Method Validation Under ICH Q2(R2): What Each Test Proves
Share
A validation package usually arrives as sixty pages with a two-page summary stapled to the front. The buyer reads the two pages, sees eight parameters marked as compliant, and moves on to price. The trouble is that "compliant" is not a finding. Each of the eight parameters answers a separate question, and a package can be complete in form while leaving the buyer's actual question untouched.
The buyer's question is almost always narrower than the guideline's scope. It is not "is this method validated" but "can this method, in this laboratory, on this instrument, measure what my specification says it measures, at the level my specification cares about". ICH Q2(R2), revised in 2023, is more explicit about that framing than the 2005 text was. It also renamed one parameter and split another, which matters because the names are what appear in the summary.
Three jobs that get conflated
Validation establishes that a method is suitable for its intended use, from first principles, with data. Verification establishes that a laboratory can run a compendial method as written. Transfer establishes that a receiving laboratory produces results comparable with those of the originating one. Three documents, three questions.
For a pharmacopoeial monograph method the ordinary requirement is verification, not full validation. Full validation is owed to in-house methods, to modified compendial methods, and to any method used outside the range the monograph states. The distinction carries commercial weight: a supplier claiming fully validated methods for a monograph assay has either done work nobody asked for or relabelled verification as validation. Either way the summary is describing something other than what the customer assumes.
The transfer case sits one step further downstream and is where most of the disputes actually happen, because two laboratories running the same validated method will still disagree by a few tenths of a per cent, and nobody measured that gap before the shipment.
Eight characteristics, eight questions
The table below is the one worth keeping. The third column is where each parameter fails quietly, which is more useful than the definition.
| Characteristic | Question it answers | How it fails quietly |
|---|---|---|
| Selectivity | Does the signal belong to the analyte? | A co-eluting impurity with the same retention time |
| Accuracy | Is the reported number close to the truth? | Spike added after extraction, excluding the loss |
| Repeatability | Do six determinations agree? | Agreement demonstrated only at high concentration |
| Intermediate precision | Do analyst, day and instrument change the result? | Reported from a single analyst on a single day |
| Linearity | Is response proportional across the range? | Regression forced through zero to improve the coefficient |
| Range | Where has the method been demonstrated? | Range narrower than the specification limits |
| Detection and quantitation limits | How little can be seen, how little can be measured? | Quantitation limit above the impurity limit |
| Robustness | Does it survive a small deliberate change? | Listed in the summary, never actually tested |
Selectivity replaced specificity
The 2023 revision retired specificity as a characteristic and put selectivity in its place, keeping specificity as the stronger claim for methods that can discriminate the analyte from everything else likely to be present. The wording change is not decoration. Selectivity accepts that a method may respond to more than one substance provided the analyte's response can be attributed. Specificity asserts that nothing else responds at all.
For a bile-acid assay at 210 nm the honest description is selective. The detection wavelength sits where the carboxyl group absorbs, and several related substances absorb there too. Resolution from those substances is a chromatographic achievement demonstrated per run, not a property of the detector asserted once in a report. A summary that claims specificity at 210 nm without showing what was injected to challenge it has used the stronger word without the stronger evidence.
Accuracy is not recovery
Accuracy is expressed either as the difference between the mean result and the true value, or as recovery of a known addition. Both are acceptable, and they answer slightly different questions. Recovery tells you how much of a spike you found. Accuracy tells you how close the reported number is to the truth. A method can show 100.5 per cent recovery and still be biased, if the spike went in after extraction and the extraction loss was therefore excluded from the experiment.
For an assay the spike belongs before sample preparation, not after it. For an impurity method the spike belongs into the sample matrix at the specification level, because that is the level at which the method will be asked to make a decision. Spiking at a level chosen for the symmetry of the table is a common and quiet way of validating the easy part of the range.
Precision has two halves
Repeatability is one analyst, one instrument, one day, six determinations. Intermediate precision is the same method across analysts, days and, where relevant, instruments. Most summaries report the first and describe it as the second. That is why a customer's own repeat sometimes lands three tenths of a per cent away from the certificate and nobody is at fault.
The figure that matters commercially is the intermediate one, because the buyer's laboratory is not the supplier's and the gap between two laboratories is exactly what will be argued about if a result lands near a limit. A relative standard deviation of 0.3 per cent within a day and 0.8 per cent across days are both respectable for a titrimetric assay, but only the second predicts what happens in a dispute.
Range is narrower than the method's reach
Range is the interval where linearity, accuracy and precision have been demonstrated together. It is common to see a linearity study from 80 to 120 per cent of working concentration presented as though it were the range. If the impurity limit sits at 0.1 per cent and the method was validated from 0.05 to 0.15 per cent, the range is adequate. If it was validated from 0.5 to 1.5 per cent because those concentrations gave tidier peaks, the limit sits outside the demonstrated range and the summary will not say so.
One more trap. Range is not the same as linearity. A method can be beautifully linear between 0.5 and 1.5 per cent and still be unusable at 0.08 per cent, because the precision at the low end was never measured. Linearity describes the slope. Range describes where anybody has checked that the slope means something.
What to read in a validation summary
Four things, in this order. The statement of intended use, because everything downstream depends on it. The range, compared against the specification limits. The intermediate precision figure, and whether it came from two analysts. The selectivity experiment, and specifically what was injected to challenge it.
If those four hold, the rest is arithmetic. If any one is missing, the summary is describing a method rather than qualifying it, and the place to raise it is before the purchase order, when the answer is still cheap and the alternative supplier is still interested.
A third-party laboratory report can complement a validation package but cannot replace it, because validation attaches to a method and to the site that runs it. An outside laboratory's confirmation of one lot says nothing about the method's range, and range is where the arguments live.
Where this leaves a supply-chain partner
We quote against a specification that names its methods. Where a customer's method differs from ours, the bridge is a transfer study, and that takes longer than a quotation cycle. Saying so in the first exchange costs a paragraph. Discovering it after the first shipment is rejected on a 0.2 per cent difference costs a quarter.
If the validation summary in front of you does not state its intended use in one sentence, ask for that sentence before you read anything else. "Related substances in ursodeoxycholic acid" and "related substances in a bile-acid matrix" are not the same claim, and only one of them covers the drum on your dock.
Verification is not a lesser activity
There is a persistent assumption that verification is what a laboratory does when it cannot afford validation. The assumption is wrong and it causes unnecessary work. Verification asks a narrower question, deliberately: can this laboratory, with this equipment, reproduce the performance the monograph already claims? The answer is a set of results from a small number of runs, plus a statement of the conditions.
What makes verification defensible is that the reference point already exists. The monograph defines the method, the limits and usually the system suitability criteria. The laboratory's job is to show it meets them, not to establish them. A verification package that re-derives limits which the pharmacopoeia already sets has misunderstood its own scope, and it will take three times as long as it needs to.
Verification becomes inadequate the moment something changes. A different column chemistry, a modified extraction, a detection wavelength chosen for convenience: each turns the activity into a partial validation, and the partial version has to cover the characteristic that the change affects. The stability-indicating method discussion is the clearest case, because a change in selectivity there can hide a degradant while leaving every other parameter intact.
Where the raw data sits, and why it matters
A validation summary is a conclusion. The evidence is the chromatograms, the weighing records, the integration events and the calculations. Inspectors ask for the summary first and the raw data second, and the second request is the one that finds problems.
Three things are worth confirming before relying on a package: that the chromatograms carry the acquisition date and the instrument identifier, that integration changes carry a reason, and that the calculations in the summary can be reproduced from the raw values. All three are routine in a laboratory with a functioning data system and all three are absent in a laboratory that assembles summaries from spreadsheets.
The public frameworks that define these expectations are not hidden. The ICH quality guidelines set out the validation expectations and the data integrity expectations in separate documents that are meant to be read together. The EDQM publishes the pharmacopoeial general chapters that define the methods themselves, and the USP does the same for its own compendial sphere. A supplier citing one without the others is citing half a framework.
The practical consequence for a buyer is a single question that separates a real package from a folder: which instrument produced these chromatograms, and where are the integration events? A laboratory that can answer in one sentence has the data. One that cannot has the summary.
Frequently asked questions
Is verification enough for a pharmacopoeial method?
For an unmodified monograph method run in a laboratory that has demonstrated suitable proficiency, verification is the correct activity and full validation is not owed. The moment the method is modified, moved outside the monograph's stated range, or used for a purpose the monograph does not describe, that changes.
Does ICH Q2(R2) require robustness testing?
It treats robustness as a characteristic to be considered rather than a fixed list of experiments. In practice a robustness study is expected to name the parameters that were deliberately varied, the range of each variation, and whether the result still met the acceptance criterion. A summary that says robustness was satisfactory without naming what was varied has described an intention.
How many levels are needed to demonstrate accuracy?
Three as a working minimum for an assay, spread across the range rather than clustered at one hundred per cent. For an impurity method the levels should bracket the specification limit, because that is where the decision will be made. Six replicates at a single level demonstrates repeatability at that level, not accuracy.
Can a validation performed at another site be reused?
Not as validation. The data from the originating site is evidence about that site's instrument, analysts and reagents. A receiving laboratory needs a transfer study that compares results on the same lots, and the acceptance criterion has to be agreed before the experiment rather than after the numbers arrive.
What is the practical difference between LOD and LOQ in a report?
LOD is the level at which the presence of a substance can be inferred but not measured. LOQ is the level at which a value can be reported with defined precision and accuracy. A certificate that reports an impurity as below the limit of detection is making a weaker statement than one that reports a number, and the two should never be presented as equivalent.
Is a validation report a regulatory document?
It is part of the technical file that supports a specification, and it is one of the first documents an inspector asks for when a result is questioned. That is why the intended-use statement at the front matters more than the chromatograms at the back.