Related Substances and Unknown Impurities

Five questions, asked in order, each with a documented answer. Nothing about the sequence is exotic — an experienced analyst runs it without writing it down. The point of writing it down is what 21 CFR 211.192 actually requires: not that you reached the right conclusion, but that the record shows how.

Figure C1.1

C1.1  Is it real?

Requirement Before a peak is a scientific question it is an authenticity question. 21 CFR 211.192 obliges an investigation of “any unexplained discrepancy” or specification failure — which means the first job is to determine whether there is anything to explain at all.

Practice The diagnostic menu is not codified in a single place, but it is standard practice: a blank injection to rule out carryover and system contamination; dilution linearity, since a genuine component's area should scale with the amount injected while many artifacts do not; retention-time reproducibility across replicate injections; and, where a standard exists, a spike to confirm the peak grows by the amount added. Column bleed, mobile-phase degradation products and gradient-related ghost peaks are common false positives, and all four checks distinguish them from a real component of the sample.

An artifact confirmed and documented this way closes the question at the bench — Phase I of an investigation, described below, is exactly this kind of work, and a confirmed assignable cause is a legitimate place for it to end. What does not close it is a peak that looks unfamiliar, survives none of these checks, and is set aside anyway.

C1.2  Above the reporting threshold?

A real peak is not automatically an obligation. Arc A3 established the dose-dependent reporting, identification and qualification thresholds under ICH Q3A and Q3B — a real peak below the applicable reporting threshold creates no duty beyond noting it in the raw data. Above the reporting threshold, it must be reported. That is a separate question from whether it exceeds the specification, which is the next gate.

C1.3  Does it exceed the specification — and is that the same as “out of trend”?

Requirement Out-of-specification (OOS) has a defined meaning: a result outside the acceptance criterion in an approved specification. It is the trigger for a formal investigation.

Practice Out-of-trend (OOT) does not have that status. FDA's own guidance on OOS investigations raises it only once, in passing:

“Although the subject of this document is OOS results, much of the guidance may be useful for examining results that are out of trend.”FDA, Investigating Out-of-Specification (OOS) Test Results for Pharmaceutical Production, Revision 1, May 2022. source

OOT is an industry and compendial convention for a result that is unusual against a batch's own history while still inside the specification — not an FDA-defined regulatory term. Treat it as a signal that earns scrutiny, not as a status that by itself triggers 211.192. A result can be real, above the reporting threshold, out of trend, and still within specification — reportable, worth investigating why the trend moved, and not yet an OOS investigation. That is the right-hand exit in Figure C1.1, and it is where Module C2 picks up: a trend that keeps climbing is a reason to ask whether the specification itself remains appropriate.

C1.4  If it is OOS: the two-phase investigation

Requirement The regulatory obligation itself is unambiguous:

“Any unexplained discrepancy…or the failure of a batch or any of its components to meet any of its specifications shall be thoroughly investigated, whether or not the batch has already been distributed. The investigation shall extend to other batches of the same drug product and other drug products that may have been associated with the specific failure or discrepancy. A written record of the investigation shall be made and shall include the conclusions and followup.”21 CFR 211.192. source

Practice The regulation states the obligation; the structure most laboratories use to meet it comes from FDA's guidance, and guidance is stated FDA expectation rather than codified law — worth teaching as a distinction even though the two are, in practice, inseparable. The current version organises the work in two phases.

Phase I — laboratory investigation. The analyst and a supervisor assess whether a laboratory error caused the result: calculation mistakes, a diluted or contaminated standard, instrument malfunction, an obvious sample-preparation error. This is the same territory as C1.1's authenticity check, extended to the full analytical process. It closes here only when an assignable cause is found and documented.

Phase II — full-scale OOS investigation. Requirement

“When the initial assessment does not determine that laboratory error caused the OOS result…a full-scale OOS investigation using a predefined procedure should be conducted.” It “should be conducted by the QU and should involve all other departments that could be implicated, including manufacturing, process development, maintenance, and engineering.”FDA, Investigating Out-of-Specification (OOS) Test Results, Revision 1, May 2022. source

This is the point at which an unknown peak stops being a laboratory question and becomes a quality-unit question — not a matter of seniority, but of scope. A confirmed lab error is a fact about one measurement. An unconfirmed OOS is a fact about a batch, and 211.192's own text extends the obligation past that one batch to “other batches…and other drug products that may have been associated” with it. That clause is doing real work: it is what makes an OOS investigation a systemic question rather than a local one, and it is precisely the clause a narrow, one-batch investigation fails to satisfy.

Figure C1.2

Return to the letter. Nothing about that peak required knowledge of nitrosamines to act on correctly — everything the tree asks in Figure C1.1 was answerable with the tools in C1.1 through C1.3, before any structural work began. The finding was not that the analytical chemistry was hard. It was that the first gate was never opened.

Arc A1 drew the grid: specified or unspecified asks whether an impurity has its own acceptance criterion; identified or unidentified asks whether anyone knows what it is. Promoting an impurity from unspecified to specified — giving it its own number — is a decision, and like any decision in this course it needs a documented basis. Q6A is the guideline that governs it.

Figure C2.1

C2.1  Four inputs, one number

Requirement ICH Q6A is explicit about where the justification for a limit is supposed to come from:

“Justification for a proposed acceptance criterion…should refer to relevant development data, pharmacopoeial standards, test data for drug substances and drug products used in toxicology and clinical studies, and results from accelerated and long-term stability studies.”ICH Q6A, §3.1.2, Justification of Specifications. source

In practice that collapses to four sources, and a defensible specification draws on more than one. Batch data describes what the process actually does — informative, but descriptive rather than justificatory on its own. Toxicological qualification — safety data, or a structure-based read-across to a related compound — is what makes a limit defensible on safety grounds once the qualification threshold is reached. Compendial standards supply a floor where one exists: Arc A3 already showed that USP ⟨476⟩ sets its default “any unspecified impurity” criterion at the identification threshold, not a fixed number — the compendial input itself traces back to the same dose-dependent ICH tables. Clinical and manufacturing exposure history supplies the question a toxicologist actually needs answered: what level has a patient already been dosed with, and in what material.

Practice No single box in Figure C2.1 is sufficient by itself. A specification built from batch data alone has no safety basis — it describes the process, not the patient. A specification built from the qualification threshold alone, with no reference to what the process actually produces, may reject material that has always been safe and always been made. The number that survives an inspection is the one built from more than one box.

C2.2  Why the obvious approach is the wrong one

Requirement The most intuitive way to set a limit — look at what the validation batches produced, and set the criterion just above it — is the one Q6A specifically warns against, in language that names the mechanism, not just the conclusion:

“At the time of filing it is unlikely that sufficient data will be available to assess process consistency. Therefore it is considered inappropriate to establish acceptance criteria which tightly encompass the batch data at the time of filing.”ICH Q6A, §3.2.1(d), Impurities. source

The mechanism is the same one Module C1 just spent an entire decision tree on. A specification that tightly encompasses two or three early batches has not controlled the process — it has recorded a small sample of ordinary variation and then criminalised every future measurement that falls outside that sample. Every one of those measurements now triggers a Phase I laboratory investigation, and when Phase I correctly finds no laboratory error — because there was no laboratory error — it escalates to Phase II. The specification has manufactured a 211.192 investigation out of the process behaving exactly as it always has.

Figure C2.2

Work the numbers in Figure C2.2. At a maximum daily dose of 400 mg, the drug-substance reporting threshold is 0.05%, the identification threshold is 0.10%, and the qualification threshold is 0.15% — the 1.0 mg/day cap that mattered in Arc A's example does not bite at this dose, so these are the plain Q3A percentages. A naive specification set at NMT 0.08% — fit tightly to the two development batches at 0.05% and 0.07% — puts the very next batch, the registration validation lot at 0.09%, out of specification. Two ordinary commercial batches at 0.11% and 0.13% follow it. None of these results reflect a process failure; all three would open a 211.192 investigation that Phase I would correctly be unable to close.

A specification justified at the qualification threshold — NMT 0.15% — absorbs all five batches. The 0.13% commercial result crosses the 0.10% identification threshold, so it still carries an obligation: that impurity must now be structurally identified. It does not cross the qualification threshold, so no additional toxicological justification is required yet. The specification did its job without ever losing its safety basis, and it did so by starting from the qualification threshold in Figure C2.1 rather than from the two batches that happened to exist first.

C2.3  A specification is not fixed for the life of the product

Practice Q6A's own cross-reference (§2.5, “Limited Data Available at Filing”) anticipates that a specification set with limited data may be revised — tightened or loosened, with justification — as manufacturing experience accumulates. A specification is a live document, not a one-time calculation. Changes to an approved specification go through a defined regulatory change-control mechanism, the subject of a later module in this arc; the point to carry forward here is only that revising a specification with accumulated data and a documented rationale is normal practice, and is a different event from a batch failing an existing one.