Large-scale retrospective evaluation of regulated liquid chromatography–mass spectrometry bioanalysis projects using different total error approaches

• Large-scale retrospective evaluation of GLP projects using total error approaches.
• Comparison of different total error approaches using real data of GLP projects.
• Risks of accepting unacceptable batches existed in regulated LC–MS bioanalysis.
• Total error approach should be adopted in regulated LC–MS bioanalysis.
• Different total error approaches usually led to similar conclusion.

The current approach in regulated LC–MS bioanalysis, which evaluates the precision and trueness of an assay separately, has long been criticized for inadequate balancing of lab-customer risks. Accordingly, different total error approaches have been proposed. The aims of this research were to evaluate the aforementioned risks in reality and the difference among four common total error approaches (β-expectation, β-content, uncertainty, and risk profile) through retrospective analysis of regulated LC–MS projects. Twenty-eight projects (14 validations and 14 productions) were randomly selected from two GLP bioanalytical laboratories, which represent a wide variety of assays. The results show that the risk of accepting unacceptable batches did exist with the current approach (9% and 4% of the evaluated QC levels failed for validation and production, respectively). The fact that the risk was not wide-spread was only because the precision and bias of modern LC–MS assays are usually much better than the minimum regulatory requirements. Despite minor differences in magnitude, very similar accuracy profiles and/or conclusions were obtained from the four different total error approaches. High correlation was even observed in the width of bias intervals. For example, the mean width of SFSTP's β-expectation is 1.10-fold (CV = 7.6%) of that of Saffaj–Ihssane's uncertainty approach, while the latter is 1.13-fold (CV = 6.0%) of that of Hoffman–Kringle's β-content approach. To conclude, the risk of accepting unacceptable batches was real with the current approach, suggesting that total error approaches should be used instead. Moreover, any of the four total error approaches may be used because of their overall similarity. Lastly, the difficulties/obstacles associated with the application of total error approaches in routine analysis and their desirable future improvements are discussed.