Objectives: To compare medical expenses among regions in Japan, the "regional difference index" of National Health Insurance has been used. The index is formulated as a ratio of observed to expected numbers. However, it has large variability such as in the standardized mortality ratio (SMR) in small populations. To circumvent this problem, we propose an alternative index based on Bayesian methods.
Methods: Regional medical expenses were assumed to have a log normal distribution and be derived from the conventional regional difference index as a statistical estimator. Under the assumed distribution, we then considered a full Bayes estimator for the index. The data for 2003-2005 were used for a comparison between the proposed Bayesian index and the conventional index.
Results: Under the assumed lognormal model, we could define the conventional index as an estimator for the expenditure level in the region. We showed that it has a large variability for small populations. The proposed Bayes estimator could solve this problem.
Conclusions: The proposed index based on Bayesian inference could stably estimate the level of regional medical expenses. We therefore suggest that a more appropriate discussion of regional differences could be given using this Bayesian index than with the conventional one.