The impurity result is 0.09%. The limit is 0.10%. The recommendation says the increase from 0.04% to 0.09% across four batches is normal variation. Three other explanations fit the same numbers, and the file considers none of them.
In this article
The first explanation offered becomes the only one tested.
The explanation arrives with the number: “0.09%, within limit, normal variation.” A reviewer then does the natural thing and asks whether that is reasonable. It sounds reasonable. The check ends there, and the other explanations were never put on the table to be ruled out.
Plausible is a low bar. Many explanations are plausible for one number. The reviewer’s job is not to approve the explanation in front of them. It is to know what else could be true and whether anything on file separates them.
If the file can only show that one explanation is possible, it has not shown that it is the explanation.
List the explanations that fit the same number.
This continues the synthetic case from The Record in Practice and the previous article: the unspecified-impurity result is 0.09% against a limit of 0.10%, after 0.04%, 0.06% and 0.07% in the three batches before. No real batch, product, person or site is involved.
| Explanation | If it were true | What would tell it apart | On file in the synthetic case |
|---|---|---|---|
| Normal variation | Values scatter around a stable level. No direction over a longer history. | The results of more than four batches. | Four batches only, in rising order. Too few to separate scatter from a trend. |
| A change in the incoming material | The increase starts when a new lot of the active ingredient or an excipient enters the process. | The material lot used in each batch, and the supplier’s results for those lots. | Not reviewed. |
| Drift in the analytical method | The same sample would read differently over time. The reference standard or column history would show a change. | System-suitability values trended across the runs, reference-standard and column history, and a re-analysis of a retained sample, allowing for change on storage. | System suitability passed in each run. The trend was not looked at. |
| A change in the process | The rise coincides with a change in a process parameter, such as time or temperature. | The four batch records compared step by step. | Batch records not compared. |
The last column is the one a reviewer reads first. Of four explanations, none is supported or ruled out by the file. For one, four batches are too few to tell scatter from a trend. One has a passing run-by-run check that says nothing by itself about a trend, because a pass against the criterion does not show whether the values moved within it. Two have not been looked at. That is the real state of the evidence, and the recommendation did not say it.
Find the item that would look different.
Good evidence here is the item that reads differently under each explanation. A result that every explanation would produce does not separate them, however many of them are on file. The four batch records compared side by side would separate a process change from the other three. The supplier’s lot results would separate a material change. The system-suitability values trended over the runs, the reference-standard and column history, and a re-test of a retained sample, allowing for any change in the sample during storage, would help separate method drift.
This is also the honest way to say what is not known. “Not reviewed” is a legitimate entry. It is a decision point waiting for an owner, not a gap to be smoothed over.
Name the decision point.
A release decision made under the site’s procedure can proceed with some questions open. What the record cannot do is hide that they are. The decision belongs to the authorizing person under the site’s procedure. What the record owes them is a plain statement of what they are deciding on.
That paragraph does not say the rise is normal variation or drift. It says what has been checked, what has not, and that the release decision is the authorizing person’s under the site’s procedure. A later reviewer can disagree with the decision and still see what it was made on.
What the decision record should contain
Applied to the synthetic case, a record that preserves the reasoning reads as follows. This is a fictional example, not a template for a real decision.
| Lens | What the record contains | Question before closing |
|---|---|---|
| EvidenceWhat was known? | The four results, system suitability for each run, and a plain list of what was and was not reviewed for each explanation. | Which items would look different under each explanation? |
| JudgmentHow was the evidence evaluated? | Four explanations considered. The first, normal variation, is not the only one the evidence permits. | Was an explanation approved on plausibility, or tested against the others? |
| RiskWhat remained uncertain? | Whether a material or process change is behind the rise and what the next batch will read. | What is still open, who owns it, and by when? |
| AuthorizationWhat was decided? | The authorizing person states which explanation the decision rests on, what stays open, and the owner and date for the open checks. | Does the decision say what it is standing on? |
What the record should explain
A result that passes can be explained in more than one way. A record that explains the decision shows which explanations were considered, what would separate them, what was checked, and which one the decision rests on. It also shows what was left open and who owns it.
Make the review answer four questions.
- What else would fit the same result? List at least the explanations a colleague would raise.
- What would look different under each? Name the item on file.
- What was checked, and what was not? Say “not reviewed” where it is true.
- What does the decision rest on? State it, and name the owner and date for what remains open.
This builds on the difference between what a result shows and what is concluded from it. When the result is out of specification and a retest passes, a different set of questions applies: the retest that passed.
Regulatory context and sources
For finished pharmaceuticals, 21 CFR 211.165(a) requires appropriate laboratory determination of satisfactory conformance to final specifications for the drug product, including the identity and strength of each active ingredient, prior to release. 21 CFR 211.165(e) requires the accuracy, sensitivity, specificity and reproducibility of the test methods to be established and documented. 21 CFR 211.192 requires production and control records to be reviewed and approved by the quality control unit to determine compliance with established, approved written procedures before a batch is released or distributed. It also requires any unexplained discrepancy, and the failure of a batch or any of its components to meet any of its specifications, to be thoroughly investigated, with the investigation extended to other batches that may have been associated with it and a written record of the conclusions and followup. In the synthetic case above no result fails a specification. Whether the increase is an unexplained discrepancy under 21 CFR 211.192 is for Quality to decide under the site’s procedure; the regulations do not say which explanation fits the numbers.
Source context: Regulatory requirements: 21 CFR 211.165, Testing and release for distribution; and 21 CFR 211.192, Production record review. The batch, the results and the explanations are a synthetic case used in the ComplianceWorxs The Record in Practice; the record example and the Four Lenses review are CW editorial examples, not regulatory mandates. This article has not been reviewed by a practitioner for regulatory sufficiency. ComplianceWorxs does not provide regulatory or legal advice and does not judge whether any record or release decision is adequate.