Introduction

Stroke recurrence and associated adverse vascular outcomes are a serious but potentially modifiable health burden among stroke survivors. For those who have experienced a first stroke, risk of recurrence is alarmingly high-up to 30% within five years [1] yet evidence indicates this risk is largely modifiable [2]. Approximately 25% of the 800,000 stroke events that occur annually are recurrent events. To reduce stroke recurrence and other adverse vascular outcomes, epidemiological and clinical trial data support the effectiveness of improving health behaviors such as increasing physical activity, adopting/maintaining a healthy diet, and smoking cessation [3–5] as well as reducing key vascular risk factors including hypertension [6–8]. Despite the benefit of blood pressure control to reduce stroke risk, hypertension remains highly prevalent among stroke survivors with estimates around 70% among recent ischemic stroke survivors [2]. Strategies to reduce hypertension among stroke survivors have been suboptimal and have not focused on factors that support successful risk reduction, including self-efficacy and emotional support, accessible information and resources, as well as proper reinforcement strategies for optimal long-term risk factor management. Part of the challenge with secondary stroke is that patients and caregivers have to work concurrently on recovery, on coping with the psychosocial impact of a devastating event, and on secondary stroke prevention. A multitude of tasks are involved in managing those aspects of post-stroke care, including daily physical activity regimens, medication adherence, preparation of healthy meals, coping with depression, understanding and acting on new health information, and navigating changes in family dynamics.

Leveraging social networks may offer an effective and sustainable mechanism to enhance risk reduction. With more than three decades of evidence, it is well understood that social networks influence health behaviors [9,10] and numerous studies have demonstrated that family structure can profoundly influence health outcomes [11–13]. A 2014 American Heart Association/ American Stroke Association (AHA/ASA) review of 32 strokerelated studies found that family caregiver and dyad interventions improve physical functioning and health-related quality of life among stroke survivors and also reduce depression and improve preparedness among caregivers [14]. In the Northern Manhattan Stroke Study (NOMAS), we demonstrated that lack of social support (having fewer than three network members who visit the home) is an independent risk factor for stroke recurrence and cardiovascular events post stroke [15]. Other studies suggest that different types of social networks (friends vs. family) may differentially impact health behaviors (physical activity vs. stroke preparedness); living arrangements may be associated with vascular risk factors including hypertension; and these effects vary by gender and race-ethnicity [15–19]. There is growing evidence that integrating a patient’s social network into behavioral interventions improves the effectiveness and sustainability of vascular risk reduction strategies [14,15,20]. Our previous studies of behavioral interventions have demonstrated that the stroke event constitutes a “teachable moment,” whereby family involvement and concern about risk reduction play a critical role in motivating behavioral change [21,22].

Much of the social support work to date in stroke recovery, as evidenced by the AHA/ASA recommendations, has focused on the relationship between patient and caregiver. However, Berkman’s theory of social networks extends the network beyond the caregiver and suggests differential health-related roles for members of a larger network [9,10,23]. Members of networks might engage in companionship [24] social influence, social support, and social undermining, all of which can shape health outcomes (Figure 1). The mechanisms by which social networks influence stroke outcomes may include reinforcing behaviors, bolstering selfefficacy, or reducing stress through physiological mechanisms [25]. All of these mechanisms potentially impact the utilization of prevention behaviors. However, despite literature strongly suggesting the importance of networks as part of the prevention paradigm, little is known about the relationship between the stroke survivor’s social network and secondary prevention efforts. Using secondary data from the Discharge Educational Strategies for Reduction of Vascular Events (DESERVE) intervention, we assessed the association between social network structure and function and Systolic Blood Pressure (SBP) reduction at one-year post stroke.

 

Figure 1: Structural and functional characteristics of social networks that influence health outcomes (Adapted 25).

Adapted from Israel, B. Social Networks and Social Support. Health Behavior. 2010. Relevant examples of companionship include engaging in group exercise routines or attending doctor’s appointments. Relevant examples of social influence include sharing risk knowledge or spreading healthy behaviors. Relevant examples of social support include medication reminders or emotional support. Relevant examples of social undermining include discouraging smoking habits.

Methods

Data for this analysis is derived from DESERVE, a two-arm randomized controlled trial conducted between August 2012 and May 2016 in New York City. The methods of DESERVE have been published in detail elsewhere [26,27] Briefly, 552 mild/ moderate stroke/Transient Ischemic Attack (TIA) survivors were prospectively enrolled and randomized from four major medical centers in New York City (i.e., Langone Medical Center, Icahn School of Medicine at Mount Sinai, Columbia University Medical Center, and Bellevue Hospital Center). DESERVE assessed a skills-based, culturally tailored intervention designed to improve patient-physician communication, medication adherence, and accurate stroke risk perception through an interactive educational session, a patient-paced workbook, patient narrative videos, and intervention follow-up at 72 hours, one month, and three months. Usual care consisted of language appropriate information on risk reduction post stroke. Individuals missing blood pressure measurements at 12 months were excluded from this analysis. The study was approved by the IRBs of the participating medical centers and was registered at ClinicalTrials.gov (NCT01836354). 

As part of the DESERVE baseline enrollment, participants were asked to identify up to five persons (“alters”) with whom they discuss important matters, including health. Based on data collected about the nature of these relationships, we classified network characteristics into two groups: structural and functional (Table 1). We defined structural characteristics as being related to the composition of the network (who are the people in the network and what are their relationships to each other). Structural characteristics were defined by the alters who make up the network and included number of alters, alter(s) living in the same household, a familyonly vs. family and friend network, and education. Education level of each alter was defined as high (completed high school) or low (did not complete high school). Functional characteristics described the nature and type of interactions that define the relationship and included actions related to companionship, social influence, social support, and social undermining. Our functional network variables included the frequency of contact with each alter, likelihood of discussing health concerns, and reported relationship closeness. 

We chose variables and cut points previously identified in the literature to describe the most beneficial scenarios within social networks (Table 1). We operationalized network characteristics as a combination of the number of alters named and the number of alters who met the most beneficial scenario for each characteristic. We grouped patients into those reporting 0, 1-2, or 3-5 alters. To analyze potential associations with alters’ educational level (a structural characteristic) and functional characteristics, we further subdivided patients reporting only 1-2 alters into one of two categories for each characteristic: those patients with at least 1 alter meeting the most beneficial scenario and those patients with no alters meeting the most beneficial scenario. Similarly, we subdivided patients who reported 3-5 alters into those who reported at least 3 respondents meeting the most beneficial scenario and those who did not. Since less than 3% of the sample reported having 1-2 alters with none meeting the most beneficial scenarios, we combined the 1-2 alters group as one category in our models.

Table 1: Network variables collected on alters

 

Our primary outcome was SBP change from baseline (predischarge) to 12-months post-discharge. At each data collection time point, Blood Pressure (BP) was measured up to three times by trained coordinators or research assistants using a BP Tru automatic sphygmomanometer, as described previously [26]. Baseline BP measurements were taken at least 48-hours post stroke. At 12-month follow-up, 27.3% of BP measurements were collected in person by the study team, 66.8% came from either clinic visits where our study staff met the patient or abstracted chart records, and 5.9% were self-reported by the participant. The distribution of source of follow-up BP measurement did not differ significantly between groups. The average of the second and third BP readings was utilized as the measure for BP or the average of the first two readings if the third was missing or just the first reading if no other measurements were collected.

Covariates included participant age, race-ethnicity, gender, NIH Stroke Scale (NIHSS) at admission, education (completed high school or more education versus less than high school completion), and prior stroke history. Marital status served as a historically relevant indicator of social support. DESERVE trial arm was handled as a covariate. We collected the Center for Epidemiologic Studies Depression (CES-D) Scale to assess full depression (CES-D score ≥ 16) at 12 months following stroke, however, we were not able to use it in the dataset due to missingness at 12 months. Antihypertensive medication adherence was assessed using the self-reported 8-item Morisky Medication Adherence Scale (MMAS-8), which has been validated against blood pressure levels and shown to be valid in a sample of primarily low-income, minority patients with hypertension [28]. The scale consisted of 8 dichotomous (yes/no, with no coded as 1, indicating higher adherence) or Likert items (normalized to range between 0 and 1, with higher values indicating higher adherence). Examples of items were, “Do you ever forget to take your high blood pressure medication?” and “When you feel like your blood pressure is under control, do you ever stop taking your medicine?” 13 Recoded items were summed and ranged from 0 to 8. The summed score was dichotomized into high adherence (score of 8) and low or medium adherence (scores of 0 to 7).

Statistical analysis

Frequencies, percentages, means, and standard deviations were used to describe the distribution of our variables of interest, as appropriate. Mean SBP change across categories of social network structure and function characteristics were assessed using ANOVA. Differences in measures of social networks across additional covariates were assessed using chi-square tests. Linear regression models were used to quantify the association of SBP change with structural and functional social network variables first unadjusted, then adjusting for trial arm, age, gender, race-ethnicity, NIHSS, education, marital status, and prior stroke, and finally additionally adjusting for medication adherence. Structural and functional social network variables were stratified to explore heterogeneity by race-ethnicity. Analyses were conducted using SAS v9.4. Hypothesis tests were conducted at the two-sided 0.05 level.

Results

Ego characteristics

There were 407 participants enrolled in DESERVE with baseline and 12-month blood pressure measurements. Participants had a mean age of 64.2 (± 14.6) years and were 48.4% male. The racial and ethnic distribution was nearly equivalent with 33.4% Hispanic, 31.7% non-Hispanic black, 28.5% non-Hispanic white, and 6.4% other race-ethnicity. Approximately one-quarter had less than a high school education, 21.0% had completed high school or equivalent, and 53.1% reported having at least some college education. Slightly less than half (47.6%) were married or living with a partner; 35.3% were divorced, separated, or widowed; and 17.1% were single/never married. Marital status was more common in those under 66 years of age, and men were more likely to be married. Approximately one-third of the sample had a prior stroke or TIA. Only 3% had high medication adherence, and among those not missing elements in the depression score, almost one-third had depression. The overall mean SBP reduction at 12-month follow-up was 6.7 (± 23.8) mmHg (Table 2).

Table 2: Demographic and clinical characteristics of the DESERVE cohort of stroke/transient ischemic attack survivors (N=407), New York City, 2012-2016.

Network characteristics

In terms of structural network characteristics (Table 3), 8% of participants identified no alters, 35% one alter, 24% two alters, 13% three alters, 9% four alters, and 11% five alters.  Of those with at least one alter, 62% reported having an alter in the same household and 72% reported having a family-only network.

Of the 241 participants reporting 1-2 alters, 96% reported talking to at least 1 alter several days per week, 96% reported they were extremely or very close to at least 1 alter, 95% reported they were very likely to discuss health matters with at least 1 alter, and 68% reported having at least 1 alter with more than a high school education. Of the 133 participants reporting 3-5 alters, 71% reported talking to at least 3 alters several days per week, 80% reported they were extremely or very close to at least 3 alters, 76% reported they were very likely to discuss health matters with at least 3 alters, and 50% reported having at least 3 respondents with more than a high school education.

The mean SBP reduction between baseline and the one-year follow-up was greater across family and friend networks [mean reduction of SBP 11.32 mmHg] vs. family-only [mean reduction of SBP 5.73mmHg] (p=0.05). Compared to no weekly contact with any alter, speaking with 1-2 or 3-5 alters at least several times a week resulted in significant SBP reduction (p=0.03). We report a dose-response relationship between number of alters having more than a high school education and greater mean SBP reduction (p=0.05). Indeed, having 3-5 alters with at least 3 alters having greater than high school education was associated with a mean SBP reduction of 13.63 mmHg. Mean SBP reduction between baseline and the one-year follow-up did not differ significantly by number of alters identified (p=0.24), whether or not the alter lived in the same household (p=0.89), relationship closeness (p=0.32), or likelihood of discussing health concerns (p=0.28) (Table 3).

Results of fully adjusted network model 

In the fully adjusted model (Model 2, Table 3), having a family and friend network compared to a family-only network (p=0.01) was significantly associated with a mean SBP reduction. Compared to reporting no alters, having 3-5 alters who speak with the ego at least three days per week resulted in a significant reduction of SBP (mean SBP reduction of 13.5 mmHg, p=0.03). Participants reporting 3-5 alters, of which 3 or more had completed high school, were associated with a significantly larger reduction in SBP compared to participants reporting 1-2 alters, of which none had completed high school (mean difference 12.4mmHg, p=0.03). Additional tests showed that the trial arm assignment did not modify the effect of any network characteristic (data not shown). These results were all consistent with the additional adjustment for medication adherence which was done separately due to additional missingness of the instrument.

Table 3: Distribution of social network structural and functional characteristics in the DESERVE cohort of stroke/transient ischemic attack survivors (N=407), New York City, 2012-2016.

Black participants were more likely to have 3-5 alters identified compared to Hispanic or White participants (p=0.006) (Table 4). Whites were least likely to report no alters compared to other races; Hispanics had the highest proportion of family-only networks (p=0.017). Blacks reported a higher frequency of talking to three or more alters regularly. Education of alters was significantly different across ethnic groups (p<0.001). Whites had the highest proportion of alters with more than a high school education, and Hispanics had the lowest. Higher education of alters was associated with a significant decrease in blood pressure.

 

Table 4: Social network structural and functional characteristics stratified by race/ethnicity in the DESERVE cohort of stroke/transient ischemic attack survivors (N=407), New York City, 2012-2016.