The social networks of hospital staff: A realist synthesis

Objectives The social ties people have with one another are known to influence behaviour, and how information is accessed and interpreted. It is unclear, however, how the social networks that exist in multi-professional health care workplaces might be used to improve quality in hospitals. This paper develops explanatory theory using realist synthesis to illuminate the details and significance of the social ties between health care workers. Specifically we ask: How, why, for whom, to what extent and in what context, do the social ties of staff within a hospital influence quality of service delivery, including quality improvement? Methods From a total of 75 included documents identified through an extensive systematic literature search, data were extracted and analysed to identify emergent explanatory statements. Results The synthesis found that within the hospital workforce, an individual’s place in the social whole can be understood across four identified domains: (1) social group, (2) hierarchy, (3) bridging distance and (4) discourse. Thirty-five context-mechanism-outcome configurations were developed across these domains. Conclusions The relative position of individual health care workers within the overall social network in hospitals is associated with influence and agency. As such, power to bring about change is inequitably and socially situated, and subject to specific contexts. The findings of this realist synthesis offer a lens through which to understand social ties in hospitals. The findings can help identify possible strategies for intervention to improve communication and distribution of power, for individual, team and wider multi-professional behavioural change in hospitals.

Nurses under stress tend to look for support from nurses who are like themselves in some way, and are also experiencing stress or burnout. Nurses under stress were centrally located on the day shift, and more peripherally located on evening and night shifts. Nursing managers were not always centrally positioned in the network (varied by shift), impacting on their ability to provide support. 3 Anderson 3 2011 US Operating room staff of general surgical and neurosurgical specialities large teaching hospital in US Staffing and case data from all operations in 2008 Core-periphery structure, Coreness Centrality: Closeness, Betweeness, Eigenvector Both specialties showed a core and periphery structure, more pronounced in neurosurgery. Core staff tended to work on longer cases, earlier in the day, and in larger teams. Nurses were disproportionately core staff. 4 Altalib 4 2017 US Neurologists and other health care providers to patients with epilepsy in US.
Patient-sharing data from electronic health records of patients with epilepsy/convulsions.
Node degree Adjusted node degree Connectivity increased over time, with networking within and between facilities. 5 Bae 5 (Bae et al., 2017) 2017 US Oncology nurses in an acute care hospital in the US Roster method to explore patterns of assistance at work and demographic electronic survey.
Centrality: In-degree, Out-degree, Incloseness, Out-closeness, Betweenness, Eigenvector, Shortest path in, Shortest path out Other: Density, Reciprocity, Transitivity Help was perceived to be given more than received. Shift (day >night) and education level (less for degree holders) were significantly associated with amount of help received by each nurse. Nurses working overtime provided more help to colleagues.
6 Barrot 6 2013 US Anethesiologists in charge of purchasing equipment at a specific hospital, as identified as manufacturers contact person Survey in which asked to name up to 6 people in work-related network, and the connections those people might have with one another Centrality Clustering of purchasing anesthesiologists Network overall between purchasing anesthesiologists at different hospitals was fragmented, however there was one main component, and one central node. Some individual purchasing anesthesiologists were isolated 7 Benham-Hutchins 7 ) 2010 US Staff involved in patient transfers (hand-offs) between departments in an acute care hospitals in the US.
Questionnaires completed by staff about communication activity between themselves and colleagues during five patient hand-offs.
Individual (provider): betweenness centrality, closeness centrality, eigenvector centrality and total degree centrality Network (handoff): betweenness centralization and hierarchy measures Information exchange during hand-offs were found to be nonlinear, unpredictable and shaped by the information needs of the providers. Patient handoffs often coincided with shift changes, adding further complexity to communication. Ring method used to identify contacts in 1 st , 2 nd and 3 rd social rings of closeness.

Journal of Health Services Research & Policy
Closeness of rings Nurses wanted a larger network. Nurses trusted those colleagues in the closest rings more, who were likely to be peers. They were more likely to discuss major decisions or emotional support with those in the closer ring. Trust and respect were linked to those in closer rings. 12 Chan Relatively infrequent medication advice-seeing was seen, with professional group cliques. Reciprociity of advice-seeking was 30%. Hierarchy (not based on seniority) was seen, where key individuals were sought for advice, as well as gatekeepers. 16 Creswick 16 2015 Australia Staff of renal and respiratory ward of Australian teaching hospital.
Roster method -for every colleague respondents asked to indicate frequency of medication advice seeking from that colleague. Roster method used in face-to-face interview to identify "highly productive working relationships". Measures of distress and burnout and social support also collected.

In-degree Reciprocity
Both network position and perception of social support are important (and not interchangeable) for reducing workplace distress. Perception of social support is also important for reducing burnout 26 Heiligers In-degree, out-degree, reciprocation Use of equipment was associated with out-degree and reciprocal nominations. In-degree nominations were not associated with use of equipment. Visualisation of sociogram When controlled for sales calls and individual characteristics of the physician, earlier adoption was associated with network and selfreported leadership. Physicians with lower leadership scores were found to be more influenced by peers. More central physicians and those who prescribed more (and therefore more credible), were found to be the most influential in the network. Sub-groups were seen of physicians of different ethnic backgrounds, with a single broker between the two groups. 32 Iyengar 32 2015 US Physicians prescribing a new drug associated with potential high risk in three large cities in the US Name generator method used as part of a postal survey to identify discussion and referral ties (conducted by a drug company). In addition prescribing data collected for 17 months after the new drug was introduced.
In-degree centrality Discussion and referral ties found to influence the trial of a new drug by physicians (particularly those with low self-reported leadership), but not repeat prescribing of that drug. Close colleagues (i.e. working in the same institution) have a peer influence on prescribing, moderated by status, which increases with time. 33 Janmano 33 2018 Thailand Staff of a district hospital in Thailand, which included male and female general medical wards, an intensive-care unit (ICU), and a private medical ward.
Name generator as part of interview on medication advice-seeking. UCINET software used in analysis. Data on medication error reporting also collected.
In-degree, out-degree, Betweenness centrality Similar out-degree across cadres. Pharmacists and physicians were most central in network (in-degree). However different centralities for different aspects of medication chain. Females and those with lower degrees were more likely to report errors. In-degree centrality was the most influential characteristic for reporting medication errors in the hospital inpatient medication system Name generator method used (only 10 staff included, but 25 more named). UCINET software used in analysis.
Density, reciprocity, centrality Group density found to be very low, suggesting difficulty in group working and processes of achieving consensus, as well as limited support on problem-solving. 37 Lindberg 37 2013 US Hospital-based outpatient haemodialysis unit in the US.
A social networks survey was conducted before and after the intervention, in addition to other methods of process and outcome data collection.
Intervention was a quality improvement collaborative established, involving 21 haemodialysis facilities. This intervention included "positive deviance" team qualityimprovement activities to reduce hospitalacquired infections.
Connectivity, inclusion, reach, centralisation Following the intervention, staff social networks were found to be larger, and the network for collaborative ties more centralised. 38 Llupia 38 2016 Spain Healthcare workers from two hospital departments following an influenza vaccination campaign at a hospital in Spain Name generator method during face-to-face interviews focusing on who individuals talked to or shared activities with during the vaccination campaign.
Dyadic analysis: sender, receiver, mutuality Links were more likely between healthcare workers from the same profession, sex, age, and department, but not between healthcare workers who shared the same vaccination behaviour. Older and vaccinated HCWs were more likely to be named by interviewees. 39 Lurie 39 2009 US Staff working in intensive care unit in hospital in the US.
Observation of team rounds, followed by a survey using roster method to identify frequency of communication with colleagues and patients' family members.
Visual inspection of sociograms Two different teams were observed for one unstable and one stable patient. The staff networks around the two patients were observed to be very different, with a physician being central in the first team, which is denser with the family more connected (unstable patient), and nursing staff central in the second, which is less dense with the family more peripheral 40 MacPhee 40 2000 US Nurses working flexible and traditional schedules in a tertiary referral paediatric hospital in the US.
Name generator with closeness rings diagram used as part of a survey. Emotional support and satisfaction variables also measured.
Position of alter in social ring structure Flexible and traditional nurses were found to report similar social networks. Colleagues who were closest and provided the most support to nurses were peers, rather than managers. Name generator used in questionnaire, including tie strength for advice ties. UCINET software was used. Attitudes towards evidence-based medicine were also measured.

Physician dyads
Homophily was seen to result in inter-physician collaboration, particularly the same field of specialization, co-authorship, affiliation with the same organizations, and similarities in attitude toward evidence-based medicine. Name generator (with roster list for reference) used to identify colleagues with whom an individual discussed professional matters. Propensity to adopt evidencebased medicine was also measured.
Ego-network constraint indicator (controls: size, number of ties in egonetwork) Physicians who had highly constrained interpersonal networks were less likely to report adopting evidence-based medicine. The presence of structural holes in networks were associated with a greater likelihood of a physician using evidence-based medicine. 45 Mascia 45 2013 Italy Physicians employed in six hospitals in an Italian local health authority (LHA).
Name generator used in questionnaire, including tie strength for advice ties. UCINET software was used. Attitudes towards evidence-based medicine were also measured.

Network Coreness score, derived from centrality and interconnectedness
Those with less self-reported use of evidence-based medicine were more central in the network than those using evidence-based medicine. 46 Mascia 46 2015 Italy Physicians working across six hospital sites in an Italian local health authority (LHA).

Study used data from Mascia 2011(a), 2011(b) and 2013
Multiple regression-quadratic assignment procedure (MR-QAP), using physician dyads Professional (speciality) homophily was found to be of stronger influence than institutional homophily. Doctors graduating less recently were seen to show fewer collaborative ties. Doctors holding managerial roles did not collaborate with others holding similar roles Participant observation of 'primary participants' for a minimum of 3 hrs, looking for intersections/interruptions ('secondary participants'). Two-perspective method used to examine each interruption. Social network of interruptions constructed from data using Gephi software.
In-degree, out-degree, standardised indegree, standardised out-degree, weighted in-degree, weighted outdegree, sociometric status, generalised in-degree centrality, generalised outdegree centrality, network density, modularity The in-charge nurse role was the most frequently interrupted. However the bedside nurse role initiated most interruptions, and interrupted every other role in the ICU. Team-leader and in-charge roles had high in-degree values, and had sole access to keys for the restricted medications cupboard, which the bedside nurse was unable to access without. Nursing roles were highly connected. 48 Meltzer 48 2010 US Physicians working in the general inpatient medicine services at a hospital in the US.
Roster method to identify frequency and quality of interactions with physician colleagues.
Degree centrality, net degree, density, betweenness centrality Sub-groups of physicians by speciality, gender, self-reported hospitalist status, and research activity were seen. More connected physicians communicated with other more highly connected physicians. Non-redundant connections held by a specific group is explored. 49 Menchick 49 2010 US Physicians working at 6 academic medical centres of different sizes in the US Medication advice-seeking networks constructed for high-esteem and lowesteem hospitals. Structurally equivalent positions were compared between high and low esteem hospitals.
Esteem: In-degree centrality, adjusted by number of physicians in hospital (population size) Within high-esteemed hospitals, physicians from higher esteemed medical schools, and those doing less clinical work were held in higher esteem. In lower-esteemed hospitals, physicians who read across journals and engaged in more patient care were more highly esteemed. Stigma was associated with graduating from a highesteemed medical school but now working in a low-esteemed hospital. 50 Myers 50 2016 US Surgical teams in a large academic teaching hospital Dataset of surgical procedures, sharpsrelated percutaneous blood and body fluid exposures (BBFE) and index of collaboration, based on social network measures applied to surgical team data.
Team stability index score Incidence of BBFE reduced with greater team collaboration score. Not much absolute reduction in real terms, but was significant for non-suture needle device related BBFE (and large numbers in analysis) Degree centrality, closeness centrality, eigenvector centrality Divergent strategic behaviour is undertaken more by those who are more central in the social structure, and also by those who are boundary spanners (though not exclusively). Socialisation with diverse people allows one to 'learn' innovation. 54 Patterson 54 2013 US Emergency department staff in a hospital in the US.

Journal of Health Services Research & Policy
Roster method using a survey was used to construction social networks of communication about problem-solving, medication advice, and socialising.

Network density Network centralisation
In-degree centrality Variation was seen in communication over time, shift, and for different purposes. General problem solving communication was also associated with medication advice-seeking and socialising (however these two were only loosely linked themselves). 55 Pinelli 55 2015 US Patient discharges from a medical unit at a university teaching hospital in the US.
Electronic medical records for seven patient discharges were used and providers were asked with whom they communicated by chain interview technique.
Density, betweenness centrality Junior doctors were most central in the discharge process, which was a multidisciplinary event.
56 Rangachari 56 2008 US Purposive sample of two good-performing hospitals and two poor performing hospitals in the US. Performance was assessed specifically in regards to medical coding.
Roster method using a survey was used to construct knowledge sharing network in regards to quality measurement and hospital coding performance.

Structural equivalence analysis (SEA) and visualisation of SEA
A stark difference was seen in communication structure between good performing and poorer performing hospitals. In good performing hospitals there was better interdisciplinary communication, with senior managers taking a proactive central role, and measures taken to raise the status of medical coding in the hospital.

Reed Elliot 2012 Brazil Staff in four hospitals in Brazil
Roster method used to identify ties and tie strength. Block modelling was used in the analysis.
Concor groups/"blocks" Formal hierarchies did not predict block modelling findings. Hospitals varied as to who was in each block/core. Core-periphery patterns were seen. This is attributed by the authors to the history of the organisations and related to institutional culture. 58 Regts 58 2016 Netherlands Medical and orthopaedic nurses in four hospitals in the Netherlands.
Roster method to identify advice network of nurses Personality traits, job satisfaction and performance also measured.
Affect-based network centrality (combination of in-degree and outdegree centrality) Study found an interaction between certain personality traits (neuroticism and introversion and stable extraversion) and benefit from network position for satisfaction and performance. Degree centrality, dyadic ties The burns unit was generally well connected across different professional groups. However, Nurses tended to communicate more with other nurses. Doctors tended to be more central overall. 63 Sreeramoju 63 2018 US Healthcare staff on six medical wards at an academic hospital in the US.

Journal of Health Services Research & Policy
Intervention study to improve patient safety culture related to infection prevention. Three wards were assigned to a social change intervention, implemented over nine months. A social network survey was conducted at 6, 15 and 24 months, to identify which colleagues an individual collaborated with to prevent infections, colleagues with whom they were involved in projects, colleagues who inspire them, and colleagues with whom they would like to work in the future. Patient safety culture and rates of hospitalacquired infection were also measured.
Degree centrality Core staff in preventing hospital infections were identified as the patient's nurse, patient care technician, charge nurse, ward manager, and ward clerk. There was no deterioration in hospitalacquired infections in intervention arm. 64 Sykes 64 2011 US Doctors working in a private hospital in the US Roster method used in survey to identify advice networks. Adoption of electronic medical records was also measured.
Degree centrality, eigenvector centrality (adapted) The more central doctors were more likely to be associated with higher patient satisfaction, but were less likely to adopt electronic medical records. This was hypothesised by authors to be due to ease of alternative communication pathways and compliance with organisational norms. 65 Sykes Social networks data relating to information exchange at work and sharing of personal issues. Organisational citizenship behaviour, workplace stress and work satisfaction were also measured.
Degree centrality Work network centrality positioning was different to friendship network centrality. Centrality was associated with organisational citizenship behaviour, which also reduced work stress and increased work satisfaction. Expected reciprocity associated with organisational citizenship behaviour was discussed by authors. 68 Venkatesh 68 ) 2011 US Staff working in a hospital in the US, after the implementation of a new IT system Roster method used as part of a staff questionnaire, to identify advice networks. UCINET software used in analysis. Data on IT system use and patient satisfaction also collected.
In-group degree centrality Out-group degree centrality A different influence for centrality on IT system use was found in different professional groups. Ties among doctors had a negative association with IT use, as did ties with doctors on IT system use by other staff. This is despite the use of the IT system being associated with better patient satisfaction. 69 Wagter 69 2012 Netherlands Staff of intensive and medium care unit in a large teaching hospital in the Netherlands Roster method used to identify interprofessional learning networks.
Density, reciprocity, tie strength The underlying network structure was found to consist of two distinct networks, with a single group of nurses being isolated. Staff were found to learn from within their own profession, particularly from senior doctors. Authors comment that opportunity (e.g. physical proximity) and homophily contributed to explain the patterns seen. 70 Walton 70 2010 Canada Staff involved in ward rounds on general paediatric wards of a tertiary care, universityaffiliated children's hospital in Canada Non-participant observation of eighteen ward rounds (noting all verbal interactions). Self-administered survey also collected ratings of the primary purpose of ward round, its utility for patient care, education and team administration, and the degree of disruption caused by the presence of the observer Freeman normalised graph centrality index Ward rounds conformed to patterns, and were often dominated by the same individuals. Time taken per patient was longer at the start than at the end of the ward round. Perceptions of educational value by ward round attendees was higher than the authors' opinions. 71 West 71 1999 UK Clinical Directors of Medicine Directors of Nursing working in hospitals in England Name generator used to identify people with whom the individual had spoken to about important professional matters, and how those identified were connected to one another.
Group degree centrality, actor information centrality Directors of nursing were found to be in less dense networks, more hierarchical, with greater bridging/brokering role. Whereas medical directors were found to have denser networks which were less hierarchical, and were not seen to broker across structural holes in the same way as directors of nursing. 72 West 72

UK Clinical Directors of Medicine Directors of Nursing working in hospitals in England
Name generator used to identify people with whom the individual had spoken to about important professional matters, and how those identified were connected to one another.
Category of alters There were few doctors in the nurse managers' networks, and few non-doctors in the medical directors' networks. Both networks did however include some managers. Gender was found to be strongly associated with communication patterns, particularly amongst male doctors, who spoke to mostly other male doctors (also those similar in age and rank), but also as many male nurses as female nurses. 73 Westbrook 73 2007 Australia Staff working on three wards in a public hospital in Australia.
Roster method used to identify which colleagues individuals sought medication advice from, or socialised with.

Sociograms
Staff interacted mostly with those from the same professional groups, however some professionals were seen to occupy bridging roles, such as the pharmacist, junior doctors and ward clerk 74 Yi 74 2017 China Doctors working in outpatient department of a large children's hospital in China, which had recently implemented a new EMR system.
Roster method used to identify adviceseeking and advice-giving networks in regards to IT-related information. Data were also collected on electronic medical records usage.
Seeking-network closure Giving-network closure (Controls: seeking-network size, givingnetwork size) Contacts who were connected with each other in advice-seeking networks, but not in advice-giving networks, were more likely to use electronic medical records. Free recall method used to construct adviceseeking networks for technological/computer-related and also clinical/professional matters.

Journal of Health Services Research & Policy
In-degree centrality and Cattel's method to identify influential physicians Medical advice-seeking networks were more cohesive than technical advice-seeking. Residents tended to go to mentors for advice, which may have presented a limit to IT adoption. Women were found to seek advice from women and men to seek advice from men.

CMOC3
When a health care worker has an existing reciprocal relationship with a peer (C), they will preferentially seek advice or support from that person (O), because they trust the person (M1) and feel comfortable (M2).
"Consistent across hospitals were significant positive effects for reciprocal relationships, indicating that there are pairs of professionals that tend to consult each other (mutual)."

CMOC4
When health care workers share a representation (C), this can lead to redundant information (O1), group-think (O2) and echo (O3), due to insular communication (M).
"Physicians who were highly constrained in their interpersonal networks were less likely to selfreport adopting EBM. This finding is consistent with the theory of structural holes (Burt, 1992). …One important risk is that subjects may start to trust their group judgment more than information from the surrounding scientific world." 4 CMOC5a When a health care worker is central within a social group (C), their personal behaviours will embody group norms (O) because they will hold tightly to group identity as own (M).
"..there is a significant negative association between the physicians' propensity to use EBM [evidence-based medicine] and the coreness they exhibit in their organization…the core is formed by physicians having a significantly lower propensity towards EBM than their peers located in the peripheral part of the network." 5

Journal of Health Services Research & Policy
The social networks of hospital staff: a realist synthesis Claire Blacklock, Amy Darwin, Mike English, Jacob McKnight, Lisa Hinton, Elinor Harriss and Geoff Wong Context-mechanism-outcome configuration (CMOC) Illustrative quotes

CMOC5b
When a health care worker is peripheral within a social group (C), they may deviate in behaviour from group norms (O1), and oppose/conflict with central members of the group (O2) because they may not wholly identify with the group representation (M).
"… less-central individuals are likely exposed to fewer negative comments about the system. Thus, less-central individuals are more likely to adopt and use the new system." 6 CMOC6a When a health care worker is central within a social group (C), they will have greater power over group norms (O), because they have greater influence over group narratives and representation (M).
"Findings suggest that the more centrally located a manager is in a social cluster, the more likely that manager is to integrate new ideas into new organizational capabilities." 7 CMOC6b When a health care worker is peripheral within a social group (C), they will have little power over group norms (O), because they have little influence over group narratives and representation (M).
"…structure of this model underlies a group tightly connected physicians who interact strongly in order to exchange relevant knowledge, and a large number of less cohesive clinicians who are more likely to be connected amongst themselves than to members of the core part of the network. This result might be interpreted as a marginalization of physicians who are more prone to use EBM in their clinical practice."

CMOC6c
When a health care worker is central within a social group (C1) and has power over material resources (C2), they will have additional power over group norms (O) because they own the distribution of material resources and decision-making authority (M).
"Nurse coordinators in operating rooms often exercised a dominating, hierarchical form of power in relation to surgeons as they approved or denied access to space, time and resources for surgical operations, based on the urgency of individual cases" 8 "In the present case, high centrality occurred with department heads, such as the high connection density in Figure 2. In general, those with high work network centrality were those experienced nurses who not only controlled physical resources but were also in command of decision-making powers." 9 CMOC7a Where a health care worker requires diverse information to undertake their role effectively, e.g. senior nurse, junior doctor (C), they will become more central in the workplace information-sharing network (O), because they must frequently seek information from others (M1) and others frequently seek information from them (M).
"Centrality was not linked to gender, age and seniority in psychiatry, but was associated with job and activity." 10 CMOC7b When a health care worker is peripheral in the workplace information-sharing network (C), they risk isolation (O), because they are not actively engaged in informationseeking (M1) or information-sharing (M2) with colleagues.
"The communication network for Unit 5 has seven isolates. In addition, two of the more influential RNs (high Eigenvector Centrality) are not communicating with staff not on their shift, and there are a number of "pendants" (people with single links). The pendants are usually PCTs"