Attitudes towards reforming primary care in Belgium: social network analysis in a pluralist context

Abstract

Objectives

Health care policies are influenced by many groups which in turn influence each other. Our aim was to describe a network of nominated influential stakeholders and analyze how it affects attitudes to reforming primary care.

Methods

Face-to-face interviews were carried out in Belgium with 102 influential people. Each respondent was asked to score solutions for improving the role of general practice in the health care system and to nominate up to six other influential stakeholders. Social network and multivariate analyses were used to describe the nomination network and its effect on attitudes to reform.

Results

The network was highly centralized and homophilous (tendency to bond with people who are similar) for language groups. Despite Belgium having a strong pluralist tradition of decision making, policy makers were central to the network (average indegree = 10.8) compared to professional representatives (6.9). Respondents supported an enhanced role for general practitioners but did not support radically new policies.

Conclusion

Social network analysis contributes to understanding why health care reforms may languish in pluralistic, decentralized health care systems. The central position of a stakeholder in a network is related to perceived influence but does not favour a radical policy orientation. In addition, language-group homophily in the ‘perceived influence network’ leads to a weak coalition that only favours small-step reform.

Keywords

health policies reform primary care social network analysis stakeholder analysis

Introduction

In the OECD, since the 1980s, health care systems have undergone several reforms to improve access, efficiency, and quality of care. The pace of reforms has been slow and implementation has been inconsistent, with some well-known exceptions.¹ Some recent landmark reforms in the United States,² in Europe,³ or at the international level have shown the importance of pluralist interest groups in the reform process.⁴ Pluralism, in political science, claims that the power to bring an issue onto the agenda or to influence decision making is distributed throughout different groups.⁵ This is so in health care, with powerful interest groups such as health care insurance companies, trade unions, professional organizations, pharmaceutical companies, and health care providers’ associations. As a consequence, agenda setting, policy design, and decision making result from the interconnections of these groups. Policy network (sets of formal, institutional, and informal linkages between governmental and other actors structured around shared, if endlessly negotiated, beliefs and interests in public policy making and implementation)⁶ has thus emerged as a useful concept to analyze how different groups are linked and influence policy options.⁵ Although research has started to address the influence of such networks in public decision making,^7,8 the way they affect health care reform remains opaque. According to Rhodes’ model, a policy network is a bargaining game between conflicting actors with diverging interests and, as a consequence, it may foster incremental change.⁹ There have, however, been few empirical analyses of this assumption.

In the last decade, there has been a growing interest in health care in the role of networks of influential actors. A recent literature review identified 52 observational or interventional studies using social network analysis (SNA) in health care organizations to analyze prescribing behaviours, decision making, communication patterns, the diffusion of innovation, and social influence and to describe organizational network structures.¹⁰ Only two studies looked at influence networks. One, carried out in a UK metropolitan area, showed that managers have greater influence than professionals or academics because of their control over the policy process¹¹ while the other in Australia found that doctors shape the health policy process.¹² It remains unclear, therefore, how different network positions are associated with policy outcomes. On the one hand, influential positions may provide more clout in support of policy changes, but on the other hand, influential positions may favour more incremental changes in order to protect the balance of interests that shapes the overall decision-making process.

General practitioners (GPs) play a key role in many health care systems: they account for most patient contacts, they are key allies in cost control and in preventive care, and they help to make the system more socially equitable.^13,14 Yet, over the last decade, the situation of GPs has been deteriorating. In most countries, the median income of specialists is almost twice that of GPs, and this gap has been widening, although it was much lower or even reversed in the UK, Finland, and Switzerland.¹⁵ In some OECD countries, GPs still work in sub-optimal conditions, working on their own or with limited and inconsistent use of information technology.^13,16 So far, policies aimed at improving the GP role have not received much support from policy makers. One possible reason would be that there is no effective coalition (agreement between different groups involved) across the stakeholders involved, despite the rhetoric about the key role of the GP.

Our aim was to investigate how the structural position of people in a policy network is associated with particular attitudes to reform of primary care in Belgium, a country with a strong pluralistic tradition of decision making.¹⁷ We used a social network survey of elite stakeholders in primary care to describe the perceived influence network and to analyze the association between the stakeholders’ network position and their priorities for reform of primary care.

Method

Setting

This study was carried out in Belgium, a country with a compulsory health insurance system, covering the entire population. It has a high medical density (4 doctors per 1000 inhabitants) and high per capita health care expenditure (10.6% of GDP). Most GPs work on their own with a predominantly fee-for-service payment system without auxiliary staff. Retention in the profession is low. In terms of the OECD typology, the Belgian health care system combines basic public insurance coverage with heavy reliance on market mechanisms at the provider level.¹⁸

Design and data

A face-to-face survey of stakeholders was carried out in 2008.¹⁹ A panel of researchers and staff from the Belgian Health Care Knowledge Centre (KCE), which commissioned the survey, identified an initial set of 30 people considered to be influential in general practice organization, funding, or training. Perceived influence was defined as the capability to affect decision making: authority, legal mandate, organizational factors, budget, number of members, expertize, and network of contacts. Snowball sampling was then used, with all initial respondents asked to nominate up to five people they considered to be important, knowledgeable, or influential. All those who received more than three nominations were contacted. A full description of the stakeholder selection procedure and interviews is given elsewhere.²⁰ The network boundary was defined as stakeholders having influence or interest in GPs’ training, funding, or organization: policy makers (those involved in making policy at the national, regional, or sickness-fund level), professional organization, academia, media, and GPs’ leaders. Both Dutch-speaking and French-speaking groups were included. In total, 116 people were contacted and 102 (88%) agreed to participate. Eight respondents were removed because of a lack of key information, leaving 94 (Table 1).

Table 1.

Stakeholders nominated and participating in the survey (number of respondents, indegree, and percentage).

	Stakeholders nominated			Stakeholders participating
	Number	%	Indegree (mean)	Number	%	Indegree (mean)
GP	31	12.1	0.0	24	25.5	0.0
Media	31	12.1	3.6	5	5.3	9.4
Policy makers	86	33.6	5.6	23	24.5	10.8
Unions or professional organizations	50	19.5	4.4	14	14.9	6.9
University	35	13.7	4.2	12	12.8	6.5
Others	23	9.0	0.9	16	17.0	0.0
Total	256	100.0	3.8	94	100.0	5.0

χ²= 24, p < 0.01.

All respondents were interviewed in a face-to-face computer-assisted interview. The survey used multi-criteria analysis, designed to assess decision making between different policies, assessed on several criteria.^21–23 This involves four steps: identification of various policy options for solving a problem, identification of criteria for appraising those policies, allocating a score to each policy option on each criterion, and weighting each criterion in order to reflect its relative importance for overall decision making.

Measurements

Each respondent was presented with 23 policy options for improving general practice attraction or retention (Table 2). Reform options were classified into five categories: training, financing, work-life balance, practice organization, and governance. Each policy was scored on four criteria on a 7-point Likert scale: effectiveness, cost to society, acceptability, and patients’ freedom of choice. An example of a question on an organizational policy option rated on the effectiveness criteria was ‘What would be the effect of removing individual on-call duty on the number of GPs choosing or staying in general practice? (1 = ‘Very negative’ to 7 = ‘Very positive’)’. The full questionnaire is available online.²⁴

Table 2.

Rankings, scores and factorial loading for each policy.

Policy label	Conservative vs. Innovative	Average ranking (std)^a	Average Score^a	Loading on factor 1	Loading on factor 2
To develop a clinical activity linked with academic centres of general practice	Conservative	3.4 (1.4)	4.3 (1.3)	0.22	0.41
To better integrate general practice courses and specialty courses	Conservative	1.9 (0.8)	5.6 (0.7)	0.06	0.15
To organize compulsory internships in general practice for all medical students	Innovative	2.0 (1.0)	5.5 (0.8)	0.23	0.3
To pay GPs with a combination of capitation per patient and fee for service	Conservative	2.0 (1.0)	5.0 (0.9)	0.49	0.19
To pay GPs with a combination of salary and fee for service	Innovative	2.8 (1.4)	4.4 (1.1)	0.25	0.55
Rewarding GPs with targeted payment for the realization of objectives	Innovative	2.6 (1.6)	4.7 (1.3)	0.04	0.57
To move towards a more equitable geographical distribution, by improving the incentives to practice in low medical density areas	Conservative	2.5 (0.8)	5.2 (0.9)	0.38	0.37
Not to penalize the work of general practitioners working part-time	Innovative	2.2 (1.0)	4.4 (1.0)	0.11	−0.01
To organize local groups of professional GPs responsible for on-call duty	Innovative	2.8 (1.2)	4.3 (1.2)	0.27	0.53
To pay GPs for their continuous training activities	Innovative	3.0 (1.3)	4.1 (1.0)	0.56	−0.06
To remove individual on-call duty and replace it by a professional service like ‘SOS Médecins’	Innovative	2.2 (1.4)	4.4 (1.5)	0.27	0.37
To support the creation of local resource agencies that would promote the attraction and retention of GPs according to local needs	Innovative	4.2 (1.2)	4.2 (1.3)	0.28	0.23
To create a master’s degree in advanced nursing practice, to back up GPs	Innovative	2.5 (1.5)	5.0 (1.3)	−0.15	0.78
To financially discourage excessive or premature recourse to second-line services	Innovative	2.0 (1.2)	5.1 (1.3)	0.18	0.07
To encourage the delegation of some clinical tasks to other existing health professions (nurses, etc.)	Innovative	2.3 (1.1)	4.8 (1.2)	−0.11	0.75
To encourage the delegation of administrative work to administrative staff	Conservative	2.0 (0.8)	5.4 (0.9)	0.54	0.43
To encourage GPs to share a common infrastructure or secretariat	Conservative	2.2 (0.9)	5.5 (0.9)	0.78	0.18
To encourage GPs to work together	Conservative	1.8 (0.8)	5.7 (0.7)	0.83	−0.02
To reinforce the role of GPs in the multidisciplinary team	Conservative	1.9 (0.8)	5.6 (0.8)	0.60	0.02

GP: general practitioners.

The lower the ranking, the higher the priority; the higher the score, the higher the overall performance.

After scoring on each criterion, respondents were asked to weigh each criterion on a scale from 0 to 100. Weights were then normalized to compute an overall score for each policy. The overall score was the weighted sum of the scores for the four criteria.

We retained 19 policy options with a Cronbach’s alpha correlation ≥0.80. For those policies, we computed two variables describing the overall policy orientation, using principal component analysis. On the basis of the eigenvalue scree plot, the first two components, accounting for 24% and 12% of the total variance, were retained. The first factor (24% of the variance) addressed a wide range of policies related to fostering group practice, giving GPs more power, either as gatekeepers to secondary care or as coordinators of multi-disciplinary teams. Group practices make it possible to share infrastructure and delegate administrative tasks. The first component also facilitates the use of capitation funding mechanisms and offers better exposure to primary care practice during medical training. It was labelled ‘A stronger GP in the driver’s seat’ (Table 2).

The second factor (12% of the variance) focused on GP–nurse relationships, proposing trained nurses to whom GPs would delegate clinical and administrative tasks. On-call would become a separate activity and would no longer be an individual duty. This component also included support for targeted payment and General Practice Clinical Academic Centres (the equivalent of teaching hospitals for consultants). Hence, this component seems supportive of specialized GPs, a key issue in the management of patients with long-term conditions.

We also computed the average ranking of policies considered to be innovative because they either required a change in existing regulations (such as targeted payment) or the establishment of a new executive body. The distinction was made by the researchers and KCE staff, using an iterative approach, until all agreed on the categorization of each policy.

Structural indicators of each stakeholder’s centrality and homophily in the perceived influence network were computed.²⁵ Centrality was measured by the average indegree (number of nominations received), betweenness centrality (percentage of times that stakeholder × lies in the path between two other stakeholders), and coreness (being close to the core group). Homophily/heterophily were computed by the Krackhardt E-I index.²⁶ The E-I index measures the tendency of stakeholders to nominate similar people: it ranges from −1, when all nominations are internal to the stakeholder group (homophily), to 1, when all nominations are to external groups (heterophily). The E-I index was computed for language (Dutch and French) and interest groups (professionals, policy makers, and others).

Results

Network description

Respondents made 981 nominations of 256 people (average 3.8 per respondent). About half the nominations (470) were of other initial respondents. Table 1 compares the initial respondents (hereafter, ‘the closed network’) to those nominated (participating or not). Chi-square test indicated that the distribution of those nominated and those who participated differed. In particular, GPs were more likely to participate, while policy makers and trade union representatives were less likely to do so. For each group, participants had a higher average indegree than the complete set of nominees, indicating that the survey was successful in targeting those with higher perceived influence. This was particularly true for policy makers, as our sample had an average indegree of 10.8, as against 5.6 for all nominees. Nevertheless, of the 30 individuals who received 10 nominations or more, 11 did not participate in the study.

Within the closed network, the average indegree was 5.0: the two most nominated groups were policy makers, with an average indegree of 10.8, and trade union representatives, with 6.9, which is consistent with the pluralistic structure of decision making. The nomination density was 0.047 and had a Freeman centralization indegree score of 43%, suggesting a highly centralized network, as shown in Figure 1. The network was strongly homophilous: most nominations were within the same language group (E-I = −0.58; p < 0.01): Dutch-speaking stakeholders nominated other Dutch-speaking stakeholders; French-speaking stakeholders did likewise. However, nominations across interest groups were heterophilous (E-I = 0.30; p = 0.05).

Figure 1.

Influence nominations and reform orientation in general practice. Language group (NL for Dutch-speaking, FR for French-speaking); colour for score on factor 1 ‘A stronger GP in the driver’s seat’ (white = score ≤ 0, light grey = score (0, 0.3), dark grey = score > 0.3); shape for stakeholder groups (circle = professional groups, square = policy makers, and triangle = others); symbol size is proportional to the number of indegree. Rim width is proportional to the number of innovative policies supported; ties reflect influence nomination. Isolated nodes were removed. Graph displayed with the Kamada–Kawai algorithm and some manual adjustments.

Figure 1 displays the network of perceived influence nominations as well as the attitude towards reform. The figure highlights three interesting patterns. First, stakeholders receiving more nominations were generally policy makers and had a higher score on factor 1 (‘A strong GP in the driver’s seat’). They were also keener to support innovative policies (ring width). Second, professional stakeholders (circle) did not cluster, which implied they did not acknowledge each other as influential stakeholders. On the right-hand side of the graph, four professional stakeholders displayed less support for the main policy orientation issue, ‘A Stronger GP’. Professional stakeholders also displayed different views on the role of GPs. Third, the graph shows language group homophily (Dutch-speaking stakeholders on the left and French-speaking on the right). Moreover, language groups displayed more heterogeneity in policy orientation: the two language communities had different perspectives on the role of GPs. In sum, the figure suggests that there was no strong coalition in support of putting the GP in the driver’s seat.

We analyzed the ranking of innovative policies (the lower the ranking, the higher the priority) and the two factor scores (the higher the score, the higher the priority) according to structural indices and stakeholder groups (Table 2). Overall, there were few statistically significant associations of the three outcomes analyzed (Table 3). Innovative policies were of higher priority for stakeholders belonging to the core perceived influence network (Beta = −1.4). This core group, however, was not supportive of a clear reform orientation: being in the core was not associated either with a higher score on factor 1 or on factor 2. We also noted that individuals with higher language heterophily (language EI index) and individuals with higher betweenness (on the log scale) gave lower priority to innovative policies: being in a broker position is associated with more incrementalism in policy orientation.

Table 3.

Nodal and structural covariates policy rankings or scores (beta and 95% confidence interval from the regression).

	Innovative policies ranking^a		Factor 1 score ‘GP in the driver’s seat’^a		Factor 2 score ‘Specialized GP’^a
Covariate	Beta	95%CI	Beta	95%CI	Beta	95%CI
French-speaking stakeholders (ref = Dutch)	−0.01	−0.12, 0.09	0.45	0.04, 0.85**	−0.08	−0.49, 0.33
GPS and Professional groups (dummy)	0.11	−0.03, 0.25	−0.14	−0.67, 0.40	−0.61	−1.15, −0.08**
Policy makers (dummy)^b	0.10	−0.06, 0.25	−0.43	−1.03, 0.17	0.02	−0.58, 0.62
Coreness (%)	−1.32	−2.17, −0.46***	1.86	−1.48, 5.19	2.00	−1.35, 5.34
Indegree (nber)	0.01	0.00, 0.02	−0.01	−0.05, 0.02	−0.01	−0.05, 0.03
Betweenness (log)	0.21	0.04, 0.39**	−0.13	−0.85, 0.58	0.48	−0.22,1.19
EI index–language (−1,1)	0.11	−0.01, 0.24*	0.48	0.00, 0.96*	−0.01	−0.49, 0.47
EI index–interest group (−1, 1)	−0.03	−0.14, 0.08	−0.31	−0.74, 0.13	−0.06	−0.50, 0.38

GP: general practioners.

The lower the ranking, the higher the priority; the higher the factorial score, the higher the overall performance. GPs and professional organizations compared to all other groups.

Policy makers compared to all other groups.

***

Significant at 1%. **Significant at 5%. *Significant at 10%.

Compared to Dutch-speaking stakeholders, French-speaking stakeholders were keener to support ‘putting the GP in the driver’s seat’, while professional groups supported specialized GPs. Coefficients for policy makers were non-significant for the three dependent variables. Brokering perceived influence across language and interest groups had very few associations with policy orientation. Those brokering language groups were somewhat less likely to support innovative policies (Beta = 0.11) and more likely to ‘put the GP in the driver’ seat’ (Beta = 0.48).

Discussion

This is one of the first studies to have applied SNA to describe the perceived influence networks of those involved in health care policies. It concerns a country with a strong pluralistic decision-making tradition. The study was able to collect both perceived influence nominations and policy orientations among a group of highly influential stakeholders of primary care policies.

Three main findings arise from this study. First, in spite of the pluralistic tradition of decision making in Belgium, policy makers were the people seen as most influential in the network. Others were cited but had rather peripheral positions in the network. This is consistent with a previous study in the UK, which showed that managers and local authorities were more influential in a local public health network than public health professionals and academics.¹¹ However, our findings are different from those of a study in Australia, in which academics were the most influential.¹² One explanation may have to do with the sampling strategy: we only had a two-wave snowball strategy, as against six in the Australian study. The latter, moreover, had a broader definition of influence that included ‘shaping ideas’ or ‘initiating proposals’, whereas we had a more limited definition of influence, linked to decision making and implementation. In addition, this difference could be because our study addressed the national level, whereas the Australian study focused on the state of Victoria. The most salient structural feature of our network was the highly fragmented structure of perceived influence according to language but not according to stakeholder group; in the end, at the national level, language identity may have superseded interest-group identity in the network structure.

Second, the many policy options for primary care reform clustered around two issues: whether GPs should have more power, or whether they should become more specialized by offloading tasks to other professions such as nurses. We found that the most important stakeholders eschewed radical policy reforms on both issues and, overall, clustered around a small-step approach. Central stakeholders also disagreed within their own interest group. Hence, there was no strong advocacy coalition either among policy makers or among professional groups.

Third, we found that the association between perceived influence and policy orientation was complex. Being part of the core group of stakeholders was associated with a preference for innovation but not with a clear orientation. One possible explanation, as raised above, is that reforming general practice was a contentious topic between the two language groups of stakeholders at that time. This was later confirmed by the sixth reform of the Belgian state that was voted in 2012, which led to more decentralization of primary care services, including funding and organization. In a way, our network analysis highlighted a process that was in the making. But this remains highly speculative, as the associations were small and not always statistically significant.

The pluralistic process of decision making in Belgium does not favour a strong advocacy coalition supporting a stronger GP in the driver’s seat. One reason, as is clear in the literature, is the lack of consensus among professional groups. Group cohesion is a key factor in a coalition’s effectiveness,^4,27,28 something we observed in a pluralistic context. Conversely, the emergence of a policy network able to get reform enacted and implemented in primary care is less likely if the profession is divided. Professional groups were not able to come up with a clear-cut programme on which they have previously agreed. This lack of a strong coalition weakens the reform process.²⁹

The network analysis shows that perceived influence is shaped much more by language group homophily than by interest group homophily. To some extent, this was expected as decentralization in Belgium that has created the conditions for network segmentation according to language.¹⁷ In addition, previous social network studies showed that ethnicity is often an important determinant in tie creation and maintenance.³⁰ This suggests that a policy network might struggle to form strong coalitions when community or language fragmentation is added to group interest or training fragmentation. In particular, the perceived influence network displayed greater heterogeneity of French-speaking stakeholders towards policy approaches compared to Dutch-speaking stakeholders. Policy network fragmentation may lead to more decentralization and conversely, more decentralization increases policy network fragmentation.

Limitations

First, respondents may not have dared to express their real policy attitudes, particularly when their mandate exposed them to public opinion. Fear of public disclosure of their views may have affected their stated opinions. We found, however, no indication that policy makers had more missing information compared to other groups (mean = 0.40 versus 0.30, p > 0.50). We also regressed the indegree and betweenness centrality on the number of missing items and found no significant effects.

The second limitation has to do with the coverage of stakeholders and the resulting overall network structure: the selection of stakeholders was limited to one-third of all nominees, and some groups were under-represented compared to the initial list. Although we were able to select those with the highest number of nominations, we did not include some important people: out of the 30 most nominated stakeholders, 11 would not be interviewed, including one important trade union representative. In addition, our snowball procedure, which only included two waves, means there was the possibility that the actual network structure may differ from the one we surveyed. It might have proved more professionally oriented if we had more than two waves of data collection.

Conclusion

SNA indicators contribute to understanding why health care reforms may languish in pluralistic, decentralized health care systems. This analysis of the policy network structure revealed that the central position of a stakeholder in the network structure is related to perceived influence. However, being an influential stakeholder in such a network had mixed effects: innovative policies were more likely to be -supported but without any clear orientation and, conversely, influential individuals supported an enhanced role for the GP but did not support radically new policies. In addition, the analysis of the network structure shows a high level of language-group homophily. Stakeholders from the two linguistic groups had different views about the role of GPs. Without any clear common policy orientation, this may weaken coalitions across language groups, favour small-step reforms and eventually may reduce the efficiency of any reform introduced.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work is supported by a grant from KCE (no. 2008-90B). The funding source was involved in the study design and approved the report; the funding source was not involved in the writing or submission of this article.

References

Cutler

. Equality, efficiency, and market fundamentals: the dynamics of international medical-care reform. J Econ Lit 2003; 40: 881–906.

Thompson

. Health reform, polarization, and public administration. Publ Admin Rev 2013; 73: S3–S12.

Rico

Saltman

Boerma

WGW

. Organizational restructuring in European health systems: the role of primay care. Soc Pol Admin 2003; 37: 592–608.

Shiffman

Smith

. Generation of political priority for global health initiatives: a framework and case study of maternal mortality. Lancet 2007; 370: 1370–1379.

Hudson

Lowe

. Understanding the policy process: analysing welfare policy and practice, Bristol: The Policy Press, 2004.

Rhodes

RAW

Policy network analysis. In: Moran

Rein

Goodin

(eds). The Oxford Handbook of public Policy, London: OUP, 2008, pp. 425–447.

Daake

Anthony

. Understanding stakeholder power and influence gaps in a health care organization: an empirical study. Health Care Manag Rev 2000; 25: 94–107.

Dowding

. Model or metaphor – a critical-review of the policy network approach. Pol Stud 1995; 43: 136–158.

Marsh

Rhodes

RAW

. Policy networks in British government, Oxford: Clarendon Press, 1992.

10.

Chambers

Wilson

Thompson

. Social network analysis in healthcare settings: a systematic scoping review. PLoS ONE 2012; 7: e41911–e41911.

11.

Oliver

De Vocht

Money

. Who runs public health? A mixed-methods study combining qualitative and network analyses. J Publ Health (Oxf) 2013; 35: 453–459.

12.

Lewis

. Being around and knowing the players: networks of influence in health policy. Soc Sci Med 2006; 62: 2125–2136.

13.

Saltman

Rico

Boerma

WGW

. Primary care in the driver’s seat? Organizational reform in European primary care, Maidenhead: Open University Press, 2006.

14.

van Doorslaer

Wagstaff

van der

. Equity in the delivery of health care in Europe and the US. J Health Econ 2000; 19: 553–583.

15.

Fujisawa R and Lafortune G. The remuneration of general practitioners and specialists in 14 OECD countries: what are the factors influencing variations accross countries? OECD Health Working Paper no. 41. OECD, 2008.

16.

Schoen

Osborn

Huynh

. On the front lines of care: primary care doctors’ office systems, experiences, and views in seven countries. Health Aff (Millwood) 2006; 25: w555–w571.

17.

Beland

Lecours

. Nationalism, public policy, and institutional development: social security in Belgium. J Publ Pol 2005; 25: 265–285.

18.

Organisation for Economic Co-operation and Development. Health care systems: efficiency and policy settings, Paris: OECD, 2010.

19.

Varvasovszky

Brugha

. A stakeholder analysis. Health Pol Plan 2000; 15: 338–345.

20.

Lorant

Geerts

Duchesnes

. Attracting and retaining GPs: a stakeholder survey of priorities. Br J Gen Pract 2011; 61: 452–453.

21.

Belton

Stewart

. Multiple criteria decision analysis: an integrated approach, Boston, MA/London: Kluwer Academic, 2002.

22.

Millstone

Lobstein

. The PorGrow project: overall cross-national results, comparisons and implications. Obes Rev 2007; 8(Suppl. 2): 29–36.

23.

Wilson

Sussex

Macleod

. Prioritizing health technologies in a Primary Care Trust. J Health Serv Res Pol 2007; 12: 80–85.

24.

Lorant

Geerts

D’Hoore

. Making general practice attractive: encouraging GP attraction and retention – appendices, Brussels: Centre BHCK, 2008.

25.

Valente

. Social networks and health models, methods, and applications, Oxford: Oxford University Press, 2010.

26.

Krackhardt

Stern

. Informal networks and organizational crises: an experimental simulation. Soc Psychol Q 1988; 51: 123–140.

27.

Sandstrom

Carlsson

. The performance of policy networks: the relation between network structure and network performance. Pol Stud J 2008; 36: 497–524.

28.

Zakocs

Edwards

. What explains community coalition effectiveness? A review of the literature. Am J Prev Med 2006; 30: 351–361.

29.

Atkinson

Coleman

. Strong states and weak states – sectoral policy networks in advanced capitalist economies. Br J Pol Sci 1989; 19: 47–67.

30.

McPherson

Smith-Lovin

Cook

. Birds of a feather: homophily in social networks. Annu Rev Sociol 2001; 27: 415–444.