When the reference standard is the suspect: investigating an OOS result without invalidating good data
An out-of-specification result is a fork: the material is wrong, or the measurement is. The standard sits on both sides of it, so the first hours decide whether good data get invalidated. This works through the diagnostic patterns, the arithmetic that separates bias from uncertainty, and the four comparisons that actually discriminate.
On this page
Re-injection answers a different question
An out-of-specification result is a fork: either the material does not meet the specification, or the measurement does not describe the material. A reference standard sits on both sides of it, because it is a sample in its own right and also the ruler every other sample is measured against. That is why an investigation which opens with a re-injection usually closes without an answer.
Re-injection repeats a measurement against the same calibration, so it narrows random dispersion and cannot detect bias. A standard whose assigned content is 1.2 per cent too low returns the same 1.2 per cent error six times, with a relative standard deviation of 0.2 per cent that looks like excellent precision. The numbers agree with each other and all of them are wrong. The FDA guidance on out-of-specification results orders the work the other way round: an initial assessment of the accuracy of the laboratory data, made before the test preparations are discarded, with the hypotheses tested against those same preparations.
Three ways a standard goes wrong, and one you cannot see
The modes fail in different places, and the response of the standard itself separates them most cheaply.
| Mode | What has changed | Effect on the result | Visible in the chromatogram |
|---|---|---|---|
| Material changed | Content in the vial: water uptake, residual solvent loss, degradation, photolysis | Reads high; the weighed mass carries less analyte than the certificate states | Response per microgram falls against history; small new peaks |
| Assigned value misapplied | The number entered in the calculation: basis, salt factor, unit, transcription | Shifts by a constant factor, either direction | Clean chromatogram; response per microgram unchanged |
| Standard preparation | Incomplete dissolution, adsorption on a filter, wrong flask, solution left standing | Reads high; the standard is weaker than the calculation assumes | Standard area low for the mass taken; reprepare to confirm |
The fourth mode is a certificate that is wrong about its own material, and nothing measured inside your laboratory detects it. A single-point calibration against a misassigned standard reproduces the misassignment with perfect fidelity, more convincingly so the better the precision. Only an independent value exposes it: a second lot, a second supplier, an orthogonal method, a proficiency test. Hence the one record worth keeping across lots is response per microgram, plotted as a control chart with limits at three standard deviations of pooled injection repeatability. A point outside the limit means the material or the preparation moved; six points on one side of the centre line means something systemic moved, such as the assigned value in use or the balance.
Read the pattern before you run anything
| What the data show | Most likely | The one measurement that separates it |
|---|---|---|
| Every sample off by the same factor from the first injection | The calibration: the assigned value in use, or the standard preparation | Recompute the existing chromatograms with each candidate value |
| Standard response per microgram below its chart, new peak near the main one | The material has changed | Prepare from an unopened vial of the same lot; compare response per microgram |
| One sample fails while standard and other samples sit on target | That sample: homogeneity, spiking, dilution, vial | Re-prepare only that sample from a fresh portion |
| Failure appears progressively along the sequence | The instrument: lamp, autosampler, column, mobile phase loss | Re-inject the same preparations in reverse order |
| Retention and shape moved, area normal | The separation, not the amount | System suitability: resolution against the nearest impurity |
The pattern that costs most time is none of these: everything slightly low, every run, by an amount that changes with the analyst. That is technique in the standard preparation, and more careful reading of the same data will not find it.
The arithmetic that decides whether the standard can explain the gap
Take a standard weighed at 10.02 mg, assigned content 98.7 per cent as is, expanded uncertainty 0.4 per cent at k = 2, made to 10.00 mL.
C = 10.02 x 0.987 / 10.00 = 0.9890 mg/mL, with an expanded uncertainty of 0.0040 mg/mL, which is the 0.4 per cent.
Now suppose the analyst uses the other number on the same certificate, the 99.9 per cent purity from the chromatographic assay: C = 10.02 x 0.999 / 10.00 = 1.0010 mg/mL. The ratio 1.0010 / 0.9890 is 1.0122, so every result in the run is 1.2 per cent high.
Two conclusions follow without any new measurement. First, the size rules out an uncertainty explanation: 1.2 per cent is three times the expanded uncertainty on the certificate, so this is bias, and bias is found by comparison, never by repetition. Second, the direction says only that the value used in the calculation is above the true content of the material. A misapplied basis does that, and so does a standard that has absorbed water, so direction alone cannot separate them. Response per microgram does. If it has held its place on the chart, the material still contains what it contained and the fault is arithmetic. If it has fallen, the material changed and the arithmetic may be blameless.
Those candidates are not equal in cost. Recomputing existing chromatograms is free and settles the arithmetic before anything is prepared again. Re-injection separates the instrument. A fresh preparation from an unopened vial separates handling history. Only the independent comparison can clear a certificate, and it is the step most often skipped because it needs something from outside the laboratory.
Comparing two lots, and how many replicates the question needs
Two values, 99.1 and 98.6 per cent, each with an expanded uncertainty of 0.4 per cent at k = 2. The difference is 0.5 per cent, which sounds large for a reference standard. Each standard uncertainty is 0.2 per cent, so the standard uncertainty of the difference is 0.283 per cent and its expanded value is 0.566 per cent. The observed 0.5 per cent sits inside that interval; the two lots cannot be told apart, and a report calling the difference real is reporting an artefact of the certificates.
The paired comparison is stronger at the same cost. Dissolve both lots at the same nominal concentration and inject them alternately in one sequence. The certificate uncertainty is common to both and cancels in the ratio, leaving injection precision. At 0.3 per cent single-injection repeatability, six alternating pairs resolve about 0.35 per cent, finer than the 0.566 per cent the certificates allowed.
The general form is worth having: to detect a difference delta against single-injection repeatability s, the number of replicates should exceed (2s/delta) squared. At s = 0.3 per cent, a 1 per cent question needs one injection, 0.35 per cent needs three, 0.2 per cent needs nine. Most statements that the standard is fine rest on three injections, which answers a question at about 0.35 per cent. An investigation into a 0.15 per cent discrepancy resting on three injections has not tested its hypothesis, however many times the sequence is repeated.
What the certificate licenses, and what it does not
The instructions for use are part of the measurement, not a handling note. Pharmacopoeial leaflets are explicit: use the material as is, do not dry or desiccate before use, and where no drying conditions are stated the substance is used as received with the assigned content expressed as is. Drying a material whose content was assigned on an as-is basis does not purify it; it removes water the assigned value had already accounted for, and the material is then no longer suitable for the use it was established for.
The second half of the instruction is read less often. A pharmacopoeial reference standard is for immediate use: once the container is opened its entire content is to be used immediately, and further storage and re-use are not warranted. A vial opened six weeks ago, resealed and kept in a desiccator is therefore outside the producer statement although the batch is still valid. That is not a reason to discard it, but it is a reason to treat handling history as a candidate cause before blaming the material, and it is why the first question is what happened to this vial after opening, not whether it is in date. Validity is itself a statement made at the time of use: producers test standards periodically and issue a batch validity statement, a different object from an expiry date printed once and never revised.
Writing the decision rule before the data arrive
An investigation that picks its acceptance criterion after seeing the repeat result is not an investigation. Three things belong in the protocol before the sequence starts: the measurement treated as decisive and its acceptance interval, expressed against the combined expanded uncertainty of the two values compared; the consequence of each outcome; and the disposition of the material, which is quarantined rather than returned to the shelf while under suspicion.
If the reprepared standard agrees within the interval, the standard is not the cause, the original result stands as an out-of-specification result, and the investigation moves to the sample and the process. If it disagrees beyond the interval, the original results are invalidated for a stated cause, not quietly deleted. One boundary deserves stating: a laboratory can qualify its own working standard against a pharmacopoeial one, and can show that a lot performs as its certificate says, but it cannot re-assign the value of a producer certified material on one comparison run. Overwriting a characterised value with a single-point result removes the only independent reference in the chain.
What to record, and what to ask next time
The record that survives inspection is short: the hypothesis written before the experiment, the measurement that tested it, the acceptance interval, the outcome, and who approved it. What does not survive is a file holding six passing re-injections and no statement of what they were meant to prove. Five questions reduce the chance of repeating the work: is the assigned content as is or dried, and in what unit; is the uncertainty expanded or standard, at what coverage factor; what are the instructions after opening and is the unit single-use; is a batch validity statement issued and where is it obtained; and when a lot is replaced, is a comparison with the previous lot published.
None of these is answered by a catalogue entry. They decide whether a standard can serve as a ruler or only as a sample. The guides below cover reading the document, the uncertainty that follows from it, and the preparation steps where the same errors reappear.
Reading a certificate of analysis | uncertainty budget for a standard solution | weighing at the minimum weight | solution hold times | salt and hydrate conversion
Frequently asked
Can I dry the standard to remove moisture before weighing?
The standard is inside its validity date but every result is low. Can the standard still be at fault?
Is re-injecting the sample a valid way to resolve an out-of-specification result?
Two lots differ by 0.5 per cent and each certificate states 0.4 per cent at k = 2. Is the difference real?
How many replicate injections does an investigation need?
Can we assign our own value to a supplier standard after one comparison?
Sources and further reading
Every document below is published by the body that owns it; none of them is a summary written by us. Accessed 2026-11-16.
- FDA - Investigating Out-of-Specification (OOS) Test Results for Pharmaceutical Production, guidance for industry, May 2022 — the order of the laboratory phase: an initial assessment of the accuracy of the laboratory data before test preparations are discarded, with hypotheses about the cause tested against those same preparations
- EDQM - FAQ: are drying conditions required before use — that where no drying conditions are stated the substance is used as received, that an assigned content is expressed as is, and that a reference standard dried before use is no longer suitable for its intended use
- EDQM - Information leaflet, Ph. Eur. Reference Standard (sucrose CRS batch 2) — the instructions for use: use as is, do not dry or desiccate, and once the container has been opened the entire content is to be used immediately, with further storage and re-use not warranted
- ISO - ISO 17034:2016, General requirements for the competence of reference material producers — the requirements a producer is assessed against for competence and consistent operation, covering reference materials including certified ones, which is the basis on which an assigned value rests
- ICH - Q2(R2) Validation of analytical procedures, final version 2023, error correction 2025 — accuracy and precision as separate validation characteristics, which is why repeated injections establish dispersion and cannot establish that a value is correct
Send us a CAS number
Availability, documentation and lead time — answered by the people who handle the batch, not a ticket queue.