Process validation and equipment qualification

Validation looks like paperwork and is actually a statistical argument. Consider the two properties a sterile biologic must have: sterility and freedom from adventitious virus. Both are rare-event properties. A sterility test examines around twenty containers; if one in a thousand were contaminated it would pass almost every time. A virus assay has a detection limit, and absence below that limit is not absence. Sampling cannot reach the confidence the claim requires without consuming the batch. So the assurance must come from somewhere else — from evidence that the process, run as designed, cannot produce the defect at a meaningful rate.

What the three stages are actually for

The FDA’s 2011 process validation guidance restructured the field around a lifecycle, and the logic of the three stages is evidential rather than procedural. Stage 1, process design, establishes which product attributes matter clinically and which process parameters move them — the critical quality attribute to critical process parameter link that ICH Q8 and Q11 formalise as a design space. Without that mapping the later stages have nothing to hold constant. Stage 2, process performance qualification, demonstrates at commercial scale and with commercial equipment that the process reproducibly stays inside those limits; the number of runs is a function of process variability and prior knowledge, not the “three batches” convention it replaced. Stage 3, continued process verification, treats the process as a system under statistical control and monitors capability over time, which is where drift is actually caught.

Equipment qualification sits underneath. Installation, operational and performance qualification each answer a different question — is it what was specified, does it do what it is meant to across its range, does it do it with this product — and the third is not implied by the first two.

Where the numbers come from

Two examples show that the acceptance criteria are derived, not conventional. Cleaning validation limits used to be set by rules of thumb: a thousandth of a therapeutic dose, or ten parts per million. The EMA’s 2014 guideline on shared facilities replaced these with health-based exposure limits — a permitted daily exposure derived toxicologically from the compound’s own no-effect level. That converts an arbitrary number into a defended one, but it also transfers the whole argument onto swab recovery: a limit is meaningless if the recovery of residue from the surface has not itself been measured.

Viral clearance is the second. Clearance is claimed as the sum of log reduction values from orthogonal steps — low-pH inactivation, solvent/detergent treatment, nanofiltration, chromatographic partitioning. Adding them is legitimate only if the mechanisms are genuinely independent; two steps that both work by hydrophobic partitioning do not give a sum, and a step run near exhaustion may give far less than its validated value.

The assumption underneath everything

Both of those claims, and most others, are generated on scaled-down models — a small column, a bench filter — because you cannot spike virus into a commercial batch. The entire structure therefore rests on the scale-down model being representative, which is demonstrated by comparing operating parameters and performance, never proved outright. Validation also proves the process as it was run. A change in a raw-material supplier, a resin lot or a hold time can invalidate the argument without violating a single acceptance criterion, which is why change control, and not the validation report, is what keeps the claim true.