This study was a secondary analysis of de-identified data and did not require a human subject’s review.
Study Design
Publicly available data from the National Health and Nutrition Examination Survey (NHANES) were pooled across ten years, 2005-2014, for this study. NHANES is a national probability, repeated cross-sectional survey of US adults and children > 12 years old that assesses health and nutrition status using interviews and medical examinations [21]. Gender is assessed by interviewer’s during the household screening such that respondents are assigned “male” or “female” gender based on either physical characteristics or direct inquiry. Transgender inclusive gender identity was neither asked nor recorded. Respondents characterized as “female” during the interview comprise the sample of women in our study.
NHANES data vary across survey years such that some data (e.g., alcohol use) are not publicly available for the subsample of respondents < 20 years old at time of interview. Moreover, some sexual orientation questions (e.g., sexual identity) are not asked of women > 60 years old at time of interview. Consequently, our study sample included women aged 20 -59 who completed sexual behavior surveys. Respondents were excluded from analyses if they did not answer sexual identity, lifetime same-sex sexual behavior, tobacco use food security, and alcohol use questions. The final analytic sample included 7,015 women.
Dependent Variables
Food insecurity
NHANES uses the USDA’s US Household Food Security Survey Module (α = 0.74-0.93 [22]) to assess past 12-month food insecurity. This measure assesses food insecurity across 4 domains, including: (1) anxiety about food supplies, (2) perceptions that quality or quantity of food is not adequate, and reduced food intake by (3) adults or (4) children (if applicable). Food security is assessed using a scale of 0-10 for households without children and 0-18 for households with children. Levels of household (HH) food security are designed as “full food security” (0 points), “marginal food security” (1-2 points), “low food security” (3-5 points HH without child, 3-7 points HH with child), and “very low food security” (6-10 points HH without child, 8-18 points HH with child). Variables were recoded so that individuals were considered food insecure if scores were ≥ 3 (low or very low food security; coded 1) and food secure if scores were ≤ 2 (i.e., full or marginal food security; coded 0) [23]. For sensitivity analyses, the variables were recoded so that individuals were considered severely food insecure if scores were ≥ 6 (household without child) or ≥ 8 (household with child) (very low food security; coded 1) and food secure (i.e., full marginal, or low food security; coded 0) if scores were ≤ 5 (household without child) or ≤ 7 (household with child).
Receipt of SNAP benefits
Respondents affirming that they, or another household member, were authorized to receive or received food stamp/SNAP benefits in the past 12-months were coded as receiving SNAP (coded 1) versus those not authorized to receive/did not receive past 12-month food stamp/SNAP benefits (coded 0).
Receipt of emergency food assistance
Emergency food assistance was assessed with the question, “In the last 12 months, did [you/you or any member of your household] ever get emergency food from a church, a food pantry, or a food bank, or eat in a soup kitchen?” Respondents were coded as receiving emergency food assistance in the past 12 months (coded 1) or not receiving past 12-month emergency food assistance (coded 0).
Independent Variables
Sexual orientation was defined in terms of sexual identity and sexual behavior according to best practice [24] and previous publications [25]. Women aged 18–59 years were asked, “Do you think of yourself as heterosexual or straight (i.e., sexually attracted only to men); homosexual or gay (i.e., sexually attracted only to women); bisexual (i.e., sexually attracted to men and women); something else?” Women were also asked to report the number of women and men with whom they had engaged in sexual behavior over the life course. We defined women’s sexual orientation as follows: Women reporting heterosexual identity and lifetime sexual activity with only male partners were defined as exclusively heterosexual women (coded 0). Women identifying as lesbian and reporting any lifetime sexual activity with women were defined as lesbian women who have sex with women (lesbian WSW; coded 1). Women identifying as bisexual and reporting any lifetime sexual activity with women were defined as bisexual WSW (coded 2). Women identifying as heterosexual and reporting any lifetime sexual activity with women were defined as heterosexual WSW (coded 3).
Covariates
Summary statistics were calculated to describe demographic, socioeconomic, and psychosocial factors. Age was recoded into four categories representing respondents across emerging (18-25), young (26-35), middle (36-45), mid-late (46-59) stages of adulthood. NHANES’ original variable structure was retained for marital status (married, widowed, divorced, separated, never married, and living with partner) and race/ethnicity categories (non-Hispanic white, non-Hispanic black, Mexican American, other Hispanic, and other race including multiracial). Education level was recoded into three categories (< high school/General Education Diploma (GED), some college/Associate’s degree, college graduate or higher). In multivariable analyses, race/ethnicity was dichotomized into person of color (coded 1) and not a person of color (coded 0). Education was dichotomized into < high school/GED (coded 1) and > high school/equivalent degree (coded 0). Family poverty to income ratio was calculated by dividing family income by the Health and Human Services Poverty guidelines specific to family size, year and state [22]. For descriptive analyses, family poverty to income ratio was presented by US Census defined poverty thresholds (<100%, 100-199%, 200-299%, 300-399%, >400%). For regression analyses, family poverty to income ratio was dichotomized where respondents were considered poor (income <200% federal poverty level [FPL]; coded 1) or not poor (income ≥ 200% FPL; coded 0). For summary statistics, health insurance was defined as reporting private insurance, Medicare/Medigap, Medicaid, other public insurance, or being uninsured. In multivariable analyses, we defined health insurance coverage as private (coded 0), public (coded 1), or none/uninsured (coded 2). Alcohol use [26] and cigarette smoking [27, 28] are two psychosocial characteristics that are associated with food insecurity and are known disparities in SMW [29]. Women were defined as at-risk drinkers (coded 1) if, during the past 12 months, they reported having > 7 or more drinks per week [30]. Current cigarette smoking was defined as having smoked > 100 cigarettes ever and currently reporting smoking on either “some” or “every” day (coded 1).
Analyses
Summary statistics described the sample. We assessed differences between sexual minority and heterosexual respondents with Likelihood Ratio chi-squared test for proportions (LR X2). We then used weighted bivariate analyses with LR X2 test for proportions to investigate differences in food insecurity and food assistance resource use prevalence across diverse sexual orientation subgroups. We reported results as weighted point estimates as percentages with standard errors, associated test statistics, and p-values (Results not shown; available upon request). Simultaneous, weighted, multivariable Poisson regression models estimated prevalence ratios (PR) and 95% confidence intervals (CI) for the associations between sexual orientation and food insecurity, severe food insecurity SNAP participation, and emergency food assistance use in sexual minority women. Covariates selected a priori as potential confounders included age, race/ethnicity, income, educational attainment, health insurance coverage, risky drinking, and smoking. Multivariable analyses were adjusted for survey year to account for potential unmeasured cohort effects. Sampling weights based on the NHANES multistage design were used for all multivariable models. We used the “subpop” command for variance estimation with Taylor series linearization as per NHANES guidance [21]. STATA 16.0 (StataCorp LP, College Station, TX) was used for all analyses.