Laboratory data reveal diverse distribution patterns: Mapping individual variability for personalized laboratory medicine

Abstract

Background: Laboratory data are frequently assumed to follow a normal or log-normal distribution. If neither of these distributions fits the data, non-parametric methods are typically applied. However, a better-fitting distribution may be available, and data analysis should not be restricted to only the normal or log-normal models. In this study, we aimed to analyze the empirical distribution of measurement results obtained from repeated sampling of individuals for a group of measurands and to identify the best-fitting distribution type(s). Method: Repeated measurements of glucose, creatinine, alanine aminotransferase (ALT), total bilirubin, and calcium over a 5-week period from 41 individuals were included in the study. The distribution of the data was analyzed using 105 distribution models to identify the best-fitting one, based on analyses performed with SciPy version 1.15.3. Results: Based on the Kolmogorov-Smirnov p-values, the best-fitting distributions were logistic for glucose and Laplace-asymmetric for ALT. For the remaining measurands-creatinine, total bilirubin, and calcium-the bestfitting distribution was the Levy-stable distribution. Conclusion: Rather than relying on predefined distribution types such as the normal or log-normal, a data-driven empirical approach should be adopted for laboratory data analysis. This approach involves evaluating the actual data to identify the best-fitting distribution, without imposing theoretical assumptions that may not accurately reflect the underlying true variation.