In-Process Check: in-Process Controls in GMP Manufacturing

An In-Process Check is a documented measurement, test, examination, or observation performed during manufacturing to determine whether an intermediate material or process condition meets defined acceptance criteria. This is an operational glossary definition, not a universal statutory definition.

IPC and in-process control are common aliases. Terminology varies by site; here, a check means one discrete assessment, while the broader in-process control arrangement specifies when to check, how to evaluate the result and what action follows. A check can inform an in-process quality gate; it is not the whole decision framework.

The following distinctions describe operating purposes, not universal regulatory classifications.

Illustrative applications include tablet mass during compression, blend loss on drying, solution pH and a documented visual examination for defined defects. Suitability depends on the product, process and decision: these are examples, not a mandatory test list. A satisfactory pH result answers a pH question; it does not establish identity, purity or every other quality attribute. Likewise, an instrument-readiness check supports confidence in the measurement setup but does not establish that the material meets its acceptance criteria.

Map an in-process check to the product and jurisdiction before deciding which legal requirement it serves. The US drug and dietary supplement provisions below address related manufacturing concerns, but they are not interchangeable. A useful regulatory map identifies the applicable duty separately from the site's chosen method of carrying it out.

Binding requirements: procedures, specifications and monitoring

For drugs, the written-procedure obligation includes both establishing the procedures and following them. It concerns appropriate samples of in-process materials from each batch—not merely having a generic inspection instruction available. The provision frames these controls around assuring batch uniformity and integrity. Use the broader 21 CFR 211 context when locating this requirement within a drug manufacturing programme.

For supplements, specification setting and monitoring are connected but distinct duties. Section 111.70(a) requires a specification at any manufacturing point, step or stage where control is necessary to ensure supplement quality. Section 111.70(c)(1) makes the connection more specific: in-process specifications belong at necessary control points, steps or stages in the master manufacturing record. This is why a supplement IPC map should connect the specification to the relevant master manufacturing record, rather than leave it as an isolated test name.

Section 111.75(b) then requires monitoring of in-process points, steps or stages where control is necessary to ensure the quality of the finished supplement batch. Paragraph (b)(1) expressly connects that monitoring to determining whether in-process specifications are met. Read these duties within the dietary supplement framework, rather than importing drug terminology as though it supplied the supplement rule.

Practice: translate the duty into a decision sequence

An operational synthesis is: specification setting, execution, evaluation and an authorised decision. Define what is acceptable; perform the planned check; compare the result with the specification; then route the outcome to the person authorised by the site's procedures to decide the next action. This sequence is a design recommendation, not additional regulatory wording or a universal approval rule.

Keep the categories explicit in procedures and training: the cited CFR provisions are binding requirements within their applicable scope; official guidance is a separate category; the workflow recommendation here is practice. A software workflow can support implementation, but should not be described as the source of the legal duty.

As recommended manufacturing practice, design an in-process check backwards from the decision it supports. Use this chain: quality concern → process location → measurable attribute → method → sample plan → acceptance rule → response → decision owner. Each link should explain why the measurement is useful at that point, rather than merely adding another result to the batch record.

Separate the measured attribute from the setting an operator can adjust. For example, capsule fill mass is an attribute; a dosing-machine setting is a process parameter. A failed mass check does not, by itself, establish which setting caused the failure. Identify any relevant critical process parameter separately and define the permitted adjustment and reassessment pathway.

For US dietary supplements, 21 CFR 111.70(c)(2) requires documented grounds for why meeting in-process specifications, together with component specifications, will help ensure that product specifications are met for identity, purity, strength, composition and relevant contamination limits. This is a binding documentation requirement, not merely a recommendation to collect more measurements.

A recommended executable check specification contains:

• Characteristic, process location and reporting units.
• Sampling trigger, sample quantity and method version.
• Acceptance rule, including whether it applies to individual results, a summary statistic or both.
• Responsible role, response instructions and the owner authorised to decide the next action.

Include startup, routine-production and post-interruption checks only where the process justifies them. Explain what each trigger detects and whether its result controls an in-process quality gate. Avoid copying one sampling frequency across unrelated operations without a process-specific reason.

Keep acceptance limits distinct from earlier warning or adjustment thresholds. An early warning can prompt inspection, increased sampling or a permitted adjustment while the result still meets the established acceptance criterion. It neither changes that criterion nor makes an unacceptable result acceptable. Define the response to each threshold explicitly.

Treat an IPC as an event within the batch, not an isolated number. The following is a recommended execution and record design, independent of any particular MES or electronic batch-record product.

Recommended execution workflow

Select effective definition → Confirm readiness → Obtain specified sample → Capture observation → Evaluate criterion → Address exceptions → Obtain required human decision

At each handoff, carry the event identity forward so production, laboratory and review personnel work from the same context. Identify who owns the next action and what evidence they need. Where the procedure calls for a human decision, attach that decision to the event rather than leaving it in an unconnected message.

Recommended evidence model

Reusable check definition + specification version → Batch-specific check execution → Observation and raw-data reference → Calculated value, if applicable → Evaluation Check execution → Exception and human review records

Link each execution into the batch execution history. Useful evidence relationships include:

• Batch and step identity; check-definition and specification versions.
• Sample identity, sampling location and time; operator identity.
• Method and instrument identity; result and units; raw-data reference.
• Evaluation, exception records and review links.

Keep the original observation distinct from any calculated value and its pass/fail classification. Preserve the specification version used for each execution: a later definition change should not silently reinterpret an earlier decision. Consistent production context tagging helps connect these records without confusing repeated checks at the same step.

Maintain completion status, result evaluation and review status separately. A recorded result may still await evaluation; an evaluated result may still await review. This separation makes the outstanding action visible instead of allowing a single “complete” flag to imply that every decision has been made.

For US drug manufacturing, 21 CFR 211.110(b) requires valid in-process specifications for the characteristics concerned to be consistent with final drug-product specifications. The same paragraph requires derivation from previous acceptable process averages and variability estimates where possible, and determination using suitable statistical procedures where appropriate. Those qualifications matter: the paragraph does not prescribe a control chart for every IPC. Historical data should inform a justified decision rule, not turn an observed process average into an acceptance limit by default.

As operating practice, choose sampling times and locations to represent the variation relevant to the decision. For example, samples taken together at startup cannot describe later process drift; a single accessible blend location may miss spatial differences. Preserve time and location through the sample handoff, and retain individual results alongside summaries. Sampling frequency and coverage need a process-specific rationale; neither example below establishes a generally adequate sampling plan.

Case 1: an acceptable mean hides a failing tablet

Assume a hypothetical drug-tablet mass check assesses five tablets individually, with an inclusive acceptance range of 490–510 mg for every tablet. These limits and this sample size are example assumptions, not regulatory defaults. The measured masses are 494, 498, 502, 504 and 522 mg.

• Sum: 494 + 498 + 502 + 504 + 522 = 2,520 mg.
• Mean: 2,520 ÷ 5 = 504 mg, within 490–510 mg.
• Highest individual result: 522 mg, exceeding the upper limit by 522 − 510 = 12 mg.

The check fails its stated individual-unit rule despite the acceptable mean. Reporting only 504 mg would conceal the decisive result. Tablet mass also does not, by itself, establish content uniformity: the measurement answers a mass question, not how much active ingredient each tablet contains.

Case 2: supplement-blend loss on drying

Assume a hypothetical supplement-blend check uses a defined suitable method, a tare-corrected initial sample mass of 10.000 g, and a final mass of 9.720 g. Its predefined, inclusive acceptance range is 2.00–3.00% mass loss.

• Mass loss: 10.000 − 9.720 = 0.280 g.
• Percentage loss: 0.280 ÷ 10.000 × 100 = 2.80%.
• Decision: 2.80% is within the stated 2.00–3.00% range.

Call this result loss on drying, not a water-specific result unless the method supports that interpretation. It establishes neither ingredient identity nor potency, and says nothing directly about other unmeasured attributes.

Across repeated checks, a rising series may warrant attention even while each result meets its acceptance rule. Conversely, a stable series can remain unacceptable. Apply the threshold distinction established in check design: observing a trend and deciding acceptance are different tasks. Preserve the sequence for trend analysis without allowing a summary statistic to replace the specified decision rule.

For US drug manufacturing, 21 CFR 211.110(c) requires in-process materials to be tested for identity, strength, quality and purity as appropriate, and approved or rejected by the quality control unit during production. Retain that qualification: the provision does not prescribe every listed test for every material.

For dietary supplements, 21 CFR 111.70(c)(3) requires quality control personnel to review and approve the documentation described in paragraph (c)(2). That is the rationale explaining why meeting in-process specifications together with component specifications helps ensure the supplement’s identity, purity, strength, composition and contamination specifications are met. This documentation-approval requirement should not be read as requiring a separate QC signature on every IPC result.

In a recommended operating model, distinguish four events: check completion records that the activity occurred; evidence review evaluates its support and completeness; material disposition decides acceptance or rejection; and batch release authorises the batch’s release. Define who may make each decision. Authorised people make approval and release decisions; a calculated pass flag or software status is not a substitute.

For operational record integrity, review original observations, calculations, timestamps, changes, missing checks and unresolved exceptions—not just a summary of passing results. An audit trail review workflow can structure the examination of changes. Preserve the original value and the reason for each correction. The reviewer should be able to reconstruct what was measured, how the result was derived, what changed, and which evidence supported the decision. Record unresolved questions and route them to the authorised decision-maker.

As an integration design practice, assign authoritative ownership before connecting systems. Keep the check definition in a controlled master record with a named process owner; use MES to coordinate execution, the instrument’s acquisition record for raw measurements, the laboratory workflow for laboratory results and their status, and batch-record review for the authorised human review decision. Document which system holds each authoritative record and which systems only display copies. A transferred result should not become a second independently editable result.

Define explicit identifiers and mappings for batch, step, sample, method, specification version and units. Use production context tagging to keep those relationships attached to each measurement. Preserve source values and units alongside converted values, including the conversion rule. Record measurement time separately from transmission and receipt times so a delayed message is not mistaken for a later measurement.

Design interfaces for duplicate messages, out-of-order results, partial transfers, rejected payloads and instrument or network outages. Give each event a stable identifier that survives retries. Reconcile sent, received and accepted events through a process event log; acknowledge acceptance only after the intended transfer is complete. Quarantine rejected payloads for investigation and use controlled retries rather than silently creating replacement results. Sequence results by their event context, not arrival order.

Define a fallback and recovery procedure naming the authoritative temporary record, who may use it and who reconciles it. Before closing the interruption, reconcile temporary entries with the electronic record and resolve missing or conflicting events.

For US drug manufacturing, 21 CFR 211.110(d) requires rejected in-process materials to be identified and controlled under a quarantine system designed to prevent their use in manufacturing or processing operations for which they are unsuitable.

OOS means outside the applicable specification. OOT (out of trend) and OOE (out of expectation) are site-defined trend or expectation categories, not universal legal classifications. An adverse trend can warrant investigation even when individual results remain within specification; define triggers in the site's trend-handling procedure.

The responses below are recommended procedure design, not additional legal requirements. Assign decision authority to named roles.

Use the production timeline to identify potentially affected material, including output since the last reliable check. Preserve the original result and document containment boundaries, timing and ownership. An interim hold is a precaution pending assessment, not an authorised rejection decision. Route additional sampling, retesting, adjustment or restart through documented authorisation; record the rationale and link subsequent evidence to the initial event rather than replacing it.

As measurement-system practice, match the method’s range, resolution, repeatability and selectivity to the characteristic being checked and its acceptance boundary. A display with more decimal places does not necessarily distinguish acceptable from unacceptable material. Define sample preparation and relevant environmental conditions, such as temperature, vibration or lighting. An instrument reading alone cannot explain where the sample came from or how it was prepared.

• Mass check: an incorrect container tare distorts the reported net mass. Confirm the container and tare before weighing, and retain enough information for net-versus-gross reconciliation.
• Visual examination: an unclear reference standard lets operators classify the same appearance differently. Use a controlled reference with explicit defect criteria and suitable viewing conditions.

Before use, check instrument identity and applicable calibration or verification status; record the method version. If the instrument is later found unsuitable, identify potentially affected results using its use history and as-found calibration information. Set verification intervals and technical evaluations for the measurement system and decision risk, rather than assuming a universal calibration interval or uncertainty calculation.

Use this recommended readiness checklist to connect the controls discussed above. It is an operating-practice aid, not an additional legal requirement or a universal validation protocol.

• Confirm the product and jurisdiction scope and name the check owner.
• Identify the current specification, method and sampling rationale.
• Assign trained execution roles, with relevant training records, and confirm instrument suitability.
• Confirm result lineage back to the sample, batch, step and applicable check version.
• Define exception routing and the people authorised to decide disposition, approve recovery and release the batch.
• Assign responsibility for interface reconciliation, including missing and duplicate transfers.
• Confirm approved change handling and ownership before enabling the check for production.

Challenge the decision path

Select risk-based challenges for boundary values, unit mismatches, missing results, a failed individual result hidden by an acceptable average, duplicate transfers, interrupted execution and an unresolved exception at review. Define the expected outcome before each test, including who receives the exception and what evidence survives recovery. These are implementation examples; choose their depth according to the foreseeable failure and its consequence.

Govern changes and review trends

For changes to limits, methods, frequencies or decision rules, use change control to assess the reason and impact, identify the effective version, address open executions and preserve historical interpretation. Name the decision owner: a configuration edit is not sufficient approval to change a manufacturing decision.

Periodically review missed-check rates, recurring failures, overrides, delayed reviews and sampling coverage. Treat these as diagnostic indicators whose meaning depends on process context, production volume and risk—not as promised performance improvements. Assign follow-up ownership where patterns warrant investigation.