2.2 Study Variables
Several demographic, lifestyle and health factors from the MWHS questionnaire were included in the study. The QoL indicator was derived from Cantril’s Ladder of Life, a self-anchoring scale that measures a person's attitude toward their health [13]. The question posed in the survey was "Here is a ladder representing the ‘Ladder of Life.’ The top of the ladder represents the best possible life for you. The bottom of the ladder represents the worst possible life for you. On which step of the ladder do you feel you personally stand at the present time?” The scale ranges from 1–10, with 1 representing the “worst possible life” and 10 representing the “best possible life”. QoL was stratified by low (1–4), middle (5–7), and high (8–10) levels to control for variation in the frequency of choice in scale values.
Demographic factors of interest included were age, annual family income, marital status, education level, and employment status. Each variable was dichotomized to a binomial factor except annual income which was partitioned to a tercile variable: Low (<=$34,999), Middle ($35,000-$74,999), and High (>= $75,000). Lifestyle factors or modifiable factors that can greatly influence health included were BMI, smoking status, and drinking status [14]. BMI was calculated from measured height and weight then dichotomized based on the Center for Disease Control’s scale such that BMI ≥ 30 was considered obese and BMI < 30 was considered non-obese [15]. Drinking status was determined by responses to the following question, "In the last 12 months have you had at least 12 drinks of any kind of alcoholic beverage?", bivariate responses of yes or no were recorded.
Health factors included in the study were depression, number of comorbidities, menopausal status, hot flash experience, hormone replacement therapy use, pregnancy status, sexual activity, and sleep disturbances. Depression was measured using the Center for Epidemiologic Studies Depression Scale (CES-D) survey [16]. CES-D score was used to identify depressive symptoms in participants. CES-D scores of < 16 were considered not depressed and > 16 were considered depressed. Women were asked if they had been diagnosed with any of the following potential morbidities: diabetes, heart disease, stroke, hypertension, high cholesterol, anemia, breast cancer, ovarian cancer, uterine cancer, other cancer, epilepsy, lupus, thyroid disorder, depression, cataracts, stomach ulcer, yellow jaundice, cirrhosis of the liver, hepatitis, arthritis, allergies, asthma, hay fever, eczema, rosacea, psoriasis, or other skin disorder. The number of comorbidities, self-reported from this list, was categorized as 0–2, 3–4, or > 5 comorbidities. Menopausal status was defined by the number of periods a woman experienced at the time of reporting. Women were dichotomized based on whether they experienced “0–11” periods in the previous year, peri-menopausal, or “greater than 11” periods, pre-menopausal. Hot flashes, hormone replacement therapy use, history of pregnancy, and sexual activity were categorized into “yes” or “no” based on whether a woman experienced the factor or not. Sleep disturbances were categorized by frequency of disturbances: “never-4 times per month”, “2–4 times per week”, and “>5 times per week”.
2.3 Statistical Analysis
Ordinal logistic regression was applied to analyze the relationship between QoL and possible risk factors (demographic, lifestyle, and health factors). Assumptions for no multicollinearity and proportional odds were tested using the Brant Test function from the brant package in R [17]. Univariate models were conducted to determine the individual effects of each factor on QoL. An alpha value of 0.05 was considered statistically significant. Due to the high levels of missingness of responses and statistical insignificance in univariate analysis, drinking and hormone replacement therapy factors were omitted from multivariate analysis for both populations.
Multivariate regression models were fitted by backward, stepwise regression based on the BIC [18]. Odds ratios greater than 1 implied increased likelihood of high quality of life. Regression analysis for univariate and multivariate models was stratified by race to compare relationships of QoL within each race and later across each race. Black-white comparisons are important because the widest gap in health disparities occurs between these two populations [19]. Analyses were conducted using R statistical software version 3.5.3 [20] and the polr function in the MASS package [21].
To determine the reliability of our fitted models we performed bootstrapping by the AIC in R [22]. During the bootstrapping process, we simulated data sets for each stratified population by sampling with replacement at 100 iterations. Next we re-fitted each simulated set to determine the number of times our covariates of interest were selected for the model [23]. In multivariate models of white women, depression and income were selected 100% and 97% of the time in refitted models, respectively. Furthermore, the number of comorbidities, depression and smoking status were selected 97%, 96%, and 84% of the time in refitted models of black women, respectively (Supplemental Table 1).
Post-hoc analysis of the comorbidity-QoL relationship was examined using network analysis and determining the frequency of morbidity occurrence. Networks were constructed in R using the package igraph [24]. Nodes were represented by individual morbidities and edges were represented by co-occurrences of morbidities within each population. Larger nodes represented morbidities that occurred more frequently, and larger edges represented co-occurrences that were observed more frequently.