Continued Process Verification (CPV): the Stage 3 program that prevents surprises

The typical scene: “Show me the trends for the last 12 months”

During an inspection, when the inspector asks:

• “Trend of CPPs and CQAs”
• “How do you define and manage an Out-of-Trend?”
• “Which signals trigger a deviation before an OOS occurs?”

…what I often see are weak answers, not because data are missing, but because a structured CPV program is missing: scattered data, “annual” charts, no signal rules, no clear ownership.

The GuideGxP guide explicitly recalls the need for Continued/Ongoing Process Verification (CPV/OPV) after initial validation, with continuous monitoring and trending.

The myth to dismantle: “We completed the 3 PPQ batches, so we’re covered”

That mindset may have been “acceptable” in a more static world. Today it is risky because:

• processes change due to micro-causes such as wear, raw material lots, maintenance, and software updates
• deviations often begin as drift and only become OOS later
• the inspection expectation is to see proactive control, not only reactive response

Contrarian insight: CPV is not just there to “keep QA happy” or satisfy the inspector. When done properly, it is one of the best tools for reducing waste and repetitive deviations without adding useless controls.

Real CPV = Stage 3 Process Validation + SPC + governance

A robust CPV has three components:

• What you monitor: CPPs/CQAs plus process indicators
• How you monitor it: SPC and signal rules
• What you do when it “rings”: OOT → triage → escalation workflow

Mini-glossary (operational, not theoretical)

Stage 3 Process Validation: maintaining evidence that the process remains in a state of control.
SPC (Statistical Process Control): statistical rules that distinguish noise from signal.
OOT (Out-of-Trend): a drift signal even when you are still within specification.

The “CPV Canvas” table I recommend using (1 page, not 40)

Which control chart to use, without overcomplicating it

From a production perspective, starting with I-MR on 5–10 key parameters beats a thousand useless charts.

OOT is not an opinion: define rules and make them live

Examples of practical triggers, inspection-style:

• 1 point beyond the action limit → event
• 2 consecutive points beyond the alert limit → OOT
• 6 consecutive points increasing or decreasing → drift
• 8 consecutive points on the same side of the mean → process shift, often seen after maintenance or raw material changes

These are operational applications of the Nelson rules, or similar rules. You do not need to be mathematicians. You need to be consistent and documented.

The part that really makes the difference: data stratification, otherwise you mislead yourself

In audit, a classic question is:
“Can you show me whether the trend is the same across all lines and shifts?”

If you do not stratify, you risk averaging everything and losing the signal. I would typically stratify by:

• line / equipment train, for example Line 1 vs Line 2
• shift / crew, for example Shift A/B/C
• raw material lot / supplier
• campaign phase, for example start vs end of campaign
• maintenance intervention, pre vs post

This is often the key to turning an OOT from a “mystery” into a “plausible cause.”

Golden Batch and multivariate tools: when you want to step up without hype

If you have many correlated parameters, which is typical in complex processes or digitalized plants, you can introduce:

• the golden batch concept, meaning the expected “good” profile
• multivariate monitoring such as PCA (Principal Component Analysis) and statistics like Hotelling’s T² to detect abnormal patterns before they become nonconformities

A mistake I see often: the company buys advanced tools but never defines who actually “calls” the anomaly and what happens next. Technology without governance is just noise.

Practical shop-floor case: slow drift in compression

Scenario: tablets with hardness specification 80–110 N, still compliant, but CPV shows:

• I-MR: 7 batches trending upward toward 108–109 N
• increase in micro-rejects due to borderline friability defects

Manufacturing Manager investigation, before it becomes OOS:

• check punch and cam wear
• verify lubrication and actual compression force versus setpoint
• correlate with excipient lot, especially flowability
• verify environmental conditions such as RH that affect compressibility

Effective action: targeted maintenance plus early punch replacement, updating the maintenance plan to a condition-based approach.
Typical result: no critical deviation, fewer rejects, clean audit trail.

Operational workflow: what to do when a CPV signal is triggered

• Triage: is it a data error, a known change such as maintenance, or a real signal?
• Impact assessment: which CQA may be affected? Is batch/campaign hold needed?
• Containment action: setpoint adjustment, line stop, segregation?
• Root cause: instruments, materials, method, people, environment, using Ishikawa logic
• CAPA + effectiveness check: demonstrate that the trend returns and remains stable

What to remember

• CPV is not an annual chart. It is SPC + rules + decisions.
• Without stratification, you lose the signal and only think you are under control.
• The inspector does not ask how much data you have. They ask how you use them to prevent problems.

A realistic 6-week implementation roadmap

Week 1: select 5–10 high-risk parameters, on 1–2 pilot products.
Week 2: map data sources and ownership: MES, SCADA, LIMS, data historian.
Week 3: define charts, limits, OOT rules, and governance, including RACI.
Week 4: run the pilot and correct issues: dirty data, inconsistent definitions, gaps.
Week 5: train supervisors and establish a weekly 30-minute CPV routine meeting.
Week 6: go live and perform the first effectiveness check: were signals managed correctly?

FAQ

1) Are CPV and PQR the same thing?
No. PQR is a broad periodic review. CPV is a living program with rules and triggers designed to anticipate deviations and OOS.

2) How many parameters should I trend?
Fewer, but critical ones. Use a risk-based CPP/CQA approach, with limits and actions. “Everything” only creates noise.

3) Who should own CPV?
In a mature model: Production as operational owner, QA for governance, and MS&T for scientific support. The important point is that the owner must have the authority to trigger action.

If you want examples and operational references on continuous monitoring, trending, and integration with the quality system, the Manufacturing Manager Technical Guide is specifically designed to translate GMP expectations into practical tools.