A new approach to evaluate regression models during validation of bioanalytical assays.
Singtoroj T., Tarning J., Annerberg A., Ashton M., Bergqvist Y., White NJ., Lindegardh N., Day NPJ.
The quality of bioanalytical data is highly dependent on using an appropriate regression model for calibration curves. Non-weighted linear regression has traditionally been used but is not necessarily the optimal model. Bioanalytical assays generally benefit from using either data transformation and/or weighting since variance normally increases with concentration. A data set with calibrators ranging from 9 to 10000 ng/mL was used to compare a new approach with the traditional approach for selecting an optimal regression model. The new approach used a combination of relative residuals at each calibration level together with precision and accuracy of independent quality control samples over 4 days to select and justify the best regression model. The results showed that log-log transformation without weighting was the simplest model to fit the calibration data and ensure good predictability for this data set.