Summary
- The problem
- Every batch can meet its specification while the process drifts. A CPV review stalls on assembly when results, batch context and investigations sit in different systems, and missing or non-comparable results are easy to lose.
- Seal’s approach
- Trends stay connected to the batches, process versions, methods and materials behind them. Neil prepares the population, calculations and draft review; qualified reviewers verify the method, interpret the signal and agree the follow-up.
- What changes
- Pending and non-comparable results stay visible outside the analysis, a signal becomes a record with its own review, and the evidence carries into investigation, change control and APQR.
- Where to start
- One monitored parameter, its monitoring plan and batch history. Book a demo.
Ask what the trend shows, and what it does not.
Assay across 24 reviewed batches
Assay · % label claimYes. The last 8 results are all above the reference.
That triggers this plan’s review rule. Passing the specification doesn’t clear the signal.
Which records are included?
24 reviewed results from recipe v09. One result is pending and one batch uses v10; neither is plotted. Both remain in the source list.
The material lot changes at the same point as the higher results. That gives us a lead—not a cause. I’d check the measurement and compare the process context before drawing a conclusion.
What should each team do next?
Check the measurement
QC / analytical lead
Confirm source results, method version, standards and instrument performance. Review B-126 when its result becomes available.
Investigate the shared context
MSAT / process owner
MAT-B appears with the eight higher results. Compare material attributes, process conditions and relevant events. The timing alone does not establish a material effect.
Agree the next monitoring window
CPV lead / Quality
Document the signal assessment, any required investigation and the next review. Keep recipe v10 separate unless pooling is scientifically justified. Do not move the reference simply to clear the signal.
1Within specification is only part of the picture.
Batch release asks whether a batch meets its specification. Continued process verification asks a different question: whether the process is still performing as it did when it was validated, across a defined population of batches.¹ A result can meet its specification while the series it belongs to changes over time.
Answering that question depends on more than the numbers. Each result needs its batch, process version, method version, material lot, equipment and review status, because those determine which results are comparable. When the values sit in one system, the batch context in another and the investigations in a third, a CPV review becomes an exercise in assembly before any interpretation can begin.
Seal connects process trends to the batches, methods and materials behind them. Ask Neil to prepare the review from those records; qualified reviewers inspect its sources, decide what the trend means and agree the follow-up work.
1.1Why teams choose Seal for continued process verification
When CPV is treated as a periodic statistical exercise, results are exported from the LIMS and batch records into a spreadsheet or statistics package, charted and reported months after the batches ran. Seal draws the trend from the same records the batches and tests were executed in, so comparability, missing results and investigations stay visible, and a signal becomes a record with its own review. When the review leads to a process change, later batches are trended against the new process version, so the next period shows whether it worked.
2Define the population before interpreting the trend.
A trend is only as sound as the population behind it. Product, process version, site, scale, equipment, material, method and review status define which batches belong to a cohort. The monitoring plan states the inclusion and exclusion rules, how completeness is judged and which versions can be compared.
In the fictional example above, monitoring plan CPV-PLAN-014 covers an assay on recipe v09, analytical method AM-009 v03 and a 1,000 L scale. Twenty-four reviewed results meet the 95–105% specification. One further batch, B-126, has a pending result. Another, B-125, ran on recipe v10. Both remain visible, outside the plotted population, rather than being dropped or pooled with the rest.
Expected records stay in view even when an analysis cannot include them. A missing value is not converted to zero, and an out-
3Separate the monitoring rule from the specification.
The example plan uses a fixed 100% reference and an illustrative rule: eight consecutive reviewed results above the reference prompt review. The last eight results in the population are above it, so the rule is met, even though every result is within specification.
The reference is not a statistically estimated control limit, and the rule is not a recommendation. Your qualified team selects and verifies the limits, methods and signal rules appropriate to the real process. Calculations keep their formula, inputs, units, aggregation, precision and missing-data behaviour, and a revised method becomes a new version rather than silently replacing the earlier one.
Neil can prepare the dataset, a proposed method, the calculations and a draft interpretation. The team verifies the assumptions, population, limits and implementation before any of it is used for a CPV decision.
4Turn a signal into reviewable work.
In the example, the material lot changes from MAT-A to MAT-B at the same point as the higher results begin. That coincidence gives the investigation a lead, not a cause. No root cause is established by the chart. A significant atypical trend still warrants investigation, even when every result passes.²
A detected condition becomes a record of its own, carrying the monitored variable, the rule, the affected batches, an initial assessment and its review. The draft CPV review keeps the population, the monitoring definition, the finding and the proposed follow-up together for QC, MSAT and Quality to assess.
From there the evidence can move into the work it calls for: an investigation, a CAPA, a controlled change to the control strategy, a method or supplier review, maintenance or a change to the monitoring itself. When a change is approved, it creates an explicit boundary. Batches made afterwards form a new cohort, and the effectiveness period is defined before the results arrive rather than chosen afterwards.
5Connect the work across your systems.
Bring permitted manufacturing context, analytical results and quality records into the same review: for example orders and material lots from SAP, results from LabWare and deviations and changes from Veeva Vault. Historian and automation data can arrive with batch, phase, source, timestamp and units; laboratory results keep their sample, method, specification, instrument and review status. Scope each connection around the records, time basis, units, process phases and version references it needs. Connections depend on the source API, configuration and access.
The reviewed population, analysis definition, findings, linked investigations and open work then carry forward into the annual product quality review.³ A later dataset or method revision does not silently replace the evidence behind an earlier conclusion, so the APQR can show what was known when each decision was made.
6Bring one process to review.
Start with one monitored parameter, its monitoring plan and the batch history behind it. Agree the population rules and the sources for each record. Inspect the population, including what is pending or excluded, and review the findings with the people who own the process.
Then agree the next piece of work: an investigation, a monitoring change or a controlled process change. Expand to further parameters and products once the first review has shown what your records can support.
References
- 1FDA, Process Validation: General Principles and Practices, guidance for industry, Revision 1 (2011): defines process validation as the collection and evaluation of data, from the process design stage through commercial production, in three stages: process design, process qualification and continued process verification. FDA
- 2EudraLex Volume 4, Part I, Chapter 6, Quality Control (2014), section 6.35: out-
of- specification results or significant atypical trends should be investigated. European Commission - 3EudraLex Volume 4, Part I, Chapter 1, Pharmaceutical Quality System (2013), section 1.10 (product quality review): quality reviews of all authorised medicinal products should normally be conducted and documented annually. European Commission
AQuestions and answers
How is CPV different from batch release?
Batch release and CPV answer different questions. A result can meet its specification while a series changes over time. CPV reviews process performance and variation across a defined population; authorised reviewers still make product and process decisions.
Are the reference and specification the same thing?
No. The example specification is 95–105% label claim. Its fixed monitoring reference is 100%, with an illustrative eight-result rule. The reference is not a statistically estimated control limit. Your team selects and verifies the appropriate limits, methods and signal rules for the real process.
Is this a live analysis?
No. The chart, replies and review use fictional, fixed example records. The calculations operate on those records; no laboratory or manufacturing system is queried. The example does not establish process control, root cause or batch approval.
Can we use our existing historian, MES and LIMS?
Yes. Scope the records, time basis, units, process phases and version references for each connection. Permitted source data or supplied exports can form the review population. Connections depend on the source API, configuration and access.
What about missing results and process changes?
Keep expected records visible, even when an analysis cannot include them. The example separates one pending result and one different process version from its 24 reviewed results. Your monitoring plan defines completeness, inclusion, exclusion and comparability; missing is never silently converted to zero.
Can Neil prepare control charts and capability analysis?
Ask Neil to prepare the dataset, proposed method, calculations and interpretation for review. Your qualified team verifies assumptions, population, limits and implementation before using them for CPV decisions. This small demonstration deliberately does not report Cp or Cpk or claim a qualified control chart.
How does the review connect to APQR?
Keep the reviewed population, analysis definition, findings, linked investigations and open work with the periodic review. A later dataset or method revision should not silently replace the evidence behind an earlier conclusion.
