OBJECTIVES: The National Health Insurance system in Taiwan has implemented a capitation program recently and its payment formulas were based on prior-year expenditures. This study seeks to determine the extent of health expenditure persistence over a 2-year period and the percentages of decedents who were high users in the year of death. METHODS: This study analyzed National Health Insurance Data for a national sample of elders 65 years and older from Taiwan’s National Health Interview Survey, 2005. High users were defined as the top 10% users and the proportion of their aggregated health expenditures to total health expenditures was determined. A transition probability matrix and logit models were estimated to predict expenditure persistence over a 2-year period. RESULTS: The top 10% users accounted for 55% of total health expenditures. Of the top 10% users in 2005, 39% retained this position in 2006. But expenditure persistence was the highest among the bottom 50% users, with 77% retained their position over 2 years. The percentage of decedents who were top 10% users was 54% in the year of death and 31% in the year preceding death. Prior expenditures and comorbidity burdens were the strongest predictors of persistence. CONCLUSIONS: Taiwan's National Health Insurance capitation payment formulas based on prior expenditures do not reflect the fluctuation in expenditure persistence among the highest users so that cost percentile ranks should also be considered for payment adjustments.