How many samples should you test for batch testing HPLC monograph API lots?
Before a number is chosen, separate three quantities that are often collapsed into one. The first is the number of containers drawn from the batch. The second is the number of independent preparations made from those containers. The third is the number of replicate determinations made from each preparation. How many samples should you test for batch testing HPLC monograph API work depends on which of those three the specification constrains. An identity test may be met by a confirmed retention time plus an orthogonal mass result on a defined preparation. A purity or related-substances limit is a numerical claim about composition. A content figure is quantitative and needs its own uncertainty budget, because area-normalised purity and a mass-balance content result do not estimate the same attribute. Calling all three one HPLC sample produces a report that looks complete and cannot be interpreted. Container number is a property of the lot, not of the chromatograph. A single well-mixed bulk solution, with mixing documented, gains little from extra containers. Lyophilised research vials are discrete units. Cake mass, residual moisture and the local impurity profile can differ between vials that share a batch number. Opening one vial and running the preparation once estimates that vial. It does not estimate the between-vial standard deviation, and it cannot show whether a second vial would fail the same limit. Eckblad frames sample number as a question, not a universal count. The number is fixed by the question and by the variability relevant to that question, not by a customary panel size. Write one sentence before any vial is opened: does this lot, as a set of containers, meet the identity, purity and content limits, at a stated confidence? If that sentence cannot be written, the sample number cannot be justified. The batch record should contain that sentence, not only a percentage on the chromatogram header.
What must be defined before an HPLC sample count means anything?
A sampling plan fails when the nouns are loose. For HPLC batch testing of a monograph-style API or research peptide, define the lot, the primary sample, the laboratory sample, the test portion and the analytical solution before discussing a count. The lot is the set of containers the batch record claims to represent: a synthesis batch, a purification pool, or a fill of lyophilised vials. The primary sample is the material removed from that lot. The laboratory sample is what arrives at the bench, still traceable to container identifiers. The test portion is the weighed amount taken for a stated method. The analytical solution is the diluted preparation in the autosampler vial. Each of those objects can be called a sample in conversation. Only one is the unit the acceptance rule applies to. If a certificate reports purity without saying whether the figure is one test portion, a composite of several vials, or the mean of three preparations, the figure has no sampling meaning. Garrett's question — how many samples should be enough — treats adequacy as relative to a defined population. In a peptide lot the population is the set of containers, or the bulk from which they were filled. Enough for a retention-time identity check is not enough for a between-vial content range. Write the population in the plan: all vials of a named lot filled on a named date from a named bulk, or the retained bulk solution before fill. A line that says only sample tested by HPLC has not defined a population, so no count can be enough. Selection method belongs in the same block. A convenience pick from the front of a shipper is not a random draw and not a stratified draw across the start, middle and end of a fill. If fill order is a plausible source of heterogeneity, record the container identifiers at those positions. If equal masses are composited, record each donor identifier and the mass from each donor. Compositing is a defined operation, not a way to hide which units were opened.
Should batch HPLC use a composite or individual containers?
A composite answers a different question from a set of individual results. Equal test portions from several vials, combined and analysed once, estimate a mass-weighted mean for the donors in the composite. They do not estimate the worst vial, the range, or the proportion of vials outside a limit. If the specification is written as a lot-mean purity or a lot-mean content, a composite can be an efficient test portion, provided every donor is identified and the blend is shown to be uniform before the chromatographic run. If the specification is written as each container, or as a maximum related-substances result that must hold for any vial a laboratory might open, a composite can mask a failing unit. The mean moves towards the centre; the failing tail disappears into the blend. That distinction decides the count. For a mean, sample number is driven by the standard deviation of the mean, which falls as the number of independent donors rises, but only if those donors are measured in a way that captures between-vial variance. A composite of ten vials analysed once has used ten containers and still yields one analytical result. It does not give ten degrees of freedom. Laboratories that report n = 10 on a composite chromatogram are describing the blend size, not the number of independent HPLC results. The batch record should use two fields: containers opened, and independent determinations. Individual-container testing is the plan to use when between-vial dispersion is the risk. A small fixed set can still miss a rare defective unit. Schmettow and colleagues, writing on sampling strategies for tests of high-risk systems, argue that the number tested should follow the risk of missing important problems, not a customary small panel. For research-peptide lot release the consequence is analytical: a mis-represented lot, a purity figure that does not travel to the next vial, or a related-substances peak present in one container and absent from the certificate. State, before the sequence starts, the rule for the mean, the lowest result and any unidentified peak.
How should replicate HPLC determinations be kept apart from lot samples?
Replicate determinations and lot samples answer different uncertainty questions. Replicates from one preparation estimate run-to-run and integration precision of the method on that solution. They do not estimate vial-to-vial differences. A sequence that runs the same autosampler vial six times has one laboratory sample, however many peaks are integrated. System-suitability replicates belong in the sequence because they show that resolution, tailing, retention-time window and repeatability met the method's pre-set criteria before sample data were accepted. They are not additional containers from the batch, and they must not be averaged into the lot purity figure. A practical sequence for a monograph-style peptide HPLC run separates these layers in the sample table. Blank and diluent runs show which peaks are artefacts of the solvent or the system. Suitability replicates, typically a small fixed set stated in the method, qualify the column and the detector on that day. Bracket standards, if the method is an assay rather than an area-normalised purity run, track drift across the sequence. Only then do the batch preparations appear, each labelled with container identifier, test-portion mass, dilution and preparation identity. If the plan calls for three independent preparations of one vial, those three rows are preparation replicates. If the plan calls for one preparation from each of five vials, those five rows are lot samples. Mixing those labels makes the reported n impossible to audit. Precision data tell you whether the replicate count is wasted. If the relative standard deviation of area-normalised purity on one preparation is already far inside the gap between the result and the limit, extra replicates of that same solution add little. If the result sits close to the limit, the method's intermediate precision — different day, analyst or column, as defined in the validation — matters more than a sixth run of the same vial. Report the observed repeatability next to the sample number so a reviewer can see whether the count was spent on the dominant variance component. Croy's question — how many names should we test — is the same design problem in another field: panel size has to be justified against the precision the decision needs.
What batch-record fields show how many units were tested?
Sampling notes from outside the peptide laboratory are not monographs, and they should not be cited as if they set HPLC limits. They do state a planning fact. A 1976 reliability paper asks how many units you should check, on the view that there is money to be saved in the sampling programme. For HPLC lot work that saving is real: each extra vial is a weigh, a dilution, a run, a review and a stored data file. The opposing cost is also real: a certificate that over-states uniformity because too few containers were opened. Optimise the plan against both, in writing, rather than against a round number. A complete batch record for an Australian research-peptide lot should make the count reconstructable without asking the analyst. Minimum fields are: lot identifier and container map; population definition; selection rule; containers opened, with identifiers; test-portion masses; whether results are individual or composited; number of independent preparations; number of suitability replicates, kept separate; method identifier, column and detection wavelength; acceptance limits for identity, purity or related substances, and content if claimed; who reviewed the integration. Link the chromatogram file name to the same container identifiers. A purity percentage with no n, no container list and no statement of composite versus individual is not a sampling result. Do not treat a larger panel as a quality claim about the molecule. Sample number is evidence about the lot relative to a specification. It does not change identity, and it does not support any statement about human use. A reviewer can check whether this block is present, whether an orthogonal mass result is tied to the same containers, and whether a retained sample is identified for repeat testing if an integration is challenged. That is the practical answer to how many samples should you test: the smallest number that can support the written decision rule, documented so a second laboratory could see what was and was not represented.
Frequently asked questions
Is there a universal sample number for HPLC monograph batch testing?
No. The count depends on whether the specification constrains a lot mean, each container, or one identified preparation, and on the between-vial variability the plan must detect. A monograph names limits. It does not replace a written sampling plan. Record containers opened separately from replicate determinations of one solution, and state the decision rule the number is meant to support.
Does one HPLC chromatogram cover every vial in the lot?
One chromatogram covers the preparation that was run, from the containers that were opened and either blended or tested individually. It does not describe unopened vials. If the certificate is used as a lot statement, list container identifiers, say whether a composite was used, and state the population the result represents. Otherwise the chromatogram is a single-unit observation.
Are system-suitability replicates part of the batch sample size?
No. Suitability replicates show that resolution, tailing, retention window and repeatability met the method criteria on that sequence. They come from a suitability solution, not from additional lot containers, and they should not be averaged into reported purity or content. Keep them on a separate line of the sample table so the number of lot preparations stays clear.
When is a composite acceptable for related-substances HPLC?
A composite fits a limit written on a mass-weighted lot mean, if every donor vial is identified, weighed and shown to be blended before the run. It is a poor plan when the limit must hold for any container, because a failing vial can be diluted by conforming neighbours. Report blend size and independent determinations as separate numbers.
What should be recorded if only one vial is tested?
Record that the result represents that vial, with its identifier, test-portion mass, method identifier and comparison with the acceptance limit. Do not imply that the rest of the lot was measured. If the record still claims lot-level uniformity, state the assumption, such as a documented homogeneous bulk before fill, and identify a retained container for a repeat preparation.
References
- DOI:10.2307/1311590 — How Many Samples Should Be Taken? — BioScience — 1991
- DOI:10.70499/hdix1161 — How many samples should be enough? — EXPLORE (Newsletter for the Association of Applied Geochemists) — 2010
- DOI:10.1016/j.jbi.2013.04.007 — With how many users should you test a medical infusion pump? Sampling strategies for usability tests on high-risk systems — Journal of Biomedical Informatics — 2013
- DOI:10.1002/dir.4000020407 — How many names should we test? — Journal of Direct Marketing — 1988
- DOI:10.1016/0026-2714(76)90066-4 — How many should you check. Chances are, there's money to be saved in your sampling program — Microelectronics Reliability — 1976
Research use only
This article is provided for laboratory research and educational purposes only. Products referenced are not for human or veterinary use. ClaraScience makes no therapeutic, medical, or efficacy claims, and nothing here constitutes medical advice.
