Continued Process Verification (CPV) in Pharmaceutical Manufacturing

Continued Process Verification is ongoing assurance that a process remains in a state of control during routine US drug production, identified as Stage 3 in FDA process-validation guidance.

Operationally, CPV asks whether accumulated manufacturing experience still supports confidence in process performance. It is a longitudinal part of process validation, rather than another name for qualification or a decision to accept a particular batch.

For EU medicinal products, Annex 15 paragraph 5.28 applies its ongoing-verification provisions across traditional, continuous and hybrid validation approaches. Thus, the choice of validation approach does not define a separate meaning of lifecycle OPV.

In practice, passing batch results alone do not describe temporal stability: results can remain acceptable while moving consistently in one direction. A batch comparison overlay can help expose that pattern, but interpretation depends on manufacturing context. A CPV plan commonly defines the process-level question being monitored; approval responsibilities belong in the governance arrangements discussed later.

This article addresses pharmaceutical manufacturing. Applying its terminology or monitoring approach to food, supplements, devices or chemicals calls for a separate applicability assessment—not an automatic extension of pharmaceutical obligations. With the terminology comparison above in mind, distinguish each reference’s jurisdiction, product scope and authority before using it in a site procedure.

For pharmaceutical manufacturers using ICH Q10 across jurisdictions, section 3.2 describes a model intended to enhance elements that may already be required in part under regional GMP regulations. It lists process performance and product quality monitoring, corrective action and preventive action (CAPA), change management, and management review of process performance and product quality. The broader ICH Q10 model provides context for connecting these activities.

In practice, a useful site-level source map separates applicable legal obligations, guidance recommendations and locally selected implementation methods. For each procedure, record the relevant market and product, the reference used, and the rationale for the chosen approach. Keep FDA guidance recommendations separate from assertions about what the US drug records regulation requires; consult the wider 21 CFR 211 framework for context.

This distinction also helps during change control: reviewers can identify whether a proposed revision changes an implementation choice or needs a fresh assessment against applicable obligations.

For US drug manufacturing, FDA Stage 3 guidance recommends collecting relevant process trends and quality data for incoming materials or components, in-process material and finished products.

As a practical monitoring design, organise the inventory into quality attributes, process parameters, material characteristics and contextual signals. Select observations that answer a defined manufacturing question, using product knowledge and the control strategy. Inclusion does not make every variable a critical attribute or critical process parameter.

These are illustrative selections, not a universal monitoring list. For each selected measure, record the decision purpose, unit, sampling location, source and accountable owner in its monitoring definition. Distinguish individual sample results from batch summaries; identify whether moisture is reported on a wet or dry basis and what theoretical quantity forms the yield denominator.

Give counts an exposure denominator: deviation totals alone can rise because more batches were manufactured. Distinguish events per batch from the proportion of batches affected, since a batch can have multiple events. Preserve product, strength, batch size, material lot, equipment, campaign position and relevant change status as context. This lets reviewers ask whether a shift belongs to the process, a material population or a changed operating condition without treating every contextual field as a numerical trend.

For EU medicinal products, Annex 15 paragraph 5.31 recommends conducting ongoing process verification under an approved protocol or equivalent documents and preparing a corresponding report. For EU medicinal products, paragraph 5.30 recommends periodic review of the extent and frequency of ongoing process verification.

For US drug manufacturing, the FDA Stage 3 recommendation is to continue monitoring and sampling process parameters and quality attributes at qualification-stage levels until sufficient data support meaningful variability estimates.

Turn the objective into an operating plan

A practical CPV plan typically specifies the following operating details; this is a suggested structure, not a prescribed regulatory template.

• Population: identify product and process versions, manufacturing boundaries, inclusion and exclusion rules, and how excluded observations remain visible with their rationale.
• Sampling: explain which observations represent the population, where samples originate, and how gaps or uneven batch production affect interpretation.
• Analysis: identify methods, baseline-selection rationale, signal rules and conditions for revisiting the baseline rather than silently resetting it.
• Responsibilities: assign data preparation, technical assessment, quality review, report approval and follow-up ownership to named roles.
• Report content: define the review period, population, data completeness, findings, limitations, decisions, action owners and follow-up status.

Separate the monitoring clocks

In practice, distinguish observation frequency, signal assessment and scheduled aggregate review. Observations follow the sampling design; signal assessment has a defined response route; aggregate review examines accumulated performance and open decisions. A signal need not wait for the next scheduled report.

Justify cadence using data accrual, process risk, changes and unresolved signals rather than assigning every process the same calendar interval. Sparse production may justify reviewing each new batch while acknowledging limited trend information. A process change or unresolved signal may justify increased monitoring intensity. Record the rationale for adjustments and connect plan revisions to change control where applicable.

For US drug manufacturing, FDA Stage 3 guidance recommends that a statistician or person adequately trained in statistical process control develop the data collection plan and statistical methods and procedures. For US drug manufacturing, the same guidance recommends procedures that guard against both overreaction to individual events and failure to detect unintended process variability.

Separate acceptance from process behaviour

In statistical monitoring practice, specifications define acceptance boundaries for an attribute. Statistical control limits describe expected process behaviour estimated from a suitable baseline; they are not acceptance criteria. Operational alert or action thresholds define when a designated response begins and may reflect process knowledge or risk rather than statistical limits. A result can meet specification yet signal unusual behaviour, or breach specification without producing a control-chart signal.

Assess stability before interpreting capability as evidence of predictable performance against specifications. Capability estimates depend on the population, distribution and variation estimate; no universal capability cutoff is proposed here. Separate within-batch variation from between-batch variation. Rational subgrouping groups observations taken under sufficiently similar conditions so within-group variation provides a meaningful comparison with changes between groups. Track dispersion alongside location: an unchanged average can conceal widening variability.

Match the method to the question

• Time plots show sequence, drift and possible change points. Preserve actual production timing when gaps matter; a plot alone does not establish statistical control.
• Individual-observation charts suit ungrouped measurements when their dependence structure and variation estimate are appropriate. Repeated readings from one batch are not automatically independent observations.
• Subgroup charts compare location and dispersion across justified sampling groups. Choose the grouping around the manufacturing question, not simply the available spreadsheet layout; see statistical process control.
• An EWMA chart or CUSUM can help detect sustained shifts when that sensitivity is justified. Select settings around the shift of interest and the tolerable false-signal burden.

Define interpretation rules before use

Autocorrelation can distort conventional chart signals; non-normal distributions can make familiar limits or capability calculations misleading. Sparse production leaves baseline estimates uncertain. Changing sample sizes can change the precision of averages or rates, so limits may need to vary accordingly. Applying multiple signal rules increases opportunities for false alarms; document which rules answer a meaningful process question.

As a methodological safeguard, make a stated rationale for exclusions and baseline resets a condition of the proposed CPV plan. Do not remove inconvenient points or reset limits merely to erase signals. Treat a flag as a reason to evaluate manufacturing context, not an automatic instruction to adjust the process. Statistical output alone cannot establish cause or justify batch disposition.

These hypothetical cases illustrate practical review decisions, not regulatory procedures or site results.

In-specification assay drift

Assume one tablet strength, an unchanged assay method, comparable sampling and a previously reviewed baseline. Assay results are expressed as percent of label claim. Successive batch results rise, each above its predecessor, while remaining within specification. Assess the sequence against the monitoring plan’s selected signal rule rather than choosing a rule after seeing the chart. Record whether the sequence triggers that rule.

Check original laboratory results, calculations, transcription and batch ordering. Then examine manufacturing changes around the apparent onset, including material lots, settings and equipment interventions. For this example, the documented decision is to open an out-of-trend assessment if the selected rule is met after those checks; otherwise, record the rationale and follow-up monitoring. The chart alone establishes neither a cause nor a batch-release decision.

Pooling hides a material-related difference

Assume one granulation process, supplier identity retained for each material lot and moisture expressed as mass percent. The pooled average appears unchanged, but plotting results by supplier reveals different distributions. Retaining supplier identity as a lot attribute makes that comparison possible. Compare supplier-specific centers and spreads, each supplier’s contribution to the pooled population, and within-batch moisture patterns at comparable sampling positions.

An unchanged overall average can coexist with widening variability or offsetting subgroup shifts. Review sampling coverage and whether supplier groups coincide with different equipment, campaigns or operating conditions. Stratification identifies a question for investigation; it does not prove that supplier material caused the difference. The next action is a documented comparison of matched manufacturing conditions before attributing a cause.

Interrupted equipment-data transfer

Assume drying temperatures recorded in degrees Celsius, a missing transfer interval and recoverable source records. A timestamp-continuity check detects the gap in the monitoring dataset. Qualify the current review as incomplete for that interval and identify affected batches and calculations. Use a controlled recovery from the original records, preserving timestamps, equipment identity and recovery history across the MES–historian connection.

Reconcile overlapping records and duplicates before recalculating affected summaries and signal tests. Retain the earlier review, link the revised assessment and explain changed conclusions. If recovery fails, temperature behavior during the missing interval remains unresolved; assess alternative evidence explicitly rather than treating interpolation as an observed record.

A proposed vendor-neutral flow takes batch execution records, laboratory results, equipment histories and quality-event records through reconciliation into a controlled analytical dataset. Link batch execution history through batch, product, material-lot, equipment, sample and process-version identifiers. Use explicit relationship tables for multiple lots or equipment items rather than duplicating observations. Preserve event time separately from import time, and retain source values and units alongside normalized values, units and conversion rules.

The data-model diagram below separates identity, measurements, monitoring logic and human interpretation. Field names are proposed analytical entities, not an application schema.

A successfully imported result can still be unsuitable for pooling without its sampling location, method or population context. Production context tagging helps distinguish comparable observations from values that merely share an attribute name.

Reconciliation controls typically compare expected batches with received records, flag late arrivals and identifier mismatches, detect repeated imports using stable source-record keys, and quarantine unexpected unit changes. Assign failed-transfer diagnosis to an interface owner and acceptance of corrected mappings to the data owner. Replay only after resolving the cause, defining the affected records and confirming duplicate protection. Retain the prior dataset version and flag affected signals for reviewer reassessment. Apply these controls to the interrupted-transfer case above rather than treating restored connectivity as proof of completeness.

Process analytical technology can supply dense process observations; multivariate monitoring assesses related measurements together rather than treating each trace independently. They are analytical options to consider within the proposed monitoring design. Align timestamps and process milestones before comparing trajectories: startup, steady operation and shutdown can have different relationships between measurements. Training data should represent the intended products, equipment configurations and operating conditions. Before interpreting an output, check whether the current batch falls within that model domain; unfamiliar conditions can make an apparently precise score unreliable.

A golden batch is a chosen empirical reference, not a universal target or, by itself, an approved acceptance criterion. Selecting only unusually smooth batches can bias comparisons and label legitimate process variation as abnormal. Document why the reference represents the intended operation, and watch for model drift as materials, equipment or process behaviour change. When a multivariate signal appears, examine contributing measurements, their timing and the operating phase, then compare them with batch events. Use that interpretable follow-up to guide investigation rather than treating the score as a diagnosis.

As implementation practice, test the CPV analytical workflow against its intended manufacturing question, not merely whether software runs successfully. Link these checks to the site's computerized system validation approach:

• Trace known records through extraction, field mapping, unit conversion, population selection, calculations and report generation; compare outputs with independently established expected results.
• Challenge boundary cases, missing values, rejected results and method changes. Check that exclusions remain visible and that changed methods are not silently pooled.
• Test permissions for changing mappings, analysis settings and review status. Confirm that reports retain the population, units and analysis version actually used.

Successful execution demonstrates that a workflow completed; it does not establish that the configured population and calculations answer the intended manufacturing question. Include a manufacturing reviewer in that assessment.

For reproducibility, preserve source references, transformation versions, analysis versions, review status and reasons for corrections or exclusions. Retain the dataset used for each conclusion, or a reproducible snapshot with its selection rules, so a reviewer can reconstruct the earlier report rather than query only today's corrected records.

Correcting an erroneous source value is different from excluding an inconvenient observation. Preserve the original value and correction rationale; justify exclusions scientifically rather than by their effect on the trend. Connect this history to audit trail review.

A CPV signal is an observation needing evaluation, not a confirmed cause. In a proposed governance workflow, signal assessment determines whether a deviation investigation is appropriate; a CAPA decision separately determines whether corrective or preventive action is justified.

1. Check data validity: confirm the observation against source records, calculation logic and manufacturing context before interpreting it.
2. Assess product and process impact: identify potentially affected batches, quality attributes and the possibility that the signal extends beyond the reviewed population.
3. Assign investigation ownership: name an accountable person, involve relevant technical specialists and quality personnel, and set a review date.
4. Document the signal disposition: record the interpretation, uncertainty, rationale and links to any deviation investigation or CAPA decision.
5. Select proportionate follow-up: additional sampling, focused investigation, closer monitoring or controlled process changes, with an owner and effectiveness criteria.

Escalate unresolved signals, recurring minor changes and evidence that the monitoring plan no longer represents the process. Handle baseline changes through change control, retaining the previous baseline and justification rather than resetting limits to absorb deterioration. Assess whether additional qualification work is warranted; not every signal calls for revalidation. Check effectiveness against predefined observations after implementation, and reopen the assessment if concerns persist.

Keep batch approval and release with authorized people. Software does not make a site compliant or replace accountable quality decisions.

For EU medicinal products, Annex 15 paragraph 5.32 recommends ongoing process verification throughout the product lifecycle to support validated status documented in the Product Quality Review.

Use these practical questions to review CPV readiness, not to certify compliance. Look for reviewable artifacts rather than assurances that a control exists.

• Does the scope document define the monitored population, boundaries and exclusions?
• Are accountable owners assigned for data preparation, statistical interpretation and approval?
• Are selected measures, units and denominators documented?
• Does the sampling rationale explain coverage across the defined population?
• Are monitoring documents approved, versioned and available to reviewers?
• Can data joins be traced to source records and relevant production context?
• Are calculations tested against expected results, including missing-data and duplicate-record cases?
• Are signal rules justified for the population and method?
• Do review records capture interpretation, decisions and unresolved questions?
• Does each open action have an owner, with an effectiveness check where appropriate?
• Is there a documented route for reassessing the plan after new knowledge or process changes?

Finish with a reconstruction test: for a selected signal, can a reviewer identify the observations, method version, interpretation and decision? Use the earlier statistics, integration and governance discussions to assess gaps.