It is necessary to judge the significance of changes in laboratory test results, especially in health screening. For this purpose, the reference change value (RCV) was proposed, which is the (1- α) 100% confidence limit of differences between any two measurements: RCV = √2 z(α) x CV(I), where CV(I) represents intra-individual CV and α = 0.05. However, RCV is not commonly employed because: (1) it assumes constant CV(I) regardless of test levels, and (2) it often results in conservative judgements about the changes due to the blind use of 95% as its confidence probability (CP). Recently, we evaluated the level dependency of CV(I) in common laboratory tests and sought an appropriate CP for computing RCV through systematic analysis of a large long-term health-screening database. The dataset used contained data from approximately 14,000 individuals who underwent annual health-checks repeatedly. None of them were taking any medications or showed unnatural changes in BMI. The level of dependency of CV(I) was clearly observed for test items which showed skewed, logarithmic normal distributions but not for those with normal distributions, indicating the need to compute RCV according to the test level for the former items. To assess the choice of practical CP for RCV, we introduced a metabolic syndrome score (sMS) derived by logistic regression analysis. AsMS was used as an external criterion of changes in the nutritional status of individuals in relation to changes in laboratory tests. We evaluated the sensitivity and specificity of RCV at various CPs for detecting significant changes in ΔsMS. The analysis revealed that CP of 80-90% for computing RCV markedly enhanced the sensitivity of detecting ΔsMS without compromising the specificity. We provide a table listing appropriate RCVs for typical levels of common screening tests obtained from the analysis.