Multilevel analysis of individual heterogeneity and discriminatory accuracy (MAIHDA) is a leading quantitative approach for intersectionality-informed health research, but most applications analyse binary or cross-sectional outcomes, ignoring event timing. We applied a multilevel survival (shared frailty) model within the MAIHDA framework to examine intersectional disparities in time-to-diagnosis of hypertension.
Using 2019 Korean Community Health Survey data (n=228 632), we defined intersectional strata by sex, education, income and residential area. Three survival specifications were implemented: accelerated failure time (AFT), parametric proportional hazards (PHs) and semi-parametric Cox PH models, each with stratum-level random intercepts (shared frailty terms). Between-stratum variance was summarised with the variance partition coefficient (VPC) where estimable and proportional change in variance quantified fixed-effect contributions. Stratum-specific random effects were compared across model types to assess ranking stability.
Between-stratum variance was small overall (AFT VPC: 1.8%), but several strata deviated markedly from the grand mean. Strata with low education and low income were diagnosed earlier than average, while high-education, low-income strata were diagnosed later. Geographic context modified these effects. Time-to-diagnosis patterns often diverged from prevalence patterns. Across models, random effect estimates and ranks were highly correlated (Spearman’s >0.97), though some middle-ranked strata shifted by up to six positions.
Applying a multilevel survival (shared frailty) model within MAIHDA enables examination of when disparities emerge, not just whether they exist. This approach retains MAIHDA’s interpretability while leveraging time-to-event data, offering advantages in settings with incomplete follow-up or irregular observation windows.