What do locational factors contribute to the explanation of regional variation in officebased

Transcription

1 What do locational factors contribute to the explanation of regional variation in officebased physicians? WIC Verena Vogt 1, Leonie Sundmacher 2 1 Department of Health Care Management, Technische Universität Berlin 2 Department of Health Services Management, Ludwig-Maximilians-Universität, Munich

2 Background The regional distribution of physicians is a persistent policy concern in many health care systems There are a number of public policies that aim aimed at attracting physicians to high need regions and discourage physicians from locating in over-supplied areas Despite the total number of physicians increased, there are significant inequalities in the regional distribution of physicians There is considerable literature showing that the regional distribution of officebased physicians is not only explained by varying health care needs of the population It is also determined by factors such as employment opportunities for spouses, quality of education or the financial attractiveness of a region 2

3 Key factors in physicians choice of practice location Individual characteristics(u.a. Daniels et al. 2007; Dussault et al. 2012) Age at graduation Sex Education Professional factors (u.a. Bildeau & Leduc 2003) Specialisation Working hours Regional factors (u.a. Van Lerberghe et al. 2003; Kazanjian et al. 1996) Conditions of employment (also for spouses) Economic incentives (income potential) Quality of school education Cultural factors Need (morbidity, population with PHI) It remains unclear, however, how much of the variation in physician density is explained by each of these determinants. 3

4 Objectives The aim of this analysis is twofold: 1. Quantifying regional variation in physician density in Germany 2. Decompose this variation into the contribution of each determinant to the measured variation 4

5 Method 1. Regional variation in physician density is quantified using the GINI-Index: k 1 G = 1 (Y i+1 +Y i )(X i+1 X i ) i=0 2. The relative importance of the explanatory variables is estimated using regression based decomposition Based on a linear regression model Variable selection: backward elimination (AIC) 5

6 Variables Variables Mean (SD) Data source (year) Financial attractiveness Share of PHI insured 13,46 (4,32) BVA (2010) Supply of care for surrounding regions (GP/specialists) Infrastructure 0,45 (0,49) / 0,35 (0,48) Versorgungsatlas.de (2010) Care rate for young children 23,81 (14,01) INKAR (2010) Major regional centres 0,39 (0,56) INKAR (2010) Travel time to regional centres (car minutes) 26,56 (17,83) INKAR (2010) Population density 519,60 (675,05) INKAR (2010) Distance to station with IC connection 22,36 (14,42) INKAR (2010) Living standard /education GDP per capita 27,58 (10,24) INKAR (2010) Household income 1548,93 (199,31) INKAR (2010) Share of high qualified employees 8,23 (3,80) INKAR (2010) 6

7 Variables ( continued) Variables Mean (SD) Datenquelle/Jahr Overall attractiveness Tourist accomodations (per sq km) 0,21 (0,19) Regionalstatistik.de (2010) Share of recreational area 2,05 (2,46) INKAR (2010) Share of near-natural area 1,40 (2,11) INKAR (2010) Total migration balance 0,56 (4,42) INKAR (2010) Need / demand Life expectancy 79,80 (1,26) INKAR ( ) Proportion of population aged over 65 21,04 (2,23) Regionalstatistik.de (2010) Unemployment quota 7,40 (3,31) INKAR (2010) Factor for the risk factor compensation scheme (RSA) 1,01 (0,42) BVA (2010) 7

8 Regional distribution of physicians GPs per inhabitants Specialists per inhabitants 8

9 cumulative proportion of physicians Gini-Index and Lorenz curve GPs Specialists Gini-Index: 0,063 cumulative proportion of population Gini-Index: 0,254 cumulative proportion of population 9

10 cumulative proportion of physicians cumulative proportion of physicians Gini-Index and Lorenz curve Dermatologists Psychologists Gynaecologists cumulative proportion of population cumulative proportion of population cumulative proportion of population Gini-Index: 0,220 Gini-Index: 0,419 Gini-Index: 0,361 10

11 Results of linear regression models VARIABLEN GPs Specialists Dermatologists Psychologists Gynaecologists PHI insured 0.655*** 2.448*** *** - Care for surrounding areas 3.643*** 35.27*** 10.88*** *** Care rate for young children * 2.534*** 1.265*** Major regional centres 2.407*** 6.040** Travel time to regional centres *** *** - - Population density ** *** - Distance to IC station *** - GDP per capita *** * 0.761** Household income * ** - - High qualified employees *** 1.632*** *** Tourist accomodations 10.45*** 22.76*** Recreational area * ** * - Near-natural area 0.275* Migration balance *** 0.516** ** Life expectancy *** *** *** ** - Population aged over * *** ** Unemployment quota *** 1.358** RSA 47.26*** 85.89** * ** Constant 108.3*** 377.4** 307.3*** 667.9** 391.7*** Observations R-squared *** p<0.01, ** p<0.05, * p<0.1 11

12 Results of decomposition VARIABLEN GPs Specialists Dermatologists Psychologists Gynaecologists s X k s X k s X k s X k s X k PHI insured Care for surrounding areas Care rate for young children Major regional centres Travel time to regional centres Population density Station with IC connection GDP per capita Household income High qualified employees Tourist accomodations Recreational area Near-natural area Migration balance Life expectancy Population aged over Unemployment quota RSA s ε Total

13 % Percentage contribution to overall variance GPs Specialists Dermatologists Psychologists Gynaecologists Residual Financial Infrastructure Living standards Attractiveness Morbidity 13

14 Discussion Morbidity/need only play a minor role in explaining the distribution of physicians Variables that measure lucrativeness or attractiveness of a region essentially contribute to the explanation of the variance The analysis enables a clear presentation of results Limitations Based on linear regression Cross sectional data 14

15 Thanks for your attention! Verena Vogt Department of Health Care Management WHO Collaborating Center for Health Care Systems Research and Management Technische Universität Berlin Straße des 17. Juni 135, H Berlin, Germany Phone:

16 Regression-based decomposition The variance of Y of a standard regression equation can be decomposed as Y = β 0 + K K K=1 X β k + ε var Y = cov[x k β k, Y] + cov[ε, Y] k=1 Or, upon dividing through by var(y) 100% = K k=1 cov[x k β k, Y] + cov[ε, Y] var(y) K = k=1 s X k + s(ε ) 16

17 Regressions-basierte Dekomposition Where each s-weight is given by s X k = cov[xk β k,y] var(y) The weight associated with the residual is given by s ε = cov[ε,y] var(y) The s-weights K k=1 s X k sum exactly to R² Expressing the s X k in terms of their percentage contribution to R² provides the p-weights p X k = s(xk ) R 2 such that the p Xk `s sum to 100% Vgl. Fields A. (2004): Regression-Based Decompositions: A new Tool for Managerial Decision-Making 17

18 Appendix Distribution of the PHI-insured 18