Patterns of collaboration in complex networks: the example of a translational research network

Networks in healthcare are promoted as a way to foster collaboration through the organisation of people across facilities, sites and disciplines [1]. Yet our understanding of the structural features and processes within such networks and how they contribute to network outcomes is limited [2, 3]. Close-knit groups of people (clusters) within networks can be strong and positive, or negative and counterproductive structures. “Silo” is a generally negative term for a cluster implying insularity, introspection and resistance to change; silos have been shown to hinder collaboration [4‐9].

Clusters can also be strong, resilient and highly productive groupings and may hold highly specialised knowledge and expertise [10, 11]. Linking clusters together helps communication and information to permeate throughout the whole network, and can also produce innovative ideas through cross-fertilization of the expertise and knowledge held in each [12]. Quantitative examination of these groups and gaps is rare, especially in healthcare settings [5].

The purpose of this paper is to examine factors that influence collaboration among the members of a new translational research network (TRN), affecting past collaborative ties, as well as current and future collaborations. Understanding more about these patterns, especially the gaps, clusters or barriers between members, responds to the identified gap in the literature around the paucity of quantitative studies examining structural features of health networks. It also offers a practical contribution by identifying how network leaders can help leaders of networks to manage subgroups, promote strategic partnering and overall connectivity. This study forms part of a more extensive study looking at network structure and key players and their effect on network outcomes [13‐15].

Collaboration is theoretically and conceptually a complex organisational characteristic [16‐21]. It has been defined by Wood and Gray [20] as an interactive process between autonomous stakeholders who can see different aspects of a common problem. Collaboration between researchers and clinicians is considered essential in translational research. Translational research undertakes the crucial role of moving biomedical discoveries out of the highly controlled laboratory environment and applying it in the messiness and complexity of actual patient and clinical service realities [22‐24]. Expertise and understanding from both arenas through collaboration are necessary to work out ways to make this happen. Yet, the research and clinical domains seem divided by ever-widening gaps of increased specialisation, complexity and sophistication on either side [25, 26].

TRNs are a strategy to facilitate translation: setting up an administrative structure to provide funding, project officers and shared resources, and a social-professional structure that can maximize opportunities for collaboration, innovation and knowledge transfer across different disciplines, organisations, sites and specialties [25, 27, 28]. They manifest in similar ways with different titles: as Collaboratives for Leadership in Applied Health Research and Care (CLAHRCs) in the UK and as interorganisational alliances awarded funded by the Clinical and Translational Science Awards in the USA. Potential partners abound in such networks but clusters and gaps in the organisation of the network can affect the likelihood of collaboration [29, 30]. We use a social network approach to examine the collaborative ties among TRN members, forming graphs of linkages (sociograms) that can be analysed for patterns [13].

People listed as full members of the TRN as of January 2012 were asked to complete an on-line survey in March 2012, each member receiving a link to the secure survey site via personal email. Ethics approvals were obtained from the University of New South Wales (HREC: 09085) and appropriate local health network and site-specific committees. Respondents were assured of anonymity in the reporting of results (names being replaced by anonymous codes) and were required to give formal consent.

The survey was informed by interviews with 14 network stakeholders and feedback from a pilot of the survey by ten participants from equivalent clinical or research backgrounds. The survey was a whole network survey [43, 44], that is we sought answers from all members to reflect the whole network, rather than a sample of members. To maximise the response rate, three follow-up reminders were emailed to non-respondents over the next two months.

The survey established respondents’ place of work, main tasks, years of experience, and if they had had previous involvement in translational research. Workplaces were grouped into three sites - Central, Satellite and Peripheral - based on both geographic proximity and professional proximity; i.e. having administrative links (largely NSW Local Health districts). The Central Site grouped six of the workplaces which were all within a one-kilometre radius of each other. The Satellite Site grouped five workplaces which were close to one another but 16 kilometres from the Central Site. The Peripheral Site grouped the remaining seven scattered workplaces that had no administrative or organisational links with the Central Site. Each respondent was categorised as a “clinician”, “researcher”, or “clinician-researcher” based on their reported key tasks, place of work and preferred title.

The second section of the survey asked social network questions (Table 1). Each question provided a roster of members’ names, job titles and primary place of work as an aid to memory. Past, Current and Future intended network graphs (sociograms) were constructed from these questions (i.e. each question produced a separate sociogram with the same respondents as the nodes but different configurations of ties between them) and the type of tie (e.g. work colleague, known by reputation) and current strength of the relationship was also given. Questions sought to differentiate between a working relationship (e.g. shared care of patients or worked in the same lab) which may involve little individual choice of partners, and collaborative relationships (e.g. co-investigators on a funded project, participating in a clinical trial or audits) which involve negotiation and choice. Strength of tie sought to quantify the tie in some way as working with someone once in 1982 but not again cannot be compared with someone who has worked continuously with someone else over the last two years. The category “strong, enduring relationship” was included to describe situations such as in medical training where a strong mentoring relationship early in one’s career pervades all future work regardless of frequency of contact. Feedback from the pilot survey with clinicians and researchers from equivalent contexts found this an intuitive and useful category.

Social network answers were analysed using UCInet v6 [45] to determine each sociogram’s characteristics and to determine the influence of the two grouping categories on patterns of clustering. We measured density, number of network components, External-Internal (E-I) indices, clustering coefficients, and network diagrams were generated in NetDraw [46]. E-I index measures the number of ties within a defined group compared to the number of ties to outside of the group and tests for a pattern of links compared to links made randomly (α = 0.05). This shows whether the grouping category has an influence on ties or not. Categories considered were site (geographic proximity), and professional group (professional proximity). Clustering coefficients average the densities of each member’s immediate contacts and when compared with overall network densities can give another network level indication of local clustering. Comparisons were also made across Past, Current and Future Collaboration sociograms considering members’ past involvement in translational research, and strength and type of tie.

The response rate for the survey was 76.5% (52/68). A comparison of respondents and non-respondents showed that they were similar in gender distribution (χ ² (1, n = 68) 1.85, p = 0.17) and representation from Central, Satellite and Peripheral Sites (χ ² (2, n = 68) 5.27, p = 0.072). The lowest response rate came from the peripheral sites.

Sixty-three per cent of respondents worked at a Central site, 27.8% of members at a Satellite site and 9.2% were at a Peripheral site. Thirty eight per cent of respondents were classified as clinicians, 53.7% researchers and 7.4% clinician-researchers with joint clinical and academic positions. Based on survey responses, most members had been involved in translational research in the past (80.8%).

Sociograms are shown in Figures 1(a) – (c) and their parameters are summarised in Table 2. There was a difference in the way respondents interpreted the social network questions shown by only 39.7% reciprocity for the question “With whom have you collaborated” and 60.1% for the question “With whom have you worked”. Reciprocity measures whether ties are mutually acknowledged: a reciprocated tie is one in which A nominates B as a tie and B nominates A. For collaboration and working ties we expected this to be much higher. Aggregating collaboration ties with the work ties lifted reciprocity to 80.2%, so the Past Collaboration sociogram was constructed by combining these two ties. The Past Collaboration sociogram contained 990 ties with every respondent nominating between six and 49 other members. Fifty-one per cent of these ties were described as weak or very weak and 49% were described as strong or very strong.

A pattern of clustering was demonstrated across the Past Collaboration, Current Collaboration and Future Collaboration sociograms. Clustering coefficients for each were higher than their respective network-wide densities indicating areas of higher densities within groups. There were no exclusive groups unlinked to others: component analysis showed a single component in all three sociograms (not counting isolates as groups or components here).

A summary of the hypotheses and the findings is presented in Table 3. All three of the collaboration sociograms showed patterns of collaboration based on one or more factors. Past Collaborations were significantly influenced by geographic and professional proximity. The Current and Future Collaboration sociograms were significantly influenced by geographic proximity and past collaborative ties.

Working at the same hospital or university does not necessarily mean the individuals are known to each other or that they have anything other than familiarity with the organisation in common. The influence of geographic proximity however is seen across all three collaboration sociograms. This likely reflects the practical incentive of choosing a partner who is close at hand: face-to-face meetings are easier to arrange and the will to collaborate may be fostered through serendipitous meetings in the corridor, car park or cafeteria [17]. The difficulty of transferring complex knowledge or maintaining momentum on a project is harder when geographic distance limits frequency of access. This has implications for a geographically dispersed network such as the TRN, identifying opportunities for more collaboration if strategies to bridge sites are addressed. Difficulties in running geographically dispersed collaborations include the need for adequate technology for virtual communication (e.g. teleconferences or video links), adaptation to this form of communication where social cues such body language and tone of voice can be lost or hidden, and an altered process of building trust between partners [16]. Social interaction apart from actual project negotiations have been shown to be useful in building long distance trust [47] so the inclusion of social gatherings in the TRN calendar of events would be recommended. The strengths of same site collaborations can be fostered through ready local access to resources at each of the sites.

The decrease in the within professional group ties from Past to Current and Future sociograms suggest that more members are currently taking advantage of, or are intending in the future to take advantage of, the opportunities the TRN structure provides for interdisciplinary collaboration. This is a significant finding as the gap between clinicians and researchers is seen as a major barrier in the translational research literature [40, 48].

The majority of collaborations in the Current and Future sociograms were with members who had a pre-existing working relationship. Considering the ties in the Current Collaboration sociogram that were between members with a pre-existing working relationship only 52.4% were described as strong or very strong meaning some of the pre-existing relationships that were being renewed were previously weak or very weak. The suspicion is that the structure of the network has allowed the renewal of old, weak ties and has bridged some gaps.

The number of Future intended collaborations reported was very encouraging and showed an enthusiasm for TRN activities. The number of intended ties based on reputation, or ties described as very weak or non-existent before the formation of the network were also reasons to be encouraged that the network was increasing connectivity between members that had no previous working relationship. Reminding members of their future intended collaborations after some time has elapsed and enthusiasm has waned may reinforce and realise those intentions.

People may choose to collaborate if they recognize that the other person holds expertise or access to resources that could assist them [49, 50] but learning what expertise or resources are available among members, often held collectively in clusters, can be difficult. TRN email and website features are one source of information but another is personal introduction via a go-between or broker [51]. The key roles of the TRN director and TRN manager as central actors and brokers, co-ordinating and linking together members is investigated further in separate studies [13‐15]. The star–shaped pattern of the sociograms [52], most evident in the Current Collaboration sociogram (Figure 1(b)) shows the strong influence that one or two actors can exert on a network and how vulnerable the TRN is to their loss. The TRN director had by far the most ties to other members in the Past Collaboration sociogram while the TRN manager dominates the Current Collaborations. She had systematically contacted all members and visited or interacted with over half of the membership at the time of the survey. These two members, having knowledge of the expertise of most members, are in an ideal network position to broker new collaborations. The increase in future intended collaborations between members known by reputation or having only very weak ties previously shows that members are reaching beyond their familiar associates. Whether this is through the brokerage activity of the TRN director and manager is not clear. Future surveys will show whether intended ties become actual collaborations and whether projects bear fruit.

While this study has focused on just one translational research network, albeit a complex one, our results have relevance to many similar networks and alliances that have been founded to address the research–practice gap: for example the Collaboratives for Leadership in Applied Health Research and Care (CLAHRCs) in the UK [53, 54] and the interorganisational alliances awarded funded by the Clinical and Translational Science Awards in the USA [28, 55‐57]. Many face similar issues of bridging cultural and epistemic differences between researchers and clinicians and facilitating geographic distance.

Diverse groups of people are brought together in healthcare networks in order to maximize cross-fertilization of ideas, transfer of knowledge, co-ordination of activities and access to resources. There are both strengths and opportunities to be found in clusters. Cohesive groups can function as highly productive work teams, holding and developing a store of knowledge and expertise. However, to prevent such clusters becoming islands, the gaps between them need to be recognised and ways found to bridge them. Research on this TRN has shown that patterns of collaboration are based on clustering by geographic proximity and previous collaborations. Future intended collaborations include a significant number of weak ties and ties based on other members’ reputations implying that the TRN has provided opportunities for partnership that may not have been available before. The highly centralised shape of the Current Collaboration sociogram suggests co-ordination and mediation by key network players. Geographic proximity remains a significant influence on choice of partner perhaps reflecting members’ preference for working locally.