IPCIn-Process Control

An In-Process Control (IPC) is a test, check or measurement taken on the product or the process during manufacturing — between the receipt of starting materials and the release of the finished batch — to confirm the batch is performing inside its validated control strategy. The defining word is product. An IPC asks 'is the material we are making what it needs to be at this point in the process?' That distinguishes it from an in-process verification (IPV), which asks 'did the operator perform the step the way the procedure required?' and from finished-product release testing, which asks 'is the final batch within spec?'

The categories overlap in everyday speech but separate cleanly in a well-designed batch record. The IPV proves the dispense happened with the right material, the right scale and the right witness. The IPC proves the blend that resulted from that dispense has the assay, blend uniformity and moisture content the process needs at that point. Both are required; both live in the BMR; neither substitutes for the other.

Every major regulated-manufacturing framework requires in-process testing, even where the phrase 'in-process control' is not used. The common requirement is that the manufacturer has identified the critical points in the process, defined what must be measured at each, set the acceptance criteria in advance, and documented the result before the batch progresses past that point.

Two themes unite the citations. First, the IPC plan is required to exist in writing before the batch starts — not derived from inspector questions after a deviation. Second, the criteria for acceptance must be established in advance: 'we will measure assay at the granulation step and accept 95.0–105.0% on a single sample' is an IPC; 'we measure assay sometimes' is not.

IPCs cluster at the points where a deviation, if missed, will be costly or impossible to correct downstream. The exact list depends on the process, but the design principle is universal: place an IPC immediately before a step that locks the batch into a state the next step cannot undo.

Solid-dose pharmaceutical (tablet) example

• Dispensing: tare-verified weights, material identity scan (this is an IPV with a paired IPC for assay-adjusted charge calculation).
• Granulation: granulation endpoint (torque, power, particle size), loss-on-drying after drying.
• Blending: blend uniformity by stratified sampling, sieve analysis, bulk density.
• Compression: average and individual tablet weight, hardness, thickness, friability, disintegration.
• Coating: weight gain, dissolution after coating, visual defects.
• Packaging: fill count, leak test, label verification, lot/expiry print check.

Biologic example

• Cell expansion: viable cell density, viability, glucose, lactate, dissolved oxygen, pH.
• Harvest: turbidity, titer by quick assay.
• Purification: A280 protein concentration, conductivity at chromatography pool, bioburden after every open step.
• Fill-finish: pre-fill bioburden, post-fill weight check, container-closure integrity test (CCIT).

Food / dietary supplement example

• Receiving: identity verification, moisture for hygroscopic ingredients, gluten screen for allergen-free claims.
• Mixing: assay or actives uniformity, moisture, microbial bioburden screen.
• Encapsulation/tableting: fill weight, content uniformity, disintegration.
• Packaging: torque on closures, fill count, label and 2D-barcode print verification.

ICH Q9 risk-management thinking has moved IPC design away from 'we test everything we can' to 'we test what controls a CQA that has no other line of defence.' The modern IPC plan is the output of a process-knowledge exercise: each critical quality attribute is mapped to the process parameters that influence it, each parameter is rated for criticality and detectability, and IPCs are placed where a critical parameter is not already controlled by a validated engineering control or upstream qualification.

1. List every CQA for the product (assay, uniformity, dissolution, sterility, identity, appearance).
2. For each CQA, list the CPPs that influence it — from a fishbone or a DoE.
3. Rate each CPP for severity, probability and detectability under current controls.
4. Where current controls are weak, place an IPC at the point where the parameter manifests.
5. Set acceptance criteria from the design space (ICH Q8), not from historical averages.
6. Define the sampling plan: how many, how often, how stratified.
7. Define what happens when the IPC fails (hold, abort, investigate, regrade).
8. Validate the IPC method and the sampling plan against ICH Q2 and USP <1010>.

An IPC result is only as good as the sample that produced it. Two sampling errors dominate FDA observations: insufficient sample size to detect the variation present in the batch, and non-stratified sampling that systematically misses parts of the batch. Both are statistical, not procedural, problems — and both are addressable by an honest sampling-plan design.

Sample size

For attributes data (pass/fail) the AQL framework in ANSI/ASQ Z1.4 gives defensible plans. For variables data (assay, weight, hardness) the sample size is driven by the required confidence and the expected process variation; for blend uniformity, USP <905> and the 2003 PQRI draft drive 10 or more stratified samples at top/middle/bottom of the blender.

Stratified sampling

Pulling 10 tablets from the same chute is not a sample of the batch — it is a sample of the chute. Stratified sampling pulls at start/middle/end of the run, from multiple ports of the blender, from each filling head of a multi-head packager, to give the batch-level inference the IPC is supposed to provide.

Acceptance limits vs action limits

Acceptance limits define pass/fail. Action limits sit tighter and trigger investigation or process adjustment before the batch fails. Modern IPC plans operate on both: action limits drive real-time control, acceptance limits drive disposition.

An IPC outside its acceptance limit is an out-of-specification (OOS) result and triggers FDA's 2006 OOS Guidance investigation pathway, even though the IPC is not a release test. An IPC inside its acceptance limit but outside its historical control-chart envelope is an out-of-trend (OOT) result and triggers the OOT investigation. Both pathways must be defined in advance; both must distinguish lab error from manufacturing-process error before deciding the batch fate.

1. Pause: the next step cannot start while the IPC is under investigation.
2. Lab-error check: re-pull, re-test by the same analyst, then by a second analyst.
3. If lab-error confirmed: invalidate the original result, document, proceed with second result.
4. If lab-error not confirmed: open formal investigation, treat result as valid, determine manufacturing root cause.
5. Disposition: continue (with rework if needed), reject, or escalate to MRB for use-as-is justification.

The traditional IPC is a sample pulled by an operator, walked to a lab, analysed, and returned as a number an hour later. PAT (Process Analytical Technology, FDA 2004) replaced this round trip for many parameters with in-line or at-line sensors that produce the IPC result inside seconds: NIR for blend uniformity, Raman for content uniformity, in-line LOD probes for granulation endpoint, on-line UV-Vis for column elution profiles. The advantage is not faster paperwork — it is feedback control: the process can be adjusted while it is still on, before the next sub-lot is in trouble.

PAT-based IPCs change the batch-record evidence: instead of a single number from a single sample, the BMR carries the continuous data stream that informed the in-line decision. The validation effort is heavier (ICH Q2 for the method, ICH Q14 for analytical-procedure development, GAMP 5 for the system) but the inspection story is simpler: the evidence is the process, not a sample of the process.

• IPC pulled but not recorded contemporaneously — the value handwritten on a glove and copied into the BMR an hour later.
• Sampling plan written for a smaller batch size than the one running; not re-validated when scale changed.
• Acceptance limits 'updated' silently to match drifting historical data, with no change control or risk reassessment.
• IPC investigation closed without ruling out manufacturing root cause when the lab cause could not be confirmed.
• Stratified sampling defined in the SOP but not honoured on the floor (operator pulls a convenience sample at one location).
• IPC method not requalified after instrument or reagent change.
• IPC results trended in a spreadsheet that is not subject to Part 11 audit trail.
• No defined action on action-limit breach — the action limit was decorative.
• IPC values backdated when the original record was 'lost'.

In V5 the IPC plan is a property of the MMR, not of the operator memory or the local lab spreadsheet. When the MMR is approved, V5 generates the sampling schedule, attaches the acceptance and action limits to each IPC step, and binds each step to the LIMS test method. At execution time the batch record forces the IPC to be entered before the next step can begin — and the data flows automatically into the SPC chart, the CPV dataset and the deviation system if it breaches.

• Acceptance and action limits come from the validated control strategy, not from a free-text field.
• Sampling location, time and sample size are mandatory at recording, not optional.
• PAT and at-line instrument readings stream into the BMR with full provenance.
• OOS/OOT investigations open automatically with linked evidence (raw data, instrument log, prior trend).
• Trending dashboards are role-scoped: production sees real-time, QA sees trend, regulators see audit trail.
• Method requalification triggers raise change-controls before the IPC can be used on a new batch.