Description:
Introduction Healthcare expenditure, a common input used in health systems efciency analyses is afected by population age structure. However, while age structure is usually considered to adjust health system outputs, health expenditure and other inputs are seldom adjusted. We propose methods for adjusting Health Expenditure per Capita (HEpC) for population age structure on health system efciency analyses and assess the goodness-of-ft, correlation, reliability and disagreement of diferent approaches. Methods We performed a worldwide (188 countries) cross-sectional study of efciency in 2015, using a stochastic frontier analysis. As single outputs, healthy life expectancy (HALE) at birth and at 65 years-old were considered in different models. We developed fve models using as inputs: (1) HEpC (unadjusted); (2) age-adjusted HEpC; (3) HEpC and the proportion of 0-14, 15-64 and 65+years-old; (4) HEpC and 5-year age-groups; and (5) HEpC ageing index. Akaike and Bayesian information criteria, Spearman's rank correlation, intraclass correlation coefcient and information-based measure of disagreement were computed. Results Models 1 and 2 showed the highest correlation (0.981 and 0.986 for HALE at birth and HALE at 65 years-old, respectively) and reliability (0.986 and 0.988) and the lowest disagreement (0.011 and 0.014). Model 2, with ageadjusted HEpC, presented the lowest information criteria values. Conclusions Despite diferent models showing good correlation and reliability and low disagreement, there was important variability when age structure is considered that cannot be disregarded. The age-adjusted HE model provided the best goodness-of-ft and was the closest option to the current standard.