Does Whom Patients Sit Next to during Hemodialysis Affect Whether They Request a Living Donation? : Kidney360

Journal Logo

Original Investigations: Transplantation

Does Whom Patients Sit Next to during Hemodialysis Affect Whether They Request a Living Donation?

Gillespie, Avrum1; Fink, Edward L.2; Gardiner, Heather M.3; Gadegbeku, Crystal A.1; Reese, Peter P.4; Obradovic, Zoran5

Author Information
Kidney360 2(3):p 507-518, March 2021. | DOI: 10.34067/KID.0006682020

Abstract

Key Points

  • Patients on hemodialysis formed relationships with the other patients in the clinic whom they sat next to and had similar transplant behaviors.
  • Participants who requested a living donation formed more relationships within the clinic, and discussed transplantation with each other.
  • Our study identifies health-behavior homophily that can be used for future behavioral interventions involving the hemodialysis clinic social network.

Introduction

In-center hemodialysis clinics are conducive to the formation of patient social networks because patients are treated in a group setting. Patients are seated together for several hours thrice weekly (123–4) and, when people are near each other repeatedly, they often form relationships (5,6). The set of these relationships (links) is called a social network (Table 1) (7). Understanding the factors that lead patients to form a social network within the hemodialysis clinic is a research priority because patients’ social networks have been shown to improve and worsen chronic diseases (5,6,891011121314–15). Furthermore, social-network interventions improve outcomes in these chronic diseases (8,9,14,15).

Table 1. - Glossary of terms
Term Definition
Social-network theories
Social contagion “The spread of affect or behaviour from one crowd participant to another; one person serves as the stimulus for the imitative actions of another.” (16) This includes sharing information, imitating behaviors, and enforcing norms.
Homophily The tendency for people to seek out, or be attracted to, those who are similar to themselves.
 Sociality The tendency for people to form relationships, commonly referred to as extroverted.
 Clustering The tendency for network members to share mutual relationships. For example, if member A is linked to member B and member C, it is likely that members B and C are also linked. This is also known as transitivity.
 Centralization The tendency for a few members to have many links while most other members have one or two links. For example, if member A joins a network and member C has five links and member B has one link, member A would preferentially form a link with member C. This is known as preferential attachment.
Social-network analysis
 Link A social-network term for a relationship between two network members.
 Edges Another term for a link or relationship used in graph theory.
 Degree The number of relationships (links) a network member has.
 Density How many links exist between members of a social network out of the possible number of links that could exist among members.
 Dense network In a dense network, most or all of the members are linked to the other members.
 Sparse network In a sparse network, most members are not linked to the other members.
 STERGM A separable temporal exponential random graph model analyzes the network as a multivariate observation with a link as the dependent variable. The observed network is then compared to 100,000 randomly generated Markov random graphs (networks) using maximum pseudo-likelihood estimation and Monte Carlo maximum likelihood estimates (26).
 GWESP Geometrically weighted edgewise shared partner weights the probability of two members forming a relationship on the basis of the number of relationships with other members they have in common. This parameter is used in STERGMs to approximate clustering within the network.
 GWDegree Geometrically weighted edgewise degree weights the probability of a network member forming a relationship on the basis of the number of relationships they already have (number of relationships=degree). This parameter is used in STERGMs to approximate centralization within the network.
STERGM, separable temporal exponential random graph model; GWESP, geometrically weighted edgewise shared partner; GWDegree, geometrically weighted edgewise degree.

In a prospective cohort study of a social network involving patients on hemodialysis (1), we showed that patients formed a hemodialysis-clinic social network in which they discussed health and kidney transplantation, and they completed more of their transplant testing if their network members also completed theirs. However, we could not differentiate whether this similarity in transplant behavior was via social contagion (Table 1) (13,14,16), where participants motivated each other to complete transplant testing, or whether participants, who were motivated to complete their testing, formed relationships with similarly motivated participants via health-behavior homophily (8,17,18) (Table 1). The completion of steps toward transplantation is not suitable for a contagion model because it is multistep process (19) that requires multiple acts and can take many months or years to complete. Social-contagion models are better suited for single behaviors that occur at a single point in time, such as asking a family member or friend to consider being a living donor, which we will refer to as requesting a living donation.

In our previous research of this cohort (1), a greater proportion of the members of the hemodialysis-clinic social network discussed kidney transplantation with other participants and requested a living donation compared with participants who were not members of the hemodialysis-clinic network (isolates). This association between network participation and living-donor requests is notable because many people are uncomfortable requesting a donation (20,21), and difficulties requesting a living donation are a significant barrier to living-donor kidney transplantation. We hypothesize that, if participants formed a relationship and discussed transplantation with someone in the hemodialysis clinic who made a living-donor request, the participants could be influenced via social contagion to request a living donation.

In this study, we first examined how seating distance and demographic characteristics were associated with patients forming a relationship. We then examined the social-network theories of sociality (some people are more social than others; Table 1) (17), health-behavior homophily, and social contagion that could explain the increase in network participants who discussed living donation with other patients and requested a living donation from family and friends. Understanding the formation and function of the hemodialysis-clinic social network should inform future interventions and could increase the rate of living-donor kidney transplantation.

Materials and Methods

Study Design, Setting, and Participants

Between August 2012 and February 2015, we conducted a prospective, observational cohort study (1) of the formation and role of social networks in a newly opened, 12-station, hemodialysis clinic in Philadelphia. Patients were eligible to participate if they had ESKD, spoke English or Spanish, and were ≥18 years old. The Temple University Institutional Review Board approved the study protocol. Written, informed consent was obtained from all participants. The clinical and research activities reported here are consistent with the Principles of the Declaration of Istanbul, as outlined in the “Declaration of Istanbul on Organ Trafficking and Transplant Tourism,” and also adhere to the Declaration of Helsinki (22,23).

Variables

The exposure variable was a participant’s exposure to other participants in the hemodialysis clinic. We had two outcome variables: (1) whether a relationship (link) was formed, and (2) whether a participant requested a living donation after joining the study. Predictors of relationship formation and requesting a living donation included shift assignment, seating distance, demographic variables (age, sex, race, and ethnicity), and health behaviors (in-center, kidney-transplant discussions with other patients, requesting a living donation before link formation). Potential confounders of relationship formation were language used (Spanish speaking only), dialysis vintage, and time within the study. Eligibility for transplant was a potential confounder for requesting a living donation.

Data Collection

The Dialysis Patient Transplant Questionnaire (DPTQ) (24) was used for data collection (Supplemental Material). The DPTQ collects demographic information, such as race and ethnicity (for which participants can select more than one option), marital status, and transplant preferences and attitudes (24). The DPTQ asks participants whether they had discussed a kidney transplant with anyone and, if so, with whom. Another item asks participants whether they had requested a living donation (yes, no, or not sure). Questionnaires were administered within 3 months of admission to the clinic, and they were repeated every 3 months. Demographic and medical data, such as age, sex, medical cause of ESKD, and eligibility for transplantation, were extracted from the patients’ medical records.

Network Identification and Conversation Measurements

We directly observed the participants’ interactions to correct for any potential recall or social-desirability bias (25) in patient self-reported social interactions. The trained research staff observed and documented patient interactions for 2 hours per day, on a weekly or biweekly basis, from the centrally located nurses’ station within the treatment area, in the waiting area, and outside the clinic while participants waited for transportation. All patient interactions in the form of verbal communication, from a simple greeting to a long conversation, were logged as indicating a relationship (link). The content of these conversations, such as whether transplantation was discussed, was obtained from the DPTQ.

This 30-month study was divided into ten 3-month periods (Figure 1). Observations within each 3-month period were compiled into a “sociomatrix” (a matrix of relationships) that coincided with the questionnaire data that were collected every 3 months. This resulted in 11 sociomatrices, including the starting point at time zero, which had no links. We assumed that, once a link had been formed, knowledge was transferred, and that the link would not be dissolved unless the participant left the clinic (Figure 2).

fig1
Figure 1.:
Formation of the network, transplant discussions, and living-donor requests over time. Panel A shows what each symbol and line represents, along with some examples of the social-network theories: social contagion, homophily, and sociality. In panel B, each square represents a time point. Times 1 through 10 are displayed. The positions of the shapes are determined on the basis of the multidimensional scaling algorithm to improve network visualization, and do not represent the seating of the participants in the clinic. Each of the ten frames represents a 3-month period. Within each frame is the social network during that period. Within the frame, participants on the Monday, Wednesday, and Friday shift (M,W,F) are represented by a triangle, and those on the Tuesday, Thursday, and Saturday shift (T,T,S) are represented by a circle. The larger shapes represent the participants who discussed transplantation in the center. Black shapes represent the participants who requested a living donation before starting the study. White shapes represent the “susceptible” participants who have not requested a living donation. If participants request a living donation after joining the study, they are considered “infected,” and their shape becomes red. A black line between two shapes represents a relationship (link). Blue arrows denote the passage of time.
fig2
Figure 2.:
Growth curves of in-center transplant discussions and requesting a living donation. In the figure key, “in-center participants” represents the total number of participants who were admitted to the clinic at that time point. “Asked before joining” represents participants who requested a living donation before joining the study. “Asked after joining” represents participants who were “infected” and requested a living donation after joining the study. “In-center transplant discussions” represent the number of participants who requested a living donation. The figure shows the growth curves of the clinic population, the number of participants who had in-center transplant discussions, and the number of participants who requested a living donation over the study period.

Shift, Seating, and Distance

The clinic’s charge nurse was blind to the network portion of this study and assigned participants to their seats and shifts—which were either a morning or afternoon shift on Monday, Wednesday, and Friday (MWF), or Tuesday, Thursday, and Saturday (TTS)—on the basis of clinical judgment, including hepatitis-B serologic status and hemodialysis prescription. Participants could request a shift assignment, but they did not choose their seat. A participant’s shift and seating assignment, and subsequent changes, were recorded as part of clinic policy. Because participants changed seats and shifts over the course of the study, participants were classified as MWF or TTS on the basis of the shift on which they spent the most time. This dialysis clinic’s 12-seat treatment area measured 44 feet (13.4 m) by 35 feet (10.7 m; Figure 3). To account for patients not always being assigned to the same seat, we calculated the mean seating distance in terms of face-to-face geodesic (shortest) distance between their two seats. We took the sum of the distances between the participants and divided it by the number of times each pair of participants was on the same shift. The maximum distance between participants on the same shift was 5.83 seats and, for patients who never sat next to each other, the distance was defined as seven seats.

fig3
Figure 3.:
Hemodialysis-unit layout and probability of forming a link on the basis of the seating distance. This figure is a representation of the seating layout of the hemodialysis clinic. There are 12 dialysis stations. The probability of a participant in seat 1 forming a relationship and discussing transplantation with a participant in each seat is written in the representation of the seat. The nurses’ station (where the observations were made) is in the center of all of the chairs. The dimensions of the treatment area are 44 feet (13.4 m) by 35 feet (10.7 m); the waiting room, not shown, is directly outside the treatment area. REF, reference.

Statistical Analyses

Descriptive Statistics

We examined the questionnaire and network variables that were associated with discussing transplantation with other participants and with requesting a living donation (Supplemental Table 1). Chi-squared and Fisher exact tests were used for the categoric variables, and ANOVA was used for continuous variables.

Modeling Link Formation

We used a separable temporal exponential random graph model (STERGM) (26) to analyze the network as a multivariate observation in which the formation of links in the network depended on the participants’ attributes and the structural processes of the networks (Table 1). We used the geometrically weighted degree (GWDegree) to model centralization, and the geometrically weighted edgewise shared partner (GWESP) to model clustering in the STERGM (27). GWDegree and GWESP have decay parameters that range between zero and one, and they were optimized to the lowest Akaike information criterion for the most parsimonious model (27). We then added hemodialysis-clinic attributes to the model, such as shift and mean seating distance between patients. Next, we examined the demographic attributes (age, sex, race, ethnicity, religion, education, dialysis vintage, and eligibility for transplant) that predicted sociality and homophily. Models were then constructed to test health-behavior homophily using in-center transplantation discussions and living-donation requests. For each model, the observed network was compared with 100,000 randomly generated Markov random graphs (networks) using maximum pseudo-likelihood estimation and Monte Carlo maximum likelihood estimates (26).

Social-Contagion Model

Our social-contagion model was a susceptible-infected contagion model (see Figure 1) (3). Participants who had not yet requested a living donation were considered “susceptible.” If participants reported requesting a living donation either before or during the study, they were considered “infected.” Once “infected,” the participants remained “infected” and could not “recover” or become “reinfected.” Figure 1 describes our contagion model in which a participant becomes “infected” after being exposed to another “infected” participant (14).

Sensitivity Analyses

We examined the STERGM diagnostics and assessed their goodness of fit by comparing simulated data with the actual data (28). To examine the robustness of the model to survey error, we performed random data manipulations. We randomly changed participants who had transplant discussions to not having discussions and vice versa for five participants (10% of the data) and then for ten participants (20% of the data), and then we re-estimated the model using these datasets. Each of these data-manipulation procedures was performed five times.

Software

The software we used included SPSS version 25 (29), for descriptive analyses, and the R packages Statnet (28), STERGM (26), and NDTV (30), for STERGM model analysis and visualization. All tests were two tailed, with P<0.05 considered to be statistically significant.

Missing Data

All questionnaires were complete because the research staff assisted participants with any incomplete questions after self-administration.

Results

Study Participation

Of the 49 patients who were eligible to participate, 46 patients participated and completed the baseline survey, and 40 patients (89%) completed at least one follow-up survey. Surveys were administered in English (78%) and Spanish (22%) (Table 2). Participants reported their race and ethnicity as White Hispanic (37%), Black (33%), non-Hispanic White (22%), and multiethnic (9%). Despite being fluent in Spanish, 44% of Hispanic participants took their survey in English. Eight participants (17%) were on the kidney-transplant waiting list before joining the study. By the end of the study, 17 participants (37%) were on the waiting list. No participants received a living-donor kidney transplant over the course of the study. The majority (91%) of participants were treated by the same nephrologist.

Table 2. - Demographic, transplant, and dialysis variables
Variables Value
Total participants, N 46
Age and sex
 Age, mean (SD) 55 (14)
 Female sex, N (%) 19 (41)
Language, N (%)
 English 36 (78)
 Spanish 10 (22)
Race, N (%)
 White 10 (22)
 Black 15 (33)
 Hispanic/Latino 17 (37)
 Multiethnic 4 (9)
Education, N (%)
 Grade 9 or less 11 (24)
 High school 27 (59)
 College or higher 8 (17)
Eligibility, wanting, and asking for a transplant, N (%)
 Eligible for transplant 37 (80)
 Wants a living-donor kidney transplant 39 (85)
 Wait-listed before joining the study 8 (17)
 Wait-listed (n=10) or received a deceased-donor transplant (n=7) by the end of the study 17 (37)
Hemodialysis shift assignment, N (%)
 Monday, Wednesday, and Friday 32 (61)
 Tuesday, Thursday, and Saturday 14 (39)
Vintage, N (%)
 <1 Year 29 (63)
 ≥1 Year 17 (37)
Nephrology providers (N=3), N (%)
 Same provider 42 (91)

Demographic and Clinical Variables Associated with Living-Donation Requests

Table 3 shows the variables associated with participants making a living-donor request to a loved one, family member, or friend. Eleven participants requested a living donation before admission to the clinic and enrollment in the study. Among the 36 participants who did not request a living donation before joining the study and were considered “susceptible,” 13 requested a living donation after joining the study. All of the participants who made a request before or after joining the study wanted a living-donor kidney transplant. Among those who did not request a living donation, 68% wanted a living donation. A greater proportion of participants who requested a living donation were on the transplant waiting list or received a deceased-donor kidney transplant by the end of the study than participants not on the waiting list (55% who requested before, versus 54% who requested after, versus 18% who never requested were on the waiting list; P=0.04; Table 3). None of the other demographic or clinical variables were associated with requesting a living donation (Supplemental Table 1).

Table 3. - Transplant waiting-list status, network participation, transplant discussions, and living-donor requests
Variables Infected Susceptible P Value a
Requested a Donation before Enrollment, n=11 (24%) Requested a Donation after Enrollment, n=13 (28%) Did Not Request a Donation, n=22 (48%)
Waiting list and transplant status
 Wants a living-donor transplant 11 (100) 13 (100) 15 (68) 0.01
 On the waiting list at beginning of study 2 (18) 4 (31) 2 (9) 0.26
 On the waiting list or received a deceased-donor transplant by the end of study 6 (55) 7 (54) 4 (18) 0.04
 Received a deceased-donor transplant 2 (18) 4 (31) 1 (5) 0.11
HD social network and discussions
 Members of the HD social network 10 (91) 11 (85) 11 (50) 0.02
 Discussed transplant with other participants 8 (73) 5 (39) 2 (9) 0.001
 Discussed transplant with staff 8 (73) 8 (62) 5 (24) 0.01
Exposure and living-donor requests
 Exposed to at least one member who discussed transplant N/A 2 (15) 8 (36) 0.26 b
 Exposed to at least one member who requested a transplant N/A 5 (39) 9 (41) >0.99
Family and friends transplant discussions
 Discussed with significant other 6 (55) 7 (54) 8 (37) 0.48
 Discussed with children 7 (64) 9 (69) 9 (41) 0.21
 Discussed with other family 8 (73) 11 (85) 13 (59) 0.27
 Discussed with friends 6 (55) 8 (61) 5 (26) 0.05
HD, hemodialysis. N/A, not applicable because they were already “infected.”
aP value is from a chi square test unless otherwise indicated.
bFisher exact test.

Network Participation; Patient-to-Patient, In-Center, Transplant Discussions; and Living-Donation Requests

Of the 11 participants who requested a living donation before admission to the clinic, ten participated in the social network (91%), and eight had in-center transplant discussions (73%). Of the 13 participants who requested a living donation after joining the study, 11 participated in the network (85%), and five had in-center transplant discussions (39%). Among the 22 participants who never requested a living donation, 11 participated in the social network (50%), and two had in-center transplant discussions (9%). These associations between participating in the network and making a living-donor request, and between having in-center transplant discussions and requesting a living donation, were significant (P=0.02 and P=0.001, respectively; Table 2).

Other People with Which Participants Discussed Transplant

A greater percentage of participants who requested a living donation before joining the study (73%) and after joining the study (62%) discussed transplantation with the hemodialysis clinic staff compared with those who did not request a living donation (24%; P=0.01; Table 2). Those who requested a living donation before (61%) and after (26%) joining the study also discussed transplantation more with friends outside of the clinic than those who did not request a living donation (55%; P=0.05).

Shift Assignment and Seating

Most participants (70%) were assigned to an MWF shift (Table 2). Because seat assignment changed for many participants, we calculated the mean (SD) seating distance between the participants, which was 4.6 (1.9) seats, with 29% of the participants never sitting side by side. Eleven percent of the interactions occurred in the waiting room between patients who were assigned to the same day but never sat together on the same shift.

The Dynamic Network and Transplant Behaviors

To understand whether the relationships in the network influenced requesting a living donation, we must first understand the dynamics of network formation. The study started with six participants and, over the course of the 30-month study period, an additional 40 participants were admitted to the clinic. A total of 14 participants left the clinic; two participants transferred out, seven received a deceased-donor transplant, and five participants died. Figure 1 shows the formation of the social network over 3-month intervals (time 0 not shown). By month 30 (time 10), there were 20 participants who formed a large network, shown in the center of the time-10 panel, which consisted of 16 MWF participants (triangles) and four TTS participants (circles). The 13 participants who were isolates are shown on the periphery of the panel. The layout in panel 10 does not represent seating distances; rather, the participants with more links are in the center of the panel (30).

Figure 1 also illustrates who had in-center transplant discussions and/or requested a transplant over the course of the study. At month 3, when there were only five participants on the MWF morning shift, the two large triangles represent two participants who had in-center transplant discussions within the clinic and had already requested a living donation (black color), and the small black triangles represent two participants who had requested a donation before starting the study but did not have in-center transplant discussions. One of the participants, represented by a small black triangle, was linked to a participant who had requested a living donation after joining the study (red triangle). This is an example of a participant who requested a living-donor transplant influencing another participant to request a living donation. By month 30, among the 20 participants who formed the connected network, nine had in-center transplant discussions, and 11 had not discussed transplantation in-center. A total of 11 members of this connected network had requested a living donation; seven members requested before the study and four did so after enrolling in the study. Figure 2 shows the increase over time in the number of participants in the clinic and the number of participants who had in-center transplant discussions and requested a living donation. The number of participants who had in-center transplant discussions and those who requested a living donation both increased until time 8, when three participants who had requested a donation left the clinic (two received a deceased-donor transplant and one died).

Modeling Network Formation, Homophily, and Health-Behavior Homophily

We examined the effects of seating distance, demographics, and transplant-discussion behaviors on network formation. Table 4 shows the results of the STERGM and the log odds of participants forming a link. Model 1 (full model) includes the structure of the network, clinic shift and seating, sociality and homophily (by age, sex, race, ethnicity, and transplant behaviors of in-center discussions), and requesting a living donation as predictors of link formation. Model 2 focuses on sociality and homophily of in-center transplant discussions, and Model 3 reports the sociality and homophily of requesting a donation. The structural variables (Table 4, model 1) indicate that, over time, the participants formed a sparse network in which not everyone was linked, relationships were clustered among groups, and a few network members had many relationships (centralization; see time 10 in Figure 1). Participants on the TTS shift were more social (sociality), being 1.4 times more likely to form a link (odds ratio [OR], 1.4; 95% CI, 1.02 to 2.06; P=0.04) as compared with participants on the MWF shift. Overall, participants were 31 times more likely to form a link (OR, 31; 95% CI, 4.4 to 213; P<0.001) with a participant on the same day assignment than a participant on a different day. Interestingly, for every seat apart, the odds of participants forming a link decreased by an OR of 0.74 (95% CI, 0.61 to 0.90; P=0.002).

Table 4. - STERGM models for link prediction
Variables Model 1: Full Model Model 2: Discussed Transplant Model 3: Requested a Donation
Structural variables, β (SEM), P value
 Edges −8.57 (1.88), <0.001 −7.80 (1.79), <0.001 −7.99 (1.88), <0.001
 GWDegree (0.25) −0.75 (0.34), 0.03 −0.91 (0.33), 0.006 −0.90 (0.34), 0.007
 GWESP (0.55) 0.80 (0.18), <0.001 0.78 (0.18), <0.001 0.77 (0.18), <0.001
Dialysis clinic variables, β (SEM), P value
 TTS (sociality) 0.37 (0.18), 0.04 0.41 (0.18), 0.02 0.33 (0.18), 0.06
 TTS (homophily) 3.42 (0.99), <0.001 3.42 (0.99), <0.001 3.25 (1.02), <0.001
 Average seating distance (per seat; Euclidean) −0.30 (0.10), 0.002 −0.28 (0.10), 0.003 −0.30 (0.10), 0.003
Patient attributes, β (SEM), P value
 Age
  Yr (sociality) 0.01 (0.01), 0.19 0.01 (0.01), 0.24 0.01 (0.01), 0.17
  Yr (homophily) −0.01 (0.01), 0.42 −0.01 (0.01), 0.31 −0.01 (0.01), 0.49
 Sex
  Female sex (sociality) −0.34 (0.32), 0.29 −0.29 (0.31), 0.34 −0.29 (0.31), 0.34
  Sex (homophily) 0.14 (0.27), 0.72 0.17 (0.38), 0.66 0.17 (0.40), 0.68
 Race
  White (sociality) REF REF REF
  White (homophily) −0.37 (1.18), 0.75 −0.46 (1.18), 0.70 −0.40 (1.16), 0.73
  Black (sociality) 0.22 (0.49), 0.64 0.11 (0.48), 0.82 0.22 (0.48), 0.64
  Black (homophily) 0.37 (0.69), 0.60 0.34 (0.68), 0.62 0.32 (0.68), 0.64
  Hispanic (sociality) −0.93 (0.46), 0.04 −0.96 (0.46), 0.04 −0.93 (0.46), 0.04
  Hispanic (homophily) 2.31 (0.69), <0.001 2.31 (0.66), <0.001 2.19 (0.68), 0.001
  Multiethnic (sociality) 1.21 (0.53), 0.02 0.97 (0.52), 0.06 1.45 (0.52), 0.03
  Multiethnic (homophily) −0.21 (1.33), 0.87 0.01 (1.38), 0.99 −0.43 (1.34), 0.75
Transplant attributes, β (SEM), P value
 Discussed transplant with other patients (sociality) 0.23 (0.29), 0.41 0.41 (0.24), 0.08
 Discussed transplant with other patients (homophily) 0.66 (0.35), 0.06 0.64 (0.31), 0.04
 Requested a donation (sociality) 0.49 (0.26), 0.06 0.49 (0.24), 0.04
 Requested a donation (homophily) −0.23 (0.34), 0.50 −0.03 (0.34), 0.93
Model factors
 Null deviance 14,187 (df, 10,234) 14,187 (df, 10,234) 14,187 (df, 10,234)
 Residual deviance 474 (df, 10,213) 479 (df, 10,215) 478 (df, 10,215)
 AIC, BIC 517, 669 517, 654 516, 653
This is the STERGM of the variables that are associated with the formation of the hemodialysis-clinic social network. We report the β coefficients, which are the log odds of the formation of link; the SEM of the coefficients; and the P values. Edges represent the log odds of a participant forming a link with another participant, independent of the other variables, and represent the intercept of the model. GWDegree is the log odds of a participant with fewer links (i.e., lower degree) to form a new link compared with those participants with links (i.e., higher degree). GWESPs are the log odds that participants are more likely to link if they already share a link in common with another participant. These effects geometrically diminish as the number of shared links decrease. Sociality represents the propensity of a participant forming a link with any other participant. Homophily represents the propensity of a participant to form a link with another participant with the same attribute. STERGM, separable temporal exponential random graph model; GWDegree, geometrically weighted degree; GWESP, geometrically weighted edgewise shared partner; TTS, Tuesday, Thursday, and Saturday dialysis shift; REF, reference; df, degrees of freedom; AIC, Akaike information criterion; BIC, Bayesian information criterion.

We then examined homophily on the basis of age, sex, race, and ethnicity, and we found that Hispanic participants had a lower odds of forming a link (OR, 0.39; 95% CI, 0.16 to 0.97; P=0.04; Table 4, model 1) with another participant than did non-Hispanic White participants. Hispanic participants, however, were 10.1 times (OR, 10.1; 95% CI, 2.6 to 40.0; P<0.001) more likely to form a link with another Hispanic participant than with a non-Hispanic participant. Whether or not participants completed their survey in Spanish did not significantly predict link formation (Supplemental Table 2). Other participant variables that did not predict link formation via sociality or homophily were age, sex (Table 4), religion, dialysis vintage, education, and transplant eligibility (Supplemental Table 2).

We modeled health-behavior homophily by examining whether participants formed links on the basis of similar transplant-related behaviors. Participants who discussed transplantation in the center were 1.9 times more likely to form a link with another participant who discussed transplantation in the center (OR, 1.9; 95% CI, 1.03 to 3.5; P=0.04; Table 4, model 2) than a participant who did not have in-center transplant discussions. Conversely, participants who did not have in-center transplant discussions were 1.9 times more likely to form relationships with each other than with a participant who had in-center transplant discussions. Participants who requested a living donation were more social (sociality) because they were 1.6 times more likely to form a link than participants who did not request a donation (OR, 1.6; 95% CI, 1.02 to 2.6; P=0.04; Table 4, model 3).

Figure 3 demonstrates the probability that a 55-year-old participant (the mean age in the study) on the TTS shift who has in-center transplant discussions forms a link with another participant who has in-center transplant discussions and sits adjacent to them is 14%. This probability decreases to 11% if they are an additional seat apart, and 4% if they are in opposite corners of the clinic. This probability did not account for other factors, such as the participants’ race and the two structural parameters, GWESP and GWDegree. For example, if both participants were Hispanic and they had a relationship with another participant in common, the probability of forming a link would be 22%.

Describing and Modeling Living-Donation Requests as Contagious Behavior

Using a contagion model, we examined whether participants who had requested a living donation and discussed transplantation in the center influenced other participants to request a living donation (Figure 1). Over the course of the study, 13 (36%) of the 35 susceptible participants requested living donation (Figures 1 and 2, Table 3). Although most participants (85%) formed links and some (39%) discussed transplantation (Table 3), most participants (61%) requested a donation before forming a relationship. Five participants (39%) requested a transplant after being exposed to a network member who had requested a transplant, whereas only two of the participants (15%) requested a donation after discussing transplantation in the center. A similar proportion of participants who never made a living-donation request were exposed to a network member who requested a transplant and/or discussed transplantation.

Goodness-of-Fit and Sensitivity Analyses

We inspected the STERGM for goodness of fit and found no significant deviation between the simulated models and the observed data. Sensitivity analyses show the robustness of our transplant-discussion homophily model (Supplemental Table 3), because the positive trend to transplant-discussion homophily persisted even if 10% and 20% of the participants incorrectly answered the questionnaire about discussing transplantation.

Discussion

Patients on hemodialysis form relationships with other patients who sit close to them, especially if they are of similar ethnicity and have similar transplant behaviors. The increase of in-center transplant discussions and living-donation requests by the members of the hemodialysis-clinic social network was not because of social contagion. Instead, participants who requested a living donation were more social, more likely to form a relationship, and preferentially formed relationships with other participants who discussed transplantation as a function of health-behavior homophily.

Requesting a living donation did not spread like a social contagion; however, health-behavior homophily is a significant finding because it can be used to create future network interventions within the hemodialysis clinic to improve access to living donor transplantation. Health-behavior homophily improves the efficacy of network interventions (31) because people are more willing to adopt a behavior if they know that people similar to them have also adopted that behavior (32). These social-network interventions should cause patients to spread the intervention from one to another via their relationships and the intervention will go “viral” via social contagion, and its adoption will grow exponentially (32,33). The hemodialysis clinic staff can facilitate this intervention because many of the participants who request a living donation discuss transplantation with the staff.

One example of a clinic-based network intervention would be to train participants who discuss transplantation in the center to help each other navigate the transplant process, discuss the best ways to request a living donation, and identify other potential living donors if they have already made a request (34). Furthermore, if those who have not discussed transplantation with other patients are seated among a majority of other participants who discuss positive transplant information, this could induce homophily-based communication (5,6). Seating and shift assignment seem to be very influential in forming relationships because only approximately 10% of the interactions occurred in the waiting room. Future research is necessary to identify how to tailor this transplant-network intervention to the ethnic homophily exhibited by the Hispanic patients.

Ethnic homophily is not uncommon (8,17,18) and can be attributed to similar beliefs, behaviors, experiences, appearances, and language. Speaking only Spanish may have contributed to ethnic homophily; however, it was not a significant variable in our homophily model, probably because English-speaking Hispanic participants were also fluent in Spanish. Thus, Hispanic, Spanish-speaking participants would preferentially form relationships with other Hispanic participants who only spoke Spanish and with those who were bilingual. In our previous study of network participation in this cohort (1), we found no ethnic differences in network participation. This is because, although Hispanic participants were less likely to form links, they were likely to form a link with another Hispanic participant if present on the shift, and Hispanic participants were present on all shifts.

Unlike previous research (1), which found sex differences in network participation, we did not find sex differences in the formation of links. However, the use of a dynamic longitudinal model, which is superior at link prediction to our previous cross-sectional model (1), is underpowered at predicting isolates (26). More research is needed to understand the factors associated with not participating in the hemodialysis social network, especially given that fewer isolates requested a living donation as compared with network participants, and social isolation is a maladaptive coping mechanism (35).

This study is not without its limitations. The major limitation is that this is a single-center study with a small sample. The sample size was small because this longitudinal, observational cohort study was set in a newly opened, 12-station clinic that started with only a few patients. This allowed us to observe the formation of new networks and the appearance and spread of behaviors (17). Also, because of the size of the clinic, most patients were under the care of a single nephrologist, which avoided differences in nephrologists’ practices toward transplantation (36).

Another limitation is that we may not have observed all participant interactions in the clinic, as evidenced by one participant who reported discussing transplantation in the center but was not observed interacting with other participants. Further, documenting simple communication behavior may have identified relationships that did not result in information transfer, especially because the length of conversations was not assessed here. To understand the content of conversation, we relied on self-reported data on the basis of participants’ recall of requesting a transplant or discussing transplantation with other participants. We assume that participants discussed positive transplant information; however, it is possible that participants also shared negative experiences, because none of the participants received a living-donor transplant. Additionally, although our questionnaire was designed to avoid desirability bias, bias is always possible when people are repeatedly asked the same question (25). To address these limitations, a multicenter study is currently underway, with special attention to what participants discuss and with whom.

In conclusion, our study demonstrates the influence of seating assignment and transplant-related health behaviors on social-network formation among patients on in-center dialysis, and demonstrates there is potential for social-network interventions to improve health behaviors as a function of sociality and health-behavior homophily.

Disclosures

C.A. Gadegbeku reports receiving research funding from Akebia and Vertex; being a scientific advisor or member of the American Society of Nephrology Council; and having consultancy agreements with Fresenius Kidney Care as medical director. H.M. Gardiner reports receiving honoraria from National Institutes of Health (NIH) and Public Health Management Corporation; and being a governing councilor for the Public Health Education and Health Promotion Section of the American Public Health Association. P.P. Reese reports having ownership interest in various equities, but none which are health related or related to his research; being a coprincipal investigator for investigator-initiated demonstration trials, funded by AbbVie and Merck, which involve transplantation of hepatitis C virus (HCV)–infected organs into recipients who are negative for HCV, followed by HCV treatment; being an associate editor for American Journal of Kidney Diseases; receiving honoraria for talks at academic centers or academic consortia, including Brigham and Women’s Hospital, Brown University, Cedars-Sinai, Health and Human Services, Massachusetts General Hospital, National Kidney Foundation, Northwestern University, and University of Pittsburgh; being a coprincipal investigator for studies of medication adherence (to any statin) funded by CVS Caremark and the NIH; providing volunteer ethics consultation to eGenesis, related to patient selection and education; being a volunteer for United Network for Organ Sharing, including in the role of past chair and current member of the ethics committee; having consultancy agreements with VAL Health – identification of patients with CKD and behavior change strategies; and providing legal consultation for an individual and not a company or institution. All remaining authors have nothing to disclose.

Funding

A. Gillespie is currently funded by the National Institute of Diabetes and Digestive and Kidney Diseases grant K23 DK111943. This research as also supported by a Norman S. Coplon Satellite Healthcare Foundation unrestricted grant.

Supplemental Material

This article contains supplemental material online at http://kidney360.asnjournals.org/lookup/suppl/doi:10.34067/KID.0006682020/-/DCSupplemental.

Supplemental Table 1. Demographic, transplant and dialysis variables and living donor requests.

Supplemental Table 2. STERGM model for link prediction.

Supplemental Table 3. The effects of randomly reversing the result of transplant discussions on the transplant parameter.

Acknowledgments

A. Gillespie would like to thank the Norman S. Coplon Satellite Healthcare Foundation for supporting this research with an unrestricted grant.

Author Contributions

E.L. Fink, C.A. Gadegbeku, H.M. Gardiner, A. Gillespie, and P.P. Reese wrote the original draft; E.L. Fink, C.A. Gadegbeku, H.M. Gardiner, Z. Obradovic, and P.P. Reese were responsible for validation; E.L. Fink, C.A. Gadegbeku, A. Gillespie, and P.P. Reese reviewed and edited the manuscript; E.L. Fink, C.A. Gadegbeku, Z. Obradovic, and P.P. Reese were responsible for resources; E.L. Fink, H.M. Gardiner, A. Gillespie, and Z. Obradovic were responsible for methodology; E.L. Fink, H.M. Gardiner, and Z. Obradovic provided supervision; E.L. Fink and A. Gillespie were responsible for visualization; C.A. Gadegbeku, H.M. Gardiner, A. Gillespie, Z. Obradovic, and P.P. Reese were responsible for project administration; H.M. Gardiner and A. Gillespie were responsible for data curation; H.M. Gardiner and Z. Obradovic were responsible for software; A. Gillespie and Z. Obradovic conceptualized the study and were responsible for funding acquisition and investigation; and all authors were responsible for formal analysis and approved the final version of the manuscript.

See related editorial, “Social Networks in the Dialysis Unit: Can They Influence Patient Behavior?,” on pages .

References

1. Gillespie A, Fink EL, Traino HM, Uversky A, Bass SB, Greener J, Hunt J, Browne T, Hammer H, Reese PP, Obradovic Z: Hemodialysis clinic social networks, sex differences, and renal transplantation. Am J Transplant 17: 2400–2409, 2017 https://doi.org/10.1111/ajt.14273
2. Arthur T: The role of social networks: A novel hypothesis to explain the phenomenon of racial disparity in kidney transplantation. Am J Kidney Dis 40: 678–681, 2002 https://doi.org/10.1053/ajkd.2002.35672
3. Browne T: The relationship between social networks and pathways to kidney transplant parity: Evidence from black Americans in Chicago. Soc Sci Med 73: 663–667, 2011 https://doi.org/10.1016/j.socscimed.2011.06.031
4. Ladin K, Hanto DW: Understanding disparities in transplantation: Do social networks provide the missing clue? Am J Transplant 10: 472–476, 2010 https://doi.org/10.1111/j.1600-6143.2009.02963.x
5. Festinger L: Informal social communication. Psychol Rev 57: 271–282, 1950 https://doi.org/10.1037/h0056932
6. Caldeira GA, Patterson SC: Political friendship in the legislature. J Polit 4: 953–975, 1987 https://doi.org/10.2307/2130779
7. Watts DJ, Strogatz SH: Collective dynamics of ‘small-world’ networks. Nature 393: 440–442, 1998 https://doi.org/10.1038/30918
8. Daw J, Margolis R, Verdery AM: Siblings, friends, course-mates, club-mates: How adolescent health behavior homophily varies by race, class, gender, and health status. Soc Sci Med 125: 32–39, 2015 https://doi.org/10.1016/j.socscimed.2014.02.047
9. Valente TW, Zogg JB, Christensen S, Richardson J, Kovacs A, Operskalski E: Using social networks to recruit an HIV vaccine preparedness cohort. J Acquir Immune Defic Syndr 52: 514–523, 2009 https://doi.org/10.1097/QAI.0b013e3181acff91
10. Christakis NA, Fowler JH: The spread of obesity in a large social network over 32 years. N Engl J Med 357: 370–379, 2007 https://doi.org/10.1056/NEJMsa066082
11. Rosenquist JN, Murabito J, Fowler JH, Christakis NA: The spread of alcohol consumption behavior in a large social network. Ann Intern Med 152: 426–433, W141, 2010 https://doi.org/10.7326/0003-4819-152-7-201004060-00007
12. Christakis NA, Fowler JH: The collective dynamics of smoking in a large social network. N Engl J Med 358: 2249–2258, 2008 https://doi.org/10.1056/NEJMsa0706154
13. Moreno JL: Who Shall Survive? A New Approach to the Problem of Human Interrelations, New York, Nervous and Mental Disease Publishing Co., 1934 https://doi.org/10.1037/10648-000
14. Centola D, Macy M: Complex contagions and the weakness of long ties. Am J Sociol 113: 702–734, 2007 https://doi.org/10.1086/521848
15. Kim DA, Hwong AR, Stafford D, Hughes DA, O’Malley AJ, Fowler JH, Christakis NA: Social network targeting to maximise population behaviour change: A cluster randomised controlled trial. Lancet 386: 145–153, 2015 https://doi.org/10.1016/S0140-6736(15)60095-2
16. Lindzey G, Aronsson E, editors: Handbook of Social Psychology: Group Psychology and the Phenomena of Interaction, Hillsdale, NJ, Lawrence Erlbaum Associates, 1985
17. Shalizi CR, Thomas AC: Homophily and contagion are generically confounded in observational social network studies. Sociol Methods Res 40: 211–239, 2011 https://doi.org/10.1177/0049124111404820
18. McPherson M, Smith-Lovin L, Cook JM: Birds of a feather: Homophily in social networks. Annu Rev Sociol 27: 415–444, 2001 https://doi.org/10.1146/annurev.soc.27.1.415
19. Sullivan C, Leon JB, Sayre SS, Marbury M, Ivers M, Pencak JA, Bodziak KA, Hricik DE, Morrison EJ, Albert JM, Navaneethan SD, Reyes CM, Sehgal AR: Impact of navigators on completion of steps in the kidney transplant process: A randomized, controlled trial. Clin J Am Soc Nephrol 7: 1639–1645, 2012 https://doi.org/10.2215/CJN.11731111
20. Waterman AD, Rodrigue JR, Purnell TS, Ladin K, Boulware LE: Addressing racial and ethnic disparities in live donor kidney transplantation: Priorities for research and intervention. Semin Nephrol 30: 90–98, 2010 https://doi.org/10.1016/j.semnephrol.2009.10.010
21. Gillespie A, Hammer H, Bass SB, Ouzienko V, Obradovic Z, Urbanski M, Browne T, Silva P: Attitudes towards living donor kidney transplantation among urban African American hemodialysis patients: A qualitative and quantitative analysis. J Health Care Poor Underserved 26: 852–872, 2015 https://doi.org/10.1353/hpu.2015.0087
22. The declaration of Istanbul on organ trafficking and transplant tourism. Istanbul summit April 30–May 2, 2008. Nephrol Dial Transplant 23: 3375–3380, 2008
23. World Medical Association: World Medical Association Declaration of Helsinki. Ethical principles for medical research involving human subjects. Bull World Health Organ 79: 373–374, 2001
24. Gillespie A, Hammer H, Lee J, Nnewihe C, Gordon J, Silva P: Lack of listing status awareness: Results of a single-center survey of hemodialysis patients. Am J Transplant 11: 1522–1526, 2011 https://doi.org/10.1111/j.1600-6143.2011.03524.x
25. Bernard H, Killworth P, Kronenfeld D, Sailer L: The problem of informant accuracy: The validity of retrospective data. Annu Rev Anthropol 13: 495–517, 1984 https://doi.org/10.1146/annurev.an.13.100184.002431
26. Krivitsky PN, Handcock MS: tergm: Fit, simulate and diagnose models for network evolution based on exponential-family random graph models, The Statnet Project. R package version 3.5.2, 2018. Available at: https://CRAN.R-project.org/package=tergm. Accessed September 20, 2020
27. Hunter DR: Curved exponential family models for social networks. Soc Networks 29: 216–230, 2007 https://doi.org/10.1016/j.socnet.2006.08.005
28. Hunter DR, Handcock MS, Butts CT, Goodreau SM, Morris M: ergm: A package to fit, simulate and diagnose exponential-family models for networks. J Stat Softw 24: nihpa54860, 2008 https://doi.org/10.18637/jss.v024.i03
29. SPSS Inc: SPSS: Version 25, Chicago, IL, SPSS Inc., 2017
30. Bender-deMoll S, Morris M: ndtv: Network Dynamic Temporal Visualizations. R package version 0.12.3, 2018. Available at: https://CRAN.R-project.org/package=ndtv
31. Centola D: An experimental study of homophily in the adoption of health behavior. Science 334: 1269–1272, 2011 https://doi.org/10.1126/science.1207055
32. Valente TW: Network interventions. Science 337: 49–53, 2012 https://doi.org/10.1126/science.1217330
33. Rogers EM: Diffusion of Innovations, 4th Ed., New York, Free Press, 1995
34. Traino HM, West SM, Nonterah CW, Russell J, Yuen E: Communicating about choices in transplantation (COACH). Prog Transplant 27: 31–38, 2017 https://doi.org/10.1177/1526924816679844
35. Lunsford SL, Simpson KS, Chavin KD, Hildebrand LG, Miles LG, Shilling LM, Smalls GR, Baliga PK: Racial differences in coping with the need for kidney transplantation and willingness to ask for live organ donation. Am J Kidney Dis 47: 324–331, 2006 https://doi.org/10.1053/j.ajkd.2005.10.018
36. Ayanian JZ, Cleary PD, Keogh JH, Noonan SJ, David-Kasdan JA, Epstein AM: Physicians’ beliefs about racial differences in referral for renal transplantation. Am J Kidney Dis 43: 350–357, 2004 https://doi.org/10.1053/j.ajkd.2003.10.022
Keywords:

clinical nephrology; clinical epidemiology; end stage kidney disease; hemodialysis; homophily; kidney transplantation; living donor; social contagion; social networks; survey research; transplantation

Copyright © 2021 by the American Society of Nephrology