Study design, area, and data collection method
Using a cross-sectional design, we investigated the health-seeking behaviors, prevalence of health problems, and related financial burden in households in Bolikhamxay province over the past 3 months. According to the latest data from this province, its total population is 273,691; it covers an area of 14,863 km2 (i.e., 5739 mi2), with a population density of 18/km2 [21]. It shares borders with Xiengkhouang Province to the northwest, Vietnam to the east, Khammouane Province to the south, and Thailand to the west. Bolikhamxay province consists of 7 districts (Pakxanh, Thaphabath, Pakkading, Borikhan, Viengthong, Xaychamphone, and Khamkeuth); it has 1 provincial hospital, 6 district hospitals, and 40 health centers [22].
Researchers working in the Lao Tropical and Public Health Institute were trained in interviewing procedures and conducted questionnaire-based interviews with the sampled households from January 15 to February 13, 2019. All interviews were conducted with the head of each household, who provided information about their family members, occasionally consulting with other family members about the same.
Sample size and household selection process
The sampling unit was a household, defined as a group of individuals living together, usually comprising parents and children, and sometimes including grandparents and uncles. To calculate the acceptable sample size for this study, we referred to a previous study on CHE conducted in Vietnam, a neighboring country with a similar political system, which showed a 17.4% CHE prevalence (20% threshold) in 2004 [23]. We also considered a margin of error of ±5, significance at the 99% confidence level, and Z = 2.57. The total number of households in Bolikhamxay province was 54,738—a number obtained by dividing the total population (273,691) [21] by the average number of members in a household (5) [21]. This yielded a sample size of 384 households. To address the possibility of withdrawal, approximately 25% more households were added to the sample, resulting in a final sample size of 480 households.
$$\textbf{Sample}\ \textbf{size},\textbf{n}=\textbf{N}\ast \frac{\frac{{\boldsymbol{Z}}^{\textbf{2}}\ast \boldsymbol{p}\ast \left(\textbf{1}-\boldsymbol{p}\right)}{{\boldsymbol{e}}^{\textbf{2}}}}{\left[\boldsymbol{N}-\textbf{1}+\frac{{\boldsymbol{Z}}^{\textbf{2}}\ast \boldsymbol{p}\ast \left(\textbf{1}-\boldsymbol{p}\right)}{{\boldsymbol{e}}^{\textbf{2}}}\right]}$$
where the margin of error is ±5, significant at the 95% confidence level, Z = 1.96, the number of households in the province is 54,738, and the estimated proportion of households incurring CHE is 17.4%.
Using a stratified systematic method, the households were sampled as follows. First, we selected three districts, including the provincial capital district (Pakxan). The other two were chosen based on their distance (in km) from Pakxan; the first, Thaphabad, was 50 km away, while the second, Pakkading, was 70–100 km away. As there is only one medical facility at the provincial level in Pakxan, we classified districts into three categories (near, halfway, and far from the capital), assuming that distance from the capital would likely have a strong impact on healthcare access for patients who needed advanced medical intervention. Second, based on the National Statistical Office’s definition of rural areas, we divided the villages in each district into three groups (urban village, rural village with road, and rural village without road). From each group of villages, we randomly selected five villages in each district: one urban village, two rural villages with roads, and two rural villages without roads. We adopted this approach because we assumed that urbanization and the availability of roads would greatly affect general healthcare access within each district. Third, we randomly selected 32–33 households from each village, using the village registration book, with equal intervals on the lists. To ensure the minimum sample size acceptable for this study, upon not acquiring sufficient valid responses from the first group of selected households, we analyzed a novel batch of interviews with the next households on the list, and this procedure was repeated until we reached an acceptable number of valid responses from 32 households per village.
Definitions
Chronic disease
Prior research has posited that differences in the conceptualization of the term “chronic disease” occur largely due to the research data and the research field of lead authors [24]. As we wanted to focus our investigation on all diseases that involved relatively longer treatment periods (to analyze their possible financial burden), regardless of disease course, we chose to simplify the definition of chronic disease. In this study, a chronic disease refers to a disease that a patient has been diagnosed with for more than 3 months at the time of the survey [25]. In this study, we classified diseases into chronic and non-chronic, according to the duration of the condition. For example, if patients had an NCD but they were cured and stopped receiving treatment within the last 3 months, said disease was not categorized as chronic.
Catastrophic health expenditure (CHE)
This is defined as an event in which a household’s medical expenditure exceeds a certain threshold according to the household’s capacity to pay. In this study, to calculate CHE, we adopted the proportionality of income approach [26], where we considered the total monthly OOP spending as a proportion of monthly income, or the proportion of household OOP spending for healthcare greater than the CHE configured in pre-specified proportions. For this study, the data on total household income and total health expenditure per month were obtained from the questionnaire. The total health expenditure per month was calculated via the following procedure. First, in the questionnaire, we asked each household head about the health services their household used and the associated costs in the 3 months preceding the interview. Further, we explained that total health expenditure was defined as the average health expenditure per month, which comprises the sum of all expenses related to medicines, transportation to and from health facilities, consultation and treatment costs, laboratory tests and diagnostic fees, hospitalization fees, cost of visits to traditional healers, and other health-related expenditures during the last 3 months, which was then divided by three. Subsequently, CHE was calculated as the average health expenditure per month in the household divided by household’s total monthly income; the latter value was also obtained during the interview.
Our literature review showed a lack of consensus regarding the thresholds indicating CHE; specifically, we observed thresholds varying from 5 to 40% of total household income [26,27,28,29,30,31]. While there is no consensus regarding the CHE threshold, in this study, we employ the threshold proposed by Xu et al., who define health expenditure as catastrophic if a household’s financial contributions to health equal and/or exceed 40% of non-food expenditure or capacity to spend [29]. However, Rashidian et al. contend that the appropriate cut-off points for the proportion of OOP health expenses to total expenditure and the proportion of health expenditure to ability to pay is 20% [30]. Considering these two previous studies, we set a healthcare expenditure of 20% of the total household income as the threshold for CHE and an expenditure of 40% as the threshold for serious CHE.
Questionnaire structure and independent variables
The questionnaire was developed by drawing on a number of studies on CHE, healthcare utilization, and chronic disease, as well as previous household surveys [26, 31,32,33,34,35,36,37]; the relevant items were adapted to the Lao PDR’s context. We divided the questions into seven sections: a) household composition and demographic characteristics (total monthly income, educational level of the household head, and distance between the household and the nearest medical facility); b) self-reported health problems of household members in the last 3 months (including a variety of health issues ranging from mild to severe, including injuries); c) health-seeking behavior (comprising self-medication and healthcare services—including medical consultations—availed by household members during the preceding 3 months); and e) health services availed and associated costs in the 3 months preceding the interview. Total health expenditure was determined as the sum of all spending on medicines, transportation to and from health facilities, consultation and treatment costs, laboratory tests and diagnostic fees, hospitalization fees, cost of visits to traditional healers, and other health-related expenditures during the last 3 months—converted to 4 weeks, following a previous study [26]. To finalize the questionnaire, a pilot study was carried out with 100 households in Khammouane province (which borders the south of Bolikhamxay Province) to evaluate the accuracy, rigor, and communicability of the questionnaire.
Following previous studies, households were categorized into four groups, in descending order of income, as richest, rich, poor, and poorest. The education level of the head of the household was categorized as under or above primary education. Household size was defined as the number of individuals in a household (< 5 people or ≥ 5 people). Place of residence was defined as the distance from the nearest health facility treating patients with non-severe illnesses (< 5 km or ≥ 5 km) and the provincial-level referral facility treating patients with more severe illnesses (< 10 km or ≥ 10 km). The type of facility visited was defined as public or private. The type of illness suffered was defined as chronic or non-chronic.
Data analysis
SPSS™ version 25 (IBM Corp., Armonk, NY, USA) software was used for statistical analysis. A descriptive analysis was undertaken to understand household background, occurrence and type of health problems, health-seeking behavior (for households and individuals, separately), and household OOP payments. Chi-square statistics were used to compare non-chronically and chronically ill patients’ demographic variables and the different types of facilities used when incurring CHE. Mann-Whitney t-tests were used to compare OOP expenses between the different types of facilities. A logistic regression (logit) model was used to predict the probability of CHE. Based on the literature, we first assumed that the type of illness and treatment episodes had an impact on households with CHE. We expected that chronic illness, hospitalization, treatment in a private facility, and treatment in a provincial-level referral hospital would be associated with high healthcare expenditure.
The second group of variables comprised household characteristics, which included household size, educational level of the head of household, and distance from the nearest health facility as well as the nearest provincial-level referral hospital. We also included households’ economic status [37,38,39]. All these variables were entered into the logit model, using the forward stepwise entry function in SPSS; if the probability of a variable’s score statistic was < 0.05, it was included; conversely, it was removed if the probability was > 0.1. The stepwise entry-removal of the various explanatory variables allowed the identification of those that had a statistically significant influence on the probability of determining CHE. The following variables were included in the model as categorical variables: a) households that incurred expenditure for treating chronic illness and hospitalization (20 and 40% threshold), b) households in which at least one member visited a private facility (20% threshold), c) income groups (20 and 40% thresholds), d) household distance from the nearest provincial-level referral hospital (40% threshold), and e) household size (20 and 40% threshold). The probability of CHE was calculated using the logit model [40], and the model’s goodness-of-fit was assessed via the Hosmer-Lemeshow test [41].
