When Clinicians Group Together: A Systematic Scoping Review of Clustering in Patient-Sharing Networks

When Clinicians Group Together: A Systematic Scoping Review of Clustering in Patient-Sharing Networks | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article When Clinicians Group Together: A Systematic Scoping Review of Clustering in Patient-Sharing Networks Alina Denham, Porooshat Dadgostar, Qiuyuan Qin, Sule Yilmaz, Reza Yousefi Nooraie This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4437662/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 25 Feb, 2026 Read the published version in BMC Health Services Research → Version 1 posted 12 You are reading this latest preprint version Abstract Background . Improvements in patient safety and quality of care can be achieved by improvements in clinicians’ teamwork, coordination and communication. Growing research examines the structure and dynamics of clinician networks using social network analysis. Such networks can have clusters of healthcare professionals within them, but systematized knowledge on these clusters is lacking. Our goal was to review the evidence on determinants and characteristics of healthcare professional clustering in patient-sharing networks and their associations with patient outcomes. Methods. We searched for English-language peer-reviewed studies published up until January 4, 2021 using PubMed and EMBASE and an existing scoping review on patient-sharing by DuGoff et al (2018). We performed a systematic scoping review in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines. We performed title and abstract screening and full-text screening to identify studies that used social network analysis to examine relationships between patient-sharing network clusters and health outcomes. From the twelve eligible studies, we extracted study information such as study design and setting, population, patient-sharing definition, network measures, clustering definition, health outcomes, and reported associations. Results. The studies varied considerably in definitions and measures of patient-sharing relations, definitions and structural measures of network clusters, settings, study population, and health outcomes. The general patterns indicate that busier physician networks (i.e., networks with more connections among physicians) are associated with worse health outcomes and better-connected physician networks are associated with better health outcomes. Conclusion. The majority of existing studies are exploratory. Rigorous theoretical grounding, interventional studies, and mixed-methods studies would help to strengthen patient-sharing research and advance our understanding of how patient-sharing clustering affects patient outcomes. patient-sharing social network analysis network clusters physician networks health outcomes Figures Figure 1 INTRODUCTION Patient-sharing network analysis is a novel approach to reveal the patterns and dynamics of relationships among healthcare professionals (HCPs) using existing administrative data (e.g., insurance claims) or electronic medical records.[ 1 , 2 ] Patient-sharing relations are usually assumed when multiple HCPs provide care to a patient, hence are an indirect indicator of provider communication, referral, collaboration, and care coordination. Increasing healthcare integration and the use of technology facilitate the process of referral and coordination.[ 3 ] Naturally, integrated care models involve a high level of patient-sharing; and patients, particularly those with complex needs, often receive care from multiple HCPs. Improving the processes of patient-sharing between HCPs is a key component of safe and high-quality care.[ 4 ] In the last decade, there has been growing interest in analyzing the structure and dynamics of patient-sharing networks using social network analysis.[ 2 , 3 , 5 ] Social network analysis allows studying the patterns of relationship among HCPs (actors of the network), using visual mapping of relationship networks as well as various structural measures of network composition, existence and characteristics of clusters, relationship strength, actors’ roles and positions, etc.[ 6 ] Social network analysis of patient-sharing networks can be used to describe behavioral dynamics in a healthcare system, to analyze care coordination, and team-based care. It can also be used to examine the process and impact of social influence among network actors (e.g., adopting new clinical guideline by peer influence).[ 7 ] While some of these questions require collecting rich data from the network actors themselves, most network analysis studies of HCP networks rely on administrative data such as insurance claims and hospital discharge records, given their large scale, ease of analysis, and natural connection to patient outcomes. Better communication and coordination among HCPs should positively affect patient care and ultimately patient outcomes.[ 4 , 8 ] Therefore, many patient-sharing network studies examine how network characteristics affect patient outcomes. However, a previous scoping review of patient-sharing network studies did not find consistent patterns in the relationships between network measures and patient outcomes, noting a substantial range of data sources, network measures and statistical approaches to estimation.[ 2 ] One important, yet less studied, aspect of patient-sharing network analysis is studying how HCP groups, clusters, or communities are formed within networks, what these clusters look like, and whether clustering among HCPs affects patient outcomes. Network clusters are groups of actors who are better connected to each other, with looser connections to others outside the cluster.[ 9 ] Studying network clusters provides valuable insights to the meso-level patterns and dynamics of connectivity, beyond the dyadic relations between individuals. The notion of clustering in human networks has been studied for decades. Humans join clusters to maximize the benefit and safety of a group,[ 10 ] and then adhere to cluster norms and cultures,[ 11 ] which themselves reshape over time. One consistent theme for the formation of clusters among professionals is homophily (or the tendency to connect to others with similar characteristics). Some studies have shown that HCPs are more likely to refer patients to colleagues that they personally know (through common demographic backgrounds or education,[ 12 ] or common activities such as journal clubs, etc.[ 13 ]), are geographically closer (e.g., in the same building or area),[ 12 ] have facilitated communication means (e.g., through face to face communication or electronic health records),[ 13 , 14 ] share insurance plans, or are affiliated with the same health systems.[ 12 , 15 ] Studying clusters of HCP networks is important to understand how healthcare is delivered and how healthcare systems can improve patient outcomes via changes in structures and processes. Such understanding can be developed by examining how these clusters are formed (Why do some HCPs prefer to group together?) and how clusters affect HCP behavior (Do HCPs adopt the normative practice of the cluster?) and patient outcomes (Do patients benefit differently if their HCPs belong to the same cluster or particular clusters?). Our systematic scoping review aims to analyze the existing evidence on HCP clustering and its relationship with patient outcomes in patient-sharing network analysis studies. METHODS Identifying relevant literature We followed the five-step framework proposed by Arksey and O’Malley for scoping reviews.[ 16 ] Previously, DuGoff et al (2018)[ 2 ] conducted a systematic review of patient-sharing networks using a structured search strategy. We expanded and updated their search strategy and systematically searched PubMed and EMBASE databases in January 2021. Our detailed search strategy is provided in the Appendix. Selecting studies We included peer-reviewed articles in English published up until January 2021 that examined the clustering of HCPs in patient-sharing networks and reported at least one patient outcome, either at individual or systemic levels. HCP network clusters could be defined by regional (e.g., a catchment area or a city) or institutional (e.g., a hospital) boundaries, or determined by cohesive patterns (e.g., community detection algorithms) in patient-sharing networks. The latter means that HCPs tended to share patients within a cluster more than sharing outside the cluster. To be eligible for inclusion, a published study had to employ network analysis, examine clustering of providers, examine at least one health outcome, and be published in English. We excluded ongoing studies, protocols, and conference abstracts. Abstracts were screened for eligibility by pairs of reviewers. Sixty-six selected articles underwent a full-text screening: each article was reviewed by two blind independent reviewers (articles were divided between two pairs of reviewers) and agreement on each article was achieved. Discrepancies were discussed and resolved in meetings. During full-text screening, the study of HCP clusters and a focus on patient outcomes were ascertained and articles that did not meet these criteria were excluded (Fig. 1 ). Extracting and charting To assess the methodological quality of included studies, we applied the New Castle-Ottawa quality assessment scale for cross-sectional[ 17 ] cohort and case-control studies. All included studies used patient records to determine HCP clusters as well as outcome measures, making a few items of the scale inapplicable to our study context. We simplified the scale and focused on the study design and the existence and nature of comparison groups and follow-ups. The included studies were divided between pairs of reviewers. Within pairs, reviewers independently extracted the following data from each article: population, study design, source of data, patient-sharing definition, patient-sharing measure(s), patient-sharing timeframe, clustering definitions, patient outcome(s), and reported associations between patient-sharing clustering measures and patient outcome(s). Each pair then reached agreement; discrepancies were discussed and resolved in meetings. Data synthesis and analysis We assessed for clinical heterogeneity, presented as diversity in terms of patient populations, HCPs, definition of patient sharing, measures of clustering, and outcome assessment. Given these methodological and clinical heterogeneities, we summarized the study findings narratively. We descriptively summarized the characteristics of studies, methods of clustering, and outcome measures. We qualitatively synthesized the association between clustering and health outcomes. We summarized the main findings of each study, as presented by the authors, provided a narrative summary of findings, and thematically categorized the findings across studies, following a thematic analysis approach.[ 18 ] The team met regularly to discuss each study and the process of the formation of themes. RESULTS Review process and retrieval We identified 3,034 citations from the Bibliographic search in PubMed and EMBASE (Fig. 1 ) and 32 citations from DuGoff et al (2018).[ 2 ] We removed 21 duplicate records and excluded 2,979 ineligible abstracts. We reviewed the full texts of 66 papers and excluded 54 for reasons listed in the PRISMA diagram (Fig. 1 ). We included 12 empirical studies that assessed HCP clustering and patient outcomes. The characteristics of included studies are summarized in Table 1 . Table 1 Characterization of the Studies Included in the Review Study Quality Assessment Data Source Sample Size Setting Definition of Clusters Ties and Actors Barnett et al, 2012 Cross-sectional study Comparison group: no Follow up: no The study adjusts for hospital characteristics (number of hospital beds, number of physicians assigned to a hospital, proportion of PCPs among assigned physicians, mean shared patient volume per physician at a hospital, urban location, teaching status, ownership, nurse full time equivalents per 1000 inpatient days, percentage of admissions from Medicare and Medicaid patients). Medicare claims, Part A and Part B, 2006 2.6 million Medicare patients, 61,146 physicians, 528 hospitals, 51 HRRs Hospitals in 50 randomly sampled HRRs and Boston, the United States Hospitals. Physicians who each had a significant encounter with one or more common patient. A significant encounter was defined as face-to-face visits or meaningful procedures with a value of at least 2 RVUs. Casalino et al, 2015 Cross-sectional data over one year Comparison group: no Follow-up: no The study controls for patient characteristics (age, sex, dual eligibility status, race/ethnicity, number of chronic conditions) and physician characteristics (U.S. trained, sex, age, board-certified). Medicare claims Part B outpatient files, 2008 782,595 patients, 54,202 physicians, 386 physician practice communities . Hospitals; 5 states in the United States (OH, PA, TN, WA, WI) Clusters (termed physician practice communities, or PPCs) were defined as "smaller networks of physicians who share patients with each other much more often than they share patients with anyone else". PPCs were identified using community detection algorithms. Physicians who provided outpatient care. Donker et al, 2012 Cross-sectional study over 8 years Comparison group: England vs the Netherlands Follow-up: no The study controls for cluster specific mean MRSA rates. The NHS Hospital Episode Statistics (HES) data (England) and the Dutch National Medical Registry data (Netherlands), April 2006 - March 2007. England: 7,420,219 patients admitted to 146 hospitals, for a total of 12,929,171 admissions Netherlands: 1,676,704 patients admitted to 98 hospitals, for a total of 2,611,452 admissions Acute care hospitals in England, hospital regional clusters in England, England vs the Netherlands Hospitals were grouped into hospital regional clusters, based on the weights of the connections between all hospitals, using a community detection algorithm. Hospitals are the actors and patient sharing happened as a result of patient referrals from one hospital to another. DuGoff et al, 2018 Longitudinal study Comparison group: no Follow-up: 1 and 2 years The study controls for the hospital characteristics (median number of physician pairs per primary care physician (PCP), median number of co-occurrences, PCPs and specialist supply, hospital beds, total number and characteristics of Medicare fee-for-service beneficiaries, and region of the country). Medicare claims, 2012–2014. 142,016 PCPs connected to at least one other physician in 2012, representing 12,190,803 physician pairs HRRs; the United States Clusters are HRRs. Referrals between PCPs and specialists. Hollingsworth et al, 2016 Cross-sectional study over 4 years Comparison group: no Follow-up: 60 days post-surgery The study adjusts for hospital and hospital service area characteristics as well as the number of physicians. Medicare Provider Analysis and Review (MedPAR), 2008–2011 251,630 Medicare beneficiaries who underwent CABG. 466,243 physicians, 1,186 health systems Hospitals; the United States Clusters are health systems. Physicians directly involved in the care of patients with coronary artery bypass graft 30 days prior and 60 days after admission. Moen et al, 2016 Cross-sectional study Comparison group: Two HRRs with difference in rates of receiving implantable cardioverter defibrillator (ICD) therapy Follow-up: no The study controls for patient characteristics (age, sex, race). Medicare Part B claim data, 2008 Gary, IN: 10,350 cardiovascular patients,481 physicians South Bend, IN: 9,653 cardiovascular patients, 639 physicians. Two HRRs; Indiana, United States Physicians and hospitals within two adjacent HRRs with disparate adherence to clinical guidelines regarding patient selection for ICD therapy. Physicians who provided care to patients with cardiovascular disease. Pollack et al, 2012 Observational and retrospective Comparison group: no Follow-up: no The study adjusts for patient clinical (Gleason score, tumor stage, PSA results, comorbidity) and sociodemographic characteristics (age, race, community-level income, marital status). SEER-Medicare data from three cities, 2004–2005 4,520 men 2420 doctors Three cities; the United States Community detection algorithm (Girvan-Newman) to identify group of physicians with more frequent patient sharing in each city. Separate analysis was done in each city. Physicians who provided care to prostate cancer patients, classified based on their role in the diagnosis, providing care, or administered the cancer treatments, within a 12-month window after the diagnosis. Pollack et al, 2014 Retrospective, observational cohort study Comparison group: no Follow-up: one year The study controls for patient-level and community-level characteristics (comorbidities, race, ethnicity, census tract or zip level median income, urologist surgical volume). SEER Medicare data, 2004–2005 For network construction: 13,465 men with prostate cancer without metastatic disease. For analyses: 2,677 men in 5 cities. The sample in each city ranged in size from 270 to 1,224 men. Physician networks in five cities; the United States Network subgroups were defined by Girvan-Newman algorithm (also called "community structure"); each doctor was assigned to a subgroup. Patients, however, may have doctors that were assigned to multiple subgroups. For the purposes of the analyses, patients were assigned to the network subgroup of the urologist who performed their prostatectomy. Urologists who provided care to the same patients with localized prostate cancer who underwent radical prostatectomy. Stein et al, 2017 Cross-sectional study Comparison group: no Follow-up: no The study controls for county characteristics including the total number of providers in the provider communities seen by the patient, the total number of patients in the provider communities seen by the provider, and state-fixed effects. Medicaid claims data, 2008–2009 29,611 Medicaid enrollees 12 states; the United States Provider communities based on the frequency with which different providers were treating the same patients, all of whom had been diagnosed with opioid use disorders. Communities were defined using modularity maximization community detection. Providers who cared for the same patients with opioid use disorder during 2009. Uddin et al, 2012 Cross-sectional study over four years and two months. Comparison group: no Follow-up: no One health insurance company’s claims, Jan 2005 – Feb 2009 2229 patients,* 85 hospitals Hospitals; Australia Physician collaboration networks were hospitals. Physicians who provided care to hip replacement patients during their hospitalization period. Uddin et al, 2015 Cross-sectional data over 4 years and 2 months Comparison group: no Follow-up: no One health insurance company’s claims, January 2005 - February 2009 2352 patients, 2229 physicians, 85 hospitals. Hospitals; Australia Community detection algorithm to determine communities of doctors within each physician collaboration network (i.e. hospital) Physicians who provided care to hip replacement patients during their hospitalization. Uddin 2016 Cross-sectional data “over 5 years”** Comparison group: no Follow-up: no The study controls for patient-level characteristics (age, gender, comorbidity index (Charlson-Deyo index)). One health insurance company’s claims “over 5 years”** 2352 patients, 2229 physicians, 85 hospitals Hospitals; Australia Community detection firefly algorithm[ 33 ] optimization approach: maximization of internal links and minimization of external links Physicians who provided care to hip replacement patients during their hospitalization Notes: SEER = The Surveillance, Epidemiology, and End Results, a program of the National Cancer Institute. MRSA = Methicillin-resistant Staphylococcus aureus, a hospital-acquired infection. HRR = Hospital referral region. RVU = relative value unit, a measure of work or effort performed on a patient that is correlated to the revenue potential.[ 46 ] NHS = The National Health Services. CABG = coronary artery bypass graft. PSA = prostate specific antigen. * This article states that data from 2229 patients were used, which may be a typo, because two other studies that use the same data (Uddin et al, 2015; Uddin 2016) list 2229 physicians and 2352 patients. ** This article states “over a period of five years” but does not specify which years; it appears that the data source is exactly the same as in Uddin et al, 2012 and Uddin et al, 2015, which is Jan 2005 – Feb 2009, which is four years and two months. Study setting and data sources Eight studies were conducted in the United States,[ 19 – 26 ] three in Australia,[ 27 – 29 ] and one in the UK.[ 30 ] The settings of seven studies were hospitals and emergency departments,[ 20 , 24 , 25 , 27 – 30 ] 3 studies were hospital referral region,[ 21 , 22 , 26 ] and two studies were geographical regions (e.g., cities).[ 19 , 23 ] Five studies used Medicare data[ 20 – 22 , 24 , 26 ] and two used Medicare-SEER linked data.[ 19 , 23 ] One used Medicaid data,[ 25 ] three used Australian insurance organization data,[ 27 – 29 ] and one used the National Health Service (NHS) Hospital Statistics data.[ 30 ] Study design Nine studies were cross-sectional.[ 20 – 22 , 24 , 25 , 27 – 30 ] Of those, three were repeated cross-sectional studies.[ 24 , 29 , 30 ] Three studies were longitudinal.[ 19 , 23 , 26 ] Patient-sharing networks, actors, and ties Network actors were HCPs[ 19 – 29 ] and hospitals.[ 30 ] Patient-sharing definition varied across studies. Patient-sharing relations between pairs of HCPs or hospitals were defined as caring for the same patient within a specific time frame, based on care context, a care episode, or professional relationships. Four studies based their definition on care episodes, e.g., during a hospitalization or an ambulatory care episode;[ 20 , 27 – 29 ] five studies were based on clinical conditions like cardiovascular disease, prostate cancer, opioid use disorder, and life-threatening chronic disease;[ 19 , 21 , 23 – 25 ] one study was based on direct referrals from primary care physicians to specialists,[ 26 ] and one study was based on transfers between hospitals.[ 30 ] One study did not specify care setting or clinical conditions.[ 22 ] Three studies used a specific time window of providing care for the same patient as the definition of patient-sharing relations.[ 23 , 24 , 26 ] Definitions of clusters Seven studies used community detection algorithms to identify HCP clusters.[ 19 , 20 , 23 , 25 , 27 , 29 , 30 ] Community detection algorithms divide network nodes based on their relationship patterns into sets to maximize within-group and minimize between-group ties.[ 20 ] It is used to identify denser connected clusters within a larger network. Three studies[ 19 , 23 , 29 ] used the Girvan-Newman community detection algorithm, which relies on modularity maximization. It involves an iterative removal of high betweenness ties with the highest number of shortest paths between nodes passing through them (in other words, identifying the network weak points).[ 19 ] These ties probably bridge loosely connected clusters in the network. Girvan-Newman is one of the traditional community detection methods and is best for smaller networks.[ 31 ] Two studies[ 25 , 30 ] use the Fast Greedy algorithm. This method is based on the notion of local modularity.[ 32 ] It starts with considering individual nodes as their own clusters, then adding nodes to the clusters one at a time in a greedy approach (examining local optimization) to create new clusters, aiming to maximize network modularity. Merging nearby clusters continues until there is no improvement in modularity.[ 31 , 32 ] One study[ 20 ] used an algorithm called the Blondel model or the Louvain algorithm.[ 31 ] Similar to the Fast Greedy algorithm, this hierarchical clustering method starts with separate nodes and adds nodes to clusters in a greedy way to optimize modularity. To accommodate for larger network size, this model iteratively aggregates new clusters into “super-nodes” in a new network, and then repeats this procedure with the new aggregated network until there is no improvement in modularity. One paper[ 27 ] used the algorithm introduced by Amiri et al.[ 33 ] The Amiri algorithm is an evolutionary technique[ 34 ] that learns from previous steps, aiming for maximizing within-cluster relations and minimizing between-cluster relations at the same time. Four of these studies applied a multi-level approach, identifying collaboration networks based on hospitals[ 27 , 29 ] or cities,[ 19 , 23 ] and then applying a community detection algorithm to each sub-network separately. Five studies did not apply any community detection algorithm and a priori defined clusters as hospitals, medical practices, clinics, health systems, or hospital referral regions.[ 21 , 22 , 24 , 26 , 28 ] Measures of cluster characteristics Cluster structure The size of clusters (number of providers in each cluster) was the most commonly reported measure.[ 19 , 22 , 23 , 27 , 29 ] Uddin et al. (2015)[ 29 ] assessed the physician-to-patient ratio in physician communities. Studies used different measures to assess collaboration between HCPs within a cluster. Network density was used most to describe the connectivity of actors (e.g., physicians or hospitals). Hollingsworth et al. (2016)[ 24 ] used the bipartite clustering coefficient to measure in the tendency to form dense groups in clusters (health systems). Casalino et al. (2015) calculated the mean adjusted valued degree of physicians within the community.[ 20 ] Cluster composition Pollack et al. (2012) measured the percentage of patients and providers in large clusters (defined as clusters of more than 50 physicians).[ 23 ] Casalino et al. (2015)[ 20 ] and Pollack et al. (2012)[ 23 ] assessed the proportion of primary care physicians (PCPs) and urologists in each cluster, respectively. DuGoff et al. (2018)[ 26 ] used the median number of physician pairs per PCP in a cluster to evaluate the breadth of PCP network. Actors’ position The position of actors in clusters was assessed in three studies.[ 20 , 22 , 28 ] Casalino et al. (2015)[ 20 ] measured the betweenness centrality of PCPs and cluster-level relative betweenness centrality by dividing the PCPs betweenness centrality (an indicator of their intermediary role) by the average betweenness centrality of other HCPs to evaluate the bridging ability of PCPs. Barnett et al. (2012)[ 22 ] measured degree centrality for PCPs and relative degree centrality using similar approach to evaluate the popularity of PCPs. Uddin et al. (2012)[ 28 ] assessed both betweenness centrality and degree centrality of actors in network clusters, to identify central actors. Cluster dynamics The persistence of connections over time was assessed in one study.[ 26 ] DuGoff et al. (2018)[ 26 ] evaluated persistence of ties between providers by assessing whether the ties were present across years, and calculated the proportion of ties that persisted in each cluster. Behavior in clusters Studies also used cluster-level measures in relation to physician behaviors. Stein et al. (2017)[ 25 ] assessed the percentage of individuals prescribed opioid analgesics in each cluster to understand how provider communities differed in prescription patterns. Measures of cluster characteristics are defined in Table 2 . Table 2 Measures of Cluster Characteristics. Structural measures Explanation Studies CLUSTER LEVEL Cluster size The total number of individuals (actors) in the cluster Uddin et al, 2016; Pollack et al, 2014; Barnett et al, 2012; Uddin et al, 2015; Pollack et al, 2012. Density The proportion of all possible connections that actors may have (actual connections divided by all possible connections) Uddin et al, 2016; Uddin et al, 2012; Moen et al, 2016; Donker et al, 2012. Bipartite clustering coefficient The tendency of individuals (actors) to form clusters. Hollingsworth et al, 2016. Distance The average distances between all pairs of actors in a cluster Uddin et al, 2012. Strength of ties The number of patients shared by pairs of HCPs (actors) Pollack et al, 2014; Casalino et al, 2015; Stein et al, 2017; DuGoff et al, 2018. Persistence of ties The proportion of ties that persisted overtime. DuGoff et al, 2018. Physician-to-patient ratio The average number of HCPs per patient Uddin et al, 2015. Physician specialty characteristics The proportion of physicians with certain specialties (e.g. PCP) in the cluster Casalino et al, 2015; Pollack et al, 2012; Stein et al, 2017; DuGoff et al, 2018. ACTOR LEVEL Degree centrality The number of connections that an individual has. At cluster level the average centrality of actors was calculated. Sometimes the relative average centrality of PCPs compared to other doctors was calculated for each cluster. Uddin et al, 2012; Barnett et al, 2012. Betweenness centrality The extent to which an individual mediates indirect connections between others who are not connected to each other. At cluster level, the average betweenness of all actors was calculated. In one study, the ratio of the average betweenness of PCPs over specialists was calculated. One study calculated the variations in betweenness centrality among actors in a cluster. Casalino et al, 2015; Uddin et al, 2012. Association of clustering with patient outcomes As the above sections elaborate, the included studies are heterogeneous in terms of study design, study setting, population, network and clustering measures, and statistical approaches used to estimate associations of these measures with patient outcomes. The findings from these studies on associations of clustering with patient outcomes are summarized in Table 3 . Four papers assessed service use and intensity (such as emergency department (ED) visits, hospitalization days).[ 22 , 24 , 26 , 27 ] Four papers assessed readmissions.[ 24 , 26 , 28 , 29 ] Four papers examined costs and spending in health care.[ 22 , 27 – 29 ] Two studies examined provider choice of treatment,[ 21 , 23 ] one study focused on opioid and benzodiazepine prescriptions,[ 25 ] and one study focused on a hospital-acquired infection.[ 30 ] Below we synthesize these findings narratively, by grouping study findings based on the identified patterns. Table 3 Associations between Network Cluster Measures and Patient Outcomes. Study Summary of findings Interpretation of findings Health care use and intensity Barnett et al, 2012 For the average-sized urban hospital, an increase of one standard deviation (SD) in the median adjusted degree of doctors in the hospital network (implying connection to more other doctors for each patient) was associated with 17.4% more hospital days and 23.8% more physician visits in the last two years of life. One SD increase in the relative centrality of PCPs to other doctors (implying their prominence in the network) within hospital networks was associated with 8.6% fewer physician visits and 14.7% fewer medical specialist visits. Hospitals with doctors with more connections have more intensive care. Hospitals with primary care-centered networks have less intensive care. Uddin, 2016 Inclusion of cluster size and density as random effect variables in models predicting the length of stay for patients undergoing total hip replacement improved the performance of the model. There was a significant variation among clusters in terms of the association between physician networks and length of stay. This supports the importance of cluster-level characteristics on patient-level relationships. DuGoff et al, 2018 An additional physician connected to each PC physician in an HRR at baseline is associated with 1.2 more emergency department (ED) visits per 1,000 Medicare enrollees the following year. There was a non-significant positive association after 2 years. A 10% increase in tie persistence for 2 and 3 years (i.e., number of ties that persisted after 2 and 3 years in an HRR) is associated with 57 and 33 fewer ED visits per 1,000 Medicare beneficiaries, respectively. Hospital referral regions (HRRs) where PC physicians had more physician connections were more likely to have higher ED visit rates after one year. HRRs with more persistent physician relationships is associated with fewer ED visits. Hollingsworth et al, 2017 High level of clustering in physician teamwork (measured by the bipartite clustering coefficient) around CABG episodes was associated with a 24.6% lower 60-day ED visit rate, relative to a low level of physician teamwork. Lower postoperative ED visit in health systems in which physicians worked together in tightly knit groups during CABG episodes. Quality of care: readmissions Uddin et al, 2012 higher density, higher average distance, and lower betweenness centralization (variation in betweenness centrality among physicians) in physician collaboration networks (PCNs; i.e., hospitals) were associated with lower readmissions rates (time frame for readmission not mentioned) among total hip replacement patients Lower readmissions rates in PCNs with higher level of connectedness between physicians, and a more even distribution of mediatory connections (smaller betweenness inequality) Uddin et al, 2015 PCNs (hospitals) with more physician communities (identified through community detection) and fewer physicians per community had lower 28-day readmission rates Hospitals that are comprised of more small communities (corresponding to better communication and information sharing in smaller clusters) had lower readmission rates.. Hollingsworth et al, 2017 High level of clustering in physician networks (measured by the bipartite clustering coefficient) around CABG episodes was associated with 24.4% lower readmission rate, relative to a low level of physician teamwork. Lower readmission rates in health systems in which physicians worked together in tightly knit groups during CABG episodes. DuGoff et al, 2018 30-day readmission rates were not associated with more average physician ties (i.e., additional physician pairs per PC physician in an HRR) or with two-year persistence in these physician ties (i.e., number of ties that persisted after 2 years in an HRR). - Quality of care: other outcomes Pollack et al, 2014 A significant variation among physician clusters and the rates of complications following radical prostatectomy: 30-day surgical complications, late urinary complications, and long-term incontinence. As for subgroup characteristics, average urologist centrality was significantly associated with the outcomes in some cities, but not always in the same direction. Network subgroups (clusters) may explain a portion of the observed variation in surgical complications. - Casalino et al., 2015 Physician practice community (PPC) fixed effects were jointly statistically significant when modeling ambulatory care-sensitive hospital admission (ACSA) rates. PPCs with a 1 SD higher percentage of primary care physicians had 4.7% higher ACSA rates. PPCs with a 1 SD increase in physicians’ mean adjusted value degree had 5.0% higher ACSA rates. No evidence of an association between ACSA rates and other PPC characteristics: percentage of US-trained physicians, percentage of board-certified physicians, primary care centrality ratio (i.e., the mean betweenness centrality of primary care physicians divided by the mean centrality of specialists), or PPC size. PPCs were associated with ACSA rates, i.e., PPCs within networks matter. PPCs with larger proportion of primary care physicians had higher ACSA rates. PPCs in which physicians share patients with more other physicians had higher ACSA rates. - Hollingsworth et al, 2017 High level of clustering in physician teamwork (measured by the bipartite clustering coefficient) around CABG episodes was associated with a 28.4% lower mortality rate, relative to a low level of physician teamwork. Lower mortality rates in health systems in which physicians worked together in tightly knit groups during CABG episodes. Costs and spending Uddin et al, 2012 Higher hospitalization costs among total hip replacement patients in PCNs with lower density, higher distance, and higher betweenness centralization. Higher hospitalization costs among total hip replacement patients in PCNs with lower level of connectedness between physicians and higher inequity in distribution of mediatory connections Barnett et al, 2012 One SD increase in the adjusted degree of doctors in the hospital network (implying connection to more doctors for each patient) was associated with a 17.8% increase in total Medicare spending, along with over 20% increases in spending on imaging and tests. One SD increase in the relative centrality of PCPs to other doctors was associated with 6.0% lower total Medicare spending, 9.2% lower spending on imaging, and 12.9% lower spending on tests. Hospitals with doctors with more connections have higher costs. Hospitals with primary care-centered networks have lower costs. Uddin et al, 2015 Higher hospitalization costs in PCNs (hospitals) with a higher physician-to-patient ratio. The association may indicate redundant physician visits to patients in hospitals with more physicians per patient. Uddin, 2016 Inclusion of cluster size and density as random effect variables in models predicting the hospitalization cost for patients undergoing total hip replacement improved the performance of the model. There was a significant variation among clusters in terms of the association between physician networks and hospitalization cost. This supports the importance of cluster-level characteristics on patient-level relationships. Provider choice of treatment Pollack et al, 2012 The differences in prostatectomy between different urologist clusters were statistically significant. Clustering of urologists caring for patients with prostate cancer were associated with the likelihood of prostatectomy. Moen et al, 2016 Associations between the use of implantable cardioverter defibrillators (ICDs), which is supported by clinical guidelines, and hospital centrality measures (betweenness centrality, closeness centrality, eigenvector centrality of hospitals in an aggregated network) were not statistically significant at 0.05 level. Including ICD-capable physician node strength and closeness centrality as covariates without hospital-level centrality measures had the largest impact on reducing the HRR effect. - The variation in adherence to clinical guidelines is primarily driven by relationships between physicians. Opioid and benzodiazepine prescriptions Stein et al, 2017 Patients were more likely to receive opioid analgesic, benzodiazepine, and combined if they visited more provider communities (ORs 1.11, 1.16, 1.15, respectively). There was variation among prescription patterns across provider communities. In the provider communities in the highest quartile of the opioid analgesic, benzodiazepine, or combined prescribing, 2.5-5 times more patients received this prescription than in the lowest quartile. Receiving treatment from prescribers in multiple provider communities was positively associated with receiving both drugs. It may be that jointly managing care becomes more difficult as prescribers from more communities are involved in a patient’s care and do not have history and collaboration channels. Also it might be the sign of doctor shopping across communities. Provider community norms influence providers’ prescribing behavior Hospital-acquired infections Donker et al, 2012 MRSA (Methicillin-resistant Staphylococcus aureus) incidence rate of an individual hospital was positively associated with the number of patients it shared (through referrals) with other hospitals within its cluster. Hospitals from larger clusters had a higher MRSA incidence rate than hospitals from smaller clusters. More connected hospitals have higher MRSA rates. Larger clusters have a higher chance of experiencing a founding event (MRSA introduction and dispersal in one hospital). Clusters matter Pollack et al. (2014), Moen et al. (2016), and Stein et al. (2017) found a significant variation across cohesive natural clusters in terms of practice patterns (e.g., prostatectomy, adherence to guidelines, and analgesic prescription).[ 19 , 21 , 25 ] Pollack et al. (2014) and Casalino et al. (2015) found a significant variation across cohesive natural clusters in terms of surgical complications and preventable admission rates, respectively.[ 19 , 20 ] The disadvantages of connection to more peers Barnet et al. (2012), DuGoff et al. (2018), and Casalino et al. (2015) found that a larger number of connections among physicians in clusters was associated with more adverse outcomes (including more intensive care, higher costs, higher ED visits, hospital admissions, and preventable hospital admissions).[ 20 , 22 , 26 ] Uddin et al. (2015) also found that in hospitals where physicians formed more small cohesive communities (rather than a large, interconnected network) there were lower readmission rates.[ 29 ] They interpreted this as higher complexity of collaboration and data sharing when physicians interact with more peers to care for the same patients, which adversely affected their quality of care. The position of PCPs Barnet et al. (2012) and Casalino et al. (2015) assessed the relative position of PCPs (compared to other specialties) in a cluster.[ 20 , 22 ] Barnet et al. (2012) found that the average centrality of PCPs in a cluster was associated with positive outcomes (care intensity and cost),[ 22 ] while Casalino et al. (2015) did not find an association between relative centrality of PCPs and preventable admissions.[ 20 ] Casalino et al. (2015) and DuGoff et al. (2018) reported that the larger number of PCPs in a cluster and larger networks of PCPs were associated with adverse outcomes (such as more preventable hospital admissions and ED visits).[ 20 , 26 ] The findings of two latter studies align with the previous theme of findings regarding the disadvantages of larger networks (i.e., being connected to more peers). Advantages of tight-knit communities DuGoff et al. (2018) found that more persistent connections among physicians in a cluster were associated with a better outcome (fewer ED visits).[ 26 ] Uddin et al. (2012) and Hollingsworth et al. (2017) showed that more cohesive connections among physicians (less average betweenness, higher clustering coefficient) in a cluster were associated with better outcomes (e.g., readmission rates, ED visits, hospitalization costs).[ 24 , 28 ] Also, Uddin et al. (2015) found that the formation of cohesive communities among physicians in hospitals improved readmission rates among hip replacement patients.[ 29 ] DISCUSSION We reviewed studies that examined clustering of HCPs in patient-sharing networks. The studies varied considerably in terms of definitions and measures of patient-sharing relations, definitions and structural measures of network clusters, settings, study populations, and health outcomes. Most studies[ 19 – 25 , 27 – 29 ] defined patient-sharing networks based on all HCPs providing care to the same patients, while some studies specifically focused on patient-sharing across and within specialties or hospitals.[ 26 , 30 ] The majority of the studies did not limit the geographic or systemic boundaries of the network as long as the care was provided to a select group of patients. In contrast, others limited the network boundaries to specific health systems or geographic regions. The diversity of networks and cluster definitions may reflect different behavioral and contextual dynamics. The motivation and constraints of HCPs to share certain patients with their peers and to form informal or formal clusters of care are moderated by various contextual and systemic factors as well as the nature of their practice. For example, physicians within a healthcare system or a hospital have more monetary and organizational incentives to share patients. Also, patient-sharing between a PCP and a specialist mainly presents as referrals, while patient-sharing among different PCPs is probably due to patients’ request to have a second opinion, changing their family doctor, or changes in insurance plans. To identify provider clusters, a few studies used community-detection algorithms (mostly variations of Girvan-Newman or modularity), while a few others used organizational boundaries (e.g., hospital). Community detection algorithms identify clusters based on the behavioral patterns of HCPs in sharing patients with peers, while clustering based on organizational structure does not take the behavior into account. Even though a common assumption is that the providers within a healthcare system are more likely to share patients, we know little about the factors determining cluster formation and the extent of overlap between clusters based on practice patterns and formal organizational structures. Variation in study concepts and methods aside, it is challenging to summarize the findings of the studies to inform policies, for two reasons. First, the studies were not clear in terms of the theoretical underpinnings for both the choice of network measures and the potential mechanisms of the effects of clustering on health outcomes. The rationale for the use of specific network measures was rarely provided. Given the wide range of network measures, it is key to ground analytical choices in the social network theories, to meaningfully interpret the findings. Also, few studies drew on behavioral, organizational, or health services theoretical models for how specific network characteristics are associated with cluster formation and its impact on health outcomes. In other words, most studies either had no clear hypotheses, or fell short to generate or support their hypotheses based on theory. Evidently, the majority of this literature is still exploratory. Second, the use of administrative data, despite the strengths listed earlier, has its disadvantages. Policies and interventions informed by provider network analyses could focus on increasing provider collaboration, effective patient care coordination, or knowledge sharing, all of which require providers to communicate with each other and share resources. However, administrative data do not allow to measure actual collaboration and communication, which can only be rigorously measured through primary data collection. Despite the heterogeneity of studies and their limitations, there are a few general themes in study findings. First, more connections between physicians (e.g. more HCPs in one cluster or being connected to more peers to care for a patient) in patient-sharing clusters are associated with worse health outcomes and higher spending. Second, better connectedness of physicians, tight-knit communities, and persistence of physician ties are associated with better outcomes and less costly care. The quality of the relationships, rather than quantity, seems to lead to better patient outcomes. Theoretically, better connectivity between network actors would suggest better communication and teamwork[ 35 ] and would thus be associated with improved outcomes.[ 36 ] Prior research has shown a positive impact of network cohesion on outcomes for network actors, including better social capital[ 37 ] and support.[ 38 , 39 ] In small cohesive networks, the connections are stronger and more intimate.[ 40 ] However, strong connectivity may limit the diversity, access to new knowledge resources, and innovation.[ 41 ] Tie persistence could be the result of satisfaction with the connection based on prior experience of patient referral and communication, implying that the strategically chosen network ties may last longer and lead to better care coordination.[ 26 ] Larger and denser networks, on the other hand, are also probably more difficult to manage by physicians, who have busy schedules, especially when the infrastructure for referral and communication is not optimal. It is also more difficult to develop personal relationships with peers in a larger cluster. The studies did not report differences in the themes of findings among pre-defined clusters (e.g., hospital, HRR) vs. naturally occurring cohesive clusters (i.e., identified through community detection). This implies that the dynamics of relationships (such as tie persistence and manageable size) associate more with better outcomes than factors determining the local formation of clusters, such as organizational boundaries or preference to connect. We offer several recommendations for future research on patient-sharing and health outcomes. First, we recommend that researchers use social network theories to define networks and clusters and choose network measures in a rigorous way. Second, researchers should use theories from social and health sciences to argue for the expected effects of physician networks on outcomes. Next, while studies using administrative data have benefits, qualitative and mixed methods studies would help assess providers’ motivation to share patients with colleagues as well as examine relationships between network patterns and provider perspectives on patient-sharing. Regarding clustering within networks, qualitative studies could further illuminate the determinants and dynamics of cluster formation, from the viewpoint of network actors. Lastly, interventional studies are needed in order to produce high-quality systematic evidence on the causal mechanisms and development of network interventions. Understanding the mechanisms and testing interventions to improve cluster formation among HCPs have implications for system integration models in healthcare. Integrated systems, which generally have a high level of efficiency in patient-sharing, may produce better health outcomes through various mechanisms, including better care coordination, team-based care, streamlined communication and value-based payment. Interventions designed to promote clustering among healthcare providers (e.g., team-based care[ 42 – 45 ]) would help disentangle network-related effects on health outcomes. Our findings are subject to several limitations. First, our findings are based on a qualitative analysis of the main themes in included studies and due to study inconsistencies, we were not able to pool the results for a meta-analysis. Our findings should not be interpreted as evidence of effects. Second, even though we assessed the quality of the studies, we did not use our assessment to exclude any studies. Third, errors in the data collection and extraction are possible. We tried to minimize errors by having two independent reviewers for each study and using structured tools for data extraction. Fourth, all studies under review report some statistically significant findings and potential publication bias may introduce bias into our summary. CONCLUSIONS Our systematic scoping review revealed that studies focused on patient-sharing network clusters varied substantially in key characteristics and approaches. More rigorous theoretical grounding, the use of interventional, qualitative and mixed-methods studies, and development of interventions to improve the composition and dynamics of HCP clusters will advance our understanding of the social dynamics of healthcare and development of better integrated health systems. Declarations Acknowledgments The authors thank Dr. Chinmayee Katragadda and Ms. Zhi Pan, who contributed to the data extraction. Funding No funding to declare. Authors’ contributions All authors contributed to the conceptualization. AD, PD, QQ, and SY performed the screening, data extraction, and analysis, under supervision of RYN. All authors contributed to the drafting the manuscript and approved the final text. Ethics approval and consent to participate Not applicable Consent for publication Not applicable Availability of data and material The authors confirm that all data generated or analyzed during this study are included in this published article. Competing interests None to declare. References Kesternich E, Rank O: Beyond patient-sharing: Comparing physician- and patient-induced networks . Health Care Management Science 2022, 25 (3):498-514. DuGoff EH, Fernandes-Taylor S, Weissman GE, Huntley JH, Pollack CE: A scoping review of patient-sharing network studies using administrative data . Transl Behav Med 2018, 8 (4):598-625. Clarke JM, Warren LR, Arora S, Barahona M, Darzi AW: Guiding interoperable electronic health records through patient-sharing networks . npj Digital Medicine 2018, 1 (1):65. Cunningham FC, Ranmuthugala G, Plumb J, Georgiou A, Westbrook JI, Braithwaite J: Health professional networks as a vector for improving healthcare quality and safety: a systematic review . BMJ Qual Saf 2012, 21 (3):239-249. Barnett ML, Landon BE, O'Malley AJ, Keating NL, Christakis NA: Mapping physician networks with self-reported and administrative data . Health Serv Res 2011, 46 (5):1592-1609. De Brún A, McAuliffe E: Social Network Analysis as a Methodological Approach to Explore Health Systems: A Case Study Exploring Support among Senior Managers/Executives in a Hospital Network . International Journal of Environmental Research and Public Health 2018, 15 (3):511. Hu H, Yang Y, Zhang C, Huang C, Guan X, Shi L: Review of social networks of professionals in healthcare settings—where are we and what else is needed? Globalization and Health 2021, 17 (1):139. Agency for Healthcare Research and Quality: Care Coordination Measurement Framework . In: Care Coordination Atlas. edn. Rockville, MD; 2014. Emmons S, Kobourov S, Gallant M, Börner K: Analysis of Network Clustering Algorithms and Cluster Quality Metrics at Scale . PLoS One 2016, 11 (7):e0159161. Levine SS, Kurzban R: Explaining clustering in social networks: towards an evolutionary theory of cascading benefits . Managerial and Decision Economics 2006, 27 (2-3):173-187. Ehlert A, Kindschi M, Algesheimer R, Rauhut H: Human social preferences cluster and spread in the field . Proceedings of the National Academy of Sciences 2020, 117 (37):22787-22792. Zeltzer D: Gender Homophily in Referral Networks: Consequences for the Medicare Physician Earnings Gap . American Economic Journal: Applied Economics 2020, 12 (2):169-197. Kierkegaard P, Owen-Smith J: Determinants of physician networks: an ethnographic study examining the processes that inform patterns of collaboration and referral decision-making among physicians . BMJ Open 2021, 11 (1):e042334. Dossett LA, Hudson JN, Morris AM, Lee MC, Roetzheim RG, Fetters MD, Quinn GP: The primary care provider (PCP)-cancer specialist relationship: A systematic review and mixed-methods meta-synthesis . CA: A Cancer Journal for Clinicians 2017, 67 (2):156-169. Linde S: The formation of physician patient sharing networks in medicare: Exploring the effect of hospital affiliation . Health Economics 2019, 28 (12):1435-1448. Arksey H, O'Malley L: Scoping studies: towards a methodological framework . International Journal of Social Research Methodology 2005, 8 (1):19-32. Herzog R, Álvarez-Pasquin MJ, Díaz C, Del Barrio JL, Estrada JM, Gil Á: Are healthcare workers’ intentions to vaccinate related to their knowledge, beliefs and attitudes? a systematic review . BMC Public Health 2013, 13 (1):154. Clarke V, Braun V: Thematic Analysis . In: Encyclopedia of Critical Psychology. edn. Edited by Teo T. New York, NY: Springer New York; 2014: 1947-1952. Evan Pollack C, Wang H, Bekelman JE, Weissman G, Epstein AJ, Liao K, Dugoff EH, Armstrong K: Physician social networks and variation in rates of complications after radical prostatectomy . Value Health 2014, 17 (5):611-618. Casalino LP, Pesko MF, Ryan AM, Nyweide DJ, Iwashyna TJ, Sun X, Mendelsohn J, Moody J: Physician Networks and Ambulatory Care-sensitive Admissions . Med Care 2015, 53 (6):534-541. Moen EL, Austin AM, Bynum JP, Skinner JS, O'Malley AJ: An analysis of patient-sharing physician networks and implantable cardioverter defibrillator therapy . Health Serv Outcomes Res Methodol 2016, 16 (3):132-153. Barnett ML, Christakis NA, O'Malley J, Onnela JP, Keating NL, Landon BE: Physician patient-sharing networks and the cost and intensity of care in US hospitals . Med Care 2012, 50 (2):152-160. Pollack CE, Weissman G, Bekelman J, Liao K, Armstrong K: Physician social networks and variation in prostate cancer treatment in three cities . Health Serv Res 2012, 47 (1 Pt 2):380-403. Hollingsworth JM, Funk RJ, Garrison SA, Owen-Smith J, Kaufman SA, Pagani FD, Nallamothu BK: Association Between Physician Teamwork and Health System Outcomes After Coronary Artery Bypass Grafting . Circ Cardiovasc Qual Outcomes 2016, 9 (6):641-648. Stein BD, Mendelsohn J, Gordon AJ, Dick AW, Burns RM, Sorbero M, Shih RA, Liccardo Pacula R: Opioid analgesic and benzodiazepine prescribing among Medicaid-enrollees with opioid use disorders: The influence of provider communities . J Addict Dis 2017, 36 (1):14-22. DuGoff EH, Cho J, Si Y, Pollack CE: Geographic Variations in Physician Relationships Over Time: Implications for Care Coordination . Med Care Res Rev 2018, 75 (5):586-611. Uddin S: Exploring the impact of different multi-level measures of physician communities in patient-centric care networks on healthcare outcomes: A multi-level regression approach . Scientific Reports 2016, 6 (1):20222. Uddin S, Hossain L, Kelaher M: Effect of physician collaboration network on hospitalization cost and readmission rate . Eur J Public Health 2012, 22 (5):629-633. Uddin S, Kelaher M, Piraveenan M: Impact of Physician Community Structure on Healthcare Outcomes . Stud Health Technol Inform 2015, 214 :152-158. Donker T, Wallinga J, Slack R, Grundmann H: Hospital networks and the dispersal of hospital-acquired pathogens by patient transfer . PLoS One 2012, 7 (4):e35002. Christensen A, Garrido LE, Guerra-Peña K, Golino H: Comparing community detection algorithms in psychological data: A Monte Carlo simulation . In . : PsyArXiv; 2020. Clauset A: Finding local community structure in networks . Phys Rev E Stat Nonlin Soft Matter Phys 2005, 72 (2 Pt 2):026132. Amiri B, Hossain L, Crawford JW, Wigand RT: Community Detection in Complex Networks: Multi–objective Enhanced Firefly Algorithm . Knowledge-Based Systems 2013, 46 :1-11. Mahajan A, Kaur M: Various Approaches of Community Detection in Complex Networks: A Glance . International Journal of Information Technology and Computer Science 2016, 8 :35-41. Barak-Ventura R, Richmond S, Hasanyan J, Porfiri M: On the relationship between network connectivity and group performance in small teams of humans: experiments in virtual reality . Journal of Physics: Complexity 2020, 1 (2):025003. O'Reilly CA, & Roberts,: Task group structure, communication, and effectiveness in three organizations. Journal of Applied Psychology 1977, 62 (6):674-681. Lin N: Building a network theory of social capital. Social capital 2017:3-28. Kawachi I, Berkman LF: 290Social Capital, Social Cohesion, and Health . In: Social Epidemiology. edn. Edited by Berkman LF, Kawachi I, Glymour MM: Oxford University Press; 2014: 0. Martí J, Bolíbar M, Lozares C: Network cohesion and social support . Soc Networks 2017, 48 :192-201. Tortoriello M, Reagans R, McEvily B: Bridging the Knowledge Gap: The Influence of Strong Ties, Network Cohesion, and Network Range on the Transfer of Knowledge Between Organizational Units . Organization Science 2012, 23 (4):1024-1039. Ruef M: Strong ties, weak ties and islands: structural and cultural predictors of organizational innovation . Industrial and Corporate Change 2002, 11 (3):427-449. Goodwin N, Dixon A, Anderson GM, Wodchis WP: Providing integrated care for older people with complex needs Lessons from seven international case studies . In : 2014 ; 2014. Rumball-Smith J, Wodchis WP, Koné A, Kenealy T, Barnsley J, Ashton T: Under the same roof: co-location of practitioners within primary care is associated with specialized chronic care management . BMC Family Practice 2014, 15 (1):149. Haj-Ali W, Moineddin R, Hutchison B, Wodchis WP, Glazier RH: Physician group, physician and patient characteristics associated with joining interprofessional team-based primary care in Ontario, Canada . Health Policy 2020, 124 (7):743-750. Gutberg J, Evans JM, Khan S, Abdelhalim R, Wodchis WP, Grudniewicz A: Implementing Coordinated Care Networks: The Interplay of Individual and Distributed Leadership Practices . Med Care Res Rev 2022, 79 (5):650-662. Silich BA, Yang JJ: Measuring emergency physicians' work: factoring in clinical hours, patients seen, and relative value units into 1 metric . West J Emerg Med 2012, 13 (2):176-180. Additional Declarations No competing interests reported. Supplementary Files SupplementaryFile.docx Cite Share Download PDF Status: Published Journal Publication published 25 Feb, 2026 Read the published version in BMC Health Services Research → Version 1 posted Editorial decision: Revision requested 21 May, 2025 Reviews received at journal 12 May, 2025 Reviewers agreed at journal 02 May, 2025 Reviews received at journal 30 Jul, 2024 Reviewers agreed at journal 28 Jul, 2024 Reviewers agreed at journal 25 Jul, 2024 Reviewers agreed at journal 23 Jul, 2024 Reviewers invited by journal 09 Jul, 2024 Editor invited by journal 21 May, 2024 Editor assigned by journal 21 May, 2024 Submission checks completed at journal 20 May, 2024 First submitted to journal 17 May, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4437662","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":308843247,"identity":"03d80c8f-0987-46d5-a960-085b23a0b950","order_by":0,"name":"Alina Denham","email":"","orcid":"","institution":"Stony Brook University","correspondingAuthor":false,"prefix":"","firstName":"Alina","middleName":"","lastName":"Denham","suffix":""},{"id":308843248,"identity":"45f57a76-d4f3-4bbb-a0f4-89637e28423d","order_by":1,"name":"Porooshat Dadgostar","email":"","orcid":"","institution":"University of Rochester Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Porooshat","middleName":"","lastName":"Dadgostar","suffix":""},{"id":308843249,"identity":"a4f8063b-884d-4c7b-8b38-31746205d861","order_by":2,"name":"Qiuyuan Qin","email":"","orcid":"","institution":"University of Rochester Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Qiuyuan","middleName":"","lastName":"Qin","suffix":""},{"id":308843250,"identity":"77805893-38a2-4f82-ba53-979ec7efb240","order_by":3,"name":"Sule Yilmaz","email":"","orcid":"","institution":"University of Rochester Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Sule","middleName":"","lastName":"Yilmaz","suffix":""},{"id":308843251,"identity":"4d9a3092-0765-465b-85cd-b6a61a97fb88","order_by":4,"name":"Reza Yousefi Nooraie","email":"data:image/png;base64,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","orcid":"","institution":"University of Rochester Medical Center","correspondingAuthor":true,"prefix":"","firstName":"Reza","middleName":"Yousefi","lastName":"Nooraie","suffix":""}],"badges":[],"createdAt":"2024-05-17 15:34:23","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4437662/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4437662/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12913-025-13893-1","type":"published","date":"2026-02-25T15:58:43+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":57723139,"identity":"aa98d31d-6173-4b52-ba2f-f0c49d6579f2","added_by":"auto","created_at":"2024-06-04 19:13:55","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":53953,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePRISMA flowchart of study selection process.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-4437662/v1/0b305e443944ca521e30cba4.png"},{"id":103765620,"identity":"b34430f8-f665-4a4b-8403-1110f52cb0ce","added_by":"auto","created_at":"2026-03-02 16:05:58","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3033228,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4437662/v1/01d304fd-d314-4b04-b1e4-34781ea7b44c.pdf"},{"id":57723138,"identity":"f1baa8c5-3789-40af-9821-b176751f0c5c","added_by":"auto","created_at":"2024-06-04 19:13:54","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":15207,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryFile.docx","url":"https://assets-eu.researchsquare.com/files/rs-4437662/v1/a8971808b6313ff0777ea944.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"When Clinicians Group Together: A Systematic Scoping Review of Clustering in Patient-Sharing Networks","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003ePatient-sharing network analysis is a novel approach to reveal the patterns and dynamics of relationships among healthcare professionals (HCPs) using existing administrative data (e.g., insurance claims) or electronic medical records.[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] Patient-sharing relations are usually assumed when multiple HCPs provide care to a patient, hence are an indirect indicator of provider communication, referral, collaboration, and care coordination. Increasing healthcare integration and the use of technology facilitate the process of referral and coordination.[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] Naturally, integrated care models involve a high level of patient-sharing; and patients, particularly those with complex needs, often receive care from multiple HCPs. Improving the processes of patient-sharing between HCPs is a key component of safe and high-quality care.[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eIn the last decade, there has been growing interest in analyzing the structure and dynamics of patient-sharing networks using social network analysis.[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e] Social network analysis allows studying the patterns of relationship among HCPs (actors of the network), using visual mapping of relationship networks as well as various structural measures of network composition, existence and characteristics of clusters, relationship strength, actors\u0026rsquo; roles and positions, etc.[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] Social network analysis of patient-sharing networks can be used to describe behavioral dynamics in a healthcare system, to analyze care coordination, and team-based care. It can also be used to examine the process and impact of social influence among network actors (e.g., adopting new clinical guideline by peer influence).[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e] While some of these questions require collecting rich data from the network actors themselves, most network analysis studies of HCP networks rely on administrative data such as insurance claims and hospital discharge records, given their large scale, ease of analysis, and natural connection to patient outcomes.\u003c/p\u003e \u003cp\u003eBetter communication and coordination among HCPs should positively affect patient care and ultimately patient outcomes.[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e] Therefore, many patient-sharing network studies examine how network characteristics affect patient outcomes. However, a previous scoping review of patient-sharing network studies did not find consistent patterns in the relationships between network measures and patient outcomes, noting a substantial range of data sources, network measures and statistical approaches to estimation.[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eOne important, yet less studied, aspect of patient-sharing network analysis is studying how HCP groups, clusters, or communities are formed within networks, what these clusters look like, and whether clustering among HCPs affects patient outcomes. Network clusters are groups of actors who are better connected to each other, with looser connections to others outside the cluster.[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] Studying network clusters provides valuable insights to the meso-level patterns and dynamics of connectivity, beyond the dyadic relations between individuals. The notion of clustering in human networks has been studied for decades. Humans join clusters to maximize the benefit and safety of a group,[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e] and then adhere to cluster norms and cultures,[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] which themselves reshape over time. One consistent theme for the formation of clusters among professionals is homophily (or the tendency to connect to others with similar characteristics). Some studies have shown that HCPs are more likely to refer patients to colleagues that they personally know (through common demographic backgrounds or education,[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] or common activities such as journal clubs, etc.[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]), are geographically closer (e.g., in the same building or area),[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] have facilitated communication means (e.g., through face to face communication or electronic health records),[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e] share insurance plans, or are affiliated with the same health systems.[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eStudying clusters of HCP networks is important to understand how healthcare is delivered and how healthcare systems can improve patient outcomes via changes in structures and processes. Such understanding can be developed by examining how these clusters are formed (Why do some HCPs prefer to group together?) and how clusters affect HCP behavior (Do HCPs adopt the normative practice of the cluster?) and patient outcomes (Do patients benefit differently if their HCPs belong to the same cluster or particular clusters?). Our systematic scoping review aims to analyze the existing evidence on HCP clustering and its relationship with patient outcomes in patient-sharing network analysis studies.\u003c/p\u003e"},{"header":"METHODS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eIdentifying relevant literature\u003c/h2\u003e \u003cp\u003eWe followed the five-step framework proposed by Arksey and O\u0026rsquo;Malley for scoping reviews.[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] Previously, DuGoff et al (2018)[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] conducted a systematic review of patient-sharing networks using a structured search strategy. We expanded and updated their search strategy and systematically searched PubMed and EMBASE databases in January 2021. Our detailed search strategy is provided in the Appendix.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eSelecting studies\u003c/h2\u003e \u003cp\u003eWe included peer-reviewed articles in English published up until January 2021 that examined the clustering of HCPs in patient-sharing networks and reported at least one patient outcome, either at individual or systemic levels. HCP network clusters could be defined by regional (e.g., a catchment area or a city) or institutional (e.g., a hospital) boundaries, or determined by cohesive patterns (e.g., community detection algorithms) in patient-sharing networks. The latter means that HCPs tended to share patients within a cluster more than sharing outside the cluster. To be eligible for inclusion, a published study had to employ network analysis, examine clustering of providers, examine at least one health outcome, and be published in English. We excluded ongoing studies, protocols, and conference abstracts.\u003c/p\u003e \u003cp\u003eAbstracts were screened for eligibility by pairs of reviewers. Sixty-six selected articles underwent a full-text screening: each article was reviewed by two blind independent reviewers (articles were divided between two pairs of reviewers) and agreement on each article was achieved. Discrepancies were discussed and resolved in meetings. During full-text screening, the study of HCP clusters and a focus on patient outcomes were ascertained and articles that did not meet these criteria were excluded (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eExtracting and charting\u003c/h2\u003e \u003cp\u003eTo assess the methodological quality of included studies, we applied the New Castle-Ottawa quality assessment scale for cross-sectional[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] cohort and case-control studies. All included studies used patient records to determine HCP clusters as well as outcome measures, making a few items of the scale inapplicable to our study context. We simplified the scale and focused on the study design and the existence and nature of comparison groups and follow-ups.\u003c/p\u003e \u003cp\u003eThe included studies were divided between pairs of reviewers. Within pairs, reviewers independently extracted the following data from each article: population, study design, source of data, patient-sharing definition, patient-sharing measure(s), patient-sharing timeframe, clustering definitions, patient outcome(s), and reported associations between patient-sharing clustering measures and patient outcome(s). Each pair then reached agreement; discrepancies were discussed and resolved in meetings.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eData synthesis and analysis\u003c/h2\u003e \u003cp\u003eWe assessed for clinical heterogeneity, presented as diversity in terms of patient populations, HCPs, definition of patient sharing, measures of clustering, and outcome assessment. Given these methodological and clinical heterogeneities, we summarized the study findings narratively. We descriptively summarized the characteristics of studies, methods of clustering, and outcome measures. We qualitatively synthesized the association between clustering and health outcomes. We summarized the main findings of each study, as presented by the authors, provided a narrative summary of findings, and thematically categorized the findings across studies, following a thematic analysis approach.[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] The team met regularly to discuss each study and the process of the formation of themes.\u003c/p\u003e \u003c/div\u003e"},{"header":"RESULTS","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eReview process and retrieval\u003c/h2\u003e \u003cp\u003eWe identified 3,034 citations from the Bibliographic search in PubMed and EMBASE (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) and 32 citations from DuGoff et al (2018).[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] We removed 21 duplicate records and excluded 2,979 ineligible abstracts. We reviewed the full texts of 66 papers and excluded 54 for reasons listed in the PRISMA diagram (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). We included 12 empirical studies that assessed HCP clustering and patient outcomes. The characteristics of included studies are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCharacterization of the Studies Included in the Review\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQuality Assessment\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eData Source\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSample Size\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSetting\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eDefinition of Clusters\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eTies and Actors\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBarnett et al, 2012\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCross-sectional study\u003c/p\u003e \u003cp\u003eComparison group: no\u003c/p\u003e \u003cp\u003eFollow up: no\u003c/p\u003e \u003cp\u003eThe study adjusts for hospital characteristics (number of hospital beds, number of physicians assigned to a hospital, proportion of PCPs among assigned physicians, mean shared patient volume per physician at a hospital, urban location, teaching status, ownership, nurse full time equivalents per 1000 inpatient days, percentage of admissions from Medicare and Medicaid patients).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMedicare claims, Part A and Part B, 2006\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.6\u0026nbsp;million Medicare patients, 61,146 physicians, 528 hospitals, 51 HRRs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHospitals in 50 randomly sampled HRRs and Boston, the United States\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHospitals.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePhysicians who each had a significant encounter with one or more common patient. A significant encounter was defined as face-to-face visits or meaningful procedures with a value of at least 2 RVUs.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCasalino et al, 2015\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCross-sectional data over one year\u003c/p\u003e \u003cp\u003eComparison group: no\u003c/p\u003e \u003cp\u003eFollow-up: no\u003c/p\u003e \u003cp\u003eThe study controls for patient characteristics (age, sex, dual eligibility status, race/ethnicity, number of chronic conditions) and physician characteristics (U.S. trained, sex, age, board-certified).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMedicare claims Part B outpatient files, 2008\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e782,595 patients, 54,202 physicians, 386 physician practice communities\u003c/p\u003e \u003cp\u003e.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHospitals; 5 states in the United States (OH, PA, TN, WA, WI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eClusters (termed physician practice communities, or PPCs) were defined as \"smaller networks of physicians who share patients with each other much more often than they share patients with anyone else\". PPCs were identified using community detection algorithms.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePhysicians who provided outpatient care.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDonker et al, 2012\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCross-sectional study over 8 years\u003c/p\u003e \u003cp\u003eComparison group: England vs the Netherlands\u003c/p\u003e \u003cp\u003eFollow-up: no\u003c/p\u003e \u003cp\u003eThe study controls for cluster specific mean MRSA rates.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThe NHS Hospital Episode Statistics (HES) data (England) and the Dutch National Medical Registry data (Netherlands), April 2006 - March 2007.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eEngland: 7,420,219 patients admitted to 146 hospitals, for a total of 12,929,171 admissions\u003c/p\u003e \u003cp\u003eNetherlands: 1,676,704 patients admitted to 98 hospitals, for a total of 2,611,452 admissions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAcute care hospitals in England, hospital regional clusters in England, England vs the Netherlands\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHospitals were grouped into hospital regional clusters, based on the weights of the connections between all hospitals, using a community detection algorithm.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eHospitals are the actors and patient sharing happened as a result of patient referrals from one hospital to another.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDuGoff et al, 2018\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLongitudinal study\u003c/p\u003e \u003cp\u003eComparison group: no\u003c/p\u003e \u003cp\u003eFollow-up: 1 and 2 years\u003c/p\u003e \u003cp\u003eThe study controls for the hospital characteristics (median number of physician pairs per primary care physician (PCP), median number of co-occurrences, PCPs and specialist supply, hospital beds, total number and characteristics of Medicare fee-for-service beneficiaries, and region of the country).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMedicare claims, 2012\u0026ndash;2014.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e142,016 PCPs connected to at least one other physician in 2012, representing 12,190,803 physician pairs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHRRs; the United States\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eClusters are HRRs.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eReferrals between PCPs and specialists.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHollingsworth et al, 2016\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCross-sectional study over 4 years\u003c/p\u003e \u003cp\u003eComparison group: no\u003c/p\u003e \u003cp\u003eFollow-up: 60 days post-surgery\u003c/p\u003e \u003cp\u003eThe study adjusts for hospital and hospital service area characteristics as well as the number of physicians.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMedicare Provider Analysis and Review (MedPAR), 2008\u0026ndash;2011\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e251,630 Medicare beneficiaries who underwent CABG. 466,243 physicians, 1,186 health systems\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHospitals; the United States\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eClusters are health systems.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePhysicians directly involved in the care of patients with coronary artery bypass graft 30 days prior and 60 days after admission.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMoen et al, 2016\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCross-sectional study\u003c/p\u003e \u003cp\u003eComparison group: Two HRRs with difference in rates of receiving implantable cardioverter defibrillator (ICD) therapy\u003c/p\u003e \u003cp\u003eFollow-up: no\u003c/p\u003e \u003cp\u003eThe study controls for patient characteristics (age, sex, race).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMedicare Part B claim data, 2008\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGary, IN: 10,350 cardiovascular patients,481 physicians\u003c/p\u003e \u003cp\u003eSouth Bend, IN: 9,653 cardiovascular patients, 639 physicians.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTwo HRRs; Indiana, United States\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePhysicians and hospitals within two adjacent HRRs with disparate adherence to clinical guidelines regarding patient selection for ICD therapy.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePhysicians who provided care to patients with cardiovascular disease.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePollack et al, 2012\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eObservational and retrospective\u003c/p\u003e \u003cp\u003eComparison group: no\u003c/p\u003e \u003cp\u003eFollow-up: no\u003c/p\u003e \u003cp\u003eThe study adjusts for patient clinical (Gleason score, tumor stage, PSA results, comorbidity) and sociodemographic characteristics (age, race, community-level income, marital status).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSEER-Medicare data from three cities, 2004\u0026ndash;2005\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4,520 men\u003c/p\u003e \u003cp\u003e2420 doctors\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eThree cities; the United States\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eCommunity detection algorithm (Girvan-Newman) to identify group of physicians with more frequent patient sharing in each city. Separate analysis was done in each city.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePhysicians who provided care to prostate cancer patients, classified based on their role in the diagnosis, providing care, or administered the cancer treatments, within a 12-month window after the diagnosis.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePollack et al, 2014\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRetrospective, observational cohort study\u003c/p\u003e \u003cp\u003eComparison group: no\u003c/p\u003e \u003cp\u003eFollow-up: one year\u003c/p\u003e \u003cp\u003eThe study controls for patient-level and community-level characteristics (comorbidities, race, ethnicity, census tract or zip level median income, urologist surgical volume).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSEER Medicare data, 2004\u0026ndash;2005\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFor network construction: 13,465 men with prostate cancer without metastatic disease. For analyses: 2,677 men in 5 cities. The sample in each city ranged in size from 270 to 1,224 men.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePhysician networks in five cities; the United States\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNetwork subgroups were defined by Girvan-Newman algorithm (also called \"community structure\"); each doctor was assigned to a subgroup. Patients, however, may have doctors that were assigned to multiple subgroups. For the purposes of the analyses, patients were assigned to the network subgroup of the urologist who performed their prostatectomy.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eUrologists who provided care to the same patients with localized prostate cancer who underwent radical prostatectomy.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStein et al, 2017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCross-sectional study\u003c/p\u003e \u003cp\u003eComparison group: no\u003c/p\u003e \u003cp\u003eFollow-up: no\u003c/p\u003e \u003cp\u003eThe study controls for county characteristics including the total number of providers in the provider communities seen by the patient, the total number of patients in the provider communities seen by the provider, and state-fixed effects.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMedicaid claims data, 2008\u0026ndash;2009\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e29,611 Medicaid enrollees\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e12 states; the United States\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eProvider communities based on the frequency with which different providers were treating the same patients, all of whom had been diagnosed with opioid use disorders. Communities were defined using modularity maximization community detection.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eProviders who cared for the same patients with opioid use disorder during 2009.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUddin et al, 2012\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCross-sectional study over four years and two months.\u003c/p\u003e \u003cp\u003eComparison group: no\u003c/p\u003e \u003cp\u003eFollow-up: no\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOne health insurance company\u0026rsquo;s claims, Jan 2005 \u0026ndash; Feb 2009\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2229 patients,* 85 hospitals\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHospitals; Australia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePhysician collaboration networks were hospitals.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePhysicians who provided care to hip replacement patients during their hospitalization period.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUddin et al, 2015\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCross-sectional data over 4 years and 2 months\u003c/p\u003e \u003cp\u003eComparison group: no\u003c/p\u003e \u003cp\u003eFollow-up: no\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOne health insurance company\u0026rsquo;s claims, January 2005 - February 2009\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2352 patients, 2229 physicians, 85 hospitals.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHospitals; Australia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eCommunity detection algorithm to determine communities of doctors within each physician collaboration network (i.e. hospital)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePhysicians who provided care to hip replacement patients during their hospitalization.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUddin 2016\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCross-sectional data \u0026ldquo;over 5 years\u0026rdquo;**\u003c/p\u003e \u003cp\u003eComparison group: no\u003c/p\u003e \u003cp\u003eFollow-up: no\u003c/p\u003e \u003cp\u003eThe study controls for patient-level characteristics (age, gender, comorbidity index (Charlson-Deyo index)).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOne health insurance company\u0026rsquo;s claims \u0026ldquo;over 5 years\u0026rdquo;**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2352 patients, 2229 physicians, 85 hospitals\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHospitals; Australia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eCommunity detection firefly algorithm[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e] optimization approach: maximization of internal links and minimization of external links\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePhysicians who provided care to hip replacement patients during their hospitalization\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"7\"\u003eNotes: SEER\u0026thinsp;=\u0026thinsp;The Surveillance, Epidemiology, and End Results, a program of the National Cancer Institute. MRSA\u0026thinsp;=\u0026thinsp;Methicillin-resistant Staphylococcus aureus, a hospital-acquired infection. HRR\u0026thinsp;=\u0026thinsp;Hospital referral region. RVU\u0026thinsp;=\u0026thinsp;relative value unit, a measure of work or effort performed on a patient that is correlated to the revenue potential.[\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e] NHS\u0026thinsp;=\u0026thinsp;The National Health Services. CABG\u0026thinsp;=\u0026thinsp;coronary artery bypass graft. PSA\u0026thinsp;=\u0026thinsp;prostate specific antigen.\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"7\"\u003e* This article states that data from 2229 patients were used, which may be a typo, because two other studies that use the same data (Uddin et al, 2015; Uddin 2016) list 2229 physicians and 2352 patients.\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"7\"\u003e** This article states \u0026ldquo;over a period of five years\u0026rdquo; but does not specify which years; it appears that the data source is exactly the same as in Uddin et al, 2012 and Uddin et al, 2015, which is Jan 2005 \u0026ndash; Feb 2009, which is four years and two months.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eStudy setting and data sources\u003c/h2\u003e \u003cp\u003eEight studies were conducted in the United States,[\u003cspan additionalcitationids=\"CR20 CR21 CR22 CR23 CR24 CR25\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] three in Australia,[\u003cspan additionalcitationids=\"CR28\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] and one in the UK.[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] The settings of seven studies were hospitals and emergency departments,[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan additionalcitationids=\"CR28 CR29\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] 3 studies were hospital referral region,[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] and two studies were geographical regions (e.g., cities).[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eFive studies used Medicare data[\u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] and two used Medicare-SEER linked data.[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] One used Medicaid data,[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] three used Australian insurance organization data,[\u003cspan additionalcitationids=\"CR28\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] and one used the National Health Service (NHS) Hospital Statistics data.[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eStudy design\u003c/h2\u003e \u003cp\u003eNine studies were cross-sectional.[\u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan additionalcitationids=\"CR28 CR29\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] Of those, three were repeated cross-sectional studies.[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] Three studies were longitudinal.[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003ePatient-sharing networks, actors, and ties\u003c/h2\u003e \u003cp\u003eNetwork actors were HCPs[\u003cspan additionalcitationids=\"CR20 CR21 CR22 CR23 CR24 CR25 CR26 CR27 CR28\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] and hospitals.[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] Patient-sharing definition varied across studies. Patient-sharing relations between pairs of HCPs or hospitals were defined as caring for the same patient within a specific time frame, based on care context, a care episode, or professional relationships. Four studies based their definition on care episodes, e.g., during a hospitalization or an ambulatory care episode;[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan additionalcitationids=\"CR28\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] five studies were based on clinical conditions like cardiovascular disease, prostate cancer, opioid use disorder, and life-threatening chronic disease;[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan additionalcitationids=\"CR24\" citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] one study was based on direct referrals from primary care physicians to specialists,[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] and one study was based on transfers between hospitals.[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] One study did not specify care setting or clinical conditions.[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] Three studies used a specific time window of providing care for the same patient as the definition of patient-sharing relations.[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eDefinitions of clusters\u003c/h2\u003e \u003cp\u003eSeven studies used community detection algorithms to identify HCP clusters.[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] Community detection algorithms divide network nodes based on their relationship patterns into sets to maximize within-group and minimize between-group ties.[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] It is used to identify denser connected clusters within a larger network.\u003c/p\u003e \u003cp\u003eThree studies[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] used the Girvan-Newman community detection algorithm, which relies on modularity maximization. It involves an iterative removal of high betweenness ties with the highest number of shortest paths between nodes passing through them (in other words, identifying the network weak points).[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] These ties probably bridge loosely connected clusters in the network. Girvan-Newman is one of the traditional community detection methods and is best for smaller networks.[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eTwo studies[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] use the Fast Greedy algorithm. This method is based on the notion of local modularity.[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e] It starts with considering individual nodes as their own clusters, then adding nodes to the clusters one at a time in a greedy approach (examining local optimization) to create new clusters, aiming to maximize network modularity. Merging nearby clusters continues until there is no improvement in modularity.[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e] One study[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] used an algorithm called the Blondel model or the Louvain algorithm.[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] Similar to the Fast Greedy algorithm, this hierarchical clustering method starts with separate nodes and adds nodes to clusters in a greedy way to optimize modularity. To accommodate for larger network size, this model iteratively aggregates new clusters into \u0026ldquo;super-nodes\u0026rdquo; in a new network, and then repeats this procedure with the new aggregated network until there is no improvement in modularity. One paper[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] used the algorithm introduced by Amiri et al.[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e] The Amiri algorithm is an evolutionary technique[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e] that learns from previous steps, aiming for maximizing within-cluster relations and minimizing between-cluster relations at the same time.\u003c/p\u003e \u003cp\u003eFour of these studies applied a multi-level approach, identifying collaboration networks based on hospitals[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] or cities,[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] and then applying a community detection algorithm to each sub-network separately.\u003c/p\u003e \u003cp\u003eFive studies did not apply any community detection algorithm and \u003cem\u003ea priori\u003c/em\u003e defined clusters as hospitals, medical practices, clinics, health systems, or hospital referral regions.[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eMeasures of cluster characteristics\u003c/h2\u003e \u003cdiv id=\"Sec14\" class=\"Section3\"\u003e \u003ch2\u003eCluster structure\u003c/h2\u003e \u003cp\u003eThe size of clusters (number of providers in each cluster) was the most commonly reported measure.[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] Uddin et al. (2015)[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] assessed the physician-to-patient ratio in physician communities. Studies used different measures to assess collaboration between HCPs within a cluster. Network density was used most to describe the connectivity of actors (e.g., physicians or hospitals). Hollingsworth et al. (2016)[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] used the bipartite clustering coefficient to measure in the tendency to form dense groups in clusters (health systems). Casalino et al. (2015) calculated the mean adjusted valued degree of physicians within the community.[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eCluster composition\u003c/h2\u003e \u003cp\u003ePollack et al. (2012) measured the percentage of patients and providers in large clusters (defined as clusters of more than 50 physicians).[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] Casalino et al. (2015)[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] and Pollack et al. (2012)[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] assessed the proportion of primary care physicians (PCPs) and urologists in each cluster, respectively. DuGoff et al. (2018)[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] used the median number of physician pairs per PCP in a cluster to evaluate the breadth of PCP network.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eActors\u0026rsquo; position\u003c/h2\u003e \u003cp\u003eThe position of actors in clusters was assessed in three studies.[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] Casalino et al. (2015)[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] measured the betweenness centrality of PCPs and cluster-level relative betweenness centrality by dividing the PCPs betweenness centrality (an indicator of their intermediary role) by the average betweenness centrality of other HCPs to evaluate the bridging ability of PCPs. Barnett et al. (2012)[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] measured degree centrality for PCPs and relative degree centrality using similar approach to evaluate the popularity of PCPs. Uddin et al. (2012)[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] assessed both betweenness centrality and degree centrality of actors in network clusters, to identify central actors.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eCluster dynamics\u003c/h2\u003e \u003cp\u003eThe persistence of connections over time was assessed in one study.[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] DuGoff et al. (2018)[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] evaluated persistence of ties between providers by assessing whether the ties were present across years, and calculated the proportion of ties that persisted in each cluster.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eBehavior in clusters\u003c/h2\u003e \u003cp\u003eStudies also used cluster-level measures in relation to physician behaviors. Stein et al. (2017)[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] assessed the percentage of individuals prescribed opioid analgesics in each cluster to understand how provider communities differed in prescription patterns.\u003c/p\u003e \u003cp\u003eMeasures of cluster characteristics are defined in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMeasures of Cluster Characteristics.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStructural measures\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExplanation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eStudies\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCLUSTER LEVEL\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCluster size\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe total number of individuals (actors) in the cluster\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUddin et al, 2016; Pollack et al, 2014; Barnett et al, 2012; Uddin et al, 2015; Pollack et al, 2012.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDensity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe proportion of all possible connections that actors may have (actual connections divided by all possible connections)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUddin et al, 2016; Uddin et al, 2012; Moen et al, 2016; Donker et al, 2012.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBipartite clustering coefficient\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe tendency of individuals (actors) to form clusters.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHollingsworth et al, 2016.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDistance\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe average distances between all pairs of actors in a cluster\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUddin et al, 2012.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStrength of ties\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe number of patients shared by pairs of HCPs (actors)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePollack et al, 2014; Casalino et al, 2015; Stein et al, 2017; DuGoff et al, 2018.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePersistence of ties\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe proportion of ties that persisted overtime.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDuGoff et al, 2018.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePhysician-to-patient ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe average number of HCPs per patient\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUddin et al, 2015.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePhysician specialty characteristics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe proportion of physicians with certain specialties (e.g. PCP) in the cluster\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCasalino et al, 2015; Pollack et al, 2012; Stein et al, 2017; DuGoff et al, 2018.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eACTOR LEVEL\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDegree centrality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe number of connections that an individual has. At cluster level the average centrality of actors was calculated. Sometimes the relative average centrality of PCPs compared to other doctors was calculated for each cluster.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUddin et al, 2012; Barnett et al, 2012.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBetweenness centrality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe extent to which an individual mediates indirect connections between others who are not connected to each other. At cluster level, the average betweenness of all actors was calculated. In one study, the ratio of the average betweenness of PCPs over specialists was calculated. One study calculated the variations in betweenness centrality among actors in a cluster.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCasalino et al, 2015; Uddin et al, 2012.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eAssociation of clustering with patient outcomes\u003c/h2\u003e \u003cp\u003eAs the above sections elaborate, the included studies are heterogeneous in terms of study design, study setting, population, network and clustering measures, and statistical approaches used to estimate associations of these measures with patient outcomes. The findings from these studies on associations of clustering with patient outcomes are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. Four papers assessed service use and intensity (such as emergency department (ED) visits, hospitalization days).[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] Four papers assessed readmissions.[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] Four papers examined costs and spending in health care.[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan additionalcitationids=\"CR28\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] Two studies examined provider choice of treatment,[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] one study focused on opioid and benzodiazepine prescriptions,[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] and one study focused on a hospital-acquired infection.[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] Below we synthesize these findings narratively, by grouping study findings based on the identified patterns.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAssociations between Network Cluster Measures and Patient Outcomes.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSummary of findings\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eInterpretation of findings\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eHealth care use and intensity\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBarnett et al, 2012\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFor the average-sized urban hospital, an increase of one standard deviation (SD) in the median adjusted degree of doctors in the hospital network (implying connection to more other doctors for each patient) was associated with 17.4% more hospital days and 23.8% more physician visits in the last two years of life.\u003c/p\u003e \u003cp\u003eOne SD increase in the relative centrality of PCPs to other doctors (implying their prominence in the network) within hospital networks was associated with 8.6% fewer physician visits and 14.7% fewer medical specialist visits.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHospitals with doctors with more connections have more intensive care.\u003c/p\u003e \u003cp\u003eHospitals with primary care-centered networks have less intensive care.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUddin, 2016\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInclusion of cluster size and density as random effect variables in models predicting the length of stay for patients undergoing total hip replacement improved the performance of the model.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThere was a significant variation among clusters in terms of the association between physician networks and length of stay. This supports the importance of cluster-level characteristics on patient-level relationships.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDuGoff et al, 2018\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAn additional physician connected to each PC physician in an HRR at baseline is associated with 1.2 more emergency department (ED) visits per 1,000 Medicare enrollees the following year. There was a non-significant positive association after 2 years.\u003c/p\u003e \u003cp\u003eA 10% increase in tie persistence for 2 and 3 years (i.e., number of ties that persisted after 2 and 3 years in an HRR) is associated with 57 and 33 fewer ED visits per 1,000 Medicare beneficiaries, respectively.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHospital referral regions (HRRs) where PC physicians had more physician connections were more likely to have higher ED visit rates after one year.\u003c/p\u003e \u003cp\u003eHRRs with more persistent physician relationships is associated with fewer ED visits.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHollingsworth et al, 2017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh level of clustering in physician teamwork (measured by the bipartite clustering coefficient) around CABG episodes was associated with a 24.6% lower 60-day ED visit rate, relative to a low level of physician teamwork.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLower postoperative ED visit in health systems in which physicians worked together in tightly knit groups during CABG episodes.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eQuality of care: readmissions\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUddin et al, 2012\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ehigher density, higher average distance, and lower betweenness centralization (variation in betweenness centrality among physicians) in physician collaboration networks (PCNs; i.e., hospitals) were associated with lower readmissions rates (time frame for readmission not mentioned) among total hip replacement patients\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLower readmissions rates in PCNs with higher level of connectedness between physicians, and a more even distribution of mediatory connections (smaller betweenness inequality)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUddin et al, 2015\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePCNs (hospitals) with more physician communities (identified through community detection) and fewer physicians per community had lower 28-day readmission rates\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHospitals that are comprised of more small communities (corresponding to better communication and information sharing in smaller clusters) had lower readmission rates..\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHollingsworth et al, 2017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh level of clustering in physician networks (measured by the bipartite clustering coefficient) around CABG episodes was associated with 24.4% lower readmission rate, relative to a low level of physician teamwork.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLower readmission rates in health systems in which physicians worked together in tightly knit groups during CABG episodes.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDuGoff et al, 2018\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30-day readmission rates were not associated with more average physician ties (i.e., additional physician pairs per PC physician in an HRR) or with two-year persistence in these physician ties (i.e., number of ties that persisted after 2 years in an HRR).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eQuality of care: other outcomes\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePollack et al, 2014\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eA significant variation among physician clusters and the rates of complications following radical prostatectomy: 30-day surgical complications, late urinary complications, and long-term incontinence.\u003c/p\u003e \u003cp\u003eAs for subgroup characteristics, average urologist centrality was significantly associated with the outcomes in some cities, but not always in the same direction.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNetwork subgroups (clusters) may explain a portion of the observed variation in surgical complications.\u003c/p\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCasalino et al., 2015\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePhysician practice community (PPC) fixed effects were jointly statistically significant when modeling ambulatory care-sensitive hospital admission (ACSA) rates.\u003c/p\u003e \u003cp\u003ePPCs with a 1 SD higher percentage of primary care physicians had 4.7% higher ACSA rates.\u003c/p\u003e \u003cp\u003ePPCs with a 1 SD increase in physicians\u0026rsquo; mean adjusted value degree had 5.0% higher ACSA rates.\u003c/p\u003e \u003cp\u003eNo evidence of an association between ACSA rates and other PPC characteristics: percentage of US-trained physicians, percentage of board-certified physicians, primary care centrality ratio (i.e., the mean betweenness centrality of primary care physicians divided by the mean centrality of specialists), or PPC size.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePPCs were associated with ACSA rates, i.e., PPCs within networks matter.\u003c/p\u003e \u003cp\u003ePPCs with larger proportion of primary care physicians had higher ACSA rates.\u003c/p\u003e \u003cp\u003ePPCs in which physicians share patients with more other physicians had higher ACSA rates.\u003c/p\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHollingsworth et al, 2017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh level of clustering in physician teamwork (measured by the bipartite clustering coefficient) around CABG episodes was associated with a 28.4% lower mortality rate, relative to a low level of physician teamwork.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLower mortality rates in health systems in which physicians worked together in tightly knit groups during CABG episodes.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eCosts and spending\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUddin et al, 2012\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigher hospitalization costs among total hip replacement patients in PCNs with lower density, higher distance, and higher betweenness centralization.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigher hospitalization costs among total hip replacement patients in PCNs with lower level of connectedness between physicians and higher inequity in distribution of mediatory connections\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBarnett et al, 2012\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOne SD increase in the adjusted degree of doctors in the hospital network (implying connection to more doctors for each patient) was associated with a 17.8% increase in total Medicare spending, along with over 20% increases in spending on imaging and tests.\u003c/p\u003e \u003cp\u003eOne SD increase in the relative centrality of PCPs to other doctors was associated with 6.0% lower total Medicare spending, 9.2% lower spending on imaging, and 12.9% lower spending on tests.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHospitals with doctors with more connections have higher costs.\u003c/p\u003e \u003cp\u003eHospitals with primary care-centered networks have lower costs.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUddin et al, 2015\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigher hospitalization costs in PCNs (hospitals) with a higher physician-to-patient ratio.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThe association may indicate redundant physician visits to patients in hospitals with more physicians per patient.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUddin, 2016\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInclusion of cluster size and density as random effect variables in models predicting the hospitalization cost for patients undergoing total hip replacement improved the performance of the model.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThere was a significant variation among clusters in terms of the association between physician networks and hospitalization cost. This supports the importance of cluster-level characteristics on patient-level relationships.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eProvider choice of treatment\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePollack et al, 2012\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe differences in prostatectomy between different urologist clusters were statistically significant.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eClustering of urologists caring for patients with prostate cancer were associated with the likelihood of prostatectomy.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMoen et al, 2016\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAssociations between the use of implantable cardioverter defibrillators (ICDs), which is supported by clinical guidelines, and hospital centrality measures (betweenness centrality, closeness centrality, eigenvector centrality of hospitals in an aggregated network) were not statistically significant at 0.05 level.\u003c/p\u003e \u003cp\u003eIncluding ICD-capable physician node strength and closeness centrality as covariates without hospital-level centrality measures had the largest impact on reducing the HRR effect.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003cp\u003eThe variation in adherence to clinical guidelines is primarily driven by relationships between physicians.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eOpioid and benzodiazepine prescriptions\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStein et al, 2017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePatients were more likely to receive opioid analgesic, benzodiazepine, and combined if they visited more provider communities (ORs 1.11, 1.16, 1.15, respectively).\u003c/p\u003e \u003cp\u003eThere was variation among prescription patterns across provider communities. In the provider communities in the highest quartile of the opioid analgesic, benzodiazepine, or combined prescribing, 2.5-5 times more patients received this prescription than in the lowest quartile.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eReceiving treatment from prescribers in multiple provider communities was positively associated with receiving both drugs. It may be that jointly managing care becomes more difficult as prescribers from more communities are involved in a patient\u0026rsquo;s care and do not have history and collaboration channels. Also it might be the sign of doctor shopping across communities.\u003c/p\u003e \u003cp\u003eProvider community norms influence providers\u0026rsquo; prescribing behavior\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eHospital-acquired infections\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDonker et al, 2012\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMRSA (Methicillin-resistant Staphylococcus aureus) incidence rate of an individual hospital was positively associated with the number of patients it shared (through referrals) with other hospitals within its cluster.\u003c/p\u003e \u003cp\u003eHospitals from larger clusters had a higher MRSA incidence rate than hospitals from smaller clusters.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMore connected hospitals have higher MRSA rates.\u003c/p\u003e \u003cp\u003eLarger clusters have a higher chance of experiencing a founding event (MRSA introduction and dispersal in one hospital).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eClusters matter\u003c/h2\u003e \u003cp\u003ePollack et al. (2014), Moen et al. (2016), and Stein et al. (2017) found a significant variation across cohesive natural clusters in terms of practice patterns (e.g., prostatectomy, adherence to guidelines, and analgesic prescription).[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] Pollack et al. (2014) and Casalino et al. (2015) found a significant variation across cohesive natural clusters in terms of surgical complications and preventable admission rates, respectively.[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003eThe disadvantages of connection to more peers\u003c/h2\u003e \u003cp\u003eBarnet et al. (2012), DuGoff et al. (2018), and Casalino et al. (2015) found that a larger number of connections among physicians in clusters was associated with more adverse outcomes (including more intensive care, higher costs, higher ED visits, hospital admissions, and preventable hospital admissions).[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] Uddin et al. (2015) also found that in hospitals where physicians formed more small cohesive communities (rather than a large, interconnected network) there were lower readmission rates.[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] They interpreted this as higher complexity of collaboration and data sharing when physicians interact with more peers to care for the same patients, which adversely affected their quality of care.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eThe position of PCPs\u003c/h2\u003e \u003cp\u003eBarnet et al. (2012) and Casalino et al. (2015) assessed the relative position of PCPs (compared to other specialties) in a cluster.[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] Barnet et al. (2012) found that the average centrality of PCPs in a cluster was associated with positive outcomes (care intensity and cost),[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] while Casalino et al. (2015) did not find an association between relative centrality of PCPs and preventable admissions.[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] Casalino et al. (2015) and DuGoff et al. (2018) reported that the larger number of PCPs in a cluster and larger networks of PCPs were associated with adverse outcomes (such as more preventable hospital admissions and ED visits).[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] The findings of two latter studies align with the previous theme of findings regarding the disadvantages of larger networks (i.e., being connected to more peers).\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003eAdvantages of tight-knit communities\u003c/h2\u003e \u003cp\u003eDuGoff et al. (2018) found that more persistent connections among physicians in a cluster were associated with a better outcome (fewer ED visits).[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] Uddin et al. (2012) and Hollingsworth et al. (2017) showed that more cohesive connections among physicians (less average betweenness, higher clustering coefficient) in a cluster were associated with better outcomes (e.g., readmission rates, ED visits, hospitalization costs).[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] Also, Uddin et al. (2015) found that the formation of cohesive communities among physicians in hospitals improved readmission rates among hip replacement patients.[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eWe reviewed studies that examined clustering of HCPs in patient-sharing networks. The studies varied considerably in terms of definitions and measures of patient-sharing relations, definitions and structural measures of network clusters, settings, study populations, and health outcomes.\u003c/p\u003e \u003cp\u003eMost studies[\u003cspan additionalcitationids=\"CR20 CR21 CR22 CR23 CR24\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan additionalcitationids=\"CR28\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] defined patient-sharing networks based on all HCPs providing care to the same patients, while some studies specifically focused on patient-sharing across and within specialties or hospitals.[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] The majority of the studies did not limit the geographic or systemic boundaries of the network as long as the care was provided to a select group of patients. In contrast, others limited the network boundaries to specific health systems or geographic regions. The diversity of networks and cluster definitions may reflect different behavioral and contextual dynamics. The motivation and constraints of HCPs to share certain patients with their peers and to form informal or formal clusters of care are moderated by various contextual and systemic factors as well as the nature of their practice. For example, physicians within a healthcare system or a hospital have more monetary and organizational incentives to share patients. Also, patient-sharing between a PCP and a specialist mainly presents as referrals, while patient-sharing among different PCPs is probably due to patients\u0026rsquo; request to have a second opinion, changing their family doctor, or changes in insurance plans.\u003c/p\u003e \u003cp\u003eTo identify provider clusters, a few studies used community-detection algorithms (mostly variations of Girvan-Newman or modularity), while a few others used organizational boundaries (e.g., hospital). Community detection algorithms identify clusters based on the behavioral patterns of HCPs in sharing patients with peers, while clustering based on organizational structure does not take the behavior into account. Even though a common assumption is that the providers within a healthcare system are more likely to share patients, we know little about the factors determining cluster formation and the extent of overlap between clusters based on practice patterns and formal organizational structures.\u003c/p\u003e \u003cp\u003eVariation in study concepts and methods aside, it is challenging to summarize the findings of the studies to inform policies, for two reasons. First, the studies were not clear in terms of the theoretical underpinnings for both the choice of network measures and the potential mechanisms of the effects of clustering on health outcomes. The rationale for the use of specific network measures was rarely provided. Given the wide range of network measures, it is key to ground analytical choices in the social network theories, to meaningfully interpret the findings. Also, few studies drew on behavioral, organizational, or health services theoretical models for how specific network characteristics are associated with cluster formation and its impact on health outcomes. In other words, most studies either had no clear hypotheses, or fell short to generate or support their hypotheses based on theory. Evidently, the majority of this literature is still exploratory.\u003c/p\u003e \u003cp\u003eSecond, the use of administrative data, despite the strengths listed earlier, has its disadvantages. Policies and interventions informed by provider network analyses could focus on increasing provider collaboration, effective patient care coordination, or knowledge sharing, all of which require providers to communicate with each other and share resources. However, administrative data do not allow to measure actual collaboration and communication, which can only be rigorously measured through primary data collection.\u003c/p\u003e \u003cp\u003eDespite the heterogeneity of studies and their limitations, there are a few general themes in study findings. First, more connections between physicians (e.g. more HCPs in one cluster or being connected to more peers to care for a patient) in patient-sharing clusters are associated with worse health outcomes and higher spending. Second, better connectedness of physicians, tight-knit communities, and persistence of physician ties are associated with better outcomes and less costly care.\u003c/p\u003e \u003cp\u003eThe quality of the relationships, rather than quantity, seems to lead to better patient outcomes. Theoretically, better connectivity between network actors would suggest better communication and teamwork[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e] and would thus be associated with improved outcomes.[\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e] Prior research has shown a positive impact of network cohesion on outcomes for network actors, including better social capital[\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e] and support.[\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e] In small cohesive networks, the connections are stronger and more intimate.[\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e] However, strong connectivity may limit the diversity, access to new knowledge resources, and innovation.[\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e] Tie persistence could be the result of satisfaction with the connection based on prior experience of patient referral and communication, implying that the strategically chosen network ties may last longer and lead to better care coordination.[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eLarger and denser networks, on the other hand, are also probably more difficult to manage by physicians, who have busy schedules, especially when the infrastructure for referral and communication is not optimal. It is also more difficult to develop personal relationships with peers in a larger cluster.\u003c/p\u003e \u003cp\u003eThe studies did not report differences in the themes of findings among pre-defined clusters (e.g., hospital, HRR) vs. naturally occurring cohesive clusters (i.e., identified through community detection). This implies that the dynamics of relationships (such as tie persistence and manageable size) associate more with better outcomes than factors determining the local formation of clusters, such as organizational boundaries or preference to connect.\u003c/p\u003e \u003cp\u003eWe offer several recommendations for future research on patient-sharing and health outcomes. First, we recommend that researchers use social network theories to define networks and clusters and choose network measures in a rigorous way. Second, researchers should use theories from social and health sciences to argue for the expected effects of physician networks on outcomes. Next, while studies using administrative data have benefits, qualitative and mixed methods studies would help assess providers\u0026rsquo; motivation to share patients with colleagues as well as examine relationships between network patterns and provider perspectives on patient-sharing. Regarding clustering within networks, qualitative studies could further illuminate the determinants and dynamics of cluster formation, from the viewpoint of network actors. Lastly, interventional studies are needed in order to produce high-quality systematic evidence on the causal mechanisms and development of network interventions. Understanding the mechanisms and testing interventions to improve cluster formation among HCPs have implications for system integration models in healthcare. Integrated systems, which generally have a high level of efficiency in patient-sharing, may produce better health outcomes through various mechanisms, including better care coordination, team-based care, streamlined communication and value-based payment. Interventions designed to promote clustering among healthcare providers (e.g., team-based care[\u003cspan additionalcitationids=\"CR43 CR44\" citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]) would help disentangle network-related effects on health outcomes.\u003c/p\u003e \u003cp\u003eOur findings are subject to several limitations. First, our findings are based on a qualitative analysis of the main themes in included studies and due to study inconsistencies, we were not able to pool the results for a meta-analysis. Our findings should not be interpreted as evidence of effects. Second, even though we assessed the quality of the studies, we did not use our assessment to exclude any studies. Third, errors in the data collection and extraction are possible. We tried to minimize errors by having two independent reviewers for each study and using structured tools for data extraction. Fourth, all studies under review report some statistically significant findings and potential publication bias may introduce bias into our summary.\u003c/p\u003e"},{"header":"CONCLUSIONS","content":"\u003cp\u003eOur systematic scoping review revealed that studies focused on patient-sharing network clusters varied substantially in key characteristics and approaches. More rigorous theoretical grounding, the use of interventional, qualitative and mixed-methods studies, and development of interventions to improve the composition and dynamics of HCP clusters will advance our understanding of the social dynamics of healthcare and development of better integrated health systems.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eAcknowledgments\u003c/p\u003e\n\u003cp\u003eThe authors thank Dr. Chinmayee Katragadda and Ms. Zhi Pan, who contributed to the data extraction.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo funding to declare.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAuthors\u0026rsquo; contributions\u003c/p\u003e\n\u003cp\u003eAll authors contributed to the conceptualization. AD, PD, QQ, and SY performed the screening, data extraction, and analysis, under supervision of RYN. All authors contributed to the drafting the manuscript and approved the final text.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and material\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors confirm that all data generated or analyzed during this study are included in this published article.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCompeting\u0026nbsp;interests\u003c/p\u003e\n\u003cp\u003eNone to declare.\u003cbr\u003e \u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eKesternich E, Rank O: \u003cstrong\u003eBeyond patient-sharing: Comparing physician- and patient-induced networks\u003c/strong\u003e. \u003cem\u003eHealth Care Management Science \u003c/em\u003e2022, \u003cstrong\u003e25\u003c/strong\u003e(3):498-514.\u003c/li\u003e\n\u003cli\u003eDuGoff EH, Fernandes-Taylor S, Weissman GE, Huntley JH, Pollack CE: \u003cstrong\u003eA scoping review of patient-sharing network studies using administrative data\u003c/strong\u003e. \u003cem\u003eTransl Behav Med \u003c/em\u003e2018, \u003cstrong\u003e8\u003c/strong\u003e(4):598-625.\u003c/li\u003e\n\u003cli\u003eClarke JM, Warren LR, Arora S, Barahona M, Darzi AW: \u003cstrong\u003eGuiding interoperable electronic health records through patient-sharing networks\u003c/strong\u003e. \u003cem\u003enpj Digital Medicine \u003c/em\u003e2018, \u003cstrong\u003e1\u003c/strong\u003e(1):65.\u003c/li\u003e\n\u003cli\u003eCunningham FC, Ranmuthugala G, Plumb J, Georgiou A, Westbrook JI, Braithwaite J: \u003cstrong\u003eHealth professional networks as a vector for improving healthcare quality and safety: a systematic review\u003c/strong\u003e. \u003cem\u003eBMJ Qual Saf \u003c/em\u003e2012, \u003cstrong\u003e21\u003c/strong\u003e(3):239-249.\u003c/li\u003e\n\u003cli\u003eBarnett ML, Landon BE, O\u0026apos;Malley AJ, Keating NL, Christakis NA: \u003cstrong\u003eMapping physician networks with self-reported and administrative data\u003c/strong\u003e. \u003cem\u003eHealth Serv Res \u003c/em\u003e2011, \u003cstrong\u003e46\u003c/strong\u003e(5):1592-1609.\u003c/li\u003e\n\u003cli\u003eDe Br\u0026uacute;n A, McAuliffe E: \u003cstrong\u003eSocial Network Analysis as a Methodological Approach to Explore Health Systems: A Case Study Exploring Support among Senior Managers/Executives in a Hospital Network\u003c/strong\u003e. \u003cem\u003eInternational Journal of Environmental Research and Public Health \u003c/em\u003e2018, \u003cstrong\u003e15\u003c/strong\u003e(3):511.\u003c/li\u003e\n\u003cli\u003eHu H, Yang Y, Zhang C, Huang C, Guan X, Shi L: \u003cstrong\u003eReview of social networks of professionals in healthcare settings\u0026mdash;where are we and what else is needed?\u003c/strong\u003e \u003cem\u003eGlobalization and Health \u003c/em\u003e2021, \u003cstrong\u003e17\u003c/strong\u003e(1):139.\u003c/li\u003e\n\u003cli\u003eAgency for Healthcare Research and Quality: \u003cstrong\u003eCare Coordination Measurement Framework\u003c/strong\u003e. In: \u003cem\u003eCare Coordination Atlas.\u003c/em\u003e edn. Rockville, MD; 2014.\u003c/li\u003e\n\u003cli\u003eEmmons S, Kobourov S, Gallant M, B\u0026ouml;rner K: \u003cstrong\u003eAnalysis of Network Clustering Algorithms and Cluster Quality Metrics at Scale\u003c/strong\u003e. \u003cem\u003ePLoS One \u003c/em\u003e2016, \u003cstrong\u003e11\u003c/strong\u003e(7):e0159161.\u003c/li\u003e\n\u003cli\u003eLevine SS, Kurzban R: \u003cstrong\u003eExplaining clustering in social networks: towards an evolutionary theory of cascading benefits\u003c/strong\u003e. \u003cem\u003eManagerial and Decision Economics \u003c/em\u003e2006, \u003cstrong\u003e27\u003c/strong\u003e(2-3):173-187.\u003c/li\u003e\n\u003cli\u003eEhlert A, Kindschi M, Algesheimer R, Rauhut H: \u003cstrong\u003eHuman social preferences cluster and spread in the field\u003c/strong\u003e. \u003cem\u003eProceedings of the National Academy of Sciences \u003c/em\u003e2020, \u003cstrong\u003e117\u003c/strong\u003e(37):22787-22792.\u003c/li\u003e\n\u003cli\u003eZeltzer D: \u003cstrong\u003eGender Homophily in Referral Networks: Consequences for the Medicare Physician Earnings Gap\u003c/strong\u003e. \u003cem\u003eAmerican Economic Journal: Applied Economics \u003c/em\u003e2020, \u003cstrong\u003e12\u003c/strong\u003e(2):169-197.\u003c/li\u003e\n\u003cli\u003eKierkegaard P, Owen-Smith J: \u003cstrong\u003eDeterminants of physician networks: an ethnographic study examining the processes that inform patterns of collaboration and referral decision-making among physicians\u003c/strong\u003e. \u003cem\u003eBMJ Open \u003c/em\u003e2021, \u003cstrong\u003e11\u003c/strong\u003e(1):e042334.\u003c/li\u003e\n\u003cli\u003eDossett LA, Hudson JN, Morris AM, Lee MC, Roetzheim RG, Fetters MD, Quinn GP: \u003cstrong\u003eThe primary care provider (PCP)-cancer specialist relationship: A systematic review and mixed-methods meta-synthesis\u003c/strong\u003e. \u003cem\u003eCA: A Cancer Journal for Clinicians \u003c/em\u003e2017, \u003cstrong\u003e67\u003c/strong\u003e(2):156-169.\u003c/li\u003e\n\u003cli\u003eLinde S: \u003cstrong\u003eThe formation of physician patient sharing networks in medicare: Exploring the effect of hospital affiliation\u003c/strong\u003e. \u003cem\u003eHealth Economics \u003c/em\u003e2019, \u003cstrong\u003e28\u003c/strong\u003e(12):1435-1448.\u003c/li\u003e\n\u003cli\u003eArksey H, O\u0026apos;Malley L: \u003cstrong\u003eScoping studies: towards a methodological framework\u003c/strong\u003e. \u003cem\u003eInternational Journal of Social Research Methodology \u003c/em\u003e2005, \u003cstrong\u003e8\u003c/strong\u003e(1):19-32.\u003c/li\u003e\n\u003cli\u003eHerzog R, \u0026Aacute;lvarez-Pasquin MJ, D\u0026iacute;az C, Del Barrio JL, Estrada JM, Gil \u0026Aacute;: \u003cstrong\u003eAre healthcare workers\u0026rsquo; intentions to vaccinate related to their knowledge, beliefs and attitudes? a systematic review\u003c/strong\u003e. \u003cem\u003eBMC Public Health \u003c/em\u003e2013, \u003cstrong\u003e13\u003c/strong\u003e(1):154.\u003c/li\u003e\n\u003cli\u003eClarke V, Braun V: \u003cstrong\u003eThematic Analysis\u003c/strong\u003e. In: \u003cem\u003eEncyclopedia of Critical Psychology.\u003c/em\u003e edn. Edited by Teo T. New York, NY: Springer New York; 2014: 1947-1952.\u003c/li\u003e\n\u003cli\u003eEvan Pollack C, Wang H, Bekelman JE, Weissman G, Epstein AJ, Liao K, Dugoff EH, Armstrong K: \u003cstrong\u003ePhysician social networks and variation in rates of complications after radical prostatectomy\u003c/strong\u003e. \u003cem\u003eValue Health \u003c/em\u003e2014, \u003cstrong\u003e17\u003c/strong\u003e(5):611-618.\u003c/li\u003e\n\u003cli\u003eCasalino LP, Pesko MF, Ryan AM, Nyweide DJ, Iwashyna TJ, Sun X, Mendelsohn J, Moody J: \u003cstrong\u003ePhysician Networks and Ambulatory Care-sensitive Admissions\u003c/strong\u003e. \u003cem\u003eMed Care \u003c/em\u003e2015, \u003cstrong\u003e53\u003c/strong\u003e(6):534-541.\u003c/li\u003e\n\u003cli\u003eMoen EL, Austin AM, Bynum JP, Skinner JS, O\u0026apos;Malley AJ: \u003cstrong\u003eAn analysis of patient-sharing physician networks and implantable cardioverter defibrillator therapy\u003c/strong\u003e. \u003cem\u003eHealth Serv Outcomes Res Methodol \u003c/em\u003e2016, \u003cstrong\u003e16\u003c/strong\u003e(3):132-153.\u003c/li\u003e\n\u003cli\u003eBarnett ML, Christakis NA, O\u0026apos;Malley J, Onnela JP, Keating NL, Landon BE: \u003cstrong\u003ePhysician patient-sharing networks and the cost and intensity of care in US hospitals\u003c/strong\u003e. \u003cem\u003eMed Care \u003c/em\u003e2012, \u003cstrong\u003e50\u003c/strong\u003e(2):152-160.\u003c/li\u003e\n\u003cli\u003ePollack CE, Weissman G, Bekelman J, Liao K, Armstrong K: \u003cstrong\u003ePhysician social networks and variation in prostate cancer treatment in three cities\u003c/strong\u003e. \u003cem\u003eHealth Serv Res \u003c/em\u003e2012, \u003cstrong\u003e47\u003c/strong\u003e(1 Pt 2):380-403.\u003c/li\u003e\n\u003cli\u003eHollingsworth JM, Funk RJ, Garrison SA, Owen-Smith J, Kaufman SA, Pagani FD, Nallamothu BK: \u003cstrong\u003eAssociation Between Physician Teamwork and Health System Outcomes After Coronary Artery Bypass Grafting\u003c/strong\u003e. \u003cem\u003eCirc Cardiovasc Qual Outcomes \u003c/em\u003e2016, \u003cstrong\u003e9\u003c/strong\u003e(6):641-648.\u003c/li\u003e\n\u003cli\u003eStein BD, Mendelsohn J, Gordon AJ, Dick AW, Burns RM, Sorbero M, Shih RA, Liccardo Pacula R: \u003cstrong\u003eOpioid analgesic and benzodiazepine prescribing among Medicaid-enrollees with opioid use disorders: The influence of provider communities\u003c/strong\u003e. \u003cem\u003eJ Addict Dis \u003c/em\u003e2017, \u003cstrong\u003e36\u003c/strong\u003e(1):14-22.\u003c/li\u003e\n\u003cli\u003eDuGoff EH, Cho J, Si Y, Pollack CE: \u003cstrong\u003eGeographic Variations in Physician Relationships Over Time: Implications for Care Coordination\u003c/strong\u003e. \u003cem\u003eMed Care Res Rev \u003c/em\u003e2018, \u003cstrong\u003e75\u003c/strong\u003e(5):586-611.\u003c/li\u003e\n\u003cli\u003eUddin S: \u003cstrong\u003eExploring the impact of different multi-level measures of physician communities in patient-centric care networks on healthcare outcomes: A multi-level regression approach\u003c/strong\u003e. \u003cem\u003eScientific Reports \u003c/em\u003e2016, \u003cstrong\u003e6\u003c/strong\u003e(1):20222.\u003c/li\u003e\n\u003cli\u003eUddin S, Hossain L, Kelaher M: \u003cstrong\u003eEffect of physician collaboration network on hospitalization cost and readmission rate\u003c/strong\u003e. \u003cem\u003eEur J Public Health \u003c/em\u003e2012, \u003cstrong\u003e22\u003c/strong\u003e(5):629-633.\u003c/li\u003e\n\u003cli\u003eUddin S, Kelaher M, Piraveenan M: \u003cstrong\u003eImpact of Physician Community Structure on Healthcare Outcomes\u003c/strong\u003e. \u003cem\u003eStud Health Technol Inform \u003c/em\u003e2015, \u003cstrong\u003e214\u003c/strong\u003e:152-158.\u003c/li\u003e\n\u003cli\u003eDonker T, Wallinga J, Slack R, Grundmann H: \u003cstrong\u003eHospital networks and the dispersal of hospital-acquired pathogens by patient transfer\u003c/strong\u003e. \u003cem\u003ePLoS One \u003c/em\u003e2012, \u003cstrong\u003e7\u003c/strong\u003e(4):e35002.\u003c/li\u003e\n\u003cli\u003eChristensen A, Garrido LE, Guerra-Pe\u0026ntilde;a K, Golino H: \u003cstrong\u003eComparing community detection algorithms in psychological data: A Monte Carlo simulation\u003c/strong\u003e. In\u003cem\u003e.\u003c/em\u003e: PsyArXiv; 2020.\u003c/li\u003e\n\u003cli\u003eClauset A: \u003cstrong\u003eFinding local community structure in networks\u003c/strong\u003e. \u003cem\u003ePhys Rev E Stat Nonlin Soft Matter Phys \u003c/em\u003e2005, \u003cstrong\u003e72\u003c/strong\u003e(2 Pt 2):026132.\u003c/li\u003e\n\u003cli\u003eAmiri B, Hossain L, Crawford JW, Wigand RT: \u003cstrong\u003eCommunity Detection in Complex Networks: Multi\u0026ndash;objective Enhanced Firefly Algorithm\u003c/strong\u003e. \u003cem\u003eKnowledge-Based Systems \u003c/em\u003e2013, \u003cstrong\u003e46\u003c/strong\u003e:1-11.\u003c/li\u003e\n\u003cli\u003eMahajan A, Kaur M: \u003cstrong\u003eVarious Approaches of Community Detection in Complex Networks: A Glance\u003c/strong\u003e. \u003cem\u003eInternational Journal of Information Technology and Computer Science \u003c/em\u003e2016, \u003cstrong\u003e8\u003c/strong\u003e:35-41.\u003c/li\u003e\n\u003cli\u003eBarak-Ventura R, Richmond S, Hasanyan J, Porfiri M: \u003cstrong\u003eOn the relationship between network connectivity and group performance in small teams of humans: experiments in virtual reality\u003c/strong\u003e. \u003cem\u003eJournal of Physics: Complexity \u003c/em\u003e2020, \u003cstrong\u003e1\u003c/strong\u003e(2):025003.\u003c/li\u003e\n\u003cli\u003eO\u0026apos;Reilly CA, \u0026amp; Roberts,: \u003cstrong\u003eTask group structure, communication, and effectiveness in three organizations.\u003c/strong\u003e \u003cem\u003eJournal of Applied Psychology \u003c/em\u003e1977, \u003cstrong\u003e62\u003c/strong\u003e(6):674-681.\u003c/li\u003e\n\u003cli\u003eLin N: \u003cstrong\u003eBuilding a network theory of social capital.\u003c/strong\u003e \u003cem\u003eSocial capital \u003c/em\u003e2017:3-28.\u003c/li\u003e\n\u003cli\u003eKawachi I, Berkman LF: \u003cstrong\u003e290Social Capital, Social Cohesion, and Health\u003c/strong\u003e. In: \u003cem\u003eSocial Epidemiology.\u003c/em\u003e edn. Edited by Berkman LF, Kawachi I, Glymour MM: Oxford University Press; 2014: 0.\u003c/li\u003e\n\u003cli\u003eMart\u0026iacute; J, Bol\u0026iacute;bar M, Lozares C: \u003cstrong\u003eNetwork cohesion and social support\u003c/strong\u003e. \u003cem\u003eSoc Networks \u003c/em\u003e2017, \u003cstrong\u003e48\u003c/strong\u003e:192-201.\u003c/li\u003e\n\u003cli\u003eTortoriello M, Reagans R, McEvily B: \u003cstrong\u003eBridging the Knowledge Gap: The Influence of Strong Ties, Network Cohesion, and Network Range on the Transfer of Knowledge Between Organizational Units\u003c/strong\u003e. \u003cem\u003eOrganization Science \u003c/em\u003e2012, \u003cstrong\u003e23\u003c/strong\u003e(4):1024-1039.\u003c/li\u003e\n\u003cli\u003eRuef M: \u003cstrong\u003eStrong ties, weak ties and islands: structural and cultural predictors of organizational innovation\u003c/strong\u003e. \u003cem\u003eIndustrial and Corporate Change \u003c/em\u003e2002, \u003cstrong\u003e11\u003c/strong\u003e(3):427-449.\u003c/li\u003e\n\u003cli\u003eGoodwin N, Dixon A, Anderson GM, Wodchis WP: \u003cstrong\u003eProviding integrated care for older people with complex needs Lessons from seven international case studies\u003c/strong\u003e. In\u003cem\u003e: 2014\u003c/em\u003e; 2014.\u003c/li\u003e\n\u003cli\u003eRumball-Smith J, Wodchis WP, Kon\u0026eacute; A, Kenealy T, Barnsley J, Ashton T: \u003cstrong\u003eUnder the same roof: co-location of practitioners within primary care is associated with specialized chronic care management\u003c/strong\u003e. \u003cem\u003eBMC Family Practice \u003c/em\u003e2014, \u003cstrong\u003e15\u003c/strong\u003e(1):149.\u003c/li\u003e\n\u003cli\u003eHaj-Ali W, Moineddin R, Hutchison B, Wodchis WP, Glazier RH: \u003cstrong\u003ePhysician group, physician and patient characteristics associated with joining interprofessional team-based primary care in Ontario, Canada\u003c/strong\u003e. \u003cem\u003eHealth Policy \u003c/em\u003e2020, \u003cstrong\u003e124\u003c/strong\u003e(7):743-750.\u003c/li\u003e\n\u003cli\u003eGutberg J, Evans JM, Khan S, Abdelhalim R, Wodchis WP, Grudniewicz A: \u003cstrong\u003eImplementing Coordinated Care Networks: The Interplay of Individual and Distributed Leadership Practices\u003c/strong\u003e. \u003cem\u003eMed Care Res Rev \u003c/em\u003e2022, \u003cstrong\u003e79\u003c/strong\u003e(5):650-662.\u003c/li\u003e\n\u003cli\u003eSilich BA, Yang JJ: \u003cstrong\u003eMeasuring emergency physicians\u0026apos; work: factoring in clinical hours, patients seen, and relative value units into 1 metric\u003c/strong\u003e. \u003cem\u003eWest J Emerg Med \u003c/em\u003e2012, \u003cstrong\u003e13\u003c/strong\u003e(2):176-180.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-health-services-research","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bhsr","sideBox":"Learn more about [BMC Health Services Research](http://bmchealthservres.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/BHSR/default.aspx","title":"BMC Health Services Research","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"patient-sharing, social network analysis, network clusters, physician networks, health outcomes","lastPublishedDoi":"10.21203/rs.3.rs-4437662/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4437662/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003e. Improvements in patient safety and quality of care can be achieved by improvements in clinicians\u0026rsquo; teamwork, coordination and communication. Growing research examines the structure and dynamics of clinician networks using social network analysis. Such networks can have clusters of healthcare professionals within them, but systematized knowledge on these clusters is lacking. Our goal was to review the evidence on determinants and characteristics of healthcare professional clustering in patient-sharing networks and their associations with patient outcomes.\u003c/p\u003e\u003ch2\u003eMethods.\u003c/h2\u003e \u003cp\u003eWe searched for English-language peer-reviewed studies published up until January 4, 2021 using PubMed and EMBASE and an existing scoping review on patient-sharing by DuGoff et al (2018). We performed a systematic scoping review in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines. We performed title and abstract screening and full-text screening to identify studies that used social network analysis to examine relationships between patient-sharing network clusters and health outcomes. From the twelve eligible studies, we extracted study information such as study design and setting, population, patient-sharing definition, network measures, clustering definition, health outcomes, and reported associations.\u003c/p\u003e\u003ch2\u003eResults.\u003c/h2\u003e \u003cp\u003eThe studies varied considerably in definitions and measures of patient-sharing relations, definitions and structural measures of network clusters, settings, study population, and health outcomes. The general patterns indicate that busier physician networks (i.e., networks with more connections among physicians) are associated with worse health outcomes and better-connected physician networks are associated with better health outcomes.\u003c/p\u003e\u003ch2\u003eConclusion.\u003c/h2\u003e \u003cp\u003eThe majority of existing studies are exploratory. Rigorous theoretical grounding, interventional studies, and mixed-methods studies would help to strengthen patient-sharing research and advance our understanding of how patient-sharing clustering affects patient outcomes.\u003c/p\u003e","manuscriptTitle":"When Clinicians Group Together: A Systematic Scoping Review of Clustering in Patient-Sharing Networks","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-06-04 19:13:48","doi":"10.21203/rs.3.rs-4437662/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-05-21T10:59:43+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-12T04:01:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"180587170332839588641944006782516036976","date":"2025-05-02T18:23:56+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-07-30T19:05:31+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"10128098758289781692663353147209675050","date":"2024-07-28T21:54:06+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"299749451571729220687108437125900716515","date":"2024-07-25T09:19:43+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"245837337458314766919622286223802444519","date":"2024-07-23T15:18:04+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-07-09T15:45:23+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-05-21T10:49:30+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-05-21T09:55:30+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-05-20T09:45:36+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Health Services Research","date":"2024-05-17T15:33:11+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-health-services-research","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bhsr","sideBox":"Learn more about [BMC Health Services Research](http://bmchealthservres.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/BHSR/default.aspx","title":"BMC Health Services Research","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"4e877177-2b0d-4908-9e34-dc8f39bfd4bf","owner":[],"postedDate":"June 4th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-03-02T16:02:07+00:00","versionOfRecord":{"articleIdentity":"rs-4437662","link":"https://doi.org/10.1186/s12913-025-13893-1","journal":{"identity":"bmc-health-services-research","isVorOnly":false,"title":"BMC Health Services Research"},"publishedOn":"2026-02-25 15:58:43","publishedOnDateReadable":"February 25th, 2026"},"versionCreatedAt":"2024-06-04 19:13:48","video":"","vorDoi":"10.1186/s12913-025-13893-1","vorDoiUrl":"https://doi.org/10.1186/s12913-025-13893-1","workflowStages":[]},"version":"v1","identity":"rs-4437662","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4437662","identity":"rs-4437662","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}