Method Validation Basics for a Small Laboratory

Method validation is the documented demonstration that a procedure does what you intend to use it for. That is the whole definition, and the word intend carries the weight: validation is judged against a stated purpose, not against an absolute standard, which is why a method can be entirely valid for one question and useless for another. Every widely available treatment of the subject is written for a regulated site preparing a submission, and the resulting programmes are correctly scaled for that context and absurd outside it. This page sets out what the parameters mean, which of them a given method actually needs, and what a proportionate programme looks like for a laboratory of three people.

What validation claims, and what it does not

A validated method carries one claim: within a defined range, applied to a defined type of sample, this procedure yields results with characterised accuracy and precision, and is not confounded by the things likely to be present alongside the analyte. Read that carefully for the qualifications, because they are where methods fail in practice. Applied outside its range, or to a sample matrix it was never tested against, a validated method has no validated status at all — validation does not travel with the procedure, it travels with the procedure applied to a defined problem 1.

Two things validation does not do. It does not make results correct: it characterises how wrong they are likely to be, which is a different and more useful thing. And it does not remain true indefinitely. A validated method on a new instrument, with a new column, a different reagent supplier or a different analyst is a method requiring at least partial reverification, which is why routine checks exist alongside the initial exercise 2.

The purpose statement is the part small laboratories skip and the part everything else depends on. Write it in one sentence before designing any experiments: this method determines the chromatographic purity of a lyophilised peptide solid, between 90 and 100 per cent, for the purpose of confirming supplier certificate values on receipt. That sentence sets the range, tells you which parameters matter, and tells you when the method is being used outside what it was shown to do.

The parameters, defined plainly

The standard set is small and the definitions are less forbidding than the vocabulary suggests. Each answers one question about the method.

Two distinctions inside that table are worth stating separately, because collapsing them is the commonest conceptual error in validation. Accuracy and precision are independent: a method that returns 94.0, 94.1 and 93.9 per cent on a 98.0 per cent reference is superbly precise and clearly biased, and no amount of replication will reveal that, because replication measures precision only. Detecting bias needs an external reference — a certified material, a known addition, an independent method — and nothing else will do.

The second is that linearity is not the correlation coefficient. A coefficient above 0.999 is routinely quoted as proof of linearity and proves almost nothing, because it is insensitive to exactly the curvature that matters at the ends of a range. Plot the residuals instead: if they show structure rather than scatter, the relationship is not linear however good the coefficient looks 3.

Identity or quantity: which parameters you need

The parameters are not a checklist to be completed. They are selected by what the method is for, and the sharpest division is between methods answering is this the right substance and methods answering how much of it is there.

An identity method needs specificity and essentially nothing else. If a mass spectrum confirms the expected molecular mass and no other species dominates, linearity is irrelevant — the method is not producing a number to be interpreted on a scale. Effort spent demonstrating recovery for an identity test is effort wasted, and the corresponding risk is under-specifying: an identity method that cannot distinguish the analyte from a closely related sequence, a deletion product or an oxidised form has failed at its only job 1.

A quantitative method needs the full set, weighted by consequence. Accuracy and precision are load-bearing because the number will be compared against a limit. Linearity and range matter because samples will not all sit at one concentration. Specificity remains essential, since a co-eluting impurity inflates the result without any visible sign. Robustness matters most for methods that will run for years across changing columns and analysts, which describes most methods in a small laboratory better than it describes methods in a large one.

Limit tests — is the impurity below one per cent — sit between the two, and need specificity and a demonstrated detection or quantitation limit, but not full linearity across a wide range. Getting this selection right is where most of the available savings are: the parameters you can justify omitting are usually more numerous than the ones you must demonstrate 2.

Full validation, verification, and what each costs

A method you developed yourself needs full validation, because nothing is known about it beyond your own experience. A method published in a pharmacopoeia or a peer-reviewed paper has already been validated by somebody else, and what you owe is verification: a demonstration that it performs as published in your hands, on your instrument, with your samples. Verification is a fraction of the work and is entirely legitimate — the published validation is evidence, and repeating it in full discards evidence you already have 5.

The scoping decision for partial revalidation should be written down as a short rationale, not left implicit. A new column lot plausibly affects specificity and retention, and does not plausibly affect the linearity of the detector response. Recording that reasoning means the decision can be defended later, and prevents the two opposite failures: revalidating everything after every change, which nobody sustains, and revalidating nothing, which nobody notices until a result is questioned 4.

A proportionate programme, and the daily version

Proportionality is the whole point for a small operation, and it is not a euphemism for doing less than is honest. It means matching evidence to the decisions a result will support. A method screening incoming material for gross discrepancy against a certificate carries a decision of low consequence and easy reversibility, and does not need the evidence a release decision needs. Scope to the decision, write down the scoping argument, and the resulting programme will be both defensible and finishable.

1. Write the purpose statement in one sentence: analyte, matrix, range, and the decision the result will support.
2. Decide identity or quantity, and list only the parameters that decision requires. Record why each excluded parameter is not applicable.
3. Look for a published method first. If one exists, verify it rather than validating from scratch.
4. Establish specificity before anything else. If the method cannot separate the analyte from what accompanies it, the remaining parameters describe a number that means nothing.
5. Determine repeatability with six determinations at the working concentration, and intermediate precision by repeating on a different day.
6. Determine accuracy against an external reference: a certified material, a spiked recovery, or an independent method. Never infer accuracy from precision.
7. Define system suitability criteria from the validation data, using the performance actually observed rather than figures copied from a textbook.
8. Write the method as a controlled procedure, with the system suitability criteria inside it, and record the validation data as the evidence supporting it.
9. Set a trigger list for partial revalidation: new instrument, new column type, new analyst, new supplier of a critical reagent, or a drift in system suitability.

System suitability is the part that repays effort most, because it converts a one-off exercise into a continuing one. Validation says the method was capable on the days it was studied; system suitability says the system was capable on the day this sample was run. A short check before each sequence — resolution between the critical pair, repeatability across replicate injections, tailing, and response of a reference standard — gives a pass or fail before any sample result exists, which is exactly when you want to learn that the column has died.

The programme above is achievable in a small laboratory in a week or two of instrument time, and it produces something genuinely defensible: a stated purpose, a reasoned selection of parameters, evidence for each one retained, daily checks derived from that evidence, and a written trigger for revisiting it. That is not a diminished version of validation. It is validation, scoped to the questions actually being asked, which is what the term meant before it became a synonym for a filing cabinet.