A finished batch waits for its certificate of analysis. During that wait it takes up space, ties up cash, and teaches nobody anything. The third lever is to decide on the measurement already made. Lead time shortens, and so does quarantine.
PAT-INDUSTRY supports that shift in pharma, biopharma, chemicals, cosmetics and food. It happens in two steps, and the second is not the automatic sequel to the first.
Two steps, two levels of demand
Replace an in-process control
An intermediate sample sent to the laboratory to authorise the next step gives way to the in-line measurement.
- Waiting between steps removed
- Sampling bias removed with the sample
- Leaves the release dossier untouched
Replace a release test
Conformity of the finished product is established on process data and in-process attributes rather than on a final test.
- Cycle time shortened by the analytical wait
- Engages the dossier and is declared
- Assumes a demonstrated control strategy
The first step is decided internally and pays immediately. The second is a regulatory project as much as an analytical one. What a control costs, and when an in-line measurement pays for itself.
What real time release testing actually demands
Real time release testing is the ability to evaluate and ensure the quality of in-process material or finished product from process data, typically a combination of measured material attributes and process controls. The notion is defined in ICH Q8(R2) and expanded in the questions and answers common to Q8, Q9 and Q10.
What it assumes, and it is rarely the sensor:
- An established link between what is measured and the release attribute. Not a correlation seen on a few batches, a demonstrated and documented link.
- A method validated for release use, with everything that carries in robustness, domain of validity and monitoring over time.
- A defined course of action when the measurement is unavailable. A fallback test remains necessary, and its existence is part of the dossier.
- A written control strategy explaining why the set of controls retained covers the risks identified.
One example: dissolution. An NIR model built against the dissolution test predicts the profile of each tablet measured, without destroying it. The link between spectrum and release attribute is demonstrated on batches that cover the variability of the process, and the laboratory test remains the fallback. Predicting dissolution.
We describe here what the approach demands. The regulatory path is built with your regulatory affairs team and your authority; we do not presume its outcome.
Outside pharma, the same shift under other names
In chemicals, cosmetics and food there is no marketing authorisation dossier and no release test in the regulatory sense. The question moves, it does not disappear.
It becomes: what does your control plan say, and what has your customer accepted? A customer specification requiring a certificate of analysis per batch produces exactly the same wait and the same cost. Replacing it with an in-line measurement is negotiated, demonstrated, and written into the specification.
The economic reasoning is identical in every sector, and it turns on a single variable: the number of batches per year.
Frequently asked
Should real time release be the goal from the start?
Rarely. The first step — replacing an in-process control — produces a measurable gain without engaging the dossier, and it builds the data history the second one will rest on. Aiming straight at the second means asking for a demonstration before you hold the data that supports it.
What becomes of the laboratory?
It remains the reference. Models are built against its results and checked against them over time. What changes is the volume of routine samples, not the role of the laboratory, which shifts towards building and monitoring the methods.
How many batches does it take to demonstrate equivalence?
There is no universal number. What decides is coverage of the real variability: raw materials from several suppliers, seasons, shifts, equipment. Twenty batches that resemble each other demonstrate less than eight that cover the difficult cases. It is settled at scoping, against your historical data.
Which control keeps you waiting longest?
Tell us which one, and how many batches a year you run through it. Forty-five minutes are enough to place which of the two steps is within your reach.