Abstract
Taking notice of shapes of distributions of clinical laboratory values, we propose an analysis process based on the power-normal distribution as a model which treats variation of clinical laboratory values. Based on the process, we can treat the background factors comprehensively, and estimate suitable reference intervals of the clinical laboratory values. Further, we can consistently and flexibly treat outliers of the clinical laboratory values, classify healthy group and disease group, compare the reference intervals based on this model with traditional ones, and evaluate information loss which results from traditional estimation methods based on asymptotic behavior of estimates of reference limits. Practically, we applied this model to the medical examination data collected at a clinic K in 2003, and estimated reference intervals. As a result, almost all reference intervals were estimated much wider than traditional ones. This result suggests that it is risky to apply traditional masked reference intervals to relevant receivers of clinical diagnosis or actual subject group.
