This is a ready-to-use qualification protocol for a process analytical technology (PAT) system whose chemometric model output will support an in-process control decision or replace a laboratory test as part of real-time release testing (RTRT). Replace every <<FILL: ...>> placeholder, set your own document numbers and dates, and route it through your normal validation review and approval. A worked filled specimen follows. Verify each cited reference against the current source before you rely on it. This content is general educational reference, not legal or regulatory advice.
Approval page
| Role | Name | Signature | Date |
|---|---|---|---|
| Protocol author | <<FILL>> | ||
| Process engineering / PAT lead | <<FILL>> | ||
| Analytical / QC lead (reference method owner) | <<FILL>> | ||
| Quality Assurance | <<FILL>> |
1. Objective
To demonstrate that the PAT system <<FILL: system/probe name and ID>>, using chemometric model <<FILL: model name and version>>, produces results that are accurate, precise, specific, and equivalent to the qualified reference method <<FILL: reference method ID>> across the intended operating range, and that the model’s applicability domain and failure detection reliably flag conditions outside that range, before the system is used to support <<FILL: in-process control / RTRT disposition of SPECIFIC TEST>>.
2. Scope
This protocol applies to the PAT hardware (the probe or sensor), the chemometric model that converts its raw signal into a reported value, and the software that hosts the model, for the process step and product named in the header. It does not cover the instrument’s basic hardware IQ/OQ (wavelength accuracy, photometric linearity, noise), which is qualified under <<FILL: instrument IQ/OQ protocol ID>> and referenced as a prerequisite.
3. System description
| Field | Entry |
|---|---|
| Process step monitored | <<FILL: e.g. fluid-bed drying endpoint>> |
| Measured attribute | <<FILL: e.g. percent moisture>> |
| Sensor / probe technology | <<FILL: e.g. NIR, Raman, focused-beam reflectance>> |
| Model type | <<FILL: e.g. PLS regression, N latent variables>> |
| Preprocessing applied | <<FILL: e.g. second-derivative, SNV scatter correction>> |
| Host software / platform | <<FILL: name and version>> |
| Reference method it is validated against | <<FILL: e.g. USP loss-on-drying, gravimetric method ID>> |
| Intended use | <<FILL: in-process control only / RTRT replacing a specified lab test>> |
4. Prerequisites
- Instrument hardware IQ/OQ complete and current (
<<FILL: protocol ID>>). - Reference method qualified or validated and currently in routine use.
- Calibration (training) data set assembled, spanning the full intended operating range including worst-case edges, with each point’s reference-method result recorded.
- Model built and frozen under version control before this protocol executes; the version under test is recorded in section 3 and cannot change during execution without a protocol deviation.
5. Roles and responsibilities
| Role | Responsibility |
|---|---|
| Process engineering / PAT lead | Builds and documents the model; executes test cases; investigates any out-of-criteria result. |
| Analytical / QC lead | Generates and confirms reference-method results used for comparison; confirms the reference method’s own validity for this comparison. |
| Validation | Authors and executes the protocol; confirms negative-path test design; compiles the summary report. |
| Quality Assurance | Reviews and approves the protocol and the final report; approves the model version for GxP use. |
6. Acceptance criteria
| Parameter | Acceptance criterion |
|---|---|
| Accuracy (bias vs. reference method) | <<FILL: e.g. mean bias within +/- 0.1% across the calibration range>> |
| Precision (repeatability) | <<FILL: e.g. RSD <= X% on replicate readings of the same sample>> |
| Precision (intermediate, different operator/day) | <<FILL: e.g. RSD <= X%>> |
| Specificity | Model is not materially affected by <<FILL: named interferent, e.g. particle-size variation>> within the range tested |
| Equivalence to reference method | <<FILL: e.g. linear regression slope 0.9-1.1, intercept not significantly different from zero, correlation coefficient >= 0.98 across N >= paired samples>> |
| Applicability domain enforcement | Every spectrum or reading outside the defined domain is flagged and routed to a hold, not silently accepted or extrapolated |
| Negative-path detection | Every seeded failure (fouled probe, out-of-range material, wrong material) is detected and flagged in 100% of challenges |
7. Test cases
| ID | Step | Expected result | Actual result | Pass/Fail | Tester | Date |
|---|---|---|---|---|---|---|
| TC-01 | Run the full calibration set through the frozen model and compare predicted values to reference-method results across the range | Accuracy and precision criteria in section 6 met | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> |
| TC-02 | Run paired samples (PAT reading vs. reference method) at the low, mid, and high end of the operating range, N replicates each | Equivalence criteria in section 6 met at each level | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> |
| TC-03 | Introduce a known interferent <<FILL>> at a realistic level and confirm the model’s prediction does not shift beyond the specificity criterion | Specificity criterion met | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> |
| TC-04 | Present a spectrum from a different, unrelated material or an intentionally fouled probe reading | System flags the reading as outside the applicability domain and does not report a disposition value | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> |
| TC-05 | Present a reading just outside the upper and lower edges of the calibrated range | Both are flagged, not extrapolated and silently accepted | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> |
| TC-06 | Confirm raw spectral/sensor data is retained and retrievable for every reading in TC-01 through TC-05 | 100% of raw data retrievable and linked to its prediction | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> |
| TC-07 | Confirm the model version identifier is recorded against every reading generated during this protocol | Version identifier present and matches section 3 for every reading | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> |
8. Deviation handling
Record any unexpected result, including a test case that passes but shows a borderline value, on a protocol deviation form referencing this protocol number. Assess impact on every prior test case before continuing. A failure in TC-04 or TC-05 (the applicability-domain checks) blocks approval until the detection logic is corrected and the test case re-executed; it is not waivable by a risk assessment, because an undetected out-of-domain reading is the specific failure mode this protocol exists to rule out.
9. Summary and conclusion
| Field | Entry |
|---|---|
| Overall result | <<FILL: Pass / Pass with documented deviations / Fail>> |
| Model version qualified | <<FILL>> |
| Intended use approved | <<FILL: in-process control only, or RTRT replacing SPECIFIC LAB TEST>> |
| Ongoing verification schedule set | <<FILL: frequency and method, cross-reference to the monitoring SOP>> |
| Conclusion narrative | <<FILL>> |
10. Attachments
- Calibration (training) data set and reference-method results.
- Raw spectra/sensor files for every test case run.
- Model build and preprocessing documentation.
- Any deviation records raised during execution.
11. References
ICH Q13, Continuous Manufacturing of Drug Substances and Drug Products. ICH Q14, Analytical Procedure Development. ICH Q8(R2), Pharmaceutical Development (design space). USP General Chapter <1039>, Chemometrics. FDA guidance for industry, PAT, a Framework for Innovative Pharmaceutical Development, Manufacturing, and Quality Assurance (2004). 21 CFR 211.110, Sampling and testing of in-process materials and drug products.
Confirm the current version and clause numbers of each reference before issue.
Filled specimen (excerpt)
A film-coating process uses an NIR probe to predict percent coating weight gain, intended to replace the end-of-run gravimetric weight-gain check.
| Field | Entry |
|---|---|
| Process step monitored | Tablet film coating, in-process weight gain |
| Measured attribute | Percent coating weight gain |
| Sensor / probe technology | NIR, in-line probe |
| Model type | PLS regression, 4 latent variables |
| Reference method | Gravimetric weight-gain determination, Method QC-114 |
| Intended use | RTRT, replacing the end-of-run gravimetric check for routine batches |
| ID | Expected result | Actual result | Pass/Fail |
|---|---|---|---|
| TC-01 | Bias within +/- 0.1%, RSD <= 3% | Bias 0.04%, RSD 1.8% | Pass |
| TC-02 | Slope 0.9-1.1, r >= 0.98, N=36 pairs | Slope 1.02, r = 0.993 | Pass |
| TC-04 | Fouled-probe reading flagged, no value reported | Flagged as “signal quality fail,” no disposition value generated | Pass |
Conclusion: model version NIR-CM-2026-03 qualified for RTRT use replacing Method QC-114 for routine batches; ongoing verification set at 1-in-10 batches against the gravimetric method for the first six months, then re-assessed.
Common inspection findings this protocol prevents
- A PAT model used for disposition with no documented accuracy, precision, or equivalence data against the reference method it replaces.
- No defined applicability domain, so an out-of-range or anomalous reading is silently extrapolated into a confident but invalid result.
- Raw spectral data discarded once the predicted value is stored, leaving no way to reprocess a disputed result.
- Model version not recorded against individual readings, so a later retrain cannot be traced back to which batches it decided.
How to adapt this protocol
- Set your acceptance criteria in section 6 from your own process capability and the risk of the decision the model supports; do not copy another product’s numbers.
- Size the calibration and paired-comparison data sets to the statistical power you need; consult a statistician for the regression and equivalence testing approach.
- If the model will support RTRT replacing a specific laboratory test, name that test explicitly in sections 1 and 3, and keep the reference method running in parallel per your ongoing verification schedule after go-live.
- Link this protocol to your model-monitoring SOP so the ongoing verification frequency set in section 9 is actually executed after qualification.