The association between health literacy, private health insurance, and medical expenditure in South Korea | BMC Health Services Research

Study design

This is a cross-sectional study examining the associations between HL level, PHI status, and annual OOP medical expenditure using the 2021 KHPS data.

Data source

This study utilized data from the 2021 KHPS. The KHPS is a nationally representative annual survey jointly conducted by the National Health Insurance Service and the Korea Institute for Health and Social Affairs. It collects comprehensive data on healthcare utilization, medical expenses, health behaviors, and health-related perceptions from a representative sample of Korean households.

The KHPS uses a two-stage stratified cluster sampling design. The first stage stratification is based on geographic regions (17 cities and provinces) and urbanization level (urban/rural). The second stage involves systematic sampling of households within the selected sample enumeration districts. In 2021, the KHPS collected data from 14,844 individuals.

Study population

From an initial sample of 14,844 individuals, participants were excluded if they lacked age information, were under 19 years old, or over 65 years old, resulting in a subset of 7,014 eligible participants. The exclusion of individuals over 65 years was justified by the high likelihood that their enrollment in PHI, including indemnity insurance, is often denied due to health-related reasons, irrespective of their HL (the primary variable of interest). According to the General Insurance Association of Korea’s disclosure, indemnity insurance generally limits enrollment age to 65–70 years [34]. Subsequently, from this eligible group, individuals who did not respond to the HL assessment or answered ‘Don’t know’ to any of the 16 items in the questionnaire were further excluded (5,965 participants). Finally, those who did not respond to questions regarding self-rated health (SRH) and chronic conditions were also excluded from the study population. Consequently, the final study population comprised 5,469 participants.

Variables

Outcomes

As outcome variables, we used PHI status (including indemnity insurance status and number of PHI policies) and annual OOP medical expenditure. Annual OOP medical expenditure was calculated as the sum of pocket payments for outpatient visits, hospitalization, and emergency room use, excluding insurance premiums.

Main independent variable

The main independent variables were HL level and PHI status, with PHI status serving a dual role—as an outcome variable in relation to HL level and as a main independent variable in the analysis of annual OOP medical expenditure. HL was assessed using the European Health Literacy Survey Questionnaire (HLS-EU-Q16), which consists of 16 items covering various aspects of understanding and utilizing health information. Responses were scored on a 4-point Likert scale: (1) Very difficult, (2) Difficult, (3) Easy, and (4) Very easy. HL scores were calculated by summing up responses to 16 questions, where responses of ‘very difficult’ or ‘difficult’ were scored as 0 points, while ‘easy’ or ‘very easy’ were scored as 1 point. Following previous studies, the total HL scores were categorized into three levels: inadequate (0–8 points), problematic (9–12 points), and sufficient (13 points or higher) [35].

Covariates

As covariates, we considered sociodemographic factors including sex, age, education level, and household income quartile, as well as health-related factors including SRH status and number of chronic diseases. Age was treated as a continuous variable ranging from 19 to 65 years. Education level was categorized as middle school or lower, high school, and college or higher. Household income was classified into quartiles after adjustment for household size.

Health-related factors were assessed using two measures. SRH was categorized as good, fair, or poor. The number of chronic conditions was classified into three groups: none, one to two conditions, and three or more conditions. Chronic conditions were classified based on 31 disease categories provided in the KHPS, including hypertension, diabetes, chronic hepatitis, and others.

Statistical analysis

Descriptive statistics were calculated for PHI status across different categories of variables, including means, standard error, frequencies, and p-values (Table 1). Binary logistic regression models were used to estimate crude and adjusted odds ratios for indemnity health insurance enrollment associated with HL. Multinomial logistic regression was employed to analyze the relationship between HL and the number of PHI policies held, presenting both crude and adjusted results. Tables 2 and 3 both present the results of the unadjusted model and adjusted for indemnity health insurance enrollment and the number of PHI policies held.

The association between annual OOP medical expenditure, PHI status, and HL level was examined using Gamma generalized linear models with log link function (Tables 4 and 5). We constructed four models: Model 1 included only PHI status, Model 2 included only HL level, Model 3 included both PHI status and HL level, and Model 4 additionally included sociodemographic and health-related covariates.

1. Reference categories: Health literacy level (inadequate), Sex (Men), Household incomes (Q1), Educational level (lower than high school), Self-rated Health (Good), Number of chronic diseases (None) 2. All monetary values are presented in US dollars, converted from Korean won using the average exchange rate for 2021 (1,293.68 KRW/USD) 3. AIC for model 1: 67,976.63, model 2: 68,106.89, model 3: 67,961.07, model 4: 67,721.73

To address the non-normal distribution and right skewness typical of healthcare expenditure data, we conducted comprehensive model diagnostics. Using number of PHI policies as the test variable, we performed the Modified Park test to determine the appropriate distribution family and explored various modeling approaches including zero-inflated negative binomial (ZINB) models and Gamma generalized linear models (GLM) with log link function for number of PHI policies.

Based on model diagnostics, we selected the Gamma GLM with log link function as our final model. This choice was supported by better model fit (AIC = 67,721.7273) compared to the ZINB model (AIC = 69,798.3464). While ZINB models provided insights into zero-inflation patterns, the Gamma GLM was more appropriate for our research objective of analyzing total out-of-pocket medical expenditure.

All analyses were conducted using sampling weights to ensure national representativeness, except for Supplementary Table 1. The KHPS incorporates non-response adjustment based on household characteristics and post-stratification calibration. During our study, we recalibrated the sampling weights to account for our specific study population. The original sample weights were recalibrated to account for the reduced sample size after applying inclusion/exclusion criteria, while maintaining the representativeness of the target population. Complete case analysis was performed. Supplementary Table 1 presents the unweighted distributions of study variables, while all other analyses used sampling weights to ensure national representativeness. Supplementary Table 2 presents the unweighted and weighted distributions of PHI status, and other variables according to HL level. Supplementary Tables 3, 4, and 5 present the results of multicollinearity diagnostics: Cramer’s V correlation matrix of categorical variables, Point-Biserial correlations between age and categorical variables, and Variance Inflation Factors from the Gamma GLM analysis, respectively. All statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC, USA). Statistical significance was set at p < 0.05.

Ethical considerations

The study used publicly available de-identified data from the KHPS. Therefore, it was exempted from Institutional Review Board review.