Study design and data sample
This study used data from the WHO’s Study on global AGEing and adult health (SAGE), Wave-1 (2007-2010). The data were originally collected in six LMICs—China, Ghana, India, Mexico, the Russian Federation, and South Africa—to better understand the health and well-being of older adults through nationally representative samples. SAGE is designed as a multi-wave panel study. Multistage cluster sampling methods were used; the original sample consists of 35,334 people aged 50 years or older who participated in the SAGE Wave-1 initiative. Face-to-face interviews was conducted using a standardized survey instruments, set of methods, interviewer training and translation protocols in all countries. A more detailed description of the SAGE Wave-1 data has previously been published [15]. The final sample for this study comprised 33,019 people aged 50 or older in five countries, after we excluded the data from Mexico due to substantial missing values (49.7 % of data).
Outcomes of interest
The main outcome variable for this study is QoL. QoL was assessed using the 8-item WHOQoL instrument [16]. The 8-item WHOQoL—a shortened version of the WHOQoL-BREF—comprised two items from each domain of the WHOQoL-BREF (i.e., physical, psychological, environmental, and social). Participants answered each question rated on a five-point Likert scale from 1 (not at all) to 5 (completely). The overall QoL score was determined by a simple summation of the scores of the eight items and then rescaling the score from 0-100, where a higher score indicated a higher QoL. Good internal consistencies (0.72-0.85) [16, 17] and acceptable convergent validity with WHOQoL-BREF (0.61-0.77) [17] were reported across the five countries.
Independent variable
Gender was assessed as the independent variable by recording the gender of the participant (male = 0, female = 1).
Covariates
The covariates consisted of demographic variables (i.e., age, education, health insurance, income, and living environment), health-related variables (i.e., cognitive function, physical function, presence of comorbidities) and social support variables (i.e., marital status, family support, community support, social cohesion index, and living arrangements).
Sociodemographic variables included age (continuous variable), education (0 = less than primary, 1 = primary only, 2 = secondary only, 3 = high school only, 4 = college and above), and health insurance (no = 0, yes = 1). Furthermore, standardized income (continuous, provided by SAGE, with a higher score of standardized income indicating a higher income status) and living environment was assessed by a summary scale based on three dichotomized indicators related to an individual’s living environment (i.e. hard floor, piped drinking water, and durable walls). The total score ranged from 0 to 3, with higher scores indicating a better living environment.
Cognitive function was measured by five tests: forward and backward digital span tests, verbal fluency, immediate recall, and delayed recall. This set of cognition tests captured several aspects of cognitive function, including working memory. First, a z-score was generated from each test before a global cognition score was calculated by averaging the z-scores. Higher z-scores indicated better cognitive function. Physical function was assessed by using the 12-item version of the World Health Organization Disability Assessment Schedule (WHODAS) 2.0 [18]. This test is a brief assessment tool to measure physical functional limitations cross-culturally. Research examining the psychometric properties of the test supported the construct validity of the one-factor solution with various samples [19-21] and a strong internal consistency [21]. A higher WHODAS 2.0 score indicates poorer physical function. Comorbidity was defined according to the presence of arthritis (no = 0, yes = 1), hypertension (no = 0, yes = 1), and diabetes (no = 0, yes = 1).
Marital status (not married = 0, married = 1) was included as a social support variable. Received social support was defined as family support and community support. The SAGE Household survey was conducted to determine whether the participants received any financial or in-kind support from 1) family members or 2) the community. Two dummy variables were created if the respondents received any family or community support. The social cohesion index consisted of 9 questions related to the frequency of taking part in various social activities, for example attending religious services or having friends over [22]. The total social cohesion index score ranged from 9 to 45, with higher scores indicating better social cohesion. The living arrangements variable was created as a dichotomized variable if participants reported a household size of one. Lastly, a country variable was included (101 = China, 106 = India, 102 = Russia, 103 = South Africa, 104 = Ghana).
Statistical analysis
The sample characteristics of the study participants were determined, and comparisons according to country were calculated with chi-square and ANOVA tests. Moreover, a post hoc analysis for group differences was performed using the Bonferroni correction. The p-values were based on 2-tailed tests and can be considered statistically significant at p < .05. Overall 8.4% of the study participants had missing data in their QoL measure. The percentage of missing values across all covariates ranged from 1% (self-reported arthritis) to 6% (physical function). To preserve the analytical sample size, multiple imputation (mi impute mvn command in STATA) was used to account for missing data (5 imputations). A preliminary analyses produced results similar to that of the multiple imputation when using listwise deletion to address missing data.
Next, multivariate linear regression models were estimated to examine the factors influencing QoL. The first set of analyses estimated differences in QoL while adjusting for different countries. The second set of analyses estimated differences in QoL while adjusting for sociodemographic variables. The third set of analyses was based on the second set of analyses while adding health-related variables into the model. Next, the fourth set of analyses included all of the factors mentioned above as well as social support factors. Because we observed significant differences in QoL as well as other sample characteristics across the five countries, we then stratified the analyses according to country (Table 3). Next, we performed further analyses to examine the influencing factors on QoL according to gender (Tables 4 and 5). Standardized coefficient estimates were presented to assist in identifying the most influential factors. Survey weights were used in the descriptive analyses to adjust for the sampling design. For the multivariate analyses, results from the unweighted models were presented, as all multivariate analyses included variables used in the sampling weights (e.g., age and gender). This was done because including survey weights may produce biased estimates and inflated standard errors [23]. The analyses in this study were conducted using Stata version 14.2.