Overview of studies

The online appendix (table A) presents an overview of the 26 studies, including details of the study objective, participants, study dates, study context and findings. The online appendix (table B) also presents the research design and methodology of the studies, including a quality rating for each study, the data collection methods, sample size and response rate, and the type and level of analysis. Table 1 summarises the key study characteristics and table 2 presents the overall key structural findings relating to health network quality and safety. More than half (14) of the studies were published between 2004 and 2009. Half were undertaken in the USA, with hospital settings (11) representing the largest proportion. Of the 24 studies directed at health professionals, seven focused on multidisciplinary groups of clinicians, with others primarily studying single health professions (nine) or single health domain professionals (six). Four studies researched healthcare collaboratives or partnerships.

With respect to quality and safety, the studies have relevance to five of the six quality improvement dimensions identified by the Institute of Medicine,53 as indicated in the online appendix (table A, column 2): safe,30 31 34 35 41–43 49 effective,26 28 33 38 45 47 patient-centred,54 efficient,35 39 48 50–52 and equitable.32 Timeliness was not addressed. Other aspects addressed by review studies with relevance to quality included culture,55 interdisciplinary teamwork,29 36 service integration44 46 47 and stability,33 34 47 and diffusion of new practices.27 37 40

Research design and analytical approach

As detailed in the online appendix (tables A and B), two studies used ethnography, while the 24 quantitative studies included case studies, multi-case studies, and cross-sectional studies. Principal data collection methods included surveys (23 studies), interviews (4), ethnographic observation (2) and archival data collection (2). Most (19) used SNA to analyse data, with one study adding survival analysis, three studies using multiple regression, and two using other social science analyses. While 11 of the SNA studies had high response rates in the 90–100% range, the survey response rates varied from 100% or whole network samples39 47 to a low of 20%.38 The online appendix (table B, column 8) identifies the key aspects of network structure examined in each SNA study.

Based on our quality assessment criteria (see online supplement, table S3), the quality ratings of the studies are provided in the online appendix (table B). Fourteen studies (54%) were assessed as meeting all of the quality assessment criteria relevant to their study design. Ten studies (39%) met almost all of the criteria, and those criteria that were not fulfilled were thought unlikely to alter the conclusions of the study. In two studies (7%), some of the criteria were fulfilled, and those criteria that were not fulfilled were thought unlikely to alter the conclusions of the study. We did not include studies that met few or no quality assessment criteria.

The level of analysis is a key area of interest in network studies. There are three principal levels: the actor/s level, the network (or organisation) level and the inter-network (or inter-organisation) level. Seventeen studies were directed at two levels, the actors and the network (three looked at the actors and team), eight studies examined networks and one examined all three levels. Four main areas of structural relationships were studied: structural relationships within and between organisations; health professionals and social context; structure of quality collaboratives and healthcare partnerships; and structure in knowledge sharing networks. Table 2 summarises the key network features examined by individual studies, and identifies the substantive structural findings for health network quality and safety, in relation to those network features.

Structural relationships within and between organisations

Six studies examined structural relationships within and between networked organisations, including two of dementia provider networks, three of mental health provider networks (one comparing rural and urban relationships) and one of a hospital network. Carpentier et al54 used comparative case studies of seven organisations to examine relationships between the networks providing assistance to community patients with early-stage dementia in Montreal, Canada. The three interaction levels (interactions between practitioners and caregivers, internal structures and linkages between groups) were found to determine the quality of the practitioner–caregiver interface. Another Canadian case study on dementia care45 evaluated the effectiveness of four community-based dementia care networks in Ontario. SNA identified patterns of administrative and clinical exchanges among networked agencies. Exchanges between groups of agencies (cliques) within each of the four networks were more important than those between individual agencies within each network.

In seminal work examining network structure and effectiveness in the health sector, Milward and Provan46 47 used SNA to examine structural relationships in networks of provider organisations in two studies. The 1998 research included four city community mental health networks and one youth substance-abuse prevention network. Each of the four mental health networks was well integrated, based on two measures – organisational links and cooperative links. With the substance-abuse prevention network, SNA helped to identify agencies not well linked to the system. Milward and Provan47 then used SNA on the results of the initial research, along with a 4-year study of one of the four networks, to evaluate collaboration and contracting strategies. The relationship between network structure and effectiveness was mediated by the context within which services were provided, with resource munificence and network stability identified as the two contextual variables. In the initial study, network stability moderated the impact of resources. In the 2003 study, an effective network was one with enough stability to maintain its ability to manage a set of jointly produced services.

Another study in a US mental healthcare setting44 used SNA to compare two rural and four urban care systems for people with severe mental disorders (SMDs). Both rural sites had numerous coordination linkages between the two types of provider groups (SMD and other service providers) for service planning and delivery. Density scores revealed the extent of service dependency when providers coordinated care to people with SMD in rural sites compared with urban sites. Service relationships among all specialised mental health providers were more likely to occur in rural than urban areas.

A study of chief executive officers in all accredited Taiwanese hospitals38 examined the impact of hospital resources, network resources and centrality on hospital performance. Hospital resources and centrality independently affected performance, whereas network resources did not. For that setting, the authors conclude that a hospital should improve performance by exploiting its in-house resources rather than obtaining network resources externally, and should occupy a central position to create a structural niche.

Health professionals and social context

Half of the studies (13) examined health professionals and social context, including six on work climate. Chase41 used ethnography in two intensive care units (ICUs) in a US teaching hospital to analyse structure and communication patterns relating to the social context in which the process of critical care clinical judgement occurred from the nurse's perspective. With multiple clinicians involved in ICU patient-care decision-making, parallel hierarchies of nurses and of doctors allowed for checks on judgement both within and across professional lines. Also, rituals (nursing reports, physician rounds, flow sheet use) provided a context to check judgement processes.

Lurie et al55 applied SNA to three settings in one US clinical institution: team function in the ICU, the interdisciplinary composition of advisory committees, and relationships between key function directors. Researchers used SNA to compare teams on aspects of their clinical team functioning, and to show the degree of inter-disciplinarity of various clinical departments on the advisory committees. SNA identified potential problem areas with gaps in knowledge of functional roles among academic departments.

SNA was employed by Cott29 to describe the structure of three multi-disciplinary, long-term care teams in a Canadian geriatric care facility. Effects of teamwork in sharing decision-making were limited to a group of higher status health professionals other than medicine, with the clearly defined hierarchy remaining for lower status sub-disciplines. Garrett and McDaniel49 conducted a cross-sectional study of five units in a US hospital to explore the relationships of environmental uncertainty, nurse characteristics and perceived work climate with professional burnout. Environmental uncertainty and perceptions of social-work climate were associated with burnout, yet findings suggested a positive social network climate could shield workers from the negative effects of crisis.

Mossholder et al35 applied SNA and survival analysis to study healthcare employees in a large public US medical centre, examining whether structural, attitudinal and behavioural variables of a relational nature were predictive of employee turnover. Two variables, network centrality and interpersonal citizenship behaviour, predicted turnover with effects above and beyond the effects of job satisfaction. Workers forming more ties with coworkers (network centrality) became more embedded and had lower turnover, and higher interpersonal citizenship behaviour resulted in lower turnover.

A study of physicians in four German hospitals analysed the relationship between overall job satisfaction of physicians and social capital in the hospitals.48 Job satisfaction was significantly associated with professional experience, and with lower workloads. The social capital of an organisation was a significant predictor of overall job satisfaction of physicians. In a similar research area, Swedish studies50–52 investigated whether psychosocial work conditions, professional network, job support, social network and support, sick leave and salary were associated with work stress in nurses in chief manager positions and physicians in clinical director positions. For both roles, a significant association was found between exposure to high job demands and a high level of work stress. Here, available psychosocial resources, inside and outside work, did not balance the experienced work stress in nurse managers and clinical directors exposed to high work demands.

The work of nurses with flexible and traditional schedules was compared in a US case study in an urban paediatric acute-care hospital.42 Applying SNA to examine the types of social networks used by both groups of nurses and to compare their workplace socialisation, no significant differences were found in the nurses' social network composition. However, traditional nurses used significantly more peer members for different types of emotional support. MacPhee and Scott43 also applied SNA to examine workplace social support networks of rural hospital nurses, compared with the urban nurse networks. Rural nurses' networks were predominantly peer based, but managers provided significant functional supports, with rural nurses expecting more guidance from management than urban nurses.

Ethnography was employed by Ormrod et al40 in three UK NHS mental health clinics to examine how organisational practices were spread within networks of practice. Professional networks within psychiatry were found not to spread particular work practices equally to psychiatrists and their associated multi-disciplinary teams at two new sites. This was largely due to ideological differences in psychiatric practice approaches and differences in founder influence.

Webster et al36 used SNA data on advice and social relations, to examine differences in eight mental health case-management teams in a Californian (USA) county mental health system. Male supervisors were substantially more central for ‘instrumental’ (advice) relations than female supervisors, displaying an autocratic leadership style. Male supervised teams were more centralised than female supervised teams. For the ‘social’ relations, male team supervisors were more central than female supervisors. Teams with female supervisors were more centralised than those with male supervisors, yet the female leaders were not the most central team member, showing a democratic leadership style.

Significant research on the professional social networks of clinical directors of medicine and directors of nursing in UK hospitals was undertaken by West et al.37 Directors of nursing were more central in their networks than clinical directors of medicine, and their networks were more hierarchical. The networks of directors of nursing had lower density (having advantages in terms of access to information) than the clinical directors of medicine who tended to be embedded in much more densely connected networks (cliques). Doctors' networks were more egalitarian and decentralised than nurses' networks, hence change processes for the doctors need to involve group processes, rather than simply convincing individuals of the need to change.

In West and Barron's subsequent research,27 both directors of nursing and clinical directors of medicine discussed ‘important professional matters’ with others similar to themselves in terms of profession, gender, age, and seniority (homophily), with doctors being more extreme in this regard. Managers (non-clinically qualified) occupied a powerful brokerage role for both nurses and doctors, whereas nurses and doctors were rarely on each other's networks. Nursing and medicine had quite different social structures. With few informal ties between the two professions, information was unlikely to be spread between professions by informal sources.

Structure of quality collaboratives and healthcare partnerships

Four studies show the application of SNA in examining the structure and sustainability of collaboratives and partnerships. Gold et al32 applied SNA to research relationships (structure and processes) among organisations participating in a large-scale, public–private, quality collaboration among major US health plans to reduce racial and ethnic disparities in healthcare. Sponsors and support organisations, along with a few of the health plans, formed the ‘glue’ holding the collaboration together. With limited communication among health plans, if the collaboration ended without greater communication among the non-core organisations, the absence of the core would leave a very sparse network. Gold et al advised collaborative sponsors to consider both short-term and long-term goals and whether they can be pursued if a collaborative ends.

Mendel et al34 used SNA to explore the numbers and types of inter-organisational partnerships within the US patient safety domain, the changes over time in these networks, and their potential for disseminating patient safety knowledge and practices. Between 2004 and 2006, partnerships grew in all activity domains, particularly dissemination and tools development, signifying growing strength in the capacity to disseminate and implement patient safety advancements. Fragmentation of the overall partnership network decreased, and potential for information flow increased. However, network centralisation increased, suggesting vulnerability to partnership failure if key participants disengaged.

SNA was applied by Lewis et al33 in a longitudinal analysis of network structure, dynamics and sustainability in primary care partnerships in Victoria, Australia at three time points between 2002 and 2005. Although network structures changed over the 3 years, there was the continuing centrality of the independent staff employed to manage the partnerships, with their crucial role in holding partnerships together. These partnerships required long-term support, not just start-up funding.

To study whether collaboration initiatives by a local health authority in Italy between 2001 and 2004 had any effect on individual and district-level general practitioner (GP) performance on drug expenditure targets, Fattore et al39 used SNA. In terms of the GP's ability to meet expenditure targets, the social influence mechanism (the performance of peers to whom the GP was directly connected) was more relevant than the social capital mechanism (the centrality of the GP in the network). While collaborative arrangements induced more homogeneous behaviour among GPs, they did not necessarily improve the ability of GPs to meet local health authority objectives.

Structure in knowledge sharing networks

The importance of homophilous behaviour was identified by Creswick and Westbrook30 who examined how a network of staff in an Australian hospital renal ward sought medication advice. Most communication occurred within professional groups. Similarly, a study of a network of staff in an Australian hospital emergency department by Creswick et al31 found that, when seeking advice, individuals were more closely connected to colleagues in their own professional groups. Heng et al26 used SNA to explore the brokerage role of facilities management in an Australian hospital. Facilities managers actively bridged information and knowledge across different functions, filling structural holes within a communication network structure and being positioned to identify inter-disciplinary opportunities.

Although various studies discussed above identified features of network structure associated with better performance, Rangachari provided one of the few to link network structure with quality outcomes.28 This research in four large US teaching hospitals used SNA to explore the relationship between the organisational knowledge sharing structure related to quality, and hospital coding performance related to quality. Good-coding performance was associated with a knowledge sharing network structure rich in brokerage and hierarchy (with senior leaders coordinating knowledge exchange related to quality and connecting the organisation with the external environment), rather than density (with everyone connected to everyone else).