Comparison of classical and clustering based reference interval calculations for Aspartate Aminotransferase, Alanine Aminotransferase and Gamma-Glutamyl Transferase in serum

Abstract

Objective: There are fixed rules applied to determine the reference intervals (RIs) of the biochemical tests. However, these rules lack for identifying subgroups within the reference population. Therefore, we suggest the clustering method, which determines the sub-groups by taking the correlations between the variables into account in the RIs calculations. In our study, it is aimed to compare the results of the RIs based on the clusters analysis with the results of the conventional method. Methods: The individuals who applied Ankara Duzen Laboratory for the check-up with normal Ultra Sono Grafi (USG) in 2012-2014 and who have had Aspartate Aminotransferase (AST), Alanine Aminotransferase (ALT) and G-Glutamyl Transferase (GGT)

(U/L) results were included in this study. We have excluded the repeated applies of patients, only analyzed the first apply to the laboratory. Reference individuals are composed of 883 people. (610 male, 273 female, 18-70 years). Non-parametric methods were used to determine reference intervals. Fuzzy C-Means clustering method was used to identify sub-groups. Results: AST, ALT, GGT measurements for all of the check-up individuals were determined by non-parametric method for the three subgroups specified after the Fuzzy-C-Means clustering method and the entire group. According to the reference intervals obtained, the third sub-cluster derived from the group intended to be used as the reference population was observed as a cluster that is narrower, and has similar properties of the actual reference population. However, when the correlations between the tests in the sub-groups are considered, the correlations between GGT-ALT-AST have been observed to be higher while the correlation level between ALT-AST in the group proposed as a real reference population does not change. Conclusion: In the reference limit studies, instead of the determination of the reference interval for a single group designated as the reference population, we think that, the subgroups which are homogeneous within itself, heterogeneous between themselves should be set in in this group. In determining multiple sub-groups, the relationship between more than one test need to be taken into consideration, and the effect of clustering should be investigated.