Participants and data
Data for this study were taken from wave 1 of the World Health Organization’s (WHO’s) Study on global AGEing and adult health (SAGE), the most recent data available from China. SAGE is a longitudinal study for which nationally representative data were collected from adults aged ≥ 50 years from six low- and middle-income countries (China, Ghana, India, Mexico, the Russian Federation and South Africa) using a multistage, stratified cluster sampling approach. The effectiveness and high response rate of SAGE are attributable to proper planning and organization from the initiation of the study [33]. All investigators, supervisors and interviewers were trained to administer the survey in the field, introduce SAGE to the sampled households and obtain household member participation and informed consent [34]. In China, wave 1 of SAGE was implemented in 16 strata in 8 provinces/municipalities [34]. A five-stage cluster sampling strategy was used to select participants, who were contacted by telephone or in person, and about 200 investigators were involved in wave 1 data collection via face-to-face interviews between 2008 and 2010 [34]. About half of the interviews were computer assisted (CAPI), and half involved manual data recording [35]. Investigators visited the selected households and collected information about household rosters; then, the survey team completed the questionnaires at a central location (e.g. a neighbourhood office) or at respondents’ homes [34]. Each respondent received a small gift for his or her cooperation [34]. An excellent response rate was achieved (93%), similar to rates for other surveys (e.g. the China Health and Retirement Longitudinal Study) conducted among older people in China. Detailed information about the SAGE data collection procedures can be found elsewhere [34].
SAGE consists of national longitudinal studies of older people (age ≥ 50 years) in six lower- and upper–middle-income countries. The instruments and threshold age used are compatible with other large longitudinal ageing studies conducted in high-income countries, such as the US Health and Retirement Study (HRS) and the Korean Longitudinal Study on Ageing (KLoSA), enabling sound international comparisons of the ageing process, health and well-being among middle-aged and older adults [35]. The original wave 1 sample included 13,367 participants from China. We enrolled respondents aged ≥ 50 years with chronic disease (angina, arthritis, asthma, chronic lung disease, diabetes, diagnosed depression, hypertension, paralysis or stroke), leading to a final sample of 6,629 respondents. Most (n = 6194, 93.4%) older persons in the sample were aged 50–80 years; people aged 50–59 years made up the largest group (n = 2270, 34.2%), those aged 60–69 years comprised the second largest group (n = 2154, 32.5%) and only 6.6% (n = 435) of the sample was aged > 80 years. The procedure for sample selection is summarized in Figure 1.
Measures
Chronic Conditions
For self-reporting of chronic conditions, respondents were asked whether they had been diagnosed with any of the following: i) angina or angina pectoris (heart disease), ii) arthritis (or rheumatism, osteoarthritis), iii) asthma (an allergic respiratory disease), iv) chronic lung disease (emphysema, bronchitis, COPD), v) diabetes (high blood sugar), vi) depression, vii) high blood pressure (hypertension), viii) paralysis and ix) stroke. The questions were formatted as: “Have you ever been diagnosed with/told by a health care professional you have…?” Respondents provided yes/no answers. They were considered to have chronic (a) disease(s) if they answered “yes” to any of the questions.
Health behaviours
Social participation was measured using summed scores for the 9-item questionnaire developed for the SAGE [36] (Appendix 1). Items enquire about respondents’ frequency of community involvement in the past 12 months, with responses ranging from ‘never’ (1) to ‘daily’ (5). The Cronbach’s alpha value for the questionnaire in this study was 0.63. We used adequate fruit and vegetable intake as an indicator of healthy diet (insufficient, fewer than servings fruit and three servings vegetables/day; sufficient, two or more servings of fruit and three or more servings of vegetables/day] [37]. Version 2 of the General Physical Activity Questionnaire was used to measure physical activity [36]. Participants were asked to report the average number of days per week and time in which they engaged in vigorous and moderate physical activity. We recorded physical activity as sufficient or insufficient according to the WHO threshold of 150 min/week [38]. Smoking habits were assessed by asking whether participants were daily smokers (yes/no).
Outcome variables
Quality of life
Quality of life was measured using the 8-item World Health Organization quality of life measure (WHOQoL) [35]. Respondents were asked to rate their satisfaction with life in general and in different domains (e.g. finances, health and relationships) on a 5-point scale ranging from 0 (‘not at all/very poor’) to 5 (‘completely/very good’). Total scores were calculated by summing the item scores and rescaling the result to 0–100 [39]. According to previous research [40], the 8-item WHOQoL is useful for the assessment of quality of life in older populations. The Cronbach’s alpha value of the instrument in this study was 0.86.
Cognitive function
Cognitive function was measured by administering five cognitive performance tests (forward and backward digit spans, immediate and delayed verbal recall, and verbal fluency) [41]. Forward digit span was tested by asking participants to repeat progressively longer number series in the exact order in which they had been presented [41, 42]. Backward digit span was tested by asking participants to repeat such series backwards [41]. Scores (longest spans repeated) for the forward and backward digit spans ranged from 0 to 9 and 0 to 8, respectively (total possible scores, 1–17) [42]. Immediate and delayed verbal recall was measured by asking participants to read 10 words aloud and soon thereafter to recall as many words as possible in 1 minute [41]. The same test was repeated three times. Scores ranged from 0 to 10 [43]. Verbal fluency was assessed by asking respondents to name as many animals as they could in 1 minute [42]. Scores were based on the number of correctly named animals, with repeated names counted only once (range, 2–38) [42, 43]. Z scores were calculated for the five test scores, and final cognitive function scores (range, 0–100) were generated by summing these scores [41, 42].
Physical function
Physical function was measured using the activities of daily living items from version 2 of the WHO’s Disability Assessment Schedule, based on the Katz Index of Independence in Activities of Daily Living [44]. Total scores was calculated by summing scores for the following items: 1) difficulty in bathing/washing your whole body, 2) difficulty in getting dressed, 3) difficulty with getting to and using the toilet, 4) difficulty with standing up from sitting down, 5) difficulty in getting up from lying down and 6) difficulty with eating (including cutting up your food). Responses are structured by a 5-point scale ranging from 0 (none) to 4 (extreme/cannot do). The Cronbach’s alpha value for this instrument in this study was 0.89.
Potential confounders
Based on data from the literature and the availability of SAGE data, we included age (in years), gender (male/female), marital status, area of residence (urban/rural), educational level and income (by quintile) as potential confounders because they are associated both health behaviours and health outcome variables [45-51]. We dichotomized marital status as non-single (including ‘currently married’ and ‘cohabiting’) and single (including ‘never married’, ‘separated/divorced’ and ‘widowed’), and educational level as higher (completion of secondary school or more) and lower (completion of primary school or less).
Respondents’ incomes were estimated. SAGE-China used the WHO’s Bayesian post-estimation method to generate raw continuous income estimates based on income indicators such as a set of household ownership of durable goods (e.g. number of chairs), various dwelling characteristics (e.g. type of floor) and access to services (improved water, sanitation and cooking fuel) [52, 53]. Estimated income was then transformed into quintiles [53], with quintile 1 denoting the lowest and quintile 5 denoting the highest income [52, 53].
Statistical analysis
Descriptive statistics and frequencies were used to describe the study population. Correlation analysis was performed to assess relationships between background characteristics and health behaviours using the outcome measures (quality of life, cognitive function and physical function). Multivariate linear regression analyses were conducted to study associations between health behaviours (physical activity, maintenance of a healthy diet, smoking and social participation) and quality of life and health outcomes while controlling for background characteristics. We used listwise deletion of missing cases in the multivariate analyses. Analyses were performed using IBM SPSS software (version 24; IBM Corporation, Armonk, NY, USA). As the sample was large, the significance level was set at p < 0.001. All statistical tests were two sided.