Retention Samples vs Ongoing Stability: Two Promises
Share
During an audit, the auditor asks to see retention samples. The supplier opens a cupboard with rows of amber bottles, each labelled with a batch number and a date. The auditor nods, then asks a second question: what do the stability results show for last year's batches? The room goes quiet. The cupboard answered the first question and has nothing to do with the second.
These are two programmes with two purposes, and they are routinely conflated by suppliers who have one and describe it as both.
An archive and an experiment
A retention sample is material held back so that a future dispute about a specific batch can be settled by re-testing the material as it was supplied. Its value is retrospective. It exists for the day when somebody claims the batch was out of specification on arrival and the question becomes what it looked like when it left.
Ongoing stability is a programme that puts a defined sample on a defined schedule and tests it at defined intervals. Its value is prospective. It answers whether the process still produces what it produced three years ago, and whether the shelf life assigned at launch is still supportable.
The first without the second means a supplier can defend a complaint and cannot detect a drift. The second without the first means a supplier can see the drift and cannot resolve a dispute about the batch that caused it. Both are needed, and neither substitutes for the other.
What the retention requirement actually says
The requirement is about quantity, pack and duration. The quantity should be enough for at least two full specification tests, so that a dispute can be tested twice, ideally by two laboratories. The pack should be the marketed pack, or one demonstrated to be equivalent. The duration is a defined period after expiry, and the figure depends on jurisdiction and on what the quality agreement says.
Three failure modes recur.
Samples in the wrong container. Glass jars in a laboratory cupboard. The material is representative, the pack is not, and the pack is where moisture and light do their work.
Samples without a chain of custody. Nobody can say who opened the cupboard, when, or why one bottle is half empty. A sample with a broken chain is not evidence.
Samples below the required quantity. One test's worth of material means the second laboratory cannot confirm the first, and a dispute becomes an argument about method rather than about the batch.
What an ongoing programme looks like
The usual shape is one batch per year per API per pack, added to the programme and tested at the standard intervals: zero, three, six, twelve, twenty-four, thirty-six months, and annually thereafter to the retest period. Additional batches are added after a significant change, after a process deviation that could affect stability, or when a trend needs confirming rather than explaining.
Bracketing and matrixing can reduce the number of samples. They cannot reduce the obligation to have a programme. A supplier who says the product is stable and therefore needs no ongoing testing has confused a conclusion with a method.
| Element | What good looks like | What an auditor notices first |
|---|---|---|
| Batch selection | One per year, plus change-triggered additions, with the rationale written down | A gap year with no explanation |
| Pack | Marketed pack, same closure | Laboratory containers |
| Test panel | Full specification at each interval, or a justified reduced panel | Assay only |
| Trend analysis | Plots across batches for assay, impurities, loss on drying | Tables with no comparison |
| Conclusion | A written statement that the retest period remains supported | Data with no conclusion |
An illustrative trend, and what it is for
The table shows four annual batches, values illustrative.
| Year | Assay at release | Total impurities | Loss on drying |
|---|---|---|---|
| 1 | 99.9% | 0.18% | 0.35% |
| 2 | 99.8% | 0.19% | 0.38% |
| 3 | 99.7% | 0.21% | 0.40% |
| 4 | 99.5% | 0.24% | 0.44% |
Every value is inside its limit. Every value is also moving in the same direction, and four points in a row on one side of the mean is a signal that a control chart would flag. This is what ongoing stability is for: not to catch a failure, but to catch a direction before it becomes one.
A supplier without a programme cannot see this. They will notice in year six, when a batch fails, and by then the cause is somewhere in a process change made in year two that nobody remembers.
Why the customer should ask
Because the customer's own file depends on it. A pharmaceutical company qualifying a supplier is not buying one batch; they are buying an assumption that batch twenty will behave like batch one. The ongoing programme is the evidence for that assumption, and the annual report is the document that lets the customer's annual product review say something true.
For a research group or a formulator, the same logic applies in a smaller way. If the assay has drifted half a per cent over three years, the customer's formulation calculations need to know, and the certificate for the batch in hand will not tell them.
What we provide, and what we do not claim
We provide a retention sample per batch in the shipped pack, held for the period agreed in the quality agreement, with a chain-of-custody record. We provide an annual stability summary covering the batches placed on programme, the conditions, the results and the conclusion, and we say plainly when a year had no addition and why.
We do not describe a cupboard of samples as a stability programme. The distinction is small in a sales conversation and large in an audit, and it is the kind of thing that is much cheaper to get right before the auditor arrives than afterwards.
What the annual report contains when it is done properly
A stability summary is not a stack of chromatograms. It is a short document that lets a reader who was not involved reach the same conclusion as the person who wrote it, and it has a predictable structure.
Scope. Which products, which packs, which batches, and the year covered. A report that does not name the pack cannot be compared with last year's, and one that does not name the storage condition cannot be compared across markets. Where a product is sold into more than one climatic zone, the report should say which condition each arm runs at, which is the argument set out in the Zone IVb article.
Programme status. Batches added, batches completed, batches withdrawn and why. A gap year with no addition is a finding and should be described as one rather than left for the reader to notice.
Results by attribute. Assay, total impurities, individual impurities where they trend, loss on drying, and any physical attribute the specification controls. Presented as a table across batches and time points, not as a narrative.
Trend. A plot per attribute, with the specification limit drawn on it. The plot is the part that turns a compliance statement into a control statement, and it is the part most often missing.
Conclusion. A sentence that says whether the assigned retest period remains supported, and if not, what is being done.
The ICH Q7 expectations for API manufacturing and the EudraLex volume 4 annexes on sampling and stability describe the obligations this report discharges. The PIC/S inspection guides are useful for the opposite reason: they show what an inspector looks for when the report is thin, which is usually the gap between the batches on the programme and the batches released.
The cost of not having one
The cost is not the audit finding. It is the six months between the drift starting and somebody noticing.
A process that moves by a tenth of a per cent a year is invisible at release. Every batch passes. Every certificate is clean. In year five a batch fails, the investigation opens, and the root cause is a change made in year two that nobody flagged because nothing failed at the time. By then the material is in a customer's warehouse, the customer's own stability data were generated on the older profile, and the conversation has stopped being technical.
An ongoing programme would have caught the direction in year two, at a point where the fix was a parameter adjustment rather than a re-qualification. That is the entire commercial case for the programme, and it is stronger than the compliance case, which is why suppliers who run one tend to keep running it after the audit is over.
The connection to OOT trend analysis is direct: the trend lives in the batch history, and the stability programme confirms whether the trend is moving the material toward a limit or merely around a mean. The raw data behind those points has to survive long enough to be plotted, which is a separate discipline with its own failure modes.
Frequently asked questions
How long must retention samples be kept?
The requirement depends on the jurisdiction and on whether the material is an API or a starting material. For APIs, the common expectation in European GMP guidance is at least one year after expiry, and some customers contract for longer. What matters commercially is that the period is written into the quality agreement rather than inherited from whoever set up the shelf.
How many batches go into ongoing stability?
The usual minimum is one batch per year per API per pack, with additional batches when a significant change is made or when a trend needs confirming. Bracketing and matrixing can reduce the number of time points, but not the obligation to have a programme. A supplier with no ongoing programme is relying entirely on historical data and on the customer never asking.
What is the difference between a retention sample and a stability sample?
A retention sample is held so that a future dispute about a specific batch can be resolved by re-testing the material as it was supplied. A stability sample is placed on a defined programme and tested at defined intervals to generate data. The first is an archive. The second is an experiment. Storing material without testing it produces the first and not the second.
Should retention samples be stored in the marketed pack?
Yes, or in a pack demonstrated to be equivalent. The point of the sample is to represent what the customer received. A sample kept in a glass jar in a laboratory cupboard represents the material but not the pack, and the pack is where most of the interesting failures occur.
Can we see the annual stability report?
Ask, and expect a summary rather than the full dataset: the batches included, the conditions, the time points, the results for the critical attributes, and the conclusion. A supplier who refuses outright is telling you the programme is thin. A supplier who sends three cobbled-together chromatograms is telling you the same thing more slowly.
Does ongoing stability replace batch testing?
No. Every batch is tested against the specification before release. Ongoing stability tests a sample of batches over time to confirm that the specification remains achievable and that no slow drift is developing. Replacing release testing with periodic stability testing would release untested material, which is a different and much worse proposition.