Study population and design
SAGE was a longitudinal cohort study of ageing and older adults in six low- and middle-income countries (China, Ghana, India, Mexico, Russian Federation and South Africa) [10]. The study population was sourced from SAGE China Wave 1 from 2007-2010, using a probability sampling design and a five-stage cluster sampling strategy[11]. SAGE China Wave 1 consisted of 1,636 individual respondents aged 18–49 years and 13,175 respondents aged 50+ years. The response rate for the individual questionnaire was 98%, and a final total sample size of 13,175 for this analysis.
SAGE was approved by the World Health Organization's Ethical Review Board (RPC146), and local approval by the ethics review committee of the Chinese Center for Disease Control and Prevention (approval notice 200601). Each respondent signed informed consent.
Measures
Anemia
Blood hemoglobin concentrations were derived from dry blood spot samples and examined using standardized ELISA techniques at the Shanghai Municipal Centre for Disease Control and Prevention Laboratory. The World Health Organization’s (WHO) definition of anemia was used: hemoglobin less than 13g/dL for men and less than 12g/dL for women[2].
Frailty
Frailty was defined using the deficit accumulation approach. A frailty index (FI) was constructed as the proportion of deficits present out of 40 variables available in the SAGE database, including self-rated health, 9 medically diagnosed conditions, 4 medical symptoms,13 functional activities assessments, 10 activities of daily living (ADLs), body mass index (BMI), grip strength and gait speed. [12]. Individual scores ranged from 0 (no deficits) to 1 (highest level of deficits in all variables). The FI cut-off value of 0.2 was defined as approaching a frail state[12].
Other covariates
SAGE used a standardized survey instrument to collect sociodemographic information and behavioral risk factors based on the WHO STEPwise approach to Surveillance (WHO STEPS, WHO 2005). Socio-demographic variables included age, sex, education, rural/urban residence, and household wealth. Age was categorized into four groups: 50 to 59 years; 60 to 69 years; 70 to 79 years; and 80 years or older. Highest level of education completed was classified into six categories using an international classification scheme (No formal education; less than primary; primary school completed; secondary school completed; high school completed; college completed and above) for use in this analysis[13]. The household wealth was generated using an asset-based approach and included possession of assets and dwelling characteristics[14], with the resulting wealth quintiles ranging from quintile 1 (Q1, poorest) to quintile 5 (Q5, wealthiest) households.
Non-communicable disease risk factors included alcohol and tobacco consumption, poor diet and low physical activity levels. Tobacco use was classified into four groups: never smoker, not current smokers, current smokers (not daily) and current daily smokers. Alcohol consumption was categorized into four groups: never drinker, non-heavy drinkers, infrequent heavy drinkers and frequent heavy drinkers according to the number of standard drinks consumed in a given week. Physical activity was measured by the Global Physical Activity Questionnaire (GPAQ) and three categories were generated: low, moderate and high levels[15]. Diet was assessed through fruit and vegetable consumption and calculated by the number of daily servings eaten. Five or more servings were defined as sufficient daily intake (equivalent to at least 400 grams per day), fewer than five servings was categorized as insufficient[16].
Statistical methods
Statistic analyses were conducted using STATA SE version 14.1 (Stata Corp, College Station, TX). The population prevalence of anemia and frailty was calculated by using normalized weights. Weights were based on selection probability, non-response, and post-stratification adjustments. A 2-level hierarchical logistic model was used to evaluate the association between anemia and frailty using STATA command “melogit”. We also included hemoglobin concentration as a continuous variable in the model (models 3 and 4) to see if there was an association between hemoglobin concentration and frailty. Covariates of interest included age, gender, education, smoking, nutrition, physical activity. P < 0.05 from two-sided statistical tests was considered statistically significant.