Analytical method transfer: acceptance criteria you can defend

Two laboratories run the same procedure on the same material and get results two per cent apart. Neither is necessarily wrong. The method moved as a document, but the standard, the instrument's gradient delay, the column lot and the calculation basis did not move with it, and any one of those is enough to open a gap the size of a specification window. A transfer is a capability demonstration, so the work is to make those variables visible before they become a deviation.

What is actually being transferred

Three things move between units and only one of them is written down in full.

• The procedure. Written, reviewable, and the part everyone assumes is the whole transfer.
• The standard. A specific lot with a specific assigned content on a specific basis. Results are calculated against it, so it is part of the measurement, not part of the paperwork.
• The tacit knowledge. Why the gradient needs four minutes of re-equilibration, which column lot was in use during validation, why the sample is sonicated for six minutes and not fifteen. None of it is usually in the SOP, and all of it changes results.

A correct SOP run with a different standard lot and a different dwell volume can still produce a number that is internally consistent, reproducible, and two per cent away from the transferring unit's.

Four routes and the evidence each produces

The route decides what evidence exists at the end. Choosing comparative testing and then discovering that the receiving system cannot meet the method's system suitability is a planning failure: the suitability run should have happened before the protocol was signed.

When a waiver is defensible

• The procedure is compendial and the receiving laboratory performs the verification the pharmacopoeia requires before putting it into routine use.
• The receiving unit already runs the procedure and holds release or stability data on it.
• The move is a location rather than a capability: same organisation, same instruments, same analysts, documented training and comparable data history.
• The procedure was developed under a characterised operable region and the receiving laboratory's conditions sit inside it, which turns the argument from experience into a parameter comparison.

Each of these is a claim about capability, and each has to read like one, with the instruments, the analysts and the dataset named. A waiver resting on the phrase "experienced laboratory" is a finding waiting to be written at the next inspection.

Building acceptance criteria from the method's precision

Two numbers decide a comparative transfer: how far apart the two means may sit, and how precisely each mean is known. The second is routinely omitted, which is how criteria get copied from an unrelated protocol and then either fail on noise or pass a real bias.

If the method's intermediate precision is s, in the same units as the result, and each unit produces n independent results, the standard error of the difference between the two means is s x sqrt(2/n). A margin narrower than about three of those standard errors will fail on noise. A margin far wider stops detecting bias that matters. Both failures sit in the same table.

Worked example. Assay by HPLC, specification 95.0 to 105.0 per cent of label claim, intermediate precision 0.6 percentage points, six independent preparations per unit. The standard error of the difference is 0.6 x sqrt(2/6) = 0.35 percentage points. With ten degrees of freedom the 90 per cent confidence interval on the difference has a half width of 1.812 x 0.35 = 0.63. A transferring unit mean of 99.5 and a receiving unit mean of 99.1 give a difference of minus 0.40 and an interval of minus 1.03 to plus 0.23, which lies inside a two sided margin of 2.0. That is an equivalence demonstration rather than a failure to prove a difference.

Impurity methods need the same arithmetic applied to a number near the reporting threshold, where relative precision is poor. Two adjustments follow. Evaluate agreement only above a stated level, and judge relative difference with an absolute floor, because 0.02 percentage points means nothing at a 0.10 per cent level and a great deal at 2 per cent. Where the available lots do not reach the level of interest, spike them so the comparison happens at a concentration the method can actually measure.

The standard is a variable, not a constant

Content results are proportional to the assigned content of the standard the calculation is anchored to. A standard assigned 97.5 per cent where the other unit's is assigned 97.0 shifts every result by roughly 0.5 per cent relative, which is a quarter of a 2.0 margin spent before any sample is weighed. Three mismatches account for most of the surprises.

The protocol should name the lot, its assigned content and its basis, and state whether results are corrected. When the receiving unit has to use its own lot, qualify that lot against a common primary standard and run a bridging experiment: both lots against one shared sample, in one laboratory. That single run separates the standard contribution from the site contribution and answers in a day what an unplanned investigation takes a week to reach.

Readiness checks before the first transfer injection

System suitability belongs in this list rather than in the transfer data. The receiving unit should reach the method's suitability criteria in a pilot run before the protocol's data are collected. A transfer whose first measurements are also the first injections at the new site is a validation exercise with no defined outcome.

Sample plan, analysts and run order

• Three lots at minimum, bracketing the specification so accuracy is tested where it is hardest rather than only at the target.
• Two or three independent preparations per lot with duplicate injections, so preparation and injection variance can be separated instead of pooled.
• A second analyst per unit where intermediate precision is part of the claim.
• Sample identity blinded where the workflow allows it, and one run order applied at both units.
• One sample set, shipped under documented conditions, so both units analyse material from the same containers.

When the numbers do not agree

Work in this order and stop at the first explanation that accounts for the whole difference:

1. System suitability at the receiving unit, including the calculations behind it.
2. The calculation itself: units, dilution scheme, any conversion factor applied twice.
3. The standard: weighed amount, assigned content, basis, and whether drying was performed as the certificate requires.
4. Instrument parameters read against the method text, not against the local template.
5. Column and mobile phase: lot, age, pH, and whether the aqueous phase was prepared the same way.
6. Sample preparation, including adsorption losses on the filter.

Discarding an outlier or averaging a repeat run needs a rule written before the data existed. A transfer investigated backwards, from the conclusion, produces a report that will not survive the first informed question.

What the protocol and the report have to contain

• Protocol, approved before testing: samples and shipment conditions, standard lots with assigned content and basis, instruments by identifier, acceptance criteria, statistical method, run order, roles, and the deviation route.
• Raw data by reference: chromatograms, suitability results and calculations for both units, not summaries.
• A statement of what was not compared and why, which is where a waiver of a single characteristic belongs.
• A conclusion that names the route taken and the evidence it produced, so the next reader does not have to reconstruct the decision.

What to ask a supplier, in transfer terms

• Assigned content with the basis stated, and the method used to assign it.
• Chromatographic purity and the impurity profile, with the method behind both.
• Water content and residual solvents, because they move the basis rather than the purity.
• Storage condition, retest or expiry, and the condition the material was shipped under.
• Lot-specific data, so both units can be supplied from one lot; where a second lot is unavoidable, a statement of how it was bridged to the first.

None of these questions is about price or speed. They are the inputs without which a transfer result cannot be interpreted, and the certificate fields are where most of them should already be answered. A laboratory holding the documents before the protocol is signed spends the transfer fortnight measuring the method.