Data source and study population
Data were collected from the Korean Longitudinal Study of Ageing (KLoSA), which was conducted in 2006, 2008, 2010, 2012, 2014, and 2016. The KLoSA is a large-scale, longitudinal survey of the population aged 45 and older living in households selected by multistage stratified probability sampling to ensure national representativeness. It was designed to help develop policies to address health and social issues that emerged because of rapid population aging. In the 2006 baseline survey, the original sample of 10,254 respondents completed interviews by well-trained interviewers. The household response rate was 70.7% and the individual response rate within households was 75.4%. This survey was followed up with 8875, 8229, and 7813 respondents in 2008, 2010, and 2012, respectively. A refreshment sample of 920 individuals born in 1962 or 1963 was introduced in 2014 and was included in the 2014 and 2016 waves. The combined sample included 8387 respondents in 2014 and 7893 in 2016 [12]. After excluding those with missing data and those who were unable to follow up, 9263 respondents were included in the present sample.
Measures
Poverty transitions
The variable of interest was the transition of poverty status across time. We employed a relative measure of poverty, defining it as earning below 50% of the median household income based on the equivalized household. The value of the poverty line was set for each year (2006, 2008, 2010, 2012, 2014, 2016) based on data from Statistics Korea. The KLoSA contains detailed information about the different types of income that comprise aggregate income, including earned income, asset income, public transfer income, financial support, and other types of income. Total household income is the sum of the incomes of all household members living together, including the respondent. The household income reported by the representative member was assigned to all the other members such that that the total amount of household income had the same value across all household members [12]. In the present study, we used equivalized household income, which considers the square root of the number of household members. The current equivalized household income of all respondents in the sample was allocated into poverty and non-poverty groups based on the previously defined poverty line. Poverty transition was measured as change in poverty status in a previous year (Y-1) and the subsequent year (Y0). We categorized the respondents into four groups: non-poverty to non-poverty (NN, persistence of non-poverty), poverty to non-poverty (PN, exiting poverty), non-poverty to poverty (NP, transition to poverty), and poverty to poverty (PP, persistence of poverty) [13].
Frailty
We used a broader definition of frailty that includes physical phenotype and social and psychological aspects. The frailty instrument consists of items measuring weakness of grip strength, exhaustion, and social isolation. It was developed to assess the risks of adverse health outcomes such as disability, institutionalization, and mortality of older adults with high predictive validity, discrimination, and calibration ability. The validity of the frailty instrument has been reported elsewhere [14]. Weakness was measured using grip strength (< 24 kg for men and < 15 kg for women). Exhaustion was evaluated by self-reporting either the feeling that every task required effort or that they could not “get going” in the preceding week. Isolation was assessed by asking about participation in meetings or group activities. The scale scores ranged from 0 to 3 and were categorized as frail (≥ 2), pre-frail (≥ 1), and robust (0) [15]. In this study, we grouped participants into two categories: frail (≥ 2) and non-frail (≤ 1).
Covariates
Demographic, socioeconomic, and health-related factors were included in the study. The demographic variables were sex, age (45–64, 65–74, 75 or older), marital status (with spouse, without spouse), and number of household members (1, 2, 3, or more). The socioeconomic variables included educational level (elementary school or below, middle/high school, college, or above), household income (quantiles), current economic activity (active, inactive), region (metropolitan, urban, rural), and health insurance (national health insurance, medical aid). The health-related factors included smoking (yes, no), drinking (yes, no), perceived health status (healthy, average, unhealthy), and presence of chronic diseases (yes, no). Chronic diseases included hypertension, diabetes, malignant tumor, liver disease, cardiovascular disease, cerebrovascular disease, psychiatric disorders, and rheumatoid arthritis. We used indicators of individuals’ functional and cognitive status, including activities of daily living [independent, needs help/difficulty with activities of daily living (ADL)], instrumental activities of daily living [independent, needs help/difficulty with instrumental ADL (IADL)], and cognitive impairment (yes, no). Cognitive impairment was measured using the Korean Mini-Mental State Examination (K-MMSE), which includes 11 items in seven categories of cognitive functions (orientation of time and place, registration, attention and calculation, recall, language, and visual construction). The total score range from 0 to 30, and higher scores indicate better cognitive function. The validity of the K-MMSE has been reported elsewhere [16]. We followed the conventional classification criteria and categorized scores as indicating normal cognitive function (K-MMSE ≥24) and mild to severe cognitive impairment (K-MMSE ≤23). Frailty status in the previous year was included to take account of its contribution to frailty in the follow-up year.
Statistical analysis
The distribution of general characteristics was calculated at baseline. Differences in baseline characteristics between non-frail and frail respondents were determined using χ2 tests. To evaluate repeatedly measured individuals, PROC GENMOD was used to employ a generalized estimating equation (GEE) for repeated measure analysis. We evaluated whether the probability of frailty changed after poverty transitions over two consecutive years (between 2006 and 2008, 2008–2010, 2010–2012, 2012–2014, or 2014–2016). Furthermore, subgroup analyses stratified by age, marital status, current economic activity, region, presence of chronic diseases, and cognitive impairment were performed to examine the association between poverty transitions and frailty after adjusting for covariates. All analyses were conducted using SAS software, version 9.3 (SAS Institute, Cary, NC).