ICH adopted M10: Bioanalytical Method Validation and Study Sample Analysis at Step 4 on May 24, 2022, and FDA announced it as final guidance for industry via a Federal Register notice on November 7, 2022. Three years on, most validation packages get the numbers right on full and partial validation — the ±15%/±20% accuracy criteria are muscle memory for any bioanalytical group. Cross-validation is where packages go wrong, and not because the math is hard. It is where a group borrows a number M10 never gave it.
What M10 actually harmonized
Before M10, sponsors worked from a patchwork: FDA's 2018 draft bioanalytical guidance, EMA's 2011 bioanalytical method validation guideline, and separate expectations from PMDA and other regions, each with its own emphasis and, in places, its own numbers. M10 replaced that patchwork with one guideline endorsed across ICH regions, covering both chromatographic assays (LC-MS/MS and related methods) and ligand-binding assays (LBAs) used for large-molecule and biomarker work. The result is one validation vocabulary for a regulatory CMC team running a global program instead of a study-by-study translation exercise between FDA's and EMA's prior expectations — and one vocabulary the PK data supporting an IND application now has to speak consistently across every lab that touches it.
Two categories with numbers, one without
Full validation applies when a method is developed and validated for the first time. Partial validation applies to a modification of an already-validated method — a new analyst, a reagent lot change, a small volume adjustment, an instrument swap within the same platform — and its scope is meant to track the size of the change, from a single accuracy-and-precision run up to nearly a full revalidation. Both categories inherit the same numeric backbone: accuracy within ±15% of nominal (±20% at the lower limit of quantification), precision at ≤15% CV (≤20% at the LLOQ), and at least 75% of calibration standards across a minimum of six concentration levels meeting those criteria. A team that has run either category before knows these numbers without looking them up.
- Full validation. A new method, validated against the complete parameter set — accuracy, precision, selectivity, sensitivity, stability, and the rest — before first use.
- Partial validation. A modification to a validated method, scoped to the parameters the change could plausibly affect, using the same numeric criteria as full validation for whichever parameters are re-run.
- Cross-validation. A comparison between two methods, or the same method in two laboratories, whose data will be compared or pooled — evaluated by statistical bias assessment against criteria the sponsor sets and justifies for that specific comparison, not a number M10 supplies.
Cross-validation is not a fourth validation with a smaller pass/fail bar. It is a different question — do these two data sets agree well enough to use interchangeably — and M10 leaves the answer to a documented, defensible judgment call. Why a borrowed number doesn't survive a review
Where the borrowed number comes from
Cross-validation shows up whenever a program compares data across methods or sites: a method transfers to a second laboratory mid-study, a platform changes between an early-phase and a pivotal study, or a multi-region submission pools PK data generated under nominally the same method in different labs. Facing that comparison without a number in the guideline to anchor to, teams reach for the closest one they know — the incurred sample reanalysis threshold (at least 67% of reanalyzed samples within ±20% for chromatographic assays or ±30% for ligand-binding assays), or a legacy regional criterion left over from the pre-M10 patchwork. Both are real numbers from real parts of the guideline. Neither was written for cross-validation, and citing one as the acceptance criterion answers a question the reviewer did not ask: not did you hit a number, but did you justify the number you used, before you generated the data it was meant to judge. That is the same risk-based discipline ICH Q9(R1) quality risk management asks for everywhere else in the quality system — a documented rationale in place of a default.
- Classify before you validate. Decide whether the work is full, partial, or cross-validation before writing any acceptance criteria — the category changes what a number even means.
- Write the bias assessment plan first. Pre-specify the statistical approach and the acceptance criteria for the comparison, justified by the assay and the decision it supports, before the comparative data exists.
- Don't borrow ISR's number. The ≥67% within ±20%/±30% threshold governs incurred sample reanalysis, not cross-validation — keep the two separate in the validation plan and in the report.
- Size ISR into the study plan early. For bioavailability, bioequivalence, first-in-human, and renal/hepatic-impairment studies, build the 10%-of-first-1,000-plus-5% sampling into the design, not into a post hoc data request.
None of this is exotic statistics — it is a classification discipline. Full and partial validation give a team fixed numbers because the question is whether a method meets an absolute performance bar. Cross-validation asks a comparative question, and M10 answers it with a process — pre-specify the bias assessment, justify the criteria, document the judgment — rather than a number, because no single threshold fits every method pair and every lab comparison equally well. Programs that write a defensible cross-validation plan before the comparative data exists get a validation package a reviewer can follow. Programs that reach for the nearest borrowed number get a package that answers a question nobody asked, and a finding that asks the real one.
Frequently asked questions
Does ICH M10 give a numeric acceptance criterion for cross-validation?
No. Full and partial validation carry explicit numeric accuracy and precision criteria, and incurred sample reanalysis has a defined pass rate. Cross-validation is the one category ICH M10 leaves without a fixed number: the guideline calls for a statistical assessment of bias between the data sets being compared, with acceptance criteria that are scientifically justified and pre-specified for that comparison rather than a threshold copied from elsewhere in the guideline.
What is the difference between partial validation and cross-validation under ICH M10?
Partial validation evaluates a modification to an already-validated method — it can range from a single accuracy-and-precision run to nearly a full revalidation, scaled to the extent of the change. Cross-validation applies when multiple bioanalytical methods, or the same method run in multiple laboratories, generate data that will be compared or pooled within or across studies; it asks whether the data sets agree, not whether either method independently passes a validation threshold.
When does ICH M10 require incurred sample reanalysis (ISR)?
ISR applies to bioavailability and bioequivalence studies, first-in-human and other early pivotal PK studies, and studies in patients with impaired renal or hepatic function. The sample size is 10% of the first 1,000 study samples plus 5% of any samples beyond that. Results are acceptable when at least 67% of reanalyzed samples fall within ±20% of the original value for chromatographic assays, or ±30% for ligand-binding assays.
Sources & further reading
- ICH. M10: Bioanalytical Method Validation and Study Sample Analysis — Step 4 Guideline (May 24, 2022). database.ich.org
- FDA. M10 Bioanalytical Method Validation and Study Sample Analysis — Guidance for Industry. fda.gov
- Federal Register. M10 Bioanalytical Method Validation and Study Sample Analysis; International Council for Harmonisation; Guidance for Industry; Availability (Nov. 7, 2022). federalregister.gov
This article is provided for general informational purposes and reflects the regulatory landscape as of August 2026. It is not legal or regulatory advice. Confirm current guideline text and regional implementation status with ICH, FDA, or qualified counsel before acting.
