Continuous manufacturing, and what it asks of the measurement

The ICH Q7 definition of a batch still applies in continuous manufacturing. What changes is the quantity that fixes its size: § 2.2 allows three — the quantity of output material, the quantity of input material, or the run time at a defined mass flow rate. Other approaches stay open where they are scientifically justified.

A batch can also be defined as a range, between a minimum and a maximum run time. Changes inside that range are managed within the pharmaceutical quality system; beyond it, they call for a post-approval change (§ 4.3).

Two practical consequences, and they show early.

• The batch size actually intended is written before manufacturing begins.
• A quantitative metric for batch-to-batch consistency is defined within the quality system, with its acceptance criteria. Where the batch is defined by the material collected, the amount diverted relative to the amount collected is one such metric.

It characterises the time available for material transport and transformation. It is specific to the process, the composition, the material properties and the equipment design (§ 3.1.2). It is what makes material tracking possible, and it underpins the sampling and diversion strategies.

The guideline sets out how to establish it, and the conditions are demanding.

• The characterisation covers the planned operating ranges and the anticipated input material variability.
• The method leaves untouched the dynamics it measures, and it is relevant to the commercial process.
• A tracer, where one is used, has highly similar flow properties to the constituent it replaces, stays inert towards the other components, and leaves unchanged how the material interacts with equipment surfaces.
• Step testing on the composition of the stream, a quantitative change in one constituent, yields the distribution with no external tracer. It is often the easier route to defend.

A change in the dynamics carries into process characteristics such as conversion yield or impurity formation. So this is not a separate modelling exercise: it is a property of the process, on the same footing as a temperature profile.

ICH Q13 says what the characterisation has to cover. It does not say how it is obtained. The USP technical guide Process Modeling in Pharmaceutical Continuous Manufacturing does, and the sensor it puts in the figure is the one this site is about.

• The pulse input method. A tracer is introduced at once at the blender inlet. The guide describes the set-up: blender, vibratory conveyor, and a near-infrared probe mounted over the conveyor, serving as the detector — “A near-infrared (NIR) spectroscopic probe served as a detector.” The three replicates overlay each other, and it is that overlay which establishes the reproducibility of the trial.
• The step change method. Rather than an external tracer, the concentration of a blend ingredient is varied, typically the API. The guide gives the reason: it lowers the risk of introducing uncontrolled disturbances into the process dynamics. In exchange, a substantial amount of material has to be added or removed for a readable output signal.
• The whole line. Unit-operation distributions combine by convolution. The resulting line model serves, the guide writes, for “material traceability, material diversion, and real-time quality assurance” — the three subjects of the next section.

One watch-point the guide states, and worth knowing before designing the trial: the continuity assumption does not hold at the feeders. Twin-screw feeders have dead zones where material stays stagnant, and “as the filling level decreases, more stagnant material is released”. The guide reads that as a potential threat to process integrity and patient safety, through segregation and the unpredictable release of non-conforming material. A feeder is not a measurement point like any other.

USP Technical Guide, Process Modeling in Pharmaceutical Continuous Manufacturing, © 2024 The United States Pharmacopeial Convention, Rockville, MD — § 4.3 pulse input, § 4.4 step change, § 5.5 feeding, § 5.6 full-line RTD. An informational document: it carries no requirement, and says so on its own first page.

A continuous line produces material during start-up, shutdown and disturbances. The ability to detect and divert the material concerned is a characteristic of continuous manufacturing, and it belongs to the control strategy (§ 3.1.6).

The in-line measurement holds a role of its own here: it determines in real time when collection stops and when it resumes. What the diversion strategy describes in the dossier, per § 4.2:

• The criteria that trigger diversion, and those that end it.
• The basis for determining the extent of diverted material. It incorporates a justified safety margin covering the uncertainty of the residence time distribution and of the measurements.
• The conditions for resuming collection.
• The effect of diversion itself on material flow and process dynamics: diverting material changes the system being steered (§ 3.1.6).

Procedures for collection, diversion and disposition stay within the pharmaceutical quality system; it is the strategy that goes into the dossier.

This is where continuous manufacturing states a requirement that batch manufacturing left open. Section 4.2 asks that the data analysis method actually detect the disturbances and the variability in the process, and it is specific on two counts.

• Averaging runs over appropriate time intervals rather than across the entire run time. An average over the whole campaign smooths exactly what diversion needs to see.
• Those intervals are chosen by relating the PAT measurement frequency to the residence time distribution and to the process response time. Statistical sampling plans and data analysis are described and justified.

Section 3.1.5, for its part, asks that measurement interference with the process be avoided, and that the impact of physical sampling on the material stream, which can affect the state of control, be considered. What a sampling plan asks for.

That is the regulatory wording of a rule we apply on every project: the first question is the pace at which you have to decide, and the technology follows from the answer. What makes a measurement hold.

Annex V describes three cases on one example: a transient change in the flow of a loss-in-weight feeder supplying a blender, with the acceptance criterion set at ± 20 % for 80 seconds and the content required to stay between 90 and 110 % of label claim.

Where that 80-second criterion comes from: a funnel plot, which ties the magnitude and duration of a disturbance to its effect on blend content. The text reads from it that a ± 20 % change lasting under 90 seconds would not take the content outside 90-110 %, and adopts 80 seconds as the criterion for discussion. It is the tool that links the dynamics to the specification, and it is built during development.

In the first two cases the text names the same confirmatory check — a near infrared measurement at the tablet press feed frame — and the third relies on the same measurement without naming its location. That is where blend content is verified closest to the forming step. Content uniformity, and the scale it is posed at.

A criterion on the frequency of disturbances can itself be set, and it is the least expensive answer to the third case.

This is one of the most directly useful contributions of the guideline, and it carries well beyond continuous manufacturing. Section 3.1.3 writes that input materials may require evaluation and control of attributes beyond those typically considered for a material specification used in batch manufacturing.

The examples are named. For a solid dosage form: particle size, cohesiveness, adhesiveness, hygroscopicity, static charge and specific surface area of drug substances and excipients, which govern powder feeding and material flow through the system. For a chemically synthesised process: viscosity, concentration, or the multiphase nature of the feed. For a therapeutic protein: lot-to-lot variability of cell culture media and feed components.

In other words, a lot that meets its specification perfectly can still behave differently. That is also what moves a model outside its domain.

Validation requirements are the same as for batch manufacturing, and both routes stay open: traditional validation on a fixed number of batches, or continuous process verification, where real-time data demonstrate that the state of control is maintained for the run time duration (§ 4.7). The choice is justified in the dossier, on product and process understanding, system design and the overall control strategy.

Continuous manufacturing lends itself to it because it produces, continuously, the data that feed that demonstration. And since output can be changed without increasing equipment size, development knowledge is built at the scale intended for commercial manufacturing (§ 3.3).

On real-time release testing the guideline is explicit: it can be applied to one or more output material quality attributes, and it is not a regulatory requirement for continuous manufacturing (§ 4.2). Many installations deliberately stop short of it, and that is a defensible choice once it is named. The three levels of ambition.

Where it is proposed, the associated reference test method is described, and the fallback plan for when the model is not available goes into the dossier. What a continuity plan holds.