Study Design and Participants
From October 2014 through June 2016, we conducted a cross-sectional study of the prevalence of cardiometabolic disease risk in a sample of 1156 community-dwelling Somalis, age ≥ 18 years, living in the Minneapolis-St. Paul metropolitan area of Minnesota, using respondent-driven sampling (RDS) [37] to recruit participants. RDS is a chain-referral recruitment process appropriate for use in “hidden” populations or those for which no adequate sampling frame exists. The recruited social contacts of the first wave of respondents become respondents and recruiters of the second wave, and so on. Sampling continues for between 5 and 9 waves, leading to an approximation of a random sample [30, 31]. Trained ethnicity- and gender-concordant community health workers (CHWs) obtained informed consent translated into Somali language for non-English speaking participants, conducted face-to-face interviews, obtained anthropometric and BP measurements. Our study team developed the interview questions in collaboration with an advisory panel of Somali individuals from the local community. The HealthPartners Institutional Review Board approved this study.
Measurements
We used self-report interview questions to obtain most measures. Anthropometric measures included measured height, weight, waist-circumference and BP. Scales were calibrated before each measurement to reduce reader error. Height was measured by a stadiometer to the nearest 0.5 cm without footwear. Using finger-stick, blood-spot samples, we obtained (non-fasting) measures of blood-sugar and cholesterol levels.
Primary outcome - hypertension
For BP measurements, participants rested quietly, seated, for 5 minutes, and once the participant’s maximum inflation level was determined, the CHWs obtained 3 consecutive BP readings by using sphygmomanometer with an appropriately sized arm cuff. The average of 3 systolic blood pressure (SBP) and diastolic blood pressure (DBP) outputs was recorded.
We defined HTN as SBP ≥ 140 mm Hg or DBP ≥ 90 mm Hg, the subject saying “yes” to taking antihypertensive medication, or the subject reporting having been told by a health care provider that they had HTN.
Primary predictors—acculturation
Our measures of acculturation included:
- Traditional measures (Duration in USA and Enclaved geographically in one of 8 contiguous zip codes (yes, no).
- Proxy measures. Four latent acculturation subgroups: High Trust/Low English, Low Trust/Low English, High Trust/High English, and Low Trust/High English.
Identification of latent acculturation subgroups
In the context of immigrant communities, “acculturation” references the process of change and adaptation by individuals and groups who have migrated from one location and culture to another. [14, 15] As a process not directly observable, it is useful to consider acculturation as a latent variable, the values of which may be inferred from multiple observed measures. Latent class analysis (LCA) is one appropriate method for identifying latent subgroups. This method identifies exhaustive and mutually exclusive subgroups within a sample, based on evaluation of categorical input measures. We used PROC LCA in SAS 9.4 [38] to identify latent subgroups within our sample based on survey responses.
Our survey included several questions about English language use and proficiency (“Do you speak English, read or write in English, and in what language do you usually read?”), as well as several questions about trust of others (“Can most people be trusted?”, “Can you trust Somali people?”, and “Can you trust non-Somali people?”). Having coded each of these measures as a “yes or no” response, we conducted LCA using these as inputs, to identify subgroups in our sample, distinguishable based on patterns of response to these questions. Further details of the LCA analyses are beyond the scope of this paper and are being prepared for inclusion in a separate publication.
Geographic Enclave
We defined individuals as “geographically enclaved” (a 0/1 measure) based on whether, at the time of our current survey, they reported living in one of eight contiguous zip codes in Minneapolis that represented the initial residence locations for most Somali immigrants to Minnesota around the year 2000. We defined individuals living outside of those zip codes as being not enclaved.
Duration in U.S.
We asked respondents in what year they arrived in the U.S. From this, and their interview date, we computed their duration in the U.S. in years. There were 8 respondents who reporting having been born in the U.S. who we assigned missing values for this measure.
Other Covariates/Confounders
Other covariates and potential confounder variables included in our multivariate models include continuous age in years, gender (male or female), marital status (not married = 1, married or in a partnered relationship = 2, separated/widowed/divorced = 3), poverty threshold (below or above federal poverty level), employment (yes or no), education (more or less than high school education), and smoking status (current smoker, nonsmoker). Social resources as “high” for the group reporting having access to 9 or more of a possible 12 types of social resources or “low” for those reporting having access to 8 or fewer types of support, and physical activity as “active” (≥150 minutes/week) and “less active” (<150 minutes/week). Health insurance status was recorded as yes or no, based on self-report. Body mass index was calculated as weight in kilograms divided by height in square meters as a continuous variable. Interview questions included history of diabetes and history of high cholesterol and were categorized as yes or no. We defined as diabetic those with hemoglobin A1C ≥ 6.5 percent according to National Institute of Diabetes and Digestive and Kidney Diseases. We defined dyslipidemia as a ratio of Cholesterol/ High density lipoprotein (Chol/HDL ≥ 5).
To aid interpretation of our multivariate modeling results, we centered age and duration of residence (years) in the U.S. by computing separate variables that subtract the sample average age and duration in U.S. values, respectively, from everyone’s age and duration of residence in the U.S.
Statistical Analyses
Analysis of RDS data requires sample weights based on an individual’s likelihood of sample inclusion, a function of the size of their social network, since individuals with more network contacts have a higher probability of being included in the sample. We used standard weights generated by RDSAT 7.1 software to re-equalize sample inclusion probabilities across individuals in the sample [30, 31].
Because individuals who are directly linked within a social network are more likely to share characteristics, RDS data also results in non-independence, or correlated data. To account for the correlated observations inherent in RDS data, as well as the unequal sampling probabilities, we estimated the relationships between HTN and our measures of acculturation using an adaptation of Generalized Estimating Equations (GEE) with a multivariate logistic regression model. We estimated correlations between individuals who were directly connected in the respondent driven sampling referrals, using the weights from RDSAT in the GEE to account for the unequal sampling probabilities [32].
We describe the characteristics of the study population by frequency (percentage) and mean (± standard deviation). We assessed comparisons between groups using Student’s t-test and ANOVA for continuous variables and by chi-squared and/or Fisher’s exact test wherever appropriate for categorical data. We performed the descriptive analyses using SAS Version 9.4 (SAS Institute Inc., Cary, NC.). In multivariate logistic regressions we considered P-values of < 0.05 statistically significant. For our multivariate analyses we used the geeNET package (citation to https://github.com/MilesOtt/geeNET) in R [33].