Continued Process Verification: Stage 3 Never Ends

CPV is Stage 3 of process validation. FDA's 2011 guidance restructured validation into three stages, and the third has no end date.

• Stage 1, process design. Define the commercial process from development knowledge.
• Stage 2, process qualification. Demonstrate the process performs as designed. This is where PPQ batches sit, and it is what most people mean when they say "validation".
• Stage 3, continued process verification. Ongoing assurance that the process remains in a state of control during routine production.

The earlier approach treated validation as an event, typically three batches and a report. Under the 2011 guidance it is a state of control that has to be demonstrated for as long as the product is made.

EU GMP Annex 15 asks for the same thing under the name ongoing process verification.

Stage 3 requirements

A defined monitoring plan. Which parameters and attributes, at what frequency, analyzed how. Chosen from the Stage 1 and 2 understanding of what matters, not from what happens to be easy to measure.

Statistical analysis. Data accumulating in a system nobody analyses is not verification; the program has to detect drift.

A trained person reviewing it, on a defined cadence, with authority to act.

Documented action when the analysis says the process is drifting.

Control limits versus specifications

A specification is the limit beyond which product fails. A control limit, from the process's own variability, is where the process tells you it has changed.

A result inside specification but outside control limits is not an OOS. It signals that something moved, usually before any batch fails. A CPV program that only reacts at the specification has given up the early warning it exists to provide.

Capability indices such as Cpk quantify the margin. A process running at Cpk 1.0 is meeting specification and has no room, which is a finding worth having before a batch fails rather than after.

Links to the APR and change control

CPV output belongs in the annual product review, and the two are often run as one exercise. It should also inform change control: a process trending toward a limit is the reason to change something, and a change is the reason to re-examine the monitoring plan.

Common failures

A plan that was never revisited. Written at launch, monitoring parameters that turned out not to matter, missing the ones that did.

Trending without limits. Charts with no control limits cannot signal anything, and they are common.

Signals with no action. The chart showed drift for eight months and the review recorded that it was noted. Eight months of drift recorded as noted is a quality-system finding in its own right.