Kalmar Skåne Skåne Norrbotten Blekinge Gotland Jämtland Jönköping Kronoberg Norrbotten Värmland Kronoberg Jönköping Västmanland Dalarna Blekinge Västra Götaland Sörmland Västmanland Värmland Östergötland Dalarna Gotland Uppsala Örebro Örebro Västernorrland Jämtland Västerbotten Västernorrland Sörmland Västerbotten Uppsala Östergötland -40 -30 -20 -10 0 10 20 30 40 50 -40 -30 -20 -10 0 10 20 30 40 50 Specialist visits: 2000 Specialist visits: 2015 Stockholm Stockholm Skåne Gotland Uppsala Uppsala Kronoberg Skåne Östergötland Örebro Västra Götaland Västerbotten Gotland Halland Sörmland Kalmar Blekinge Blekinge Jönköping Västmanland Örebro Jönköping Halland Värmland Kalmar Dalarna Västerbotten Västra Götaland Västmanland Östergötland Värmland Sörmland Norrbotten Gävleborg Gävleborg Kronoberg Västernorrland Västernorrland Jämtland Norrbotten Dalarna Jämtland -40 -30 -20 -10 0 10 20 30 40 50 -40 -30 -20 -10 0 10 20 30 40 50 Percentage deviation from national mean Percentage deviation from national mean Fig. 2 Relative differences (percentage deviation from national mean) between region means of visits to physicians in primary care and visits to specialists respectively. Note. Zero represents the national mean number of visits, and the scale is the percentage deviation from the national mean. The national mean is heavily dependent on Stockholm’s figures as Stockholm holds about 2 million (20%) of Sweden’s population magnitude. If the capital region of Stockholm had socio-economy and supply, at least 50% of regional had the same per capita number of visits as the differences remain. lowest utilizing regions in 2015, it would correspond Our results for visits to specialists are in line with pre- to 2 million visits less in Stockholm primary care, vious findings indicating that demography and health ex- and 1.5 million visits less in specialist care (now a plain 45–55% of regional variations in health care total of 4.4 and 3.5 million visits, respectively). Even utilization [5, 19] and in health care expenditure [7, 13], after adjusting for regional mortality, demography, while our results for visits to primary physician deviate Table 2 Estimated regional variation and degree of explanation Model 1: Visits to primary physicians Model 2: Visits to specialists σ^ %of σ^ explained σ^ %of σ^ explained δ δ δ δ Unadjusted 0.1597 – 0.2152 – Adjusted for: Mortality 0.1580 1.1% 0.1761 18.2% +Demography 0.1652 −3.4% 0.1086 49.5% +Socio-economy 0.1418 11.2% 0.1025 52.4% + Supply 0.1064 33.4% 0.1069 50.3% σ^ −Estimated standard deviation of random effect δ (variation on regional level) δ i %of σ^ – percentage of regional variation explained by included covariates δ Johansson et al. BMC Health Services Research (2018) 18:403 Page 7 of 9 from these. The different measures of outcome (type of number of patients listed, in comparison to other re- expenditure or utilization) make direct comparison to gions’ reimbursement schemes . some of these studies difficult. In contrast to our find- The deviating reimbursement scheme in Stockholm, ings, Kopetsch and Schmitz  find socio-economic with some incentives for providers to encourage add- structures and supply additionally explain regional itional visits, may contribute to the presence of supplier variation in specialist visits. And for visits to primary induced demand, which traditionally has been seen as a physician, Kopetsch and Schmitz  find demography theoretical cause of regional variation in health care . and health explain a large portion of the regional But, it is very difficult to causally show the existence of variation, while we find demography and mortality are supplier induced demand without details on e.g. patient irrelevant for variation in primary care. preferences. With present study design, we cannot Our study differs from Kopetsch and Schmitz in confirm the existence of supplier induced demand, and several aspects (setting, type of data set, size of area con- we cannot rule it out either. Cutler et al.  showed sidered, measure of regional variation) which might while organizational factors matter, physician treatment explain the deviating results. It is also important to ac- preferences are more important for regional variation in knowledge the difference in using only mortality as our health care expenditure. measure of health status, while Kopetsch and Schmitz The outcome measures are based on physician visits  include a composed measure of morbidity, see fur- made in each region, irrespective of patient’s residence. ther discussions below. Our findings that mortality and This is a limitation in our analysis as there may be demography explain a larger share in specialist visits spillover effects of people consuming health care outside compared to primary care might be related to the insti- their home-region. The main reason for patients to con- tutional settings of gate keeping and waiting times, sume care outside their home-region is unplanned pri- which is applied stricter in specialist care. And our find- mary or acute care, for example due to a sudden illness ings that supply-side factors, measured as the per capita or an accident . In 2016, the ratio of specialist visits number of providers and physicians, add to the degree consumed by region-inhabitants to visits produced in of explanation in primary care but not in specialist care, the region, varied between 0.3–3.6 across the regions is probably primarily related to accessibility, as more . However, this does not seem to be associated with choices usually are available in primary care. overall regional variations in outpatient specialized care Our findings show that 50–67% of regional variation when controlling for other covariates (results available in physician visits remains unexplained by included co- on request). A previous study modeled spatial effects variates and imply that we cannot rule out that at least between the 402 German counties and found spillover part of the variations may be related to misallocation of effects to be negligible . It would have been of inter- resources or inefficiencies. But for visits to specialist we est to take into account and measure the variations show that since 50% of variations are explained by mor- within regions, such as nesting on municipal level and tality and demography at least half of the variations are nesting on hospital and primary care center level. Due not due to inefficiencies. We cannot make the same to limitations in data availability we were not able to do claim for primary physician visits since mortality and this type of analysis. demography do not explain the regional variations. Using the mortality rate along with a set of demo- The large proportion of variation remaining unex- graphic covariates to account for the regional level of plained also suggest that other omitted factors as well as health and medical need has its limitations, and can be measurement errors in included covariates may contrib- seen as a measurement error. The additional inclusion ute to the variation. Potentially important determinants of some measure of morbidity, such as self- or are for example cultural context (including social norms physician-assessed health status from surveys or insur- of health care seeking behavior), rurality, and ance records, may capture the general health status bet- supply-side specific institutional settings that differ ter. The disadvantage of self- or physician-assessed across regions. It is possible that patients and phys- health status is the risk of selection bias, as those visiting ician in a region develop a regional culture, poten- a physician are more likely to be diagnosed and to be tially due to system incentives. Stiernstedt  aware of their underlying health status. The mortality pointed out that the deviating reimbursement scheme rate alone is in a sense a crude measure of health, but it of Stockholm County Council may have had a major is an objective, comparable measure, and in our case influence on the large increase in primary physician based on quality register data of the whole population. visits in Stockholm seen from 2007 and onward. In In a random effects model we assume independence Stockholm, a larger proportion of the reimbursement between included covariates and the random intercepts, toprimarycareproviders wasattributedtothe just as the error term is assumed to be independent . number of visits and a smaller proportion to the This is a limitation in our choice of model, but given the Johansson et al. BMC Health Services Research (2018) 18:403 Page 8 of 9 need of a model that deal with dependence between ob- who found that health-adjusted pharmaceutical spending servations and that can estimate variation on different in US Medicare does not correlate with nondrug medical levels, the random effects model is our preferred choice. spending, indicating that drugs are a substitute for some A fixed effects model is not suitable for our cause since patients and a complement for others. Potential substi- it cannot measure variation between groups (regions), tution between outpatient primary and specialist care and it assesses only the within variation, eliminating all was considered in some of our earlier tests, i.e. including region-specific time-invariant factors. Using an aggre- specialist visits as an independent variable in the regres- gated panel data set, we can assess the variation on sion of primary physician visits and vice versa. However, regional level and over time, but cannot say anything being insignificant they were dropped out without affect- about the variation on individual level. Our model can- ing the degree of explanation. not answer the (normative) question of what factors are justified determinants of regional variation. Medical Conclusions need, as an acceptable reason for diverse health care use, In summary, we use an observational, longitudinal study is not easily measured objectively. From the set of covar- design with the 21 Swedish regions over 13 years and run iates included here, mortality rate along with some a random effects model dealing with dependence across demographic and socio-economic factors might together regions using region specific random intercepts. Our re- capture medical need. sults show that regional mortality and demography do not In general, coefficient estimates show expected signs, explain variation in visits to primary physicians, while for example, a higher education level is associated with a these factors explain a large portion of the variation in lower level of visits in primary care, but a higher level of visits to specialists. Having adjusted for mortality and visits in specialist care. But our results show that a larger demography, the regional level of socio-economy and es- proportion of 65–79 year olds is associated with lower pecially supply does not add to the degree of explanation levels of visits to the specialist. Both of these issues of variation in specialist visits but accounts for a large por- might be related to the location of universities and uni- tion of variation in primary physician visits, indicating versity hospitals, something we have not adjusted for. that access to care matters for utilization of primary This shows that even though demography explain a care. Overall, 70% of the (unexplained) variation in good share of the regional variation in specialist visits, it the sample is variation across regions, the remaining is not necessarily the expected association (e.g. that an being variation over time. We conclude that regional older population means more health care used). An un- variations in health care utilization cannot be solely expected finding is that regional variation does not al- explained by underlying medical need and demand, ways decrease when adding a set of covariates, true for and that this warrant further research and analyses to demography in primary visits and for supply in specialist form policy proposals. visits. With a closer look in the regression results, the proportion of 65–79 year olds have a positive association Additional files with the level of primary visits, but only variation over Additional file 1: Figure of the national mean per capita number of time is reduced in this step. This indicates that the effect visits to physicians from 2000 to 2015, primary and specialist outpatient within a region over time and the effect between regions care respectively. (PDF 56 kb) are not equal, which is an assumption in the random Additional file 2: Figure of a full overview of per capita number of effects model . Despite a significant association be- physician visits to primary care and specialists in all 21 regions (county councils) from 2000 to 2015. (PDF 64 kb) tween a given independent variable and the dependent Additional file 3: Table S1. Specification of variables and data sources. variable, the degree of explanation can differ on different (DOCX 18 kb) hierarchical levels (across regions and over time). Additional file 4: Table S2. Regression results for Model 1, dependent Potential substitution between outpatient and in- variable visits to physician in primary care. (DOCX 20 kb) patient services is not taken into account in our analysis, Additional file 5: Table S3. Regression results for Model 2, dependent so we can neither reject nor confirm that substitution variable visits to specialist. (DOCX 20 kb) plays a role. Running Pearson’s correlation between pri- mary physician visits and average number of hospital Abbreviations GRP: Gross regional product; SEK: Swedish kronor days, and between specialist visits and hospital days in our data set, we find a low, negative correlation, − 0.11 Availability of data and materials and − 0.23 respectively. Looking at some previous evi- The datasets analyzed during the current study are available in the following dence on the correlation between different type of health online databases: the Swedish municipality and region database, Statistics Sweden, the National Board of Health and Welfare and, the Swedish care services use, Welch et al.  found a positive cor- Association of Local Authorities and Regions, accessible at https:// relation between outpatient and inpatient services in US www.kolada.se, https://www.scb.se, https://www.socialstyrelsen.se, https:// Medicare spending. Another example is Zhang et al.  www.skl.se. Johansson et al. BMC Health Services Research (2018) 18:403 Page 9 of 9 Authors’ contributions 15. Bolin K, Lindgren A, Lindgren B, Lundborg P. Utilisation of physician services NJ1, NJ2 and MS designed the study. NJ1 gathered data, performed the in the 50+ population: the relative importance of individual versus analysis and drafted the manuscript. All authors contributed to writing the institutional factors in 10 European countries. Int J Health Care Finance final manuscript, and read and approved the final manuscript. Econ. 2009;9(1):83–112. 16. Baker LC, Fisher ES, Wennberg JE. Variations in hospital resource use for Ethics approval and consent to participate Medicare and privately insured populations in California. Health Aff The data included are aggregated at the regional level collected from online (Millwood). 2008;27(2):123–34. national databases, thus ethical review by the ethics committee was not 17. Chandra A, Staiger DO. Productivity spillovers in health care: evidence from necessary. the treatment of heart attacks. J Polit Economy. 2007;115(1):103–40. 18. Song Y, Skinner J, Bynum J, Sutherland J, Wennberg JE, Fisher ES. Regional Competing interests variations in diagnostic practices. N Engl J Med. 2010;363(1):45–53. The authors declare that they have no competing interests. 19. Kopetsch T, Schmitz H. Regional variation in the utilisation of ambulatory services in Germany. Health Econ. 2014;23(12):1481–92. 20. Eurostat European Commission. Regions in the European Union - Nomenclature Publisher’sNote of territorial units for statistics. In: Eurostat Methodologies & working papers. Springer Nature remains neutral with regard to jurisdictional claims in Luxembourg: Publications offices of the European Union; 2011. published maps and institutional affiliations. 21. Jakobsson N, Svensson M. The effect of copayments on primary care utilization: results from a quasi-experiment. Appl Econ. 2016;48(39):3752–62. Author details 22. Jakobsson N, Svensson M. Copayments and physicians visits: A panel data Health Metrics, Sahlgrenska Academy, University of Gothenburg, PO Box study of Swedish regions 2003-2012. Health Policy. 2016;120(9):1095–9. 463, SE-405 30 Gothenburg, Sweden. Department of Economics, Karlstad 23. OECD. Health at a glance 2015: OECD indicators. Paris: OECD Publishing; 2015. University, SE-651 88 Karlstad, Sweden. Norwegian Social Research (NOVA), 24. Wooldridge JM. Introduction to econometrics. Hampshire: Cengage Oslo, Norway. Department of Economics, Williams College, Williamstown, Learning EMAE; 2014. MA, USA. 25. Bell A, Jones K. Explaining fixed effects: random effects modeling of time- series cross-sectional and panel data. Political Science Research and Received: 12 December 2017 Accepted: 16 May 2018 Methods. 2015;3(1):133–53. 26. Weisberg S. Applied linear regression. USA: Wiley; 2005. 27. Stiernstedt G. Mest resurser - bäst sjukvård? Om effektiviteten i Stockholms References sjukvård. 2016. https://mb.cision.com/Public/79/2040427/ 1. Skinner J. Causes and consequenses of regional variations in health care. In: a73d307530832a4d.pdf Accessed 20 Apr 2017. Pauly MV, Barros PP, McGuire TG, editors. Handbook of health economics, 28. SKL. Vårdval i primärvården. 2012. Sveriges kommuner och landsting https:// vol. 2. London: Elsevier Science; 2012. webbutik.skl.se/sv/artiklar/vardval-i-primarvarden.html Accessed Jan 2017. 2. Corallo AN, Croxford R, Goodman DC, Bryan EL, Srivastava D, Stukel TA. A 29. Martinelle C, Ulriksson M. Utomlänsvård. 2016. Region Östergötland systematic review of medical practice varition in OECD countries. Health 30. National Board of Health and Welfare. Patientströmmar över Policy. 2014;114:5–14. landstingsgränsen 2018. http://www.socialstyrelsen.se/publikationer2018/ 3. OECD. Geographic variations in health care: What do we know and what 2018-2-12 Accessed 2 Apr 2018. can be done to improve health system preformance? In: OECD health 31. Welch WP, Miller ME, Welch HG, Fisher ES, Wennberg JE. Geographic policy studies. OECD publishing 2014. https://www.oecd-ilibrary.org/social- variation in expenditures for physicians' services in the United States. N Engl issues-migration-health/geographic-variations-in-health-care_ J Med. 1993;328(9):621–7. 9789264216594-en. Accessed 10 Nov 2016. 32. Zhang Y, Baicker K, Newhouse JP. Geographic variation in Medicare drug 4. Cutler D, Skinner J, Stern AD, Wennberg D. Physician beliefs and patient spending. N Engl J Med. 2010;363(5):405–9. preferences: a new look at regional variation in health care spending, Harvard Business School Working Paper; 2015. p. 15–090. 5. Finkelstein A, Gentzkow M, Williams H. Sources of geographical variation in health care: evidence from patient migration. Quart J Econ. 2016;131(4): 1681–726. 6. Skriabikova O, Pavlova M, Groot W. Empirical models of demand for out- patient physician services and their relevance to the assessment of patient payment policies: a critical review of the literature. Int J Environ Res Public Health. 2010;7(6):2708–25. 7. Cutler D, Sheiner L. The geography of Medicare. Am Econ Rev. 1999;89(2): 228–33. 8. National Board of Health and Welfare. Regionala variationer i vårdkonsumtion - En studie av skillnader mellan och inom länder. 2017. https://www.socialstyrelsen.se/Lists/Artikelkatalog/Attachments/20543/2017- 3-40.pdf Accessed 1 Apr 2017. 9. American Hospital Association. Equity in care: a toolkit for eliminating health care disparities. 2015. http://www.hpoe.org/Reports-HPOE/equity-of- care-toolkit.pdf Accessed 6 Apr 2017. 10. Reich O, Weins C, Schusterschitz C, Thoni M. Exploring the disparities of regional health care expenditures in Switzerland: some empirical evidence. Eur J Health Econ. 2012;13(2):193–202. 11. Baker LC, Bundorf MK, Kessler DP. Patients' preferences explain a small but significant share of regional variation in medicare spending. Health Aff (Millwood). 2014;33(6):957–63. 12. Sheiner L. Why geographic variation in health care spending cannot tell us much about the efficiency or quality of our health care system, Brookings paper on economic activity, Fall 2014, vol. 2014. 13. Gopffarth D, Kopetsch T, Schmitz H. Determinants of regional variation in health expenditures in Germany. Health Econ. 2016;25(7):801–15. 14. Eibich P, Ziebarth NR. Analyzing regional variation in health care utilization using (rich) household microdata. Health Policy. 2014;114(1):41–53.
BMC Health Services Research – Springer Journals
Published: Jun 4, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.
Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.
All the latest content is available, no embargo periods.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera