Before you run process performance qualification, your development and engineering batches already answer the question that PPQ is supposed to confirm: is the process capable of meeting its critical quality attributes consistently inside the proposed ranges? If the existing data show a capability index whose lower confidence bound is below 1.0, or variability whose sources are unidentified, PPQ will either fail or pass by luck. The readiness assessment is the step that turns three expensive batches from a test into a confirmation.
The number of batches is not the criterion
The FDA process validation guidance (2011) and Annex 15 both dropped the fixed "three batches" rule in favour of a justified number based on process knowledge and risk. Three remains common, but three is a sample size, not a strategy. What a regulator asks is: why this number, what does it demonstrate, and what did you know before you started.
A useful way to think about it: PPQ is a confirmation with low statistical power. Everything that can be learned from statistics has to be learned in stage 1.
What can three batches actually show?
Take a critical quality attribute with a specification of 95.0 to 105.0 % and eight development results: 99.8, 100.6, 99.1, 101.2, 100.3, 99.5, 100.9, 100.0 %. Mean 100.18 %, standard deviation 0.71 %.
Performance index Ppk = min(USL − mean, mean − LSL) / (3 × s) = 4.82 / 2.14 = 2.26. That looks excellent, until you ask how precise the estimate is. The approximate lower 95 % confidence bound for Ppk is:
Ppk − 1.645 × sqrt(1/(9n) + Ppk² / (2(n−1)))
| n | Ppk estimate | Lower 95 % bound |
|---|---|---|
| 3 | 2.26 | 0.37 |
| 8 | 2.26 | 1.25 |
| 15 | 2.26 | 1.54 |
| 30 | 2.26 | 1.76 |
With three batches, the same point estimate of 2.26 only guarantees 0.37. That is the statistical reason why three PPQ batches cannot establish capability: they can confirm what larger development data already showed.
A better question than Cpk: what fraction will conform?
A tolerance interval says, with stated confidence, that a given proportion of future units lies inside an interval. For the eight results above, a two-sided interval covering 99 % of the population with 95 % confidence runs from 96.7 % to 103.7 %. Both bounds sit inside the specification, which is a direct, defensible statement: the process, as characterised, produces conforming material with margin.
That is the statement PPQ then confirms on commercial scale, equipment and operators.
The readiness checklist that matters
1. Every CQA has a measurement that is fit for purpose. Method validated, intermediate precision known, and small compared with the specification width. If the analytical variance is a third of the total, capability is partly a method problem. 2. Critical process parameters are identified and ranked. Not a list, a ranking with evidence: designed experiments, historical data analysis or a sensitivity study that shows which parameter moves which attribute and by how much. 3. Proposed ranges are supported. The normal operating range sits inside the proven acceptable range, and the edges of the range were actually tested, not assumed. 4. Variability sources are assigned. Batch-to-batch, within-batch, operator, raw material lot. A nested or mixed model on development data shows which component dominates; that component is the one that will decide PPQ. 5. Raw material variability is covered. At least two, preferably three, lots of the critical raw materials represented in development data, and the PPQ plan deliberately spans lots. 6. Scale effects are addressed. Mixing time, heat transfer, hold times and filtration behave differently at scale. Name each scale-dependent step and state the evidence that the chosen setting transfers. 7. The sampling plan can detect what you care about. Enough units per batch, at defined positions, to see within-batch variation. A single composite sample per batch shows almost nothing about uniformity. 8. Acceptance criteria are prospective and justified. Derived from capability targets, not from the specification alone, and written before the batches run. 9. Continued process verification is planned. Control charts, parameters and frequency defined in stage 3 before stage 2 starts, so the transition is continuous.
What to do when readiness fails
The honest outcome of a readiness assessment is often "not yet". Three typical findings and their responses:
- Capability driven by a single parameter. Tighten its control or add an in-process control before PPQ, rather than hoping three batches avoid the bad end of the range.
- Unexplained batch-to-batch spread. Run the designed experiment that development skipped. A PPQ failure costs more than a four-run study.
- Analytical variance too large. Improve or re-validate the method; capability cannot exceed what the measurement can resolve.
Limits of the statement
- Capability indices assume a stable, roughly normal process. Confirm stability with control charts and normality with a plot before quoting any index; a trending process has no single capability.
- Development and engineering batches are not commercial batches. Differences in equipment, operators and raw material supply are exactly what PPQ is for.
- A readiness assessment is a prediction based on historical data. It lowers the risk of a failed PPQ, it does not remove it.
- Product-specific regulatory expectations, for example for sterile products or biologics, can require more than the statistics suggest.
Frequently asked questions
How many batches are needed for PPQ?
There is no fixed number since the 2011 FDA guidance and the current Annex 15. The number must be justified from process knowledge, risk and the variability observed in development. Three is common for well-understood processes; higher variability or limited knowledge argues for more.
Can three PPQ batches demonstrate process capability?
Not in a statistical sense. With n = 3 the lower confidence bound on a Ppk point estimate of 2.26 is only 0.37. Capability has to be established on development and engineering data; PPQ confirms it at commercial scale.
What is the difference between Cpk and Ppk in process validation?
Cpk uses the within-subgroup variation of a demonstrably stable process; Ppk uses the overall variation including shifts between batches. For PPQ data, where between-batch variation is the point, Ppk is the more honest index.
Why use a tolerance interval instead of a capability index?
A tolerance interval states directly, with confidence, what fraction of future output lies inside which limits. It makes the sample size visible instead of hiding it in a single number.
What belongs in a PPQ readiness assessment?
Evidence that CQAs and CPPs are identified and ranked, that ranges are supported and their edges tested, that variability sources are assigned, that raw material lots and scale effects are covered, that the sampling plan can detect within-batch variation, and that acceptance criteria and continued process verification are defined in advance.