Study setting and the FilaBavi surveillance system
The study was conducted in 2007 within the longitudinal demographic and health surveillance system of FilaBavi [29]. The field site operates in the rural Bavi district of Vietnam. Bavi covers an area of 410 km2, including lowland, highland and mountainous areas. Thirty percent of the land is used for agriculture and 17% is forest. Mountainous areas account for 42% of the land mass. The 2007 population was 262,763 people. Among adults over 20 years of age, the majority completed primary/secondary school (65% of men, 73% of women) and high school/higher education (34% of men, 23% of women). The rest are illiterate. Two-thirds of the population are farmers (39% of men, 57% of women) or industrial workers (31% of men, 9% of women) and the remainder are business people, students, government staff, retired persons or others.
The FilaBavi surveillance system consists of a representative sample of 67 out of 352 clusters in the district that have been selected with a probability proportional to size since 1999. In 2007, 53,927 individuals were followed by FilaBavi, and this represented 20.5% of the total district population. People aged 60 and over represented 11.5% of the total population followed by FilaBavi at the 2007 mid-year point.
Study design, sampling and sample size
The sample size was calculated by estimating a proportion in a population-based survey. Using an estimated proportion of 13% (estimation error of 2.6%) of elderly who need daily support for care in daily living in a rural area of Vietnam [30], a sample size of 2,699 elderly is required. This was adjusted for a design effect of two for cluster sampling of FilaBavi, then doubled for robustness of multivariate analysis, and allows for a 10% non-response rate. This figure is approximately equal to 50% of all people aged 60 or greater in the FilaBavi sampling frame.
Fifty per cent of households with elderly members followed by FilaBavi were randomly selected for a household cross-sectional survey. This resulted in 2,255 households with 2,968 individuals. During the survey period from July to October 2007, 166 households were excluded due to absence of the elderly. However, each of these households was replaced with the nearest unselected household that had elderly members. In total, 2,240 households with 2,873 elderly were included in the quantitative part of this study.
Qualitative data were collected to increase credibility and provide further understanding of the quantitative study. Four focus group discussions (FGD) were conducted in one commune with a socioeconomic status at the district average. The first discussion was with six elderly people, and the second discussion was with six representatives of households with older people. The discussions were organized at one village in the commune. The discussants in each group had equal numbers of men and women.The elderly who participated in the FGD belonged to different age groups. The household representatives were not elderly themselves and had different household roles (two heads, two main caregivers, and two other members).
Six or seven representatives of the key social stakeholders in elderly care participated in each of the other discussions, one at the village level and another at the communal level. These participants included individuals from the local authority, health sector, elderly association, women's union, youth union and former soldiers' union.
Study variables and information
Willingness to use and willingness to pay for different community-centric models of elderly care were examined among both older people and their household representatives in the cross-sectional survey. Willingness to use each of the models was further considered as a dependent variable in the analysis of its association with a number of independent variables, such as the elderly's need of help in ADLs and socioeconomic characteristics of older people and their households.
Individual characteristics of older people were collected from the survey. These included date of birth, sex, education, marital status, household head status, status of living with spouse, and working status such as working in one's own rice fields or not working. Variables on household economic and living conditions were extracted from the mid-2007 FilaBavi census in order to estimate the wealth index and poverty status of households. This type of census has been repeated every second year since the establishment of FilaBavi in 1999. Variables include land area, structural housing components, assets, sanitation conditions, income, expenditures and debt.
The qualitative study focused on perceived care needs of the elderly, current and expected roles of different key stakeholders, encouraging/limiting factors in providing needed care, solutions for overcoming barriers in providing the care, and expected future models of care.
Variables measurement and data collection
Using structured questionnaires, face-to-face interviews were performed by 52 trained field FilaBavi personnel at houses with elderly members. Questions included those about supports needed in ADLs, models of elderly care, individual and household characteristics of the elderly.
Three scales of ADLs were applied when measuring the daily care needs. They included Katz's basic ADLs [31] (bathing, dressing, toilet use, transferring in and out of bed or chair, urine and bowel continence, eating), instrumental ADLs (cleaning house, cooking, shopping, travelling) and intellectual ADLs (writing, reading, listening to radio, watching TV). Support needs for each activity (none, need some helps, complete dependence) were assessed, together with levels of support received (none, not enough, enough).
Three options for possible care models were described to older people and the household representatives. These included: a) a mobile team of nurses in the respondent's commune to provide home care services for the elderly at their request; b) a day care centre in the village that would be a place the elderly could visit for a period of time every day or every other day; c) a nursing centre in the commune or district where the elderly could stay for as long as needed (days, weeks or months).
The assumptions for the last two models were that food would be served, relaxation activities provided, and nursing available. For each model, the elderly and their household representatives were asked whether they would likely use the model if it was provided free-of-charge, for a fee (less than the actual cost) or actual cost. In the first two models, expected types of services were listed as choices. Willingness to pay for, and frequency of using services, were asked for each model.
Six field supervisors reviewed each completed questionnaire and randomly selected 5% for re-interview. Each questionnaire with missing or irrelevant values was returned to the field personnel for checking and completion after re-visits to the corresponding households. Double data-entry was performed using EpiData 3.1 http://www.epidata.dk to check for consistent values of each variable. Correction of data-entry errors was based on actual values from the completed questionnaires.
Assets were classified by category, eg, furniture, communication and electrical equipment, types of vehicles, agricultural machines, and cattle. These items were classified as "present" or "not present" regardless of their quantity and quality. Sanitation conditions were assessed as sources of water for drinking and cooking, type of latrine, and presence of a bathroom. All types of income (ie, agriculture, breeding, forestry, and others) were recorded and summed for the total income of a given household. The sum of daily food expenditures was multiplied by 30 days and added to the sum of other monthly expenditures to estimate total monthly household expenditures. Monthly income and expenditures were then divided by household size to generate "per capita" variables.
Using corresponding guidelines, the discussions and interviews were conducted by a main researcher and 1-2 assistant researchers who were trained and experienced with qualitative research methods. The discussions and interviews were manually noted, tape-recorded, transcribed, and translated to English.
Statistical analysis
The datasets from the present survey and the repeated census were linked and analysed using STATA 10 (StataCorp LP, College Station, TX, USA). An index was calculated for each ADL scale by summing up the score from each activity (score is 0 if no need or need some help; score is 1 if complete dependence). The basic ADL index ranges from 0 to 6. The instrumental and intellectual indices range from 0 to 4. Household wealth index was calculated as the first component for all economic variables from the census. Classification of household wealth quintile was based on the hierarchies among all FilaBavi households. Household poverty status was classified using the national poverty line for rural areas, and based on monthly per capita income being equal to VND 200,000 (USD 12.5) for 2006-2010 [32].
Distributions of study subjects by socioeconomic group, willingness to use care models, frequency of using services, and type of expected service were described using percentages and corresponding 95% confidence intervals. Willingness to pay for care services was estimated as the average monthly expenditure in VND for the elderly or their households with corresponding 95% confidence intervals. Significant differences in percentages or averages between groups of older people, or between older people and their household representatives, were identified by comparing the corresponding 95% confidence intervals.
Multivariate logistic regression analyses were performed to measure the effect of ADL indices and socioeconomic factors on elderly willingness to use care services by models of care and levels of payment. Being independent in ADL, female, aged 80 years and above, illiteracy, widowed status, living without a spouse, position as household member, not working until old age, belonging to the poorest quintile, and living above the national poverty line were used as references in the analyses. A backward stepwise procedure, with a p-value of 5% for removal, was used to identify significant factors that remained in the final multivariate model. Robust standard errors from cluster data were used for accurate estimation of the model parameters [33].
Qualitative analysis
Thematic content analysis was performed by two researchers. Only information that illustrates or explains the quantitative research results regarding care for the elderly is used in this article.
Ethical considerations
Ethical approval for the FilaBavi demographic surveillance system, including data on socioeconomic status, was given by the Research Ethics Committee at Umeå University, Sweden (reference number 02-420). The present study was also approved by the Research Ethics Committee at Hanoi Medical University (reference number 51/HMU-RB). As all selected households belonged to the sampling frame of FilaBavi DDS, and these individuals were familiar with the DSS data collection, only oral consent was required. Purposes of the study and the main contents of the interviews were briefly described, together with a commitment to keeping individual and household information confidential. The participants reserved the right to refuse to answer any question or withdraw from the interview at any time.