Mapping the Social Network Structure and Composition of Black Individuals with Rheumatic and Musculoskeletal Conditions

Abstract

Objectives:

Our objective was to describe the social networks of Black individuals with rheumatic and musculoskeletal conditions and understand the clustering of health-related behaviors to inform future community-based, peer-led interventions.

Methods:

We used an adapted Personal Network Survey for Clinical Research (PERSNET) to map the personal social networks of Black individuals with rheumatic conditions in Boston and Chicago. We used egocentric network analyses to quantify network composition, and descriptive statistics and Welch’s two-sample t-tests to assess network characteristics and behavioral concordance between participants and their network members.

Results:

We mapped the social networks of 40 individuals with rheumatic conditions in Boston and Chicago. All participants self-identified as Black, with 6 indicating Hispanic or Latino ethnicity. One participant self-identified as male. Networks had a median size of 10, with 54% of all possible ties present and an average of 5 structurally distinct connections within each network. Boston networks had a larger proportion of kin (71%) compared to those in Chicago (40%). There was greater diversity in sex and education within Boston networks. Chicago networks displayed more balanced strong and weak ties and higher prevalence of health problems within networks. Smoking and drinking behaviors clustered within networks across both sites, which may reflect patterns of behavioral homophily or social influence.

Conclusions:

Differences in network density, constraint, and diversity highlight structural and relational patterns that may influence how health behaviors develop. They also reveal network configurations that position individuals to act as bridges, introducing new information and behaviors into communities where medical mistrust and systemic inequities undermine health messaging.

Introduction

Rheumatic and musculoskeletal conditions affect nearly one in five U.S. adults, making them among the most prevalent chronic conditions nationwide.^1,2 These conditions, including connective tissue diseases (e.g. systemic lupus erythematosus (SLE)), osteoarthritis (OA), inflammatory arthritis (e.g. rheumatoid arthritis (RA) and spondyloarthropathies), and crystalline arthritis (e.g. gout), are associated with chronic pain, functional limitations, and decreased quality of life.³ Effective management often requires pharmacologic interventions, such as disease-modifying antirheumatic drugs (DMARDs), in addition to engagement in health-promoting behaviors, such as nutritious diet, physical activity, and routine preventive care.^2,4,5

However, significant racial and ethnic disparities exist in disease prevalence, treatment access, and outcomes.^4,6 Black individuals are two to three times more likely to be diagnosed with SLE compared to White individuals.⁷ In RA, Black patients experience greater functional limitations and achieve lower rates of clinical remission.⁸ These inequities are deeply rooted in social determinants of health, including racism, chronic stress, unequal access to care, and socioeconomic disadvantage.^4,6

Discrimination and historical injustices contribute to distrust in providers and poor adherence to treatment plans, limiting acceptance of new information, particularly among Black patients.⁹ Thus, trusted peers and community leaders are instrumental in shaping health beliefs. Social networks can serve as conduits for information, support, and health behavior influence.¹⁰

Health behaviors, such as physical activity, diet, and engagement in preventative care can cluster within social groups.^11–13 One study linking social network theory to health behaviors found that individuals were “significantly more likely to engage in regular exercise if a peer did,” with stronger relationship ties increasing this likelihood.¹³ Similarly, recent work found that individuals’ attitudes towards vaccination were shaped by whether they knew someone who had been vaccinated or hospitalized for a vaccine-preventable infection.¹² These findings highlight the potential of social networks to function as active channels through which behaviors, risks, and even health outcomes are transmitted.

Despite existing evidence, social network-based interventions remain underutilized in public health, and to our knowledge, have never been focused on individuals with rheumatic conditions.¹⁴ These strategies could be especially beneficial for patients with these conditions for whom lifestyle modifications are recommended by trusted national organizations and preventive care is critical due to the frequent use of immunosuppressive medications.^2,5 Improvements in diet and physical activity, for example, have been shown to reduce pain, promote better function, and improve overall disease management.^15,16

This study aimed to identify, describe and compare the personal social networks of these individuals to understand how preventive health-associated behaviors are clustered and where new information transmission about healthy behaviors may be possible. Our goal was for this work to inform future peer-driven, community-based trials that leverage personal networks to promote healthy behaviors and preventive care uptake among individuals with rheumatic conditions.

Materials and Methods

Data collection was conducted between December 2024-April 2025 in the greater Boston and Chicago areas. The protocol was approved by the Institutional Review Boards (IRBs) at Mass General Brigham, Northwestern University, Boston Children’s Hospital and Boston Medical Center. All participants provided informed consent prior to enrollment.

Inclusion/Exclusion Criteria

Eligible participants were English-speaking individuals of any sex and ethnicity who were 18–85 years of age and self-identified as Black. The study team, comprised of community leaders, patients, rheumatologists and public health researchers, focused on Black communities due to the disproportionate burden of rheumatic conditions, racial disparities in adverse outcomes, and high prevalence of mistrust due to historical and current injustices, which elevates the importance of social networks to convey health information. Participants self-reported a physician diagnosis of ≥1 rheumatic or musculoskeletal condition, including but not limited to connective tissue diseases, inflammatory arthritis, crystalline arthritis, and osteoarthritis. Participants were required to have received at least one COVID-19 vaccine dose after August 31, 2022. While COVID-19 vaccination was an eligibility requirement and informed the overarching study context, the specific aim of this component was to understand how health behaviors are clustered and shared within social networks, providing a framework to apply network interventions across multiple preventive health behaviors. Additionally, participants were required to score at least 2 out of 4 on an adapted version of the Berkman-Syme Social Network Index, which asked questions about marital status, contact with friends/family, and group affiliations to assess social connectivity.¹⁷ A score of 0 indicates high social isolation, whereas a score of 4 indicates high social integration.¹⁷ We hypothesized that a score ≥2 would indicate that individuals were sufficiently socially connected to engage with social network mapping and serve as future peer leaders for information sharing.¹⁷ We excluded individuals who were pregnant, incarcerated or institutionalized.

Recruitment

Participants were recruited through academic medical centers, including Boston Children’s Hospital, Boston Medical Center, Brigham and Women’s Hospital (BWH), and Northwestern University, as well as by community partners, including Alliance Chicago, Lupus Society of Illinois, True Alliance Center, Inc., Vital CxNs, and Women of Courage, between May and August 2024 in the greater Boston and Chicago areas. Investigators identified eligible patients, prioritizing those with involvement in disease advocacy efforts, support groups, or online health communities. Various methods were used to recruit those who were eligible, including targeted flyer distribution, recruitment through local recreational centers or businesses, and secure messaging via electronic patient portals with permission from providers.

Enrollment

Interested individuals completed a phone-based pre-screening questionnaire administered by study staff, who recorded responses in a secure REDCap database. This questionnaire included demographic and clinical information (age, sex, race, rheumatic or musculoskeletal condition diagnoses, and vaccination history, including the COVID-19 and influenza vaccines). Eligible participants then completed the Berkman-Syme Social Network Index to assess social connectedness. Those who scored ≥2 were invited to participate and provided informed consent.

PERSNET Instrument

We used an adapted version of the Personal Network Survey for Clinical Research (PERSNET), a validated tool originally based on the General Social Survey and developed for scalable social network analysis in clinical populations.¹⁸ The modified instrument began with asking the participant (“ego”) three name-generating questions:

1. “Who do you discuss personal matters with?”

2. “Who do you often socialize with?”

3. “Who provides support for your health?”

Participants listed an unlimited number of social network members (“alters”) and assessed the strength of ties between the first ten during the network inter-relater section. Finally, participants were asked to provide demographic and behavioral information about each alter, including health habits (e.g., drinking, smoking, diet, physical activity) and any known health conditions (e.g., general health issues, pain, mental/cognitive health, cardiac problems). Decreased drinking and smoking were defined as cutting back on heavy drinking or smoking habits. Regular exercise was defined as physical activity 3–4 times weekly for over the past 3 months. Healthy diet was defined as eating healthy, balanced meals regularly over the past 3 months. Our adapted version contained additional questions about vaccination status and presence of ≥1 rheumatic or musculoskeletal condition.

Data Collection

Most surveys were administered by phone by study staff (N=23 in Boston, N=9 in Chicago). Several patients opted to complete the survey independently via secure REDCap link (N=2 in Boston, N=6 in Chicago). In some cases, due to length, surveys were completed over two or three calls. All participants received a $50 electronic gift card or check following survey completion.

Analysis Plan

Descriptive statistics were used to characterize the structure and composition of participants’ social networks. Network structure included quantitative descriptions of the “arrangements of social ties” (e.g., network size and density).¹⁹ These standard network measures (including network size, density, constraint, effective size, max degree, and mean degree) were derived from participant responses about their relationships with their alters and their individual demographics.²⁰ Network compositional variables described characteristics of social network members.¹⁹ We analyzed indices of qualitative variation (IQV) for sex and education, proportion kin, and the proportion of alters with specific health behaviors or conditions (e.g. drinking, smoking, exercise, diet, cardiac and rheumatic conditions, cognitive concerns, and pain).

We created social network maps using the information obtained from the PERSNET survey for both the Boston and Chicago sites, which can be found in Supplemental Figures 1–16. Each figure shows the network maps of all participants in their respective cities. The lines indicate strong or weak relationships between egos and alters, and the shapes represent whether egos and alters engaged in healthy or unhealthy behaviors. The letters and numbers surrounding the images serve as identifiers for each individual network map. We additionally stratified the graphs by different network characteristics including health behaviors and conditions. The code was adapted from https://github.com/AmarDhand/PersonalNetworks.

We performed Welch’s two sample t-tests to evaluate the difference in mean proportion of health behaviors between egos and their alters using a significance level of p<0.05. All analyses were completed in R (Version 4.5.0).

Results

Table 1 shows the demographics of the enrolled egos by city. We screened 100 individuals and enrolled 41 participants: 25 from Boston sites and 16 from Chicago, with one Chicago participant excluded due to incomplete data. The mean age was 53.7 years (SD 14.1). Boston had a slightly older cohort, with a mean age of 54.4 (SD 14.9) compared to 52.7 (SD 13.2) in Chicago. All but one participant identified as female. Four participants in Boston identified as more than one race, and 15% of participants, all from Boston, reported Hispanic or Latino ethnicity. Most of the enrolled participants received a Berkman-Syme score of 3, with 15 of the individuals screened ineligible based on their scores.

Table 1:

Participant Demographics, by City

Characteristic	Overall	Boston	Chicago
	N = 40	N = 25	N = 15
Age- mean (SD)	53.7 (14.1)	54.4 (14.9)	52.7 (13.2)
Sex – N (%)
Female	39 (98%)	25 (100%)	14 (93%)
Male	1 (2.5%)	0 (0%)	1 (6.7%)
Race– N (%)
Black/African American	36 (90%)	21 (84%)	15 (100%)
More than one race	4 (10%)	4 (16%)	0 (0%)
Ethnicity– N (%)
Declined	1 (2.5%)	0 (0%)	1 (6.7%)
Hispanic or Latino	6 (15%)	6 (24%)	0 (0%)
Not Hispanic or Latino	33 (83%)	19 (76%)	14 (93%)
Education– N (%)
Advanced Degree	11 (28%)	5 (20%)	6 (40%)
Associate degree	5 (13%)	1 (4%)	4 (27%)
Bachelor’s Degree	9 (23%)	7 (28%)	2 (13%)
High School	15 (38%)	12 (48%)	3 (20%)
Marital Status– N (%)
Married	15 (38%)	9 (36%)	6 (40%)
Not married	25 (63%)	16 (64%)	9 (60%)
Most recent COVID-19 vaccine – N (%)
9/22 Bivalent	13 (33%)	5 (20%)	8 (53%)
9/23 Updated vaccine	27 (68%)	20 (80%)	7 (47%)
COVID Vaccine history– N (%)
One of two	3 (7.5%)	3 (12%)	0 (0%)
Two of two	37 (93%)	22 (88%)	15 (100%)
Has had COVID-19
Don’t know	1 (2.5%)	1 (4%)	0 (0%)
No	17 (43%)	10 (40%)	7 (47%)
Yes	22 (55%)	14 (56%)	8 (53%)
Berkman-Syme Index Score
2	10 (25%)	7 (28%)	3 (20%)
3	19 (48%)	12 (48%)	7 (47%)
4	11 (28%)	6 (24%)	5 (33%)
Community organization participation–N (%)	20 (50%)	11 (44%)	9 (60%)
Cut back heavy drinking– N (%)
Does not drink heavily	32 (80%)	21 (84%)	11 (73%)
No	1 (2.5%)	0 (0%)	1 (6.7%)
Yes	7 (18%)	4 (16%)	3 (20%)
Exercised regularly– N (%)	18 (45%)	10 (40%)	8 (53%)
Cut back smoking– N (%)
Does not smoke	34 (85%)	19 (76%)	15 (100%)
No	1 (2.5%)	1 (4%)	0 (0%)
Yes	5 (13%)	5 (20%)	0 (0%)
Has a healthy diet– N (%)	24 (60%)	17 (68%)	7 (47%)
Health Problem - General– N (%)	21 (53%)	14 (56%)	7 (47%)
Health Problem - Pain– N (%)	22 (55%)	12 (48%)	10 (67%)
Health Problem - Cognitive/Mental –N (%)	8 (20%)	2 (8%)	6 (40%)
Health Problem – None – N (%)	10 (25%)	8 (32%)	2 (13%)
Rheumatic or Musculoskeletal Condition
Lupus	21 (53%)	11 (44%)	10 (67%)
Osteoarthritis	17 (43%)	10 (40%)	7 (47%)
Rheumatoid Arthritis	11 (28%)	8 (32%)	3 (20%)
Fibromyalgia	7 (18%)	6 (24%)	1 (6.7%)
Other Rheumatic Conditions	5 (13%)	3 (12%)	2 (13%)
Other Inflammatory Arthritis	2 (5.0%)	1 (4.0%)	1 (6.7%)
Osteoporosis	1 (2.5%)	1 (4.0%)	1 (6.7%)
Psoriatic Arthritis	1 (2.5%)	1 (4.0%)	0 (0%)
Sarcoidosis	1 (2.5%)	1 (4.0%)	0 (0%)

mean (SD); n (%)

All participants were previously vaccinated against COVID-19. Most completed the primary Moderna or Pfizer series, though 7.5% received only one dose. Most Boston participants (80%) received the updated 2023 COVID-19 vaccine, compared to 47% in Chicago. Thirteen participants received the bivalent booster after September 2022.

Most participants were non-smokers and did not drink heavily. Chicago participants reported higher rates of regular exercise (53%) compared to those in Boston (40%). However, a greater proportion of Boston participants reported maintaining a healthy diet (68%) compared to those in Chicago (47%). Pain was the most prevalent health concern across both sites.

Network Structure

Participants named individuals with whom they share personal, social, or health-related support ties. The overall median network size was 10 alters (interquartile range [IQR] 6–12) (Table 2). Chicago’s median network size of 10 alters (IQR 7–13) was slightly larger than Boston’s, with a median of 9 (IQR 6–10) alters. Additionally, Chicago maps had a greater distribution of strong and weak ties, represented by the red and blue lines within the maps, respectively. Figures 1 and 2 illustrate the variability across maps.

Table 2:

Social Network Structure and Composition

Characteristic		Overall	Boston	Chicago
		N = 40	N = 25	N = 15
Network Size	Number of alters listed, excluding the ego19	10 (6, 12)	9 (6, 10)	10 (7, 13)
Density	Proportion of actual ties among alters out of all possible ties (network cohesion)19,31	0.54 (0.40, 0.90)	0.59 (0.32, 0.92)	0.52 (0.42, 0.87)
Unknown		1	1	0
Constraint	Measure of how interconnected an alter’s connections are19,32	0.38 (0.34, 0.53)	0.38 (0.34, 0.48)	0.38 (0.33, 0.55)
Effect Size	Number of non-redundant (independent) alters in a network	4.86 (2.55, 6.46)	4.82 (2.50, 5.77)	5.05 (2.67, 6.84)
Max Degree	Maximum number of ties held by a single alter19	5.00 (4.00, 7.00)	6.00 (4.00, 8.00)	5.00 (4.00, 7.00)
Mean Degree	Average number of ties per alter19	3.78 (2.54, 4.85)	3.75 (2.40, 5.00)	3.80 (3.00, 4.33)
IQV sex (heterogeneity)	A measure using an index where 0 indicates complete homogeneity (all network members are one class) and 1 indicates perfect heterogeneity (network members are equally divided among all classes)19	0.67 (0.48, 0.84)	0.76 (0.40, 0.88)	0.64 (0.56, 0.82)
IQV education (heterogeneity)		0.71 (0.50, 0.81)	0.79 (0.65, 0.85)	0.61 (0.24, 0.76)
Proportion of kin in network	Percentage of alters who are family members, including spouses	0.50 (0.36, 0.83)	0.71 (0.50, 1.00)	0.40 (0.20, 0.50)
SD of ages in network	Standard deviation of alters’ ages	13.0 (11.4, 16.5)	13.0 (11.6, 16.3)	13.7 (11.0, 18.0)
Unknown		1	1	0
Proportion of heavy drinkers	Percentage of alters who had not reduced heavy alcohol or tobacco use	0.13 (0.00, 0.37)	0.12 (0.00, 0.40)	0.20 (0.00, 0.33)
Proportion of smokers		0.11 (0.00, 0.25)	0.17 (0.00, 0.40)	0.00 (0.00, 0.14)
Proportion who do not exercise regularly	Percentage of alters who exercised fewer than 3–4 times per week in the three months preceding survey completion	0.40 (0.20, 0.65)	0.40 (0.20, 0.62)	0.30 (0.20, 0.67)
Proportion who have a poor diet	Percentage of alters who did not maintain a healthy diet over the prior three months	0.20 (0.00, 0.33)	0.20 (0.00, 0.30)	0.20 (0.00, 0.43)
Proportion who have a health problem	Percentage of alters experiencing general health issues, pain, cognitive or mental health, cardiac disease, or rheumatologic diagnoses	0.55 (0.30, 0.95)	0.50 (0.20, 0.67)	0.90 (0.50, 1.20)

Median (Q1, Q3); n (%)

Figure 1.

Boston social network maps

Figure 2.

Chicago social network maps

The median density was 0.54, indicating that over half of all possible alter-alter ties existed. Boston networks had a slightly higher density (0.59) than those in Chicago (0.52). This pattern is illustrated in Boston map B5, where many alters are mutually connected, forming a tightly clustered network structure. One Boston participant had missing data for density.

The median network constraint across both sites was 0.38 (IQR 0.34–0.53), suggesting that while many alters were part of tightly interconnected clusters, participants still maintained access to individuals who did not have connections with others in the network, known as nonredundant ties. This is supported by the overall median effective size of 4.86, indicating that participants had approximately five structurally distinct connections. The median maximum degree was 6.00 in Boston, reflecting the potential for certain individuals to serve as central connectors within a network.

Network Composition

Kinship ties differed notably by site: a median of 71% of alters in Boston networks were family members, compared to just 40% in Chicago (Table 2). Boston networks showed greater heterogeneity in sex (76%) and education level (79%), suggesting more diverse social ties across those dimensions.

Supplemental Figures 1–16 illustrate the social network maps for Boston and Chicago egos based on health behaviors, some of which show variation in behaviors across networks. In Chicago, 20% of alters were reported to drink alcohol heavily, compared to 12% in Boston. Maps A6 and B14 in Supplemental Figure 2 and B2 and C4 in Supplemental Figure 10 illustrate completely homogenous behaviors among egos and alters. In these maps, all individuals exhibit the same drinking behaviors, as opposed to maps A3 in Supplemental Figure 2 and A4 in Supplemental Figure 10 which show more evenly distributed behaviors. The median proportion of individuals who smoked across both sites was 11%, though Boston had a higher median (17%) than Chicago (0%). Some network maps showed discordance among egos, even when all their alters exhibited the same behavior. Map D4 in Supplemental Figure 3 shows an ego who smokes, a behavior none of their social network members share. Conversely, map A2 in Supplemental Figure 11 shows an ego in a smaller network who does not smoke, although their network members do. Boston networks contained more alters who did not exercise regularly. However, in both cities, approximately 80% of alters were reported to have healthy dietary habits, compared to 60% of egos who reported maintaining a healthy diet. Importantly, 90% of Chicago alters were reported to have at least one health problem, compared to 50% of alters in Boston.

Mean Proportion Differences

Using Welch’s two-sample t tests (Table 3), we found that networks of egos who engaged in a certain behavior tended to contain a higher mean proportion of alters with the same behavior. In Table 3, “Ego Negative Behavior” represents those who do not have the unhealthy behavior (e.g. egos who do not smoke), and “Ego Positive Behavior” represents those who do have the unhealthy behavior (e.g. egos who do smoke). Although results were not statistically significant at p<0.05, there were notably large differences observed for drinking and smoking. Among egos who drank heavily, 43% of alters reportedly drank heavily, compared to 19% of alters in networks where the ego did not drink. For smoking, 36% of alters reportedly smoked when egos smoked, whereas only 15% of alters smoked in networks where egos did not. For physical activity, 37% of alters exercised in networks where egos did not, while 45% exercised in network where egos did as well.

Table 3:

Associations Between Ego and Alter Health Behaviors

	Ego Negative Behavior	Ego Positive Behavior	p-Value*
	Mean (standard deviation)
Alter Heavy Drinking	19% (27%) (ego N=8)	43% (31%) (ego N=32)	0.064
Alter Smoking	15% (23%) (ego N=6)	36% (35%) (ego N=34)	0.206
Alter Unhealthy Diet	20% (20%) (ego N=16)	26% (28%) (ego N=24)	0.495
Alter No exercise	37% (25%) (ego N=22)	45% (34%) (ego N=18)	0.420

^*Welch’s two-sample t-tests were used to compare means across groups

Discussion

Our study offers insight into the structure and composition of the personal social networks of 40 Black individuals in two U.S. cities with rheumatic and musculoskeletal conditions. Using the PERSNET survey, we captured both the size and connectivity of participants’ networks, and an understanding of the health-related behavioral norms and demographic diversity within them.

Our recruitment strategy, which intentionally partnered with both community-based organizations and academic centers, successfully identified individuals who generally maintained robust and varied social networks. By leveraging trusted community groups alongside rheumatology clinics, we reached participants who were both socially connected and engaged in advocacy and peer support, which positions them to spread health information in future interventions. Importantly, our analysis suggests heterogeneity in network composition and structure across both sites, underscoring the need to design interventions adaptable to the unique relational patterns, cultural norms, and health needs of network members and the areas in which they live.

Compared to other clinical populations, participants in our study had relatively large social networks. Stroke survivors assessed by Dhand et al. had mean network sizes of 5–8 alters.²¹ Similarly, a Brinkhues et al. study of individuals with diabetes reported an average network size of approximately 7.²² The larger networks observed in our study may reflect the intentional recruitment of more socially connected individuals who are well positioned to influence others, selected through Berkman-Syme Index score screening. On this 0–4 scale, most participants in our study (48%) scored a 3, with scores of at least 2 required for participation. This was intentional as one key aim of this study was to develop a strategy to identify individuals with robust social networks with an eye towards a future intervention where these individuals would disseminate information through their networks.

The overall median network density was 0.54, which indicates a moderately cohesive network structure. This value is notably lower than those reported in two other clinical network studies by Dhand et al. In the first, stroke survivors had a baseline network density of 0.79, and in the second, former professional football players’ baseline network density was 0.75.^19,21 While high density networks may offer benefits, such as greater social reinforcement for health behaviors, they may also limit the flow of new information due to redundancy within the network. The variation in density within our sample (IQR 0.40–0.90) suggests that some individuals, particularly those in lower-density networks, may be better positioned to act as bridges between otherwise unconnected groups, potentially supporting the flow of new health information related to vaccine guidance, physical activity, and lifestyle modification.²³

Network constraint and effective size offer additional insights into network connectivity. The median constraint of 0.38 suggests moderate clustering and the presence of structural holes, which are network gaps that can enable access to novel information.²⁴ The median effective size was 4.86, higher than the 3.50 value reported in the aforementioned Dhand et al. study of former professional football players.¹⁹ This implies that on average, our participants had approximately five distinct network members through whom novel information could circulate, a promising feature for interventions aiming to promote health behavior change through these networks. Maps E2 in Figure 1 and C2 in Figure 2 exemplify this pattern, each displaying three alters who were disconnected from the rest of the network.

Statistical tests measuring the statistical significance of between-site comparisons were not performed due to the small sample size, however, we observed several differences between sites. Boston participants’ social networks were primarily kin-based (71%), consistent with prior descriptions of strong familial clustering in Black communities and immigrant populations.²⁵ While these kin-heavy networks may offer strong support, they may also represent more homogenous behavioral norms. Chicago networks had fewer kin ties (40%) that could reflect historical migration and neighborhood structures in the region.^26,27 Such networks may introduce a broader range of health behaviors and perspectives, creating more opportunity for behavioral influence or diffusion of new information.

The social network maps reveal visual patterns in the clustering and distribution of health behaviors across study sites. In both Boston and Chicago, behaviors such as drinking, physical activity, and healthy eating were not evenly distributed across networks, as shown in Supplemental Figures 2, 4, 5, 10, 12, and 13. Instead, they appear to cluster within certain subnetworks which may reflect patterns consistent with behavioral homophily or social influence, though those mechanisms cannot be distinguished in this study. Across both sites, healthy diets were more commonly reported among alters than regular physical activity. Additionally, there were more individuals who did not smoke than those who did, a potential reflection of public health initiatives to decrease nicotine use. However, the structural configuration of these behaviors varied by location: Chicago exhibited a more even distribution of strong and weak ties, while Boston often demonstrated high-density clustering.

In Chicago. 90% of alters were reported to have at least one health problem. This may reflect broader structural inequities and an underlying heightened burden of illness among Chicago participants, including redlining and environmental racism that increases pollution and toxin exposure.²⁸ Such disparities have implications for how health information is shared and supported within personal networks, especially when a significant portion of one’s network is also navigating complex health needs.

These visual and structural findings suggest that socially connected individuals with rheumatic conditions hold diverse social networks, reflected in differences in tie strength, demographic composition, and network interconnectedness, which may facilitate the sharing of health information and potentially influence the spread of healthy behaviors. Our study highlights that patients with robust social networks can serve as bridges between individuals with varied health behaviors and demographic traits in a future community-engaged intervention. This bridging function is relevant for peer-led interventions, which depend on individuals embedded in networks with both high connectivity and access to nonredundant ties. As Granovetter argued, behavior change spreads more effectively through networks with weak ties that can connect otherwise disconnected groups.²³ Some of our participants, especially those with lower constraint and moderate density, displayed network characteristics consistent with this bridging function that may be important for future network-based interventions.

There are several limitations to this study. First, we intentionally recruited individuals in two U.S. cities who were socially connected to understand how information may be spread through robust social networks. Thus, these results are not generalizable to individuals with baseline social isolation. Study participants were predominately female, which limits the generalizability of our findings. The higher participation of female participants could suggest differences in response or engagement that were not directly measured. This is consistent with findings from our previous work, where women compared to men were more likely to participate in a study that aimed to disseminate health information through their communities.^29,30 The small sample size also limited our ability to perform between-group comparisons and detect significant differences, however between-site comparisons were not the primary goal of this work.

Additionally, all health behavior data about network members were based on ego reports and were not verified by the alters, which may be subject to recall or social desirability bias. However, this approach has been validated in similar studies and still offers a feasible approach for large-scale community research.¹⁸ The cross-sectional nature of these data makes it difficult to determine whether observed similarities in health behaviors were impacted by social influence, selection effects, or shared environmental contexts. Lastly, we did not test the dissemination of information and how behaviors are spread through social networks in this study, as this is the goal of future work.

Network-based strategies may offer a scalable and sustainable approach to improve outcomes among patients with rheumatic conditions, particularly in communities historically marginalized by the healthcare system. Our findings provide early evidence that personal social networks could be leveraged to disseminate preventive health information and encourage behaviors that support disease management. This method has, to our knowledge, never been studied in a population of individuals with rheumatic or musculoskeletal conditions and may be a powerful tool to reduce known disparities through preventive care communication and health behavior changes. While prior research in other populations has focused on the influence social network members have on the ego, leveraging data from this study, we aim to explore the reverse: how the ego can spread health behaviors to influence their alters. This approach is promising in communities where medical mistrust and systemic inequities may have undermined health messaging and limited uptake of healthy behaviors. Identification of socially connected individuals with rheumatic and musculoskeletal conditions who model healthy behaviors will allow us to test whether information disseminated by these individuals through their networks promotes similar behaviors (e.g. preventive care uptake, physical activity, medication adherence) among their peers.

Significance and Innovations.

• This study provides the first characterization of personal social networks among Black individuals with rheumatic or musculoskeletal conditions.

• Our findings reveal structural and relational patterns that may influence how health behaviors are developed and shared within social networks.

• Participants who have access to individuals removed from network clusters, known as nonredundant ties, can serve as bridges within their networks and facilitate the dissemination of preventive health information and healthy behaviors.

• There are meaningful differences in network composition between Boston and Chicago, which highlight the importance of local and sociocultural contexts in developing network interventions.