Calibration and validation of linearity in chromatographic biopharmaceutical analysis

• Published in: Journal of pharmaceutical and biomedical analysis 1991, Vol.9 (10), p.911-918
• Main Authors: Karnes, H. Thomas, March, Clark
• Format: Article
• Language: English
• Subjects: Analysis; Biological and medical sciences; Calibration; Chromatography - methods; General pharmacology; Linear Models; linear regression; Medical sciences; Pharmaceutical Preparations - analysis; Pharmacology. Drug treatments; quality control; Reproducibility of Results; standardization; validation
• ISSN: 0731-7085; 1873-264X
• DOI: 10.1016/0731-7085(91)80022-2

Calibration in chromatographic biopharmaceutical analysis is a major determinate of method performance and many methods have been proposed to evaluate an appropriate calibration model, to determine the linear range and to evaluate the goodness of fit. Ten chromatographic bioanalytical methods have been evaluated in this work by observation of concentration—response curves, linearity plots, calculation of concentration residuals, correlation coefficients and lack of fit analysis. These methods were applied to univariant linear regression, weighted regression, polynomial regression and power fit models in order to determine the most appropriate way to establish and evaluate calibration functions. It was found that weighted linear regression provided the most appropriate calibration function for eight of the 10 methods studied, whereas unweighted regression and the power fit model proved appropriate for one each of the other two methods. The choice of calibration function was best accomplished through observation of calculated concentration residuals. Linearity and sensitivity plots were of little value for assessment of linearity through the selected calibration range if conventional (±5%) tolerance limits are employed. Validation of the calibration model can be accomplished by demonstrating the concentration residuals and the slope of the log concentration—log response plots are within reasonable tolerance limits or by lack of fit analysis. Correlation coefficients were demonstrated to be of little value for this purpose and the quadratic approach to linearity validation was in disagreement with other methods in four of the 10 methods evaluated.