Social networks of oncology clinicians as a means for increasing survivorship clinic referral

Advances in cancer diagnosis and therapy have increased long-term survival among adults, yet many survivors experience late effects that require ongoing monitoring and supportive care. Specialized survivorship clinics are recommended to address these needs, but in adult oncology such clinics are relatively new and underutilized. Effective strategies are needed to cultivate survivorship services and optimize referrals. Social network analysis (SNA) offers a framework to understand communication patterns, diffusion of innovations, and adoption of new practices among clinicians. This study aimed to identify opinion leaders and network positions associated with referrals to a newly established survivorship clinic at an NCI-designated comprehensive cancer center. The authors hypothesized that clinicians would cluster by clinical role and by referral patterns, and sought to determine personal and/or network factors associated with survivorship clinic referral to inform network-based interventions.

Pediatric oncology has a long-standing infrastructure for survivorship care, including ubiquitous survivorship clinics and evidence-based long-term follow-up guidelines, whereas adult survivorship services remain limited. National bodies (NCCN, American College of Surgeons) have prioritized adult survivorship care and survivorship care plans. Implementation of new clinical practices often follows diffusion of innovations dynamics. SNA has a history in healthcare for understanding adoption of innovations among physicians and has been used to improve guideline compliance, prescribing, and evidence-based medicine. Provider network structures are linked to quality of care. However, SNA has not been applied previously to adult cancer survivorship referral networks, highlighting a gap this study addresses.

Design and setting: Cross-sectional survey conducted June–August 2018 at USC Norris Comprehensive Cancer Center (NCI-designated comprehensive cancer center) with a survivorship clinic started in 2017 for adult survivors treated with curative intent under age 50 who received therapies associated with long-term toxicity.

Participants: 163 eligible clinicians and clinical support staff with regular, direct contact with eligible patients and roles in the referral process were identified via staff lists, rosters, and managers. Roles included physicians (medical, surgical, radiation oncologists), physician assistants, nurse practitioners, clinic nurses, nurse navigators, social workers, genetic counselors, schedulers, direct care partners, and clerical referral specialists. Purposeful recruitment ensured representation across teams and disciplines. Participation was voluntary; consent was electronic; a $10 gift card was provided. IRB approval: University of Southern California (HS-09-00673).

Data collection: A confidential online Qualtrics survey assessed: (1) awareness of the survivorship clinic (yes/no/not sure); (2) whether the respondent had ever referred patients (yes/no/not sure); (3) referrals of eligible patients in the past 12 months; and (4) estimated number of referrals in the past 12 months.

Network measures: Respondents named up to seven colleagues they go to for advice about patient care (advice network) and up to seven colleagues with whom they discuss patient care (discussion network). Directed adjacency matrices were constructed (x_ij = 1 if i nominated j). Network exposure was calculated as the proportion of nominated alters who reported referring patients to the survivorship clinic.

Analytic approach: Advice and discussion networks were analyzed separately and restricted to survey completers. Exponential Random Graph Models (ERGMs) were estimated via MCMC maximum likelihood in R (version 4.1.2). Dependent variable: presence/absence of a tie. Structural terms: edges (density), mutuality (reciprocity), geometrically weighted edgewise shared partner (Gwesp), and geometrically weighted outdegree distribution (Gwodegree). Attribute effects: role (physician; NP/PA; social worker; scheduler/other; clinic nurse as reference), awareness of the clinic, and referral behavior. Matched effects: node match on role, awareness, and referral. Additional structural terms (Gwidegree) were tested but did not converge. Multivariable logistic regression assessed factors associated with having referred patients to the survivorship clinic, including network exposure, betweenness centrality (from the discussion network), and role at the clinic (nurses as reference).

• Participation: 69/163 provided sufficient data (42% response rate). Sample composition: 31.2% schedulers/other, 29.0% physicians, 23.2% clinic nurses, 8.7% social workers, 7.3% PAs/NPs.
• Awareness and referral: 78.0% aware of the survivorship clinic; only 30.4% had ever referred patients. Among non-referrers, many were physicians, advanced practice providers, and nurses.
• Network descriptives: Average indegree was 1.6 (advice) and 2.0 (discussion). Opinion leaders by indegree (advice): two social workers (indegree 11 and 7) and a medical oncologist (8). In discussion: one of the same social workers (12) plus another social worker and a medical oncologist (each 9).
• ERGM results (both networks): Significant role homophily (node match on role; advice estimate 0.88, p<0.0001; discussion 0.79, p<0.0001). Significant Gwesp (advice 1.15, p=0.0004; discussion 1.02, p<0.0001) and Gwodegree (advice −1.47, p<0.0001; discussion −1.33, p=0.001), indicating shared partners and centralized outdegree distributions. In discussion, mutuality was significant (1.52, p<0.0001).
• Attribute effects: Social workers and NPs/PAs were significantly more likely to receive nominations (advice: social workers 0.38, p=0.0003; NP/PA 0.31, p=0.04; discussion: social workers 0.52, p=0.0002; NP/PA 0.37, p=0.004). Schedulers/other were less likely to receive nominations (advice −0.31, p=0.04; discussion −0.26, p=0.05). Awareness and referral attributes were not significant in tie formation.
• Logistic regression (discussion network): Betweenness centrality positively associated with ever referring patients (estimate 0.43, SE 0.19, p=0.025), independent of role and network exposure. Network exposure and role categories were not significant predictors. No significant predictors emerged from the advice network models.
• Interpretation: Individuals in bridging positions (high betweenness) are more likely to refer; social workers emerged as key opinion leaders in the advice network.

Findings indicate that occupational role homophily structures clinician communication, potentially constraining cross-role information flow about survivorship referrals. Although awareness of the clinic was relatively high, referrals were low and referral behavior did not appear to diffuse through the network as an attribute in the ERGMs, suggesting the need for intentional network interventions. Social workers and NPs/PAs were central targets for information exchange, with social workers specifically emerging as opinion leaders in the advice network. Clinicians occupying bridging positions (high betweenness centrality) in the discussion network were more likely to have referred patients and are well-situated to facilitate diffusion of referral behaviors across clusters. Network-informed interventions that engage both opinion leaders (e.g., social workers) and strategic brokers (high-betweenness members) may enhance communication between occupational groups and accelerate uptake of survivorship referral practices.

Considering network structure and dynamics offers a novel, practical approach to improving survivorship clinic referrals and care quality. This study identifies strong role-based homophily, highlights social workers as opinion leaders, and shows that clinicians with high betweenness centrality are more likely to refer patients. These insights can guide inclusive, network-based interventions that leverage opinion leaders and strategic brokers to disseminate information across clinician groups. Future research should test the effectiveness of such interventions in longitudinal, controlled designs and determine whether mixed-group versus role-specific strategies optimize referral uptake across diverse healthcare settings.

• Response rate was 42%, within typical ranges for physician surveys but lower than ideal; missing data limited sample size and is less ideal for ERGMs modeling ties.
• Missing demographic information for non-participants and incomplete surveys for some participants.
• Cross-sectional design precludes causal inference.
• Survey did not assess reasons for non-referral or role-specific differences in the referral process to reduce participant burden.
• Generalizability may be limited due to site-specific network dynamics at an NCI-designated center; community-based centers may have different communication patterns.
• Although centrality measures like indegree are robust to sampling/missingness, overall network missingness remains a limitation.