Accepted for/Published in: JMIR Public Health and Surveillance
Date Submitted: Jul 6, 2026
Date Accepted: Oct 1, 2026
Can Public-Statistics-Based Synthetic Personas Support Small-Area Public Health Surveillance? A Sociodemographic Benchmark Consistency Study Using CHS and KOSIS Data
ABSTRACT
Background:
Background:
Synthetic population and persona datasets can provide artificial individual-level records without exposing personal data. For surveillance use, however, they must preserve the joint profiles that define vulnerable groups, not only marginal distributions.
Objective:
This study evaluated whether a public-statistics-based Korean synthetic persona dataset is consistent with Korea Community Health Survey (CHS) and Korean Statistical Information Service (KOSIS) benchmarks for auxiliary small-area public health surveillance.
Methods:
We compared synthetic persona rows with CHS 2021-2025 survey benchmarks and selected KOSIS municipal-level indicators. Harmonized variables included region, age group, sex, marital status, living-alone status, education, economic activity, and occupation. We assessed marginal, joint, high-dimensional, and policy-relevant vulnerable profiles using percentage-point differences, total variation distance (TVD), Jensen-Shannon divergence, sparse-cell sensitivity, temporal-stability diagnostics, KOSIS regional correlations, and 500-replicate stratified cluster bootstrap intervals for main TVD and Jensen-Shannon divergence estimates.
Results:
Persona rows closely matched CHS benchmarks for basic margins, including age group (TVD 0.007; bootstrap 95% CI 0.005-0.008), living-alone status (TVD 0.008; 95% CI 0.007-0.009), sex (TVD 0.010; 95% CI 0.009-0.011), and province (TVD 0.010; 95% CI 0.009-0.011). Larger differences appeared for education (TVD 0.049; 95% CI 0.048-0.051), marital status (TVD 0.050; 95% CI 0.049-0.052), and occupation (TVD 0.183; 95% CI 0.182-0.184). Excluding occupation, TVD increased from age by sex (0.016; 95% CI 0.015-0.017) to the municipal-level age-sex-living-alone-education profile (0.118; 95% CI 0.117-0.119) and the national high-dimensional profile (0.138; 95% CI 0.137-0.139). High-dimensional discrepancies persisted after sparse-cell restrictions; for the national high-dimensional profile, TVD remained 0.222 after restricting to CHS cells with an n of 100 or more. KOSIS comparisons showed strong regional ranking consistency for aged-population share (Pearson 0.985; Spearman 0.980).
Conclusions:
The synthetic persona data captured basic demographic margins and regional aging-related vulnerability patterns, but fell short for high-dimensional profiles relevant to subgroup surveillance. They should not be used as stand-alone substitutes for observed surveillance data. A more defensible role is auxiliary use in scenario analysis, vulnerability profiling, surveillance planning, and calibration workflows when benchmark differences and uncertainty are reported.
Citation
Request queued. Please wait while the file is being generated. It may take some time.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.