Wanting care at a Korean medicine clinic and naming it as one’s main provider answer different questions. A survey may allow several preferred institutions but only one main place of care. A disagreement between those answers does not, on its own, establish that someone was unable to obtain the care they wanted.
I used older respondents’ records from the 2023 Korean Survey on Persons with Disabilities to match preference with regular-provider use within each person, then examine what changes when different preference ranks are counted. The work follows a preliminary conference poster. In the subsequent manuscript, I centered the analysis on the differing scope of the two survey questions.
Preferred institutions and a main provider
The preference item allowed respondents to rank up to three institutions where they would like periodic health-management services. The utilization item first asked about regular, continuous care and then asked those receiving it to identify their main institution. Counting a second-choice Korean medicine clinic alongside a first choice asks a different question from counting first choices alone.
I therefore kept the respondents and the provider measure fixed while changing the preference definition. One definition included Korean medicine hospitals or clinics at any of the three ranks; the other included only the first rank. This is a sensitivity analysis of definitions applied to the same survey, not a before-and-after comparison of changing use over time.
I did not subtract one population percentage from another. Each respondent was classified by both their stated preference and whether Korean medicine was their main regular provider. Someone who did not name it as their main provider might still use it occasionally or alongside another institution. The provider item cannot distinguish those possibilities.
What weighting can and cannot recover
The analysis concerns community-dwelling registered adults with disabilities aged 65 or older. I applied the supplied population weights rather than giving every respondent equal influence, and used province categories as analytical strata. Weights represent how much of the population each respondent stands for; strata enter the calculation of uncertainty.
The supplied file did not contain the original sampling-cluster or design-stratum identifiers. Respondents drawn from the same cluster can resemble one another, and that structure could not be fully represented. Applying weights does not reconstruct the complete original survey design. I kept that limitation in the manuscript so it remains visible when interpreting confidence intervals and p values.
For comparisons that considered demographic, disability and health characteristics together, I used a modified Poisson model to estimate adjusted prevalence ratios. These compare the prevalence of the response combination after accounting for the included characteristics; they are not interchangeable with odds ratios from a logistic model. I also prepared separate analyses using the alternative preference definition and a different treatment of blanks generated by questionnaire skip rules.
Reasons for non-use require another question
A difference between preference and main-provider use cannot identify whether transport, cost or the availability of institutions contributed. The survey separately asked about wanted but forgone hospital or clinic visits. I analyzed that unmet-healthcare measure separately from preference–provider discordance, rather than treating them as the same outcome. Cross-sectional self-reports also do not establish that a preference or disability characteristic caused a difference in use.
The subsequent manuscript is accompanied by the variable definitions, analysis code and aggregate tables. Individual survey records are not redistributed; a researcher repeating the analysis from source data must follow the provider’s application and use conditions. Specific estimates and their interpretation are not disclosed in this post yet.
The preliminary work presented at the conference has a separate presentation record. The subsequent manuscript was in review as of 3 September 2026. Once an online version of record and active DOI are confirmed, I plan to explain the results here as well.