Knowing a process is not knowing its nominal settings. It is knowing how far it can move away from them before the product stops conforming, and why.
That knowledge is built, not observed. PAT-INDUSTRY establishes it with development and process teams, in pharma, biopharma, chemicals, cosmetics and food.
Three domains best kept apart
What you do
The nominal settings, and the narrow band production actually lives in.
Often far narrower than the recipe allows.
What you have demonstrated
The combination of parameters and material attributes for which quality is established by data rather than assumed.
In pharma, the design space in the sense of ICH Q8.
What you do not know
Everything else. A process can run there perfectly well — until the batch where it does not.
That is where unexplained deviations live.
Robustness is the distance between the first domain and the second. A process run at the centre of a widely demonstrated space absorbs a drift in raw material. Run at the edge of a barely explored one, it produces a non-conforming batch and an investigation that goes nowhere.
The names each of these three domains carries
The vocabulary exists, and it is worth using: anyone trained in Quality by Design will look for it.
- What you do — the normal operating range, NOR. A term of professional usage: it does not appear in ICH Q8(R2), unlike the two below.
- What you have demonstrated — the design space in the sense of ICH Q8(R2), not to be confused with a proven acceptable range, PAR: a range established “while keeping other parameters constant”.
- What you do not know — beyond it lies the edge of failure, to which ICH Q8(R2) devotes section 2.4.6.
And the sentence that settles it comes from the text: “A combination of proven acceptable ranges does not constitute a design space” — ICH Q8(R2), § 2.4.5. Varying one parameter at a time yields useful knowledge, not a design space: a design space is multidimensional, and it exists only where the interactions have been explored.
How a demonstrated domain is built
Not by accumulating routine batches. Production always manufactures at the same point: a thousand identical batches demonstrate that a point works, not that a range works.
You have to vary deliberately, and in an organised way. That is what design of experiments is for: choosing the combinations to try so that effects can be separated, interactions between parameters included, with a bearable number of runs.
Two lessons recur from one plan to the next. Interactions often matter more than main effects — temperature alone explains nothing, temperature at high humidity explains everything. The useful domain is rarely a rectangle: the acceptable limits of one parameter depend on the values of the others.
In-line measurement changes the economics of this exercise. It gives a full trajectory per run instead of a single end point, so far more information for the same number of batches spent.
What it commits you to, and what it does not
In pharma the design space is defined by ICH Q8(R2) as the multidimensional combination and interaction of input variables — material attributes included — and process parameters for which quality is demonstrated. The text sets out two consequences.
Moving within that space is not considered a change. That is its industrial value: operating latitude earned once and usable afterwards. Leaving the space is a change, and falls under post-approval change procedures.
One watch-point deserves stating plainly: the design space is proposed by the manufacturer and subject to assessment and approval by the authority. We describe here what the approach demands and produce the technical elements; we do not presume the regulatory outcome, which is built with your regulatory affairs team.
Outside pharma none of this is regulated, and the value remains whole: knowing how far you can go without risking the batch is useful whether a text requires it or not.
Frequently asked
Do you need a design space to do PAT?
No. The two reinforce each other without conditioning each other. An in-line measurement can be installed on a process whose nominal settings are all that is known, and it will produce knowledge from the first batches. Conversely, a design space established without continuous measurement remains useful. The combination gives the most.
How many runs does it take?
It depends on the number of parameters retained and the interactions to separate, and that is precisely what a design of experiments serves to minimise. The order of magnitude is computed once the list of critical parameters is settled — that is, after the step described in What impacts quality, never before.
Does a space demonstrated at pilot scale hold in production?
Not directly. Transfer, mixing and heat transfer phenomena do not carry over unchanged when scale changes. The usual practice is to establish the knowledge at pilot scale, then verify in production the points that govern the scale change. An in-line measurement present at both scales makes that comparison far more direct.
Do you know how far your process can move?
If the answer is “we have never tried”, that is a useful answer. Forty-five minutes are enough to see what an instrumented design of experiments would bring on your case.