A new robust statistical model for interpretation of differences in serial test results from an individual

Background: Population-based reference intervals have very limited value for the interpretation of laboratory results when analytes display high biological individuality. In these cases, the longitudinal evaluation of individual results using the reference change value (RCV) is the recommended approach. However, the traditional model for RCV calculation requires a Gaussian frequency distribution of data and risks to overestimate the parameter if a correlation between within-subject serial measurements is present. Methods: We propose and validate an alternative non-parametric statistical model for interpretation of differences in serial results from an individual, overcoming data distribution and correlation issues. Results: After describing the traditional and newly proposed statistical models, we compared them with each other using a simulation on three specific analytes displaying different concentration distributions in biological setting. We demonstrated that when analyte concentrations followed a Gaussian frequency distribution, as in the case of glycated hemoglobin, both methods can be used equally. On the contrary, if analyte concentrations present a bimodal (e.g., chromogranin A) or skewed (e.g., C-reactive protein) distribution, the information obtained by two statistical methods is different. Conclusions: The proposed statistical approach may be more appropriate in assessing difference between serial measurements when individual data are not normally distributed.