Welcome to 2025! I hope the new year brings you joy, productivity, and progress in achieving your bioanalytical goals. One topic I hope we can reach a consensus on within our community is how to handle bioanalytical method cross-validations according to the ICH M10 guidelines. Over the holidays, I read a couple of publications that furthered the dialogue, and I thought I’d start the new year by drawing attention to this topic.
If you are familiar with ICH M10 (ICH Harmonized Guideline for Bioanalytical Method Validation and Study Sample Analysis, May 2022), you are likely aware of the extension from previous BMV guidance that addresses the assessment of bias between methods. Please refer to ICH M10 for further details on where cross-validations are required.
Typically, we are tasked with conducting cross-validations between two fully validated methods (within or between laboratories) when the data will be combined for a given study for regulatory submission and decision-making. Cross-validation requirement is not new, but previous guidance left it to the bioanalytical lab to stipulate in SOP or validation plan what criteria to use a priori. Most of us deferred to ISR criteria when comparing spiked QC (or study samples) data from both bioanalytical methods. However, multiple publications and case examples have highlighted that the ISR approach fails to identify underlying trends and biases between two methods. Essentially, it was convenient but not sufficient for the objective of a cross-validation assessment.
ICH M10 attempts to tackle this dilemma by proposing a statistical assessment of data from multiple methods to measure bias. However, it does not stipulate the acceptance criteria to judge cross-validation success. “And oh, how much as bioanalysts do we need our acceptance criteria!” Absence of acceptance criteria is counter to what we are used to in regulated bioanalysis. We are conditioned to a priori acceptance criteria to remove subjectivity in concluding our bioanalytical experiment results. However, this is where we are today with our ability to visualize bias (Bland-Altman and scatter plots) and even further quantify agreement between data sets with Deming regression and Concordance Correlation Coefficient, but we are not in agreement on what the acceptance criteria should be. The debate has reached the point: “Should we, or can we, have appropriate a priori acceptance criteria for cross-validation of bioanalytical methods?”
The two papers I’m referencing build upon the work of C. Gleason, Q. Ji, and E. Wickremsinhe (2020) and undoubtedly other publications. Both are recent papers, firstly M.S. Fjording, J. Goodman, and C. Briscoe (2024) and secondly I. Nijem, R. Elliott, J. Brumm et al. (2025) – see full references below. Both were directed to pharmacokinetic (PK) assays, but they take alternate positions when it comes to defining acceptance criteria and, just as importantly, who sets such criteria and conducts the statistical assessments.
Fjording, Goodman, and Briscoe argue that the context of the cross-validation and the resulting statistical data cannot be dissociated from the purpose of the study. As such, the clinical pharmacology and biostatistics team need to be involved in the design of the cross-validation plan and the proposed use of the data. Furthermore, they rationalize that experienced statisticians should define the appropriate statistical approach, assess any bias, draw conclusions, and even generate a separate cross-validation report. They make a strong case that the necessary statistical skills and tools are not within the realm of the typical bioanalytical laboratory and, particularly for CROs, the assessment of cross-validation data should be conducted by the sponsor or a third-party statistician.
There’s solid rationale in the Fjording paper conclusions including per the paper title that pass/fail criterion is inappropriate. However, I wonder if some sponsors outsourcing bioanalytical work will have the statistician resources on staff and require the bioanalytical provider to design and conclude the cross-validation data. This is where the proposal made by I. Nijem et al. offers an alternative and prescriptive approach to the topic.
The Nijem paper provides a standardized approach to cross-validation planning and a decision tree process for interpreting data. The approach is predicated on selecting enough samples (n>30) and concentrations that appropriately span the sample concentration range. Importantly, the approach described sets a priori acceptance criteria where initial assessment of equivalency is met if the 90% confidence interval (CI) of the mean percent difference of concentrations is within +/-30%. This is followed by determining if there are any concentration bias trends by assessing the slope in the concentration percent difference vs. mean concentration curve. Statistical processing was conducted in Microsoft Excel (with XLstat add-on software), enabling quantitative measurement of the 90% CI of the slope. This two-step approach to assessing equivalency is proposed as suitable for bioanalytical method comparison. Of course, it still leaves to debate when bias is determined what the impact on PK calculations will be. Two case studies accompanied the proposed approach with practical demonstrations of detecting concentration-related bias in one example.
I’m not a statistician, but I found the Nijem paper compelling in its ability to quantify bias. Where I was less confident was in the fundamental assumptions of the statistical treatments used and whether they hold as appropriate across bioanalytical studies and data sets in general. I hope this proposed approach is taken up for discussion by the bioanalytical community together with pharmacology/biostatisticians in a collaborative manner. We need to reach a consensus on how to conduct cross-validations of bioanalytical methods with an emphasis on interpreting the data.
Overall, I sense we are getting closer to a defendable position on managing regulated bioanalysis cross-validations that meets the intent of ICH M10 language. I look forward to hearing further input and invite your own critical review of the papers referenced.
References:
- C.R. Gleason, Q. C. Ji, E. R. Wickremsinhe; Evaluation of correlation between bioanalytical methods Bioanalysis (2020), 12(6) 419-426
- M.S. Fjording, J. Goodman, C. Briscoe; Cross-validation of pharmacokinetioc assays post-ICH M10 is not a pass/fail criterion Bioanalysis, 1-6 (2024) https://doi.org/10.1080/17576180.2024.2418284
- I. Nijem, R. Elliott, J. Brumm, L. Liu, K. Xu, R. Mendez, R. Hendricks, B. Wang, P. Siguenza; Cross validation of pharmacokinetic bioanalytical methods: Experimental and statistical design Journal of Pharmaceutical and Biomedical Analysis (2025) 116485
