Impact of calibration fitting models on the clinical value of chromogranin A.

Background: The clinical relevance of chromogranin A (CgA) concentrations depends on the analytical performance of the assay. The goal of the present study was to define the clinical involvements in CgA calibration models by evaluating the confidence intervals (Cls) for values from patients who were undergoing monitoring for disease. Methods: Thirty calibration curves for the CgA assay [immunoradiometric assay (IRMA), (CIS-BIO)] were built using linear regression (LR), and four-parameter logistic models were used to estimate Cls for patient concentrations. Results: We reported the inadequacy of the LR curve estimation procedure. We showed: 1) no evidence that the straight calibration line could fit the average responses, 2) non-constant and non-uniform variance of the replicated calibration responses. All tests performed in the analysis of variance and CI calculation for the calibration curve should be invalidated. The four-parameter logistic function yielded results for 16 curves only; this result could be due to the low number and inappropriate concentration of calibrators. This suggests that some aspects of the assay design should be reviewed. However, using the variance function estimated in this model, we could assess the CI for calibration curves and patient samples. Conclusions: We showed that the four-parameter logistic calibration model with estimated variance function should better support clinical interpretation of marker concentration changes in patients serially tested.