Conventional quality control samples a few points and judges the result. Quality by Design builds quality during manufacturing, and the measurement serves to steer it rather than to record it.
Both approaches coexist in most plants, and the second does not abolish the first. PAT-INDUSTRY supports that shift in pharma, biopharma, chemicals, cosmetics and food.
Two ways to reach the same conformity
Quality by Testing
- It establishes the reference value, and it is what releases the batch
- Snapshots, at a few chosen moments
- Most often at the end of the operation
- The verdict arrives after the decision was taken
- Between two samples the process is not observed
- Every sample carries its own sampling bias
Quality by Design and PAT
- A continuous measurement, from start to finish
- On product attributes and process parameters alike
- In-line, on-line or at-line, depending on access
- The deviation shows while you can still act
- In-line, no physical sample is taken — but the sensor’s field of view is still an increment, and its position decides representativeness
The difference is not the accuracy of the measurement, it is when it arrives. A laboratory always ends up telling you what the batch contained. It tells you after the batch is made.
What Quality by Design is not
Three misunderstandings keep coming back, and each one costs a project.
It is not a dossier. Quality by Design produces documentation, but its object is process knowledge. A thick dossier on a poorly understood process is not QbD; a few pages that genuinely explain where variability comes from is.
It is not reserved for new products. The approach was born in development and applies to a process that has been in production for fifteen years. The difference is that you then hold a history, which is an advantage rather than a handicap.
It is not the end of quality control. The laboratory remains the reference against which models are built and checked. What changes is the routine volume, not the role.
What it changes on the shop floor
The shift shows in three everyday actions.
The decision to stop. Under Quality by Testing an operation stops on a validated duration, computed with a margin that covers the slowest batch. Every other batch pays that margin. With PAT it stops when the followed quantity reaches its criterion.
Deviation investigation. Without a continuous measurement you reconstruct what may have happened from a few points and the memory of the team. With the process signature you look at the curve and see the moment it diverged from conforming batches.
Scale-up and site transfer. Comparing two processes on end-of-batch results tells you whether they give the same product. Comparing their trajectories tells you whether they manufacture in the same way — a different question, and a different robustness.
Where the term comes from, and what the texts say
Quality by Design is formalised for pharmaceuticals by ICH Q8, complemented by Q9 on risk management and Q10 on the quality system. The FDA PAT guidance, published in September 2004, is its instrumental side: it describes how measurement serves the understanding and running of the process.
And validation itself changes form. ICH Q8(R2), in its appendix, sets validation “primarily based on initial full-scale batches” against a “lifecycle approach to validation and, ideally, continuous process verification“. This continuous process verification is defined there as “an alternate approach to process validation in which manufacturing process performance is continuously monitored and evaluated“.
⚠️ Not to be confused with the FDA’s continued process verification, which is the third stage of its validation lifecycle, after process design and process performance qualification. The two expressions differ by a single word and do not designate the same thing: one replaces classical validation, the other extends it. It is a frequent confusion, and it changes what is promised in a dossier.
We keep the English term, in French as well. It is the one the professional community uses, and translating it would make the point less clear, not more.
Outside pharmaceuticals, the approach has no single founding text, and yet it is widely practised: statistical process control, design of experiments, capability. The vocabulary differs, the logic is the same.
When is a process “well understood”? The FDA definition
The FDA PAT guidance — PAT — A Framework for Innovative Pharmaceutical Development, Manufacturing, and Quality Assurance, CDER/CVM/ORA, September 2004 — sets out three conditions.
- All critical sources of variability are identified and explained.
- Variability is managed by the process itself.
- Product quality attributes can be accurately and reliably predicted over the design space established for materials used, process parameters, manufacturing, environmental and other conditions.
The text adds a sentence that works as a criterion: “The ability to predict reflects a high degree of process understanding.” Prediction is the test — it is what attests to the degree of understanding reached.
Quality by Design calls for that understanding. Measurements taken during the process are what makes it possible to establish, then to demonstrate — the thread running through the four phases of our approach.
Frequently asked
Do you have to choose between the two?
No, and presenting them as exclusive would be wrong. Almost every plant combines them: conventional release controls, plus one or two in-line measurements where they change a decision. The movement is gradual, operation by operation.
Does a continuous measurement produce too much data?
It produces a lot, and the question is not the volume but what you keep and why. Raw spectrum, pre-processed spectrum, predicted value: the three have neither the same usefulness nor the same useful life. That decision belongs at the start of the project, not to the day the disk fills up.
Does Quality by Design mean revalidating?
Adding a measurement in observation does not change the validated process. It is the decisions taken from it that may constitute a change — stopping on a measured criterion, replacing a control. The question arises when the measurement stops observing and starts acting, and it is anticipated at scoping.
On which operation are you paying a safety margin?
A fixed duration, a wait for a result, a systematic control. Forty-five minutes are enough to estimate what a continuous measurement would change there.