Our Take
Sample size alone cannot fix a broken analytical method; you need a documented control strategy that identifies and monitors the biggest sources of error from development onward.
Why it matters
Quality control labs in pharma face daily decisions about when to reject results, repeat tests, or average multiple measurements. A lifecycle-based approach anchored in uncertainty quantification and risk-driven control strategy prevents both false acceptance and wasteful retesting.
Do this week
QC manager: map your current analytical control strategy (ACS) against NIST uncertainty propagation and ICH Q14 requirements before your next method transfer, so you can identify which error sources actually matter.
The sample-size trap in QC analytics
When pharmaceutical QC labs need to reduce measurement uncertainty, the instinctive response is to increase sample size: more preparations, more measurements, more replicates. This follows from basic statistics. Standard error of the mean (SEM) decreases with the square root of n. But the article argues this is backwards reasoning. It treats a symptom (high variability) instead of root causes (systematic bias, instrument drift, sample preparation inconsistency).
The framework outlined comes from NIST Technical Note 1297 and ICH Q14 guidance. Measurement uncertainty has two components: standard error of the mean (the random noise) and bias uncertainty (the systematic offset). Simply collecting more data without understanding which component dominates wastes resources and delays root-cause identification.
The real answer: build an Analytical Control Strategy (ACS) during method development that targets the biggest contributors to error. An internal standard might cut variability more effectively than doubling your sample count. Optimizing extraction conditions might eliminate bias altogether. A lifecycle-based approach (Analytical Lifecycle Management, per USP <1220>) requires you to quantify uncertainty early, identify control points in development and performance qualification (PPQ), then monitor continuously in operation (CPV).
Out-of-spec results expose strategy gaps
The article highlights a concrete regulatory dilemma: if your method requires three sample preparations, two pass specification, and one fails, but the average passes, what do you report? The FDA's May 2022 OOS guidance states clearly: treat the average as out-of-spec and investigate. The passing results are no more likely to represent the true value than the failing one if the spread exceeds what your method's control strategy predicted.
This creates pressure. Labs either increase sample size (expensive, slow) or redesign their control strategy to reduce variability at source. A strong ACS documented during development includes acceptance criteria for the agreement between replicate preparations. When those criteria are exceeded, an investigation triggers—not because one result failed, but because your method performed outside its known behavior envelope.
The second scenario in the FDA guidance applies to replicate injections from the same vial (e.g., HPLC). If variability acceptance criteria are met, a single outlier injection does not invalidate the reported average. Again, this depends entirely on having defined the expected variability window in advance and demonstrated its validity during PPQ.
Organizations without this structure face recurring investigations, delayed release decisions, and audit findings. Those with lifecycle-aligned ACS reduce false rejects and avoid the false confidence of averaging away real problems.
Audit your method against the uncertainty framework now
If your lab uses multiple preparations or measurements, document the statistical rationale. Use ANOVA to partition variability and identify the top contributors. Determine whether increasing sample size by 1 or 2 units actually moves the needle (diminishing returns apply quickly: n goes up by 10% but SEM drops by 3%). Then redesign the control strategy to target the dominant sources.
For any method in CPV, establish sample-agreement criteria that reflect expected variability plus a safety margin. When these are exceeded, investigate. Do not assume averaging is valid until you prove the method's ACS is working as designed. If an OOS result appears alongside an agreement failure, the two are likely related—fix the process, not the paperwork.
Regulatory expectation is now clear: analytical methods require lifecycle governance, not just batch testing. Build it early or defend it repeatedly in audits.