Postdoctoral PDQC801 ⭐ 3 Credits Semester 1

Pharmaceutical Quality Systems, cGMP, Data Integrity, Laboratory Operations, and Contract-Laboratory Management

Sets the quality-system frame for the whole postdoctoral track. The regulatory hierarchy is taught first and explicitly: statute, then regulation, then compendia, then guidance, then voluntary standards. On that frame the course builds the working machinery of a peptide QC testing laboratory. Core content covers 21 CFR 211 Subparts I and J and Part 11, ICH Q9(R1) and Q10, and ALCOA+ data integrity. It also covers FDA's 2022 OOS guidance and the deviation, CAPA, and change-control system. Chromatographic data-processing governance is anchored on the lab's three Agilent 1260 HPLCs and their CDS. It covers integration rules, manual reintegration, the prohibition of trial injections, aborted sequences, repeat analysis, excluded results, and review of original electronic records with audit trails. The sampling program covers representative sampling, container selection, destructive-test allocation, transport excursions, chain of custody, retains, and reconciliation. Release reporting covers USP General Notices rounding, decision rules near limits, conditional results, and authorized signatures. A firm boundary is maintained throughout: the laboratory reports results, and the client's quality unit dispositions product. Contract-laboratory management covers quality and technical agreements, subcontracting approval, sample logistics, raw-data access, OOS notification, and responsibility for the final report. Routine operations cover scheduling, inventory, consumable lot qualification, downtime, and business continuity. The 503A, 503B, and ISO/IEC 17025 frames are taught as distinct regimes with different legal force, not as interchangeable labels. Builds directly on CHEM504 (research ethics, documentation), CHEM302 (measurement statistics), and BCHM700 (hands-on peptide analytical practice). Taken alongside PDQC800, whose product control-strategy matrix supplies the tests this quality system must govern.

📅 Studied by Skipper on 2026-10-03

📖 Core Concepts

Claude Opus 5.5

Core Concepts: Pharmaceutical Quality Systems, cGMP, Data Integrity, Laboratory Operations, and Contract-Laboratory Management

PDQC801 · Postdoctoral track · Peptide QC laboratory design (Peptide Partners: Zak + Larissa, 3× Agilent 1260, LC-MS/MS, second MS, BacT/ALERT)

Verification convention. Items marked [VERIFY] should be checked against the current official text before they go into a controlled SOP. This applies especially to USP chapters, which are revised on a rolling basis, and to FDA draft guidances. CFR section numbers and ICH element names are cited only where I am confident of them. Even so, an SOP should quote the current eCFR text directly and should not rely on these notes.

0. What This Subject Is Really About

In earlier courses you learned how to make a measurement: how to separate peptides on a C18 column, how to read an MS/MS spectrum, how to compute a confidence interval. This course is about something different. It asks how a measurement becomes a claim that a stranger can trust without being in the room.

A QC result is not just a number. It is a testimonial statement that has three parts:

  1. Capability. "This method, on this instrument, run by this person, can tell 97.9% from 98.1%." This is the job of validation, qualification, and competence.
  2. Fidelity. "What I report is what actually happened, completely, with nothing hidden." This is the job of data integrity.
  3. Scope. "Here is exactly what I am claiming, against which standard, under which decision rule, and what I am not claiming." This is the job of reporting, decision rules, and CoA language.

A helpful analogy is forensic evidence. A DNA match is worthless if the chain of custody is broken, even when the sequencing was flawless. Likewise, a chromatogram showing 98.6% purity is worthless if nobody can prove that it is the only chromatogram that was acquired for that sample. Most FDA laboratory citations are not about bad chemistry. They are about broken chains of custody around good chemistry.

The second big idea is that a Pharmaceutical Quality System (PQS) is a feedback-controlled dynamical system, not a binder of documents. In MATH102 you met first-order systems with negative feedback. A PQS works the same way:

  • Monitoring (KPIs, trending, OOT detection) is the sensor.
  • Management review is the comparator.
  • CAPA and change control are the actuator.
  • Knowledge management is the memory.

A lab with beautiful SOPs and no working feedback loop is like an open-loop controller. It drifts, and nobody notices until an inspector or a client does. ICH Q10 is, in effect, a specification for the control loop [ICH Q10, 2008].

The third idea is specific to Peptide Partners. You are designing a small testing laboratory that serves clients who operate under different legal regimes: 503A pharmacies, 503B outsourcing facilities, and possibly research or other customers. Your lab therefore needs a quality system strong enough to survive the most demanding of these frames. It also needs reporting language disciplined enough that it never claims a regulatory status it does not hold.

Every requirement in your PDQC800 control-strategy matrix comes from somewhere. Where it comes from determines what happens if you ignore it. The core skill here is tagging each requirement by both source and legal force.

TierExampleLegal forceTag
StatuteFD&C Act §501(a)(2)(B), §501(b), §503A, §503BLaw. Violation means the product is adulterated.S
Regulation21 CFR 210/211; 21 CFR Part 11Binding. Has the force of law.R
CompendiaUSP–NF monographs and General Notices; general chapters numbered <1000Enforceable through §501(b) for articles recognized in the compendium. Chapters ≥<1000> are informational unless referenced.C
GuidanceFDA OOS (2022), DI Q&A (2018), Quality Agreements (2016), ICH Q9(R1)/Q10Non-binding, but it is FDA's "current thinking". Deviating from it requires an equivalent, documented alternative.G
Voluntary standardISO/IEC 17025:2017, ISO 14644, ISO 2859-1, ILAC-G8Not law. It becomes binding through accreditation or contract.V
ContractQuality agreement and technical agreement with a clientEnforceable as contract law.K

The statutory anchor for everything in cGMP is §501(a)(2)(B). A drug is adulterated if:

"…the methods used in, or the facilities or controls used for, its manufacture, processing, packing, or holding do not conform to or are not operated or administered in conformity with current good manufacturing practice to assure that such drug meets the requirements of this chapter as to safety and has the identity and strength, and meets the quality and purity characteristics, which it purports or is represented to possess." — FD&C Act §501(a)(2)(B)

Two features of this text are surprising and worth noticing:

  1. Adulteration under §501(a)(2)(B) does not require the product to be defective. A batch with 99.9% purity is legally adulterated if the controls that produced the number were non-conforming. The law regulates the process of knowing, not just the result. This is why data-integrity failures alone generate Warning Letters.
  2. "Current" is a moving target. The "c" in cGMP means that today's expectations, including those expressed in guidance, define compliance. A practice that was acceptable in 2005, such as a standalone CDS with shared logins, is not cGMP today, even though 21 CFR 211 has barely changed. Guidance is legally non-binding but practically load-bearing.

1.1 The three quality frames

503A (pharmacy compounding)503B (outsourcing facility)ISO/IEC 17025 (lab competence)
Who is regulatedLicensed pharmacist or physician compounding for identified patientsRegistered outsourcing facilityThe testing laboratory itself, voluntarily
Exemptions from FD&C ActExempt from §501(a)(2)(B) cGMP, §502(f)(1), and §505Exempt from §502(f)(1), §505, and §582. Not exempt from cGMP.n/a
Quality standardUSP <795>/<797>, enforced mainly through state boards of pharmacy. FDA retains authority, e.g., over insanitary conditions under §501(a)(2)(A).cGMP, as interpreted in FDA's 503B cGMP guidance [VERIFY: currently a revised draft; confirm the version]Clauses 4–8 of the standard, assessed by an accreditation body
InspectorState board (FDA for-cause)FDA, routine and for-causeAccreditation body assessor
What it means for your lab dataResults support pharmacist release under USP chapters. Lab cGMP is not legally required, but the data must still be defensible.Your lab acts as an extension of the client's quality control unit. Its data are subject to 211 Subpart I/J and Part 11 expectations through the client.Competence for the specific scope listed on the accreditation certificate, and nothing more

Design consequence. Build the lab to the 503B/cGMP standard and report within a 17025-style disciplined scope. A system that survives an FDA inspection of a 503B client's contract lab will also satisfy a 503A client and a 17025 assessor. A system built only to the 503A level will fail as soon as you sign your first 503B client.

1.2 ISO/IEC 17025 is not a GMP certificate

17025 accreditation certifies technical competence for a defined scope: specific methods, matrices, and ranges listed on the certificate. It does not certify cGMP compliance, and FDA does not treat it as a substitute. Two practical corollaries follow:

  • A CoA result for a method outside the accredited scope cannot carry the accreditation mark or claim. ISO/IEC 17025 and the accreditation body's rules govern this. Some bodies require a marked, unaccredited section on the report.
  • Conversely, a cGMP-compliant lab without accreditation is fully legitimate for 503B work. Accreditation is a commercial and assurance asset, not a legal prerequisite for cGMP testing.

1.3 The peptide-specific benchmark

ICH Q3A/Q3B explicitly exclude peptides from their scope. So what is the impurity benchmark for a synthetic peptide? The closest authoritative analytical standard is FDA's 2021 guidance on ANDAs for highly purified synthetic peptides that reference rDNA-origin listed drugs (glucagon, liraglutide, nesiritide, teriparatide, teduglutide). Its key thresholds, as I understand them, are:

  • Identify and characterize any new peptide-related impurity at >0.10%, and justify why it does not affect safety or effectiveness.
  • No new specified peptide-related impurity above 0.5%.
  • Each impurity common to the RLD must be at a level ≤ that in the RLD.

[VERIFY] the exact wording and the comparator logic against the final 2021 text.

Your lab uses these thresholds as an analytical benchmark. They tell you that the related-substances method needs an LOQ comfortably below 0.10% (e.g., ≤0.05%) and specificity good enough to resolve and identify impurities at that level, with LC-MS for identity. Your lab does not claim that a client's product "meets ANDA requirements". That is a regulatory-status claim, and only an applicant and FDA can make it.

Exercise structure (Week 1 deliverable). Build one row per control-strategy requirement with these columns:

Requirement | Source citation | Tag (S/R/C/G/V/K) | Applies to 503A? | 503B? | 17025? | Evidence artifact (SOP/record) | Owner

A requirement with no source citation is either a good-science choice (label it "internal / scientific justification") or something you should drop.

2. 21 CFR 211 Subparts I and J: The Laboratory as an Extension of the Quality Control Unit

For a contract testing lab, the central idea is that you perform part of the client's quality control unit's (QCU) function under §211.22. The QCU's responsibility for approving or rejecting cannot be delegated. The testing can be. This split, "lab tests, client dispositions", runs through the whole course.

2.1 The clauses that touch the lab

ClauseCore requirementLab artifact
211.22A QCU with authority and responsibility to approve or reject. It has adequate laboratory facilities, and its procedures are written.Quality manual, organizational chart, and role-independence statement
211.25Personnel have education, training, and experience. Training is in cGMP and in their specific operations.Training matrix, competency records
211.67 / 211.68Equipment cleaning and maintenance. For automatic and electronic equipment: calibration and checks, controls so that only authorized personnel can make changes, and backup of data.Instrument qualification, CDS access control, backup and restore test
211.84Testing and approval of components, including identity testingMethods for incoming raw materials and APIs (client scope)
211.160Specifications, standards, sampling plans, and test procedures that are "scientifically sound and appropriate". Changes are drafted and approved by the QCU. Deviations are recorded and justified. Instruments are calibrated.Specification files, method SOPs, calibration program
211.165Testing and release against specifications. 211.165(e): the accuracy, sensitivity, specificity, and reproducibility of test methods must be established and documented.Validation and verification reports (linked to PDQC814 and the methods courses)
211.166A written stability program with reliable, meaningful, and specific test methodsStability protocols, trending (Week 8 OOT)
211.170Reserve samplesRetain program
211.180Records retained (generally ≥1 year after batch expiry) and available for inspectionArchive SOP, retention schedule
211.192Thorough investigation of unexplained discrepancies and failures, whether or not the batch has been distributedOOS and deviation SOPs
211.194Complete laboratory recordsWorksheets, CDS data, review records
211.198Complaint files. Complaints may trigger lab review of retains and data.Complaint-to-lab linkage procedure

2.2 The reconstruction test (211.194)

"Laboratory records shall include complete data derived from all tests necessary to assure compliance with established specifications and standards, including examinations and assays…" — 21 CFR 211.194(a)

The most useful way to think about 211.194 is as a reconstruction test. A competent stranger, given only the record, should be able to regenerate the reported value and verify every link in the chain without asking anyone. The required elements map onto that chain:

  1. Sample identity and receipt: source, quantity, lot, date received.
  2. The method used, by reference, including version.
  3. Sample weight or measure, plus identification of standards and reagents (lot, preparation, expiry).
  4. A complete record of all raw data: chromatograms, instrument identity, and for an electronic CDS, the electronic records themselves.
  5. All calculations.
  6. Results and how they compare with specifications.
  7. The analyst's signature and date.
  8. The reviewer's signature and date, confirming accuracy, completeness, and compliance.

The word "complete" is the trap. It means all data, including failed SSTs, aborted sequences, and invalidated runs. A record that is complete for the passing result but silent about a preceding failed run fails 211.194. This is exactly where data integrity (§5) and OOS (§7) begin.

2.3 Reserve samples: a worked quantity calculation (211.170)

The rule for an active ingredient is a reserve of at least twice the quantity needed for all tests required to determine conformance, excluding sterility and pyrogen tests. It is retained for 1 year past the expiry of the last lot of drug product containing the API, under specified conditions. For drug product reserves, the regulation adds a visual examination at least annually. [VERIFY] 211.170(a) vs (b) for which retention clock applies to your client's article type.

Q_reserve ≥ 2 × Σ q_i      over all release tests i, excluding sterility and pyrogen/endotoxin*

Example: peptide drug substance (lyophilized powder)
  HPLC purity / related substances  (duplicate preps, 2 × 10 mg) =  20 mg
  Assay (peptide content, 2 × 10 mg)                             =  20 mg
  Identity by LC-MS                                              =   2 mg
  Water (KF, contracted; 2 × 50 mg)                              = 100 mg
  Counter-ion (TFA/acetate by IC, contracted; 2 × 20 mg)         =  40 mg
  Σ q_i                                                          = 182 mg
  Q_reserve ≥ 2 × 182 = 364 mg   → specify 400 mg in ≥2 containers

*Whether bacterial endotoxin counts as "pyrogen testing" for this exclusion is an interpretation to confirm with the client QCU and to document in the quality agreement. Do not assume it.

3. The PQS as a Control Loop: ICH Q10, ICH Q9(R1), FMEA, and KPIs

3.1 ICH Q10's four elements, scaled to two to six people

ICH Q10 names four PQS elements:

  1. Process performance and product quality monitoring
  2. CAPA
  3. Change management
  4. Management review

It also names two enablers: knowledge management and quality risk management [ICH Q10, 2008]. Q10 itself was written for manufacturers. For a lab, "process performance" becomes method and laboratory performance: SST pass rates, invalid-run rates, control-sample trends, and OOS rates.

In a two-person lab, the hard problem is not documentation. It is independence. The fix is to define independence by role on a given record, not by headcount:

  • The person who generated a dataset may never be its reviewer or its approver.
  • If Larissa runs the analysis, Zak, or a qualified third party, reviews it, and vice versa.
  • The CDS administrator role (IT/admin) is held by someone who does not generate GMP data. If no such person exists, administrator actions are audit-trailed and reviewed periodically by the other person or by an external QA consultant.
  • QA oversight (approving SOPs, closing CAPAs) can be partially outsourced to a contract QA consultant until headcount allows. Document this in the quality manual as a deliberate design choice.

3.2 ICH Q9(R1): what changed and why it matters

Q9(R1) (Step 4, January 2023) keeps the original framework: risk assessment (identify → analyze → evaluate), control, communication, and review. It adds or clarifies four things [ICH Q9(R1), 2023]:

  • Formality is a spectrum. The level of formality (documentation, team, structured tool) should be proportional to uncertainty, importance, and complexity. Not every decision needs an FMEA. A documented rationale in a change request can be adequate for low-risk changes.
  • Subjectivity must be managed. Scoring is inherently subjective. Mitigations include defined scale anchors, multidisciplinary input, documenting assumptions, and acknowledging uncertainty.
  • Product availability is a legitimate risk dimension, because shortages harm patients. For a lab, this translates into TAT and capacity risks.
  • Risk-based decision-making and hazard identification are given more attention.

3.3 FMEA, and why the RPN is a dangerous number

RPN = S × O × D

S = severity of the failure effect           (1 = negligible … 10 = wrong release decision, undetected)
O = occurrence likelihood                    (1 = remote … 10 = almost certain)
D = detectability (inverse)                  (1 = certain to be detected before reporting … 10 = undetectable)

Range: 1–1000.
Action rule (state it in advance):
  act if RPN ≥ 100  OR  S ≥ 9 regardless of RPN  OR  S×O ≥ 40

Worked FMEA rows: peptide related-substances workflow

StepFailure modeEffectSODRPNControl / action
Sample prepMet oxidation in diluent during autosampler residenceArtificially high Met(O) impurity → false OOS, or masking of true degradation856240Solution-stability limit from validation (e.g., ≤24 h at 5 °C). Autosampler thermostat at 5 °C. Bracket with a fresh standard.
WeighingHygroscopic peptide gains water during weighingAssay bias (content), not purity665180Equilibrate in a controlled-RH area. Rapid weighing. Water by KF on the same day.
SeparationCo-elution of a D-amino acid diastereomer with the main peakPurity overstated; impurity undetected948288Specificity challenge with spiked epimer during validation. Orthogonal method (different pH or stationary phase). DAD peak purity cannot detect this, because isomers share a UV spectrum.
IntegrationAd hoc manual integration of a shoulderNon-reproducible impurity value; DI finding854160Integration SOP (§6), second-person approval, audit-trail review
DataUnreported pre-injection of a sampleTesting into compliance1037210Injection reconciliation. Ban sample "test" injections. Audit-trail review.

The counterintuitive part. S, O, and D are ordinal scales. You know 8 is worse than 4, but not that it is "twice as bad". Multiplying ordinal numbers produces an artifact. RPN = 120 can come from (S=10, O=3, D=4), a catastrophic but rare failure, or from (S=3, O=10, D=4), a frequent nuisance. These are completely different risks with the same number. This is why the action rule above includes a severity override, and why Q9(R1) emphasizes managing subjectivity.

Treat the RPN as a sorting aid, never as a measurement. If you report "RPN reduced from 288 to 72", you must also show which factor moved. Usually only D or O can move. Severity is a property of the effect, and lab controls rarely change it.

3.4 KPI dashboard definitions for management review

OTIF (on-time-in-full)    = reports issued by agreed date with all tests complete / reports due
Invalid-run rate          = runs invalidated (SST fail, instrument fault, documented lab error) / total runs
OOS rate                  = initial OOS results / total reportable results
Confirmed-OOS rate        = OOS confirmed after Phase I (no assignable lab cause) / initial OOS
Lab-error rate            = OOS invalidated with assignable lab cause / initial OOS
Right-first-time (RFT)    = datasets passing review with zero corrections / datasets reviewed
CAPA aging                = count (and list) of open CAPAs > 90 days
Calibration currency      = instruments within calibration / total in use   (target 100%)
Training currency         = analyst–SOP pairs current / required pairs     (target 100%)

Reading these as a system.

  • A rising lab-error rate is a capability problem.
  • A lab-error rate that is implausibly high is a red flag in its own right. FDA's OOS guidance states that laboratory error should be relatively rare [FDA, 2022]. If most OOS results are being invalidated as "lab error", an inspector will suspect that OOS results are being explained away.
  • A near-zero invalid-run rate in a high-volume HPLC lab is also suspicious. Real instruments fail. A suspiciously clean record suggests that failures are going unrecorded.

Escalation criteria. Pre-define what triggers an out-of-cycle management review. Examples: any confirmed OOS, any data-integrity finding, CAPA aging above 3 items, or OTIF below 85% for two consecutive months. These numbers are design choices you must justify. They are not regulatory values.

4. Document Hierarchy, Good Documentation Practice, and Training

4.1 The four-level pyramid

Level 1  Quality Manual (QM-001)          WHY / WHAT   — policy, scope, organization, clause cross-reference
Level 2  Policies & SOPs (SOP-QA-xxx…)    WHO / WHEN   — system procedures (OOS, CAPA, DI, training)
Level 3  Analytical procedures / WIs      HOW          — ATM-HPLC-001 Related Substances, WI-CDS-003
Level 4  Forms, logbooks, worksheets      EVIDENCE     — FRM-xxx, LOG-xxx, controlled worksheets

Suggested numbering: SOP-[AREA]-[NNN]-v[MM]. Area codes: QA, DI, CHEM, MICRO, ENDO, EQP, CSV, FAC, OPS. Forms inherit their parent number, e.g., FRM-QA-005-02 is form 2 of SOP-QA-005. Then any form can be traced to its governing procedure without a lookup table.

Other control rules:

  • Periodic review on a defined cycle, e.g., every 2 years. This is a common industry convention, not a regulatory number.
  • Review also on trigger: a change control, CAPA, or regulatory update.
  • Obsolete versions are withdrawn from points of use and archived with an "OBSOLETE" status.

Cross-reference matrix (deliverable). Each Level 2 SOP row lists the clauses it satisfies across 21 CFR 211, Part 11, ICH Q10, and ISO/IEC 17025. Example: the OOS SOP maps to 211.192 and 211.160(b), to Q10 CAPA, and to 17025 clauses 7.10 (nonconforming work) and 8.7 (corrective action). The matrix proves coverage, and it also exposes orphan clauses that have no owning SOP.

4.2 Good documentation practice

USP <1029> (informational) collects the practices you already know from CHEM101L, now with legal teeth [VERIFY current chapter text]:

  • Entries are contemporaneous, in indelible ink.
  • Corrections use a single line through the original, so it stays legible, with initials, date, and reason.
  • No blank fields: use "N/A" with initials.
  • No pre-dating or back-dating.
  • Worksheets are controlled: issued from a master with a unique serial number recorded in an issuance log, and reconciled on return. Reconciliation is the paper analog of injection reconciliation (§6). If you cannot account for every issued worksheet, you cannot prove that a failed result was not recorded on a sheet that was later discarded.
  • A true copy must preserve content and meaning, including metadata. For dynamic electronic records it must preserve the ability to reprocess (§5.3). The copy is certified by a documented verification, not by a stamp alone.

4.3 Training versus competence

"Read and understood" proves exposure, not competence. For any SOP whose execution affects a reported result, require a demonstrated-competency step:

  • An observed analysis with a checklist, or
  • A blind sample with known value and an acceptance criterion, e.g., within ±2% of the assigned value for assay and within the method's intermediate precision for impurities, or
  • A proficiency-test result.

ISO/IEC 17025 clause 6.2 is stricter in form than 21 CFR 211.25. It requires the lab to document competence requirements, selection, training, supervision, authorization for specific activities, and monitoring of competence. The authorization element is the key difference: there is a record saying "Larissa is authorized to perform ATM-HPLC-001 and to review data under it, from date X."

Requalification triggers:

  • A major SOP revision affecting technique
  • More than 12 months since the analyst last performed the method (a design choice)
  • A confirmed lab-error OOS attributed to technique
  • A failed proficiency test

5. Data Integrity: ALCOA+, Part 11, and the Physics of Trust

5.1 ALCOA+ as properties of a record over its whole lifecycle

AttributeMeaningTypical CDS failure
AttributableWho did it, and whenShared "Lab" or "Admin" login
LegibleReadable and permanentProprietary format that is no longer readable after a software upgrade
ContemporaneousRecorded at the time of the activityEditable PC clock; worksheet completed at end of day
OriginalThe first capture, or a certified true copyPrintout treated as the original; electronic file deleted
AccurateError-free, with edits documentedUnrecorded integration changes
CompleteAll data, including repeats and failuresOrphan injections; aborted sequences missing
ConsistentChronology and format coherentTime stamps out of order; sequence log contradicts audit trail
EnduringLasts for the full retention periodData on a local C: drive with no tested backup
AvailableRetrievable for review and inspectionBackup exists but has never been restored

PIC/S PI 041-1 (2021) and the MHRA GXP guidance (2018) frame data integrity across the data lifecycle: creation → processing → review → reporting → retention and retrieval → destruction. Controls must exist at every stage [PIC/S PI 041-1, 2021], [MHRA, 2018]. The useful insight here is that ALCOA+ is not a checklist you apply once. It is a set of invariants that must hold at every stage transition.

5.2 Part 11: the clauses you will actually configure

"Use of secure, computer-generated, time-stamped audit trails to independently record the date and time of operator entries and actions that create, modify, or delete electronic records. Record changes shall not obscure previously recorded information." — 21 CFR 11.10(e)
  • 11.10(d) Limit system access to authorized individuals. This requires unique accounts and no shared logins.
  • 11.10(e) Secure, computer-generated, time-stamped audit trails, retained at least as long as the records.
  • 11.10(g) Authority checks. Only authorized individuals can use the system, sign, alter records, or perform operations. This is role-based privilege.
  • 11.50 Signature manifestation: printed name, date and time, and the meaning of the signature (e.g., "reviewed", "approved").
  • 11.70 Signatures are linked to their records so they cannot be excised, copied, or transferred to falsify a record.
  • 11.100–11.300 Electronic-signature controls: uniqueness, identity verification, two distinct identification components for non-biometric signatures, and password and identification-code controls.

Note also 21 CFR 211.68(b): appropriate controls over computers so that only authorized personnel can make changes, plus backup data that is exact, complete, and secure. FDA cites 211.68 alongside Part 11 in nearly every CDS Warning Letter.

5.3 Static versus dynamic records: the most important idea in CDS data integrity

FDA's 2018 Q&A distinguishes two kinds of record [FDA, 2018]:

  • Static records are fixed, like a paper balance printout.
  • Dynamic records allow interaction, like a chromatogram that can be reintegrated.

A PDF or printout of a chromatogram is a static picture of a dynamic record. It cannot show what integration events were applied, whether other injections exist, or what the audit trail says. Hence:

  • The electronic record is the original. A printout cannot substitute for it, and the paper cannot be the "raw data" while electronic files are deleted.
  • Audit trails must be reviewed as part of record review, before final approval or release of the result.
  • Samples should not be used for "trial" or "test" injections to see if they pass. System suitability is demonstrated with standards, not samples.
  • Unvalidated spreadsheets used for GMP calculations are not acceptable as controlled calculation tools.

Here is a helpful analogy. A printout of a chromatogram is like a photograph of a chess position. A reviewer who sees only the photo cannot know how many games were played before this one, or whether pieces were moved. The electronic record with its audit trail is the full game record.

5.4 Warning-Letter failure modes and their signatures

Failure modeHow you detect itControl
Shared loginsA single user name appears across shifts and analystsUnique accounts; Windows and CDS identity linked
Analysts with admin rightsRole report shows analysts can delete, edit methods, or change audit-trail settingsRole segregation (§5.5)
Editable PC clockWindows permissions allow time change; time stamps go backwardLock the clock; network time sync; log clock changes
Orphan dataData files in folders not linked to any reported sequencePeriodic folder-versus-report reconciliation
Unreported "test" injectionsInjection count exceeds the count in the reported sequence; gaps in the vial logInjection reconciliation; ban on sample trial injections
Deleted or overwritten filesAudit trail shows deletion; file system shows gapsNo delete privilege for users; secure storage
Hybrid disconnectPaper worksheet value ≠ electronic result; paper signed before electronic reviewLink paper and electronic records with unique IDs; review electronic data first

5.5 Role design for the Agilent 1260 CDS

First, confirm the actual configuration. ChemStation standalone stores data in the Windows file system. Without an additional secure-storage layer, a Windows user with folder permissions can delete or overwrite files outside the application's audit trail. OpenLab CDS with a secure data-storage backend changes this picture [VERIFY: product, version, and storage model on the actual installation, per Week 6 inventory].

PrivilegeAnalystReviewerLab adminIT admin
Acquire, create sequenceYesNo (on own review)NoNo
Process with locked processing methodYesYesNoNo
Edit or approve methodsNoApprove onlyVersion controlNo
Delete dataNoNoNoOnly under change control
Configure audit trailNoNoNoYes (logged, reviewed)
Change system clockNoNoNoYes (logged)
E-sign: reviewed / approvedExecuted-byReviewed-byNoNo

Interim versus technical fixes. If the current system cannot enforce these roles technically, do two things:

  1. Interim procedural controls:
    • An injection logbook reconciled against sequence files for every run.
    • A weekly folder audit.
    • Windows-level restriction of delete rights.
    • Documented reviewer inspection of the file system.
  2. Technical remediation: migration to a Part 11-capable configuration. CSV execution is handed to PDQC814.

Procedural controls are accepted as interim measures only when they are paired with a dated remediation plan. They are a bridge, not a destination.

5.6 Audit-trail review: risk-based, not exhaustive

Reviewing every audit-trail line is impossible and pointless. Define critical events:

  • Integration parameter changes and manual integrations
  • Processing-method edits and version changes
  • Deletions
  • Sequence aborts or stops
  • Re-acquisition into the same sequence
  • Changes to sample weights or dilution factors after acquisition
  • Clock or configuration changes
  • Failed login clusters

Then set two review levels:

  • Data-level review: every GMP result before reporting, filtered to critical events for that sequence.
  • System-level review: periodic, e.g., monthly, covering admin actions, user management, and configuration.

The reviewer records a conclusion, for example "No unexplained critical events" or "Event X reviewed, justified by deviation DEV-24-012". A bare signature is not enough.

6. Chromatographic Integration and Reprocessing Governance

The key conceptual shift is that integration is part of the measurement. A related-substances result is a function of the detector signal and the integration algorithm with its parameters. The same raw signal can produce a 0.08% or a 0.12% impurity depending on how the baseline under a shoulder is drawn. If the parameters are free to vary per injection, the method has an uncontrolled degree of freedom. That is analytically unsound, and it is a data-integrity vulnerability.

6.1 The validated integration approach

  • Each method's processing method defines default events: peak width, slope sensitivity or threshold, area reject, tangent-skim rules for riders on the main-peak tail, and baseline hold or valley rules. The values are justified during validation using representative chromatograms that include the worst-case impurity profile.
  • The processing method is version-controlled and locked. Changes go through change control.
  • Parameter changes apply to the whole sequence: standards, SST, and samples alike. Per-injection changes that are not applied uniformly are prohibited, because they let the analyst treat samples differently from standards.

6.2 Manual integration: permitted, governed, and rare

Manual integration is not forbidden. Forbidding it would push it underground. It is permitted under these conditions:

  • It is used only for SOP-listed justifications, such as co-elution that the automatic algorithm cannot resolve consistently, or a baseline disturbance.
  • Before and after chromatograms are retained.
  • The reason is recorded in the audit trail.
  • A second person approves it before the result is reported.
  • It is trended: a rising manual-integration rate is a method-robustness signal that should feed a method-improvement CAPA.

6.3 Prohibited practices

  • Trial or "prep" injections of samples.
  • Unreported sequences.
  • Integrating to pass SST, e.g., adjusting the baseline to raise resolution above 1.5 or to drop tailing below 2.0.
  • Selective reprocessing, meaning reprocessing only the failing injections.
  • Deleting or overwriting.

An aborted sequence is not deleted. It is retained, its reason is documented (instrument fault, wrong vial, etc.), and the replacement sequence references it.

6.4 Injection reconciliation, the most powerful single DI control

N_acquired (audit trail / data folder for the instrument, period)
     =  N_reported (in released sequences)
      + N_invalidated (each with a deviation / documented reason)
      + N_non-GMP (system blanks, equilibration, column conditioning — logged as such)

Any residual ≠ 0  →  unexplained data  →  deviation (potential DI investigation)

This equation is a conservation law for data. Like mass balance in a synthesis, it does not tell you what happened to the missing quantity, but it proves that something did. The equation is particularly strong because it is cheap to run, objective, and hard to game unless the audit trail itself is compromised. That is why audit-trail configuration rights must be segregated (§5.5).

6.5 Repeat analysis and SST failure

  • An SST failure invalidates the run before sample results are evaluated, provided the SST fails on its own merits.
  • The run is documented and rerun after correction, and the failed run stays in the record.
  • Invalidating a run because the sample results look wrong while the SST passed is not SST handling. It is an OOS investigation (§7).

Reviewer checklist essentials:

  • Review the electronic originals, not printouts.
  • Check processing-method version against the approved version.
  • Review the audit trail for critical events.
  • Reconcile injections.
  • Check SST criteria were met with the locked integration.
  • Verify bracketing standards and sample-solution age against the validated stability window.
  • Verify calculations from raw areas.
  • Check DAD peak purity where the method specifies it, remembering its limits (§10).

7. OOS Investigations, the Barr Decision, and the Statistics of Testing into Compliance

7.1 Barr, the origin story

In United States v. Barr Laboratories (D.N.J. 1993), Judge Wolin addressed how a manufacturer must treat failing test results [United States v. Barr Laboratories, 1993]. The holdings that shaped FDA's OOS guidance, paraphrased, are these:

  • An OOS result requires a failure investigation.
  • An OOS result can be set aside only if the investigation identifies an assignable cause, i.e., lab error.
  • Retesting must follow a predefined procedure and may not continue until a passing result is obtained ("testing into compliance").
  • The USP outlier test did not then apply to chemical assays.
  • Averaging can hide variability that the individual results reveal.

Read the decision itself rather than secondary summaries before you quote it in an SOP.

7.2 Why testing into compliance is a statistical crime

Connect this to the multiple-testing problem from MATH201. Suppose a lot's true purity is 97.9%, which genuinely fails NLT 98.0%. The method's intermediate precision is σ = 0.15% absolute. A single result rounds to ≥98.0, and therefore "passes", if the measured value is ≥97.95.

P(single result passes) = P(X ≥ 97.95) = 1 − Φ((97.95 − 97.90)/0.15) = 1 − Φ(0.33) ≈ 0.37

If you retest until a pass, with k independent attempts:
P(at least one pass in k) = 1 − (1 − 0.37)^k
   k = 1 : 0.37
   k = 3 : 1 − 0.63³ ≈ 0.75
   k = 5 : 1 − 0.63⁵ ≈ 0.90

A failing lot has a 90% chance of producing a "passing" result if you allow five tries and report the pass. This is exactly the "p-hacking" logic you know from statistics. It is the quantitative reason the retest plan must be predefined (number of retests, who runs them, how results are evaluated), and why all results must be reported in the investigation.

7.3 Phase I: the laboratory investigation

Phase I is performed promptly by the analyst and supervisor, before any retest [FDA, 2022]. The checklist covers:

  • Calculation errors, checked from raw areas.
  • Standard and sample preparation: weights, dilutions, volumetric glassware, and standard potency and lot.
  • Instrument performance and SST, including trends within the run.
  • Column history.
  • Integration, checked against the locked method.
  • Solution age versus the validated stability window.
  • Reagents and mobile-phase preparation.
  • Analyst observations recorded during the run.

Hypothesis testing on the original prepared solution is allowed only when it is predefined, e.g., re-injection to test an autosampler or injection fault. It does not replace the original result unless it demonstrates an assignable cause.

The assignable-cause standard is documented, demonstrable evidence of a lab error. "Probable analyst error" is not enough.

7.4 Phase II: full-scale investigation

  • Phase II is triggered when Phase I finds no assignable lab cause.
  • In a contract lab, this is where the client interface begins. The client's QCU owns the manufacturing investigation and the disposition. Your lab owns the laboratory portion and must notify the client promptly. The quality agreement should state the timeline, e.g., within 1 business day of the initial OOS.
  • Retest plan: a predefined number of retests on the same sample (same homogeneous material), by a different analyst where possible. All results are reported.
  • Resampling (new sample from the lot) is allowed only with documented justification, such as evidence that the original sample was not representative or was compromised. It is not a second chance.

7.5 Averaging and outliers

  • Averaging is legitimate only where the method defines the reportable value as an average, e.g., the mean of duplicate preparations. Never average an OOS individual with passing retests to produce a passing mean.
  • Where the individual results matter, as in uniformity-type tests, variability itself is the information.
  • Outlier tests are generally inappropriate for discarding results from chemical assays. The FDA OOS guidance discusses their limited applicability, and USP <1010> discusses outlier tests as statistical tools. [VERIFY current <1010> wording.] A statistical outlier is a reason to investigate, never a reason to discard without an assignable cause.

7.6 OOT: catching the drift before the failure

OOT (out-of-trend) results are within specification but inconsistent with prior behavior. For routine data such as control samples or SST metrics, use an individuals (I-MR) chart:

Center line:  x̄
MR_i = |x_i − x_{i−1}|,   MR̄ = mean moving range
UCL/LCL = x̄ ± 2.66 · MR̄          (2.66 = 3/d₂, d₂ = 1.128 for n = 2)

Western Electric rules (signal if any):
 1. One point beyond 3σ
 2. Two of three consecutive points beyond 2σ on the same side
 3. Four of five consecutive points beyond 1σ on the same side
 4. Eight consecutive points on the same side of the center line

For stability data, use regression of the attribute against time with a prediction interval for a new observation. A point outside the 95% (or predefined) prediction interval is OOT. Detailed treatment is in PDQC831.

ŷ(x₀) ± t_{α/2, n−2} · s · √[ 1 + 1/n + (x₀ − x̄)² / Sxx ]

Note the "1 +" term. The prediction interval is for a new observation, not the confidence interval for the mean. Using the narrower confidence interval inflates the OOT rate, a common mistake.

7.7 Mock investigation frame: 97.8% vs NLT 98.0%

97.8 is unambiguously OOS. No rounding argument applies.

  1. Freeze the data and notify the supervisor and client per the quality agreement.
  2. Phase I checklist. Peptide-specific hypotheses:
    • Did Met oxidation occur in solution (Met(O) peak increased relative to the reference)?
    • Was the sample solution beyond its stability window?
    • Did a new early-eluting peak appear only in this injection (carryover)?
    • Did integration events differ from the locked method?
  3. If there is no assignable cause, move to Phase II. Execute the predefined retest plan, e.g., n = 2 independent preparations by a second analyst, and report all results.
  4. Use LC-MS to identify the impurity responsible: deamidation (+0.984 Da), oxidation (+16 Da), a deletion sequence, or a diastereomer (same mass, so chromatography plus a reference standard is needed).
  5. Report to the client. The client's QCU dispositions the lot.

8. Deviations, Root Cause, CAPA, Effectiveness Checks, and Change Control

8.1 Classification and timelines

ClassDefinition (lab context)Example
MinorNo potential impact on a reported result or data integrityLogbook entry made one hour late, with the data otherwise intact
MajorPotential impact on result validity; requires impact assessmentStandard used one day past its retest date
CriticalActual or likely impact on a reported result or release decision, or a data-integrity breachUnreported injections discovered; result already reported to a client
  • Planned deviations are approved before execution and are rarely appropriate for test methods.
  • Unplanned deviations are recorded when they occur.
  • Example design timelines: initiated within 1 business day, closed within 30 days, or extended with a documented justification. These are internal design choices, not regulatory numbers.

8.2 Root cause: why "analyst error" is almost never the answer

Useful tools:

  • Ishikawa / 6M: Man, Machine, Method, Material, Measurement, Mother Nature (environment).
  • 5 Whys, which is fast but can follow one branch only.
  • Fault-tree analysis, which uses Boolean logic and suits multi-cause events.

"Analyst error" is a description of where the failure surfaced, not why it occurred. Keep asking why:

  • Why could the analyst make this error?
  • Was the SOP ambiguous?
  • Was the worksheet missing a field?
  • Was the dilution scheme error-prone, for example a 1:3 then 1:7 serial dilution with no calculation check?
  • Was the analyst under schedule pressure from a capacity problem?

"Retraining" as the corrective action presumes a knowledge deficit. Most errors by trained people are system-design failures. A recurrence after retraining is the classic inspector signal of an inadequate root cause.

8.3 CAPA and the surprising weakness of effectiveness checks

CAPA actions should be SMART: specific, measurable, achievable, relevant, and time-bound. The effectiveness check must be defined at CAPA approval, with the metric, window, and pass criterion fixed in advance.

Now the counterintuitive statistics. A common criterion is "zero recurrence in 20 runs". What does that actually prove? If you observe 0 events in n trials, the one-sided 95% upper bound on the true event rate p is:

(1 − p)^n = 0.05   →   p_upper = 1 − 0.05^{1/n} ≈ 3/n      ("rule of three")

n = 20  → p_upper ≈ 0.14  (exact: 1 − 0.05^{0.05} ≈ 0.139)
n = 60  → p_upper ≈ 0.05

If the pre-CAPA rate was 10% and nothing changed:
P(0 recurrences in 20 runs | p = 0.10) = 0.9^20 ≈ 0.12

So "zero in 20" would occur 12% of the time even if the CAPA did nothing. That is weak evidence. Size the window to the baseline rate. To claim an improvement from a 10% baseline with reasonable confidence, you need roughly 30 or more clean runs, since 0.9^30 ≈ 0.04. Alternatively, pair a recurrence metric with a leading indicator such as worksheet RFT. This is the kind of reasoning that impresses an auditor, because most effectiveness checks are underpowered without anyone noticing.

8.4 Change control

The workflow runs:

  1. Request
  2. Classification
  3. Risk assessment, with Q9(R1)-proportional formality
  4. Impact assessment on validated methods (is revalidation or verification needed?), qualified instruments (is requalification needed?), and SOPs and training
  5. Client notification per the quality agreement. Some changes require prior client approval, e.g., method changes, site moves, and subcontractor changes.
  6. Implementation
  7. Verification
  8. Closure

Changing a column brand on a related-substances method is a change. Upgrading CDS software is a change. Moving an HPLC across the room is a change, because it may require at least an operational qualification check.

9. Sampling and Sample Management: Garbage In, Defensible Garbage Out

No analytical precision can fix an unrepresentative sample. Sampling error often exceeds analytical error, and it is invisible in the chromatogram.

9.1 Peptide bulk powders

USP <1097> covers bulk powder sampling principles [VERIFY current chapter]. Peptide-specific hazards:

  • Hygroscopicity: the peptide gains water on opening, so sample in low RH.
  • Electrostatic charge: lyophilized peptides cling to spatulas and walls, so use anti-static tools and ionizers.
  • Segregation: in practice this is a smaller issue for single-component lyophilizates than for blends, but still a concern for lots built from multiple sublots.

A common container-sampling heuristic for a homogeneous lot is n = √N + 1 containers. For N = 25 containers, that gives 6. This is a convention without a statistical guarantee. It assumes homogeneity across containers and must be justified, or replaced with a statistically based plan if containers come from different sublots.

9.2 Vialed product

For attribute inspection of vialed product, use ISO 2859-1 / ANSI/ASQ Z1.4. The lot size gives a code letter, and the code letter plus the AQL gives the sample size and accept/reject numbers. Remember that the AQL describes the producer's risk. It is not a guarantee that the accepted lot's defect rate is at or below the AQL.

9.3 Containers

  • Peptides adsorb to glass and to some plastics. Hydrophobic and cationic sequences are the worst.
  • Low-bind polypropylene versus Type I glass should be decided from data: a recovery study at the working concentration.
  • Protect Trp- and Met-containing peptides from light, and use headspace control (inert gas) where oxidation is a known pathway.

9.4 Vial-allocation table (example, sterile peptide vial lot)

UseVialsBasis
Chemistry (ID, purity/RS, assay)4Duplicate preps; 2 vials pooled per prep
Water (contracted)2Method requirement
Counter-ion (contracted)2Method requirement
Endotoxin (contracted)3Method plus inhibition/enhancement check
Sterilityper compendial table [VERIFY USP <71> minimum quantities against fill volume and lot size]Compendial
Retains (211.170, 2× non-sterility tests)≥ 2 × (4+2+2+3) = 22Regulatory minimum; confirm endotoxin inclusion with the client
Investigation reserve6Enough to support one full OOS retest plan

9.5 Receipt, chain of custody, reconciliation

  • USP <1079> frames storage and transport risk.
  • Use temperature loggers with predefined receipt acceptance criteria, e.g., no excursion beyond 2–8 °C for longer than the client-defined allowable time. Excursions go to the client for disposition, because the lab does not decide product impact.
  • A chain-of-custody form records every transfer of custody.
  • Physical segregation: quarantine, then in test, then retained.
N_received = N_consumed + N_retained + N_destroyed (+ N_in-test, until closure)
Residual ≠ 0  →  deviation

This is the third conservation law in this course, after injection reconciliation and worksheet issuance. The pattern is deliberate. Every countable thing in a GMP lab should balance.

10. Reporting, Rounding, Decision Rules, and the Scope of a Claim

10.1 USP rounding (General Notices 7.20)

Carry unrounded values through all calculations. Round only the final reportable value to the number of decimal places in the specification limit, then compare. The rule rounds up when the next digit is ≥5 and down when it is <5 [VERIFY current GN 7.20 text].

Example 1:  computed purity 98.04%,  spec NLT 98.0%
            round to one decimal → 98.0   →  conforms

Example 2:  computed purity 97.96%,  spec NLT 98.0%
            round → 98.0   →  conforms   (surprising to many — and correct under compendial convention)

Example 3:  impurity 0.105%,  spec NMT 0.10%
            round to two decimals → 0.11  →  does not conform

Example 3b: impurity displayed as "0.105" but unrounded value is 0.10496
            round → 0.10  →  conforms
            → the displayed value is not the reportable basis; the unrounded value is.

Example 3b is why spreadsheets and CDS custom calculations that round intermediate steps are data-integrity and accuracy hazards. Rounding twice can flip a decision.

10.2 Decision rules: compendial simple acceptance vs ISO/IEC 17025 guard bands

The two frames differ:

  • Compendial frame: USP tests are applied with simple acceptance. The rounded result is compared directly to the limit. The method's validation and SST carry the burden of measurement reliability.
  • ISO/IEC 17025 frame: clause 7.8.6 requires that when the lab states conformity, it applies a documented decision rule that accounts for the risk from measurement uncertainty, and states that rule on the report unless it is inherent in the requested specification. ILAC-G8:09/2019 and JCGM 106:2012 describe guard-banded rules [ILAC-G8, 2019], [JCGM 106, 2012].
Lower tolerance limit T_L (e.g., 98.0%).  Expanded uncertainty U = k·u (k = 2).
Guard band w (often w = U).   Acceptance limit A_L = T_L + w.

Example: u = 0.15%, U = 0.30%  →  A_L = 98.3%
  Measured 98.1%:
    Simple acceptance           → conforms
    Guarded (w = U) binary rule → conformity not demonstrated / "fail" per agreed rule

Probability of false acceptance (PFA) at the acceptance limit, normal model:
  measured = T_L       → PFA ≈ 50%
  measured = T_L + 2u  → PFA ≈ 2.3%

The beautiful and uncomfortable insight is that a result sitting exactly on the limit is a coin toss. Simple acceptance silently accepts up to 50% specific false-acceptance risk for results at the limit. This is not wrong. It is a convention, compensated for by specifications set with process knowledge and by method validation. But you should know you are accepting it.

For a compendial specification, follow the compendial convention unless the client agrees otherwise. Agree the rule in writing in advance, in the quality or technical agreement. Choosing a rule after seeing the result is the reporting equivalent of testing into compliance.

10.3 Conditional reporting

  • Pending: an incomplete test is listed as "pending", never omitted silently.
  • Result with deviation: the result is reported with a reference to the deviation and an impact statement.
  • Subcontracted result: identified as such, with the provider named or identifiable per the agreement (ISO/IEC 17025 clauses 6.6 and 7.8; [VERIFY exact sub-clause]).
  • Below LOQ: report as "<LOQ value", e.g., "<0.05%", and state the convention for totals. For example, impurities below the reporting threshold are excluded from total impurities. Document this convention in the method.

10.4 Scope-of-claim language

The lab states that "the sample, as received, was tested by the methods listed, and the results conform / do not conform to the specification provided by the client [spec ID, version]." The client's QCU dispositions the lot.

Prohibited words on a CoA include "GMP-released", "pharmaceutical-grade", "safe for use", "FDA-approved", and "meets ANDA requirements". Each asserts a regulatory status or a product-level conclusion the lab cannot make.

Signature meanings must be explicit, and must conform to 11.50 if the signature is electronic. Amended reports carry a new unique identifier, reference the superseded report, state the reason for amendment, and are re-authorized by an authorized signatory.

11. Contract-Laboratory Management: You Can Outsource the Work, Not the Responsibility

The core principle of FDA's 2016 Quality Agreements guidance: the owner remains responsible for the quality of the product. Contracting a function does not transfer accountability. [FDA, 2016]

Peptide Partners will be both a contract lab to its clients and an owner of outsourced testing for capabilities it does not yet have: KF, IC, ICP-MS, endotoxin, GC-HS, AAA/qNMR. You must design both sides of the agreement.

11.1 RACI matrix (excerpt)

ActivityClient QCUPeptide Partners labSubcontract lab
Specification ownershipA/RCI
Method validation / transfer / verificationARR (own methods)
OOS lab investigation (Phase I)I (notify ≤1 business day)A/RR for its results
Lot dispositionA/R——
Data retentionARR
Change notificationA (approve)RR
AuditsR (audits lab)R (audits subcontractor)C

11.2 Technical agreement content

  • Method status: compendial, verified, transferred, or validated.
  • Sample shipping and storage conditions, and excursion handling.
  • Turnaround time (TAT).
  • Data-package contents: raw electronic data or certified true copies, audit-trail extracts for the relevant sequences, and SST and chromatograms.
  • Retention period, aligned to 211.180 and the client's needs.
  • Decision rule (§10.2).
  • Subcontracting permission and flow-down: the subcontractor is bound by the same DI, OOS, and change-notification requirements.

11.3 Oversight

  • A qualification audit before first use, with a checklist covering QMS, DI, OOS handling, instrument qualification, and training.
  • A scorecard: TAT, invalid-run rate, OOS-notification timeliness, and data-package completeness. Compute these the same way as your own KPIs (§3.4), so the system is consistent.
  • Periodic raw-data review, e.g., pull one dataset per quarter and review the audit trail.
  • Requalification triggers: an FDA 483 or Warning Letter at the subcontractor, a confirmed DI issue, two late OOS notifications, or an ownership or site change.

11.4 Outsourcing bridge as a purchase decision

Each capability gap is an investment decision with evidence. Compare annual contract cost (volume × price + shipping + oversight labor + TAT cost) against in-house cost (capital amortization + qualification + consumables + staff time + proficiency testing).

KF is often the first purchase to justify. It is cheap, high-volume, and water content drives the peptide content calculation. Water is also sensitive to transit time and packaging, so in-house testing reduces excursion risk. ICP-MS rarely justifies itself at low volumes.

12. Laboratory Operations, Capacity, and Business Continuity

12.1 Capacity model: a worked example

Related-substances method: 60 min gradient + 10 min re-equilibration = 70 min/injection

Sequence for 4 lots (duplicate preps):
  blank 1 + SST 6 + sensitivity (LOQ) 1 + samples 8 + bracketing stds 2  = 18 injections
  18 × 70 min = 21.0 h   + initial column equilibration 1.0 h = 22.0 h

Available instrument time: 24 h × availability 0.80 (maintenance, failures) ≈ 19.2 h
→ one 4-lot RS sequence does NOT fit in one instrument-day; split to 3 lots or run over 2 days
  (and check solution stability covers the full residence time!)

Three 1260s: ≈ 57.6 instrument-h/day ≈ 2.6 four-lot RS sequences/day theoretical

Now the review bottleneck. If one sequence takes about 2.5 h of reviewer time (including audit trail and reconciliation), and the reviewer has about 5 productive review hours per day alongside other duties, review capacity is about 2 sequences per day. That is less than instrument capacity.

Little's law (WIP = throughput × lead time) says that once arrivals exceed the slowest stage's capacity, the queue in front of it grows without bound. In small labs, TAT failures are almost always review-limited, not instrument-limited. Buying a fourth HPLC would not improve OTIF. A second qualified reviewer would.

Note also the solution-stability coupling: a 22 h sequence means the last sample sits in the autosampler for about 22 h. The validated solution-stability window must cover that.

12.2 Inventory and lot qualification

  • Set min/max levels and use FEFO (first expiry, first out).
  • Prepared solutions carry assigned expiry dates justified by stability data.
  • Lot qualification before a new lot enters GMP use:
    • New column lot: SST on reference standard meets criteria, with resolution of the critical pair.
    • HPLC solvents and TFA: blank gradient shows no ghost peaks above, e.g., 50% of the reporting threshold in the impurity window.
    • Vials and caps: blank or diluent injection after a soak period to check for extractables.

TFA quality matters especially for peptides. Degraded TFA contributes baseline artifacts at low wavelength (~214–220 nm), right where peptide-bond detection lives.

12.3 Downtime and continuity

  • Keep an instrument failure log.
  • For in-flight samples, the impact assessment asks whether the solution is still within its stability window, and whether a sequence can be resumed or must be re-prepared.
  • Reroute only to an equivalent qualified instrument, or to a qualified contract lab under agreement.
  • Test that CDS backups can actually be restored. Schedule a restore test, because an untested backup is a hope, not a control (211.68(b)).
  • Validate UPS sizing and the CDS behavior on power loss: what happens to an in-progress sequence.
  • Key-person dependency. If Larissa is the only qualified analyst and reviewer for LC-MS/MS, the lab has a single point of failure. The control is cross-training, plus a contract-lab fallback written into business continuity.
  • Address single-source supplier risk for critical columns and reference standards.

Connections to Peptide Chemistry

Almost every quality-system decision in this course is shaped by peptide-specific chemistry you learned in CHEM202, BCHM602, BCHM603, and BCHM613:

  • Purity ≠ content.
    • HPLC area% purity (e.g., 98.5%) and net peptide content (e.g., 75–85% w/w, after water and counter-ions such as TFA or acetate) are different reportables.
    • The CoA must name which value is reported and how it was calculated. Mixing them up is one of the most common peptide QC reporting errors.
    • This is also why KF and IC, currently outsourced, are tied to the assay result: if their data are late or wrong, the content value is wrong.
  • SPPS impurity families define specificity.
    • From BCHM602, the main families are deletion sequences, truncations, insertions, incomplete deprotection adducts, aspartimide and its β-Asp/D-Asp products, and racemization (D-epimers).
    • Diastereomers are the hardest case. They are isobaric, so MS cannot tell them apart, and they are UV-identical, so DAD peak purity also fails.
    • Only chromatographic selectivity, demonstrated with spiked reference material, proves specificity. This is the strongest argument against relying on peak purity as a specificity claim.
  • Solution chemistry creates false OOS results.
    • Met oxidation, Asn deamidation (+0.984 Da, pH- and temperature-dependent), aggregation, and adsorption all happen in the vial after preparation.
    • Solution-stability data are therefore the first line of Phase I hypotheses. They also define the maximum length of a sequence.
  • The 0.10% threshold drives method capability.
    • The 0.10% figure from the 2021 peptide ANDA guidance is not a lab specification. It is a requirement on the method: an LOQ around 0.05%, resolution of critical pairs, and LC-MS identification capacity.
    • Your triple quad and second MS are therefore not only research tools. They are the identification arm of the related-substances control strategy.
  • Integration governance is hardest for peptides.
    • Peptide related-substances chromatograms contain many closely eluting species on the main-peak tail.
    • This is where tangent-skim rules, valley-drop versus perpendicular decisions, and manual integration temptations concentrate. The integration SOP is not generic. It is written around these chromatograms.
  • QbD connection. Yu et al. (2014) frame quality as designed in through understanding of critical quality attributes and their sources of variability [Yu et al., 2014]. Your FMEA and control-strategy matrix are the laboratory-scale expression of that philosophy. Every impurity attribute is traced back to a synthetic origin and forward to a method that can see it.

Common Misconceptions to Watch Out For

  1. "ISO/IEC 17025 accreditation means we're GMP-compliant." No. It certifies competence within a listed scope. cGMP is a separate legal frame, and FDA does not accept accreditation in place of it.
  2. "Our software is Part 11 compliant." Software cannot be compliant. Only a configured, validated system used under procedures can be. A Part 11-capable CDS with shared logins is non-compliant.
  3. "The printout is the raw data." For a CDS, the electronic record, including metadata and audit trail, is the original [FDA, 2018].
  4. "Manual integration is prohibited." It is permitted under governance. What is prohibited is undocumented, unjustified, or selectively applied integration.
  5. "An OOS result means the batch fails." Not necessarily. Under Barr, the result stands unless an assignable cause is found, and disposition belongs to the client's QCU after a full investigation. The opposite error, "retest until it passes", is far more dangerous.
  6. "We can use a Grubbs or Dixon test to drop the high replicate." Outlier tests are generally inappropriate for invalidating chemical assay results. They flag; they do not excuse.
  7. "Retraining fixed it." Retraining addresses knowledge deficits. Most lab errors are system-design failures, and recurrence after retraining is an inspection red flag.
  8. "Zero recurrences proves the CAPA worked." Only if the observation window has statistical power. Zero in 20 is consistent with an unchanged 10% failure rate about 12% of the time.
  9. "RPN 120 vs RPN 100: the first is a bigger risk." RPNs are products of ordinal scores. Equal RPNs can describe completely different risks. Always apply a severity override.
  10. "97.96% fails NLT 98.0%." Under USP rounding conventions it conforms, because it rounds to 98.0. And "0.105% passes NMT 0.10%" is false: it rounds to 0.11 and fails. Round once, at the end, from unrounded values.
  11. "Guard bands are always required." Under ISO/IEC 17025 the decision rule must be documented and agreed. For compendial specifications, simple acceptance is the convention. Neither frame lets you choose the rule after seeing the result.
  12. "503A clients don't need GMP-quality data." The legal frame differs, but a patient receives the product either way, state boards and FDA can still act, and a lab with two quality standards will eventually misapply the weaker one. Run one system at the higher standard.
  13. "Outsourcing endotoxin testing transfers responsibility." The owner remains responsible [FDA, 2016]. You must qualify, contract, monitor, and review the subcontractor's data.
  14. "Low invalid-run and OOS rates mean a good lab." Implausibly low rates are an inspection signal that failures are going unrecorded. Credible metrics show real, explained variation.
  15. "Backups are done nightly, so data are enduring." Not until a restore has been tested and documented.

Primary reading to pair with these notes: McDowall (2019) chapters on data governance and the data lifecycle; McDowall (2017) on CDS validation and role configuration; ISPE GAMP Records & Data Integrity Guide (2017) on risk-based audit-trail review; Nally (2016) on QC-unit structure; the Barr opinion (1993) in full; and FDA's 2022 OOS guidance alongside it, to see how the court's reasoning became regulatory expectation.

⚙️ Mechanisms & Applications

GPT-6 Astra

Mechanisms & Applications — From a Peptide Measurement to Defensible Quality Evidence

Skipper’s central connection: in analytical chemistry, I learned to ask whether a measurement estimates the intended property. In a pharmaceutical quality system, I must also demonstrate which sample was measured, under which authorized conditions, what else happened, and why the final conclusion follows from the complete record. A correct number with an unverifiable history is not adequate release evidence.

The quality system is therefore part of the measurement system. It controls the transformations from material to sample, sample to signal, signal to result, and result to decision. This is especially important for peptides: an apparently excellent purity result can coexist with incorrect identity, low peptide content, excessive water or counter-ion, or an unmeasured microbiological risk.

Working boundary: numerical thresholds and timelines identified below as examples are proposed laboratory controls—not universal regulatory requirements. Verify applicability and wording against current official regulations, USP–NF, FDA guidance status, and the laboratory’s contracts before issuing controlled documents. Peptide Partners’ exact CDS configurations and several instrument capabilities remain unconfirmed.

1. Regulatory Mapping: Determine Which Rules Govern the Evidence

The first mechanism is not an analytical procedure; it is an applicability decision. A laboratory cannot design appropriate records, validation, or reporting controls until it knows the intended use of its data.

  1. Identify the material and activity. Is this an API, finished drug product, compounded preparation, stability sample, development sample, or investigation sample?
  2. Identify the client’s regulatory setting. Conventional drug manufacture, qualifying section 503A compounding, and section 503B outsourcing are not interchangeable.
  3. Identify each requirement’s source. Separate statute, regulation, applicable compendial requirement, guidance, voluntary standard, and contractual commitment.
  4. Translate the requirement into evidence. For example, “complete laboratory records” becomes an identifiable sample, approved method, preparation records, original data, calculations, review, and traceable reporting.
  5. Assign responsibility. Specify who supplies the specification, qualifies the method, investigates failures, retains records, and makes the disposition decision.
Control-strategy row =
quality attribute → method → specification → intended use
→ authority/source → applicability → evidence → responsible owner
  • FD&C Act section 501(a)(2)(B): cGMP adulteration is about inadequate manufacturing controls, not merely a failed final assay. Passing a test does not cure a deficient quality system.
  • Section 501(b): compendial strength, quality, and purity requirements require attention to the applicable official compendium and statutory qualifications.
  • Section 503A: qualifying compounded drugs are exempt from specified provisions, including section 501(a)(2)(B). This is not an exemption from all federal adulteration provisions or applicable state requirements.
  • Section 503B: qualifying outsourcing facilities are not exempt from cGMP. Do not describe an FDA draft or interim enforcement approach as replacing the underlying statute and regulations.
  • ISO/IEC 17025:2017: accreditation concerns demonstrated laboratory competence within an accredited scope. It is not “cGMP certification,” and neither accreditation nor a CoA establishes that a product is legally marketable.

API versus finished-product distinction: ICH Q7 supplies an important GMP framework for API activities; 21 CFR Part 211 addresses finished pharmaceuticals. Do not copy finished-product requirements into an API agreement without considering applicability. Conversely, a laboratory serving finished-product release cannot dismiss the client’s applicable Part 211 requirements merely because it is an independent contractor.

For synthetic peptides, FDA’s 2021 guidance on certain ANDAs provides a useful example of a demanding impurity-comparison framework. Its discussion of new impurities above 0.10% is context-specific; it is neither a universal peptide impurity limit nor proof that a tested peptide qualifies for that regulatory pathway.

Application: the same related-substances chromatogram may support development, API control, or finished-product release. The chemistry can be identical while the required validation, approval, investigation, retention, and reporting obligations differ. Sources: FD&C Act §§501, 503A, 503B; 21 CFR Parts 210–211; ICH Q7; ISO/IEC 17025:2017; FDA synthetic-peptide ANDA guidance, 2021.

2. The Pharmaceutical Quality System as a Feedback-Control Mechanism

ICH Q10 is easier to understand through the control systems encountered in physiology. A sensor alone does not maintain homeostasis. The system needs a target, monitoring, an error signal, an intervention, and feedback demonstrating that the intervention worked.

Approved process
    ↓
Performance and product-quality monitoring
    ↓
Deviation, OOS, OOT, complaint, or adverse trend
    ↓
Investigation and impact assessment
    ↓
Correction + corrective/preventive action
    ↓
Controlled implementation and effectiveness check
    ↓
Management review and resource decisions

Correction and corrective action differ. Replacing a leaking autosampler seal repairs today’s instrument. Determining why the seal failed prematurely, identifying affected analyses, and establishing an evidence-based maintenance interval addresses recurrence. “Retrain the analyst” is not a root cause unless an actual competency deficit explains the event.

For Zak and Larissa, the system should be small enough to operate but strong enough to prevent self-approval. One person may hold several roles, but cannot independently review their own analytical work. Qualified external support may be necessary for independent review, quality oversight, IT administration, and absence coverage.

FMEA example: loss of a peptide impurity during sample preparation. Adsorption can preferentially remove a hydrophobic impurity, producing a falsely favorable purity result even when the parent-peptide peak appears stable.

Example laboratory scoring scale: 1–5 for each factor

RPN = Severity × Occurrence × Detectability

Detectability = 5 means unlikely to be detected before reporting.

Illustrative initial assessment:
S = 5, O = 3, D = 4
RPN = 5 × 3 × 4 = 60

After supported container/recovery studies and suitable controls:
S = 5, O = 2, D = 2
Residual RPN = 20

The severity has not decreased: an incorrect release-supporting result remains consequential. Only the estimated likelihood and detectability have changed. An example action rule might be “RPN ≥40 requires action; every severity-5 hazard requires explicit review regardless of RPN.” RPN is an ordinal prioritization tool—not a physical quantity, calibrated failure probability, or substitute for judgment. Document scoring assumptions and uncertainty, as emphasized by ICH Q9(R1).

Management dashboard: define every numerator, denominator, reporting period, and exclusion. Distinguish invalid runs from OOS results and confirmed OOS results.

Invalid-run rate = invalid runs / all initiated runs × 100%

Right-first-time review =
packages accepted without correction / packages reviewed × 100%

Example:
6 invalid runs / 120 initiated runs = 5.0%
54 first-pass approvals / 60 reviewed packages = 90.0%

A falling OOS rate is not automatically improvement: it can also reflect suppressed reporting. Pair performance metrics with audit-trail findings, investigation quality, overdue actions, and client-notification performance. Sources: ICH Q10; ICH Q9(R1); 21 CFR 211.22 and 211.192.

3. Controlled Documents and Competence: Making the Intended Process Reproducible

A document hierarchy resembles executable software: the quality manual defines architecture, SOPs define workflows, analytical procedures define specific operations, and completed records demonstrate what actually executed.

  • Level 1: quality manual—scope, organization, independence, regulatory applicability, and governance.
  • Level 2: policies and SOPs—OOS, data integrity, change control, sample management, reporting, and outsourcing.
  • Level 3: analytical procedures and work instructions—specific peptide assays, CDS operations, equipment checks, and calculation instructions.
  • Level 4: controlled forms and completed records—worksheets, logs, investigation records, and training evidence.
  1. Assign a unique identifier, owner, version, approval, and effective date; for example, QA-SOP-008 for OOS investigations.
  2. Assess changes for training, method validation, equipment qualification, software, and client impact.
  3. Complete required training before the revised process becomes effective.
  4. Remove obsolete point-of-use copies while preserving historical versions needed to interpret older records.
  5. Authorize personnel by demonstrated competence for the actual task.

“Read and understood” may suffice for a minor administrative revision. It does not establish competence in manual integration or an unfamiliar peptide extraction. Those require observed performance, acceptable calculations, recognition of failure modes, and documented authorization.

Record observations contemporaneously. Correct paper records without obscuring the original entry; identify the person, date, and reason as appropriate. Use justified “not applicable” entries rather than unexplained blanks. In electronic systems, corrections must remain traceable. Never reconstruct a missing original record as though it had been contemporaneously created.

Sources: 21 CFR 211.25 and 211.194; ISO/IEC 17025:2017, clause 6.2; USP <1029>.

4. Data Integrity: Preserve the Experiment, Not Just Its Printed Answer

A chromatographic result is a data object plus context: detector signal, sample identifiers, acquisition method, processing method, integration events, user identities, timestamps, calculations, and relevant audit trails. A PDF usually cannot preserve all the functions and relationships of the dynamic electronic record.

This resembles structural biology: a picture of a refined protein is not a substitute for diffraction data, processing decisions, and refinement evidence.

  1. Map the actual configuration. Record the CDS product/version, licenses, storage architecture, instrument connections, Windows identities, permissions, time settings, interfaces, and backup paths. Do not infer software capability from “Agilent 1260.”
  2. Map the data lifecycle. Follow sample receipt through acquisition, processing, review, reporting, archival, and authorized destruction.
  3. Test ALCOA+ controls. Can the record be attributed? Is it contemporaneous? Are original data and metadata complete, consistent, enduring, and available?
  4. Separate privileges. Analysts need analytical permissions—not unrestricted deletion, clock modification, audit-trail disabling, or administrative access.
  5. Challenge failure modes. Test aborted acquisitions, network loss, reintegration, restoration, and unauthorized-access attempts under approved protocols.
  6. Review the relevant audit trails. Examine changes affecting reportable results before approval; document the reviewer’s conclusion and resolution of unexplained events.
  7. Demonstrate retrieval and restoration. A backup is useful only if it can reconstruct complete, readable, usable records with their relationships intact.
Example injection reconciliation:

48 acquired injections =
6 blanks + 8 standards + 30 routine sample injections
+ 2 authorized investigation injections + 2 aborted acquisitions

Every acquisition must map to:
sample/purpose → sequence → original file → processing history
→ reported result or documented explanation

Important distinction: backup supports recovery; archival supports controlled long-term preservation and retrieval. Neither a successful file-copy message nor a checksum alone proves restoration of a usable CDS record. A restore challenge should verify raw signals, metadata, methods, audit trails, access, and accurate retrieval or reprocessing in a controlled environment.

Part 11 applicability depends on the electronic records and signatures used to satisfy FDA requirements. There is no instrument-independent “Part 11 compliance switch”: configuration, validation, procedures, training, and governance all matter. Interim injection logs and second-person reconciliation can reduce risk, but cannot reliably compensate for a system that permits undetectable deletion. Such a gap may require restricting release-supporting work or outsourcing it until remediation.

Sources: 21 CFR 11.10, 11.50, 11.70, and 11.100–11.300; 21 CFR 211.68 and 211.194; FDA Data Integrity Q&A, 2018; PIC/S PI 041-1; McDowall, 2017 and 2019; ISPE GAMP Guide: Records and Data Integrity, 2017.

5. Chromatographic Integration: Turning a Signal into an Accountable Result

Integration is a measurement model, not cosmetic editing. Peak boundaries and baseline construction determine the area attributed to each component. For a peptide impurity near its limit, a small baseline change may alter the conformance conclusion.

Uncorrected area percentage:
Impurity i (%) = A_i / ΣA_j × 100

If RRF_i is defined as:
(response per concentration of impurity i) /
(response per concentration of reference component),

then corrected contribution is proportional to:
A_i / RRF_i

The method must define the included peaks, exclusions,
response-factor convention, and denominator.

Biochemical connection: peptide UV responses depend on wavelength, peptide bonds, aromatic residues, and impurity structure. An area-percent result is not automatically a mass-percent result. Nor does “98.0% chromatographic purity” establish “98.0% peptide content by mass.” Water, salts, residual solvents, and poorly detected impurities break that equivalence.

  1. Establish the processing approach during method development and validation. Define expected peaks, integration settings, reporting thresholds, permitted baseline events, and rules for unresolved components.
  2. Acquire using the approved sequence. Account for conditioning, blanks, standards, SST, samples, and required bracketing checks.
  3. Apply processing consistently. Do not tune individual injections until they pass. A necessary method-level change requires an impact assessment and consistent treatment of comparable affected data.
  4. Use manual integration only under controlled exceptions. Retain original and revised integrations, a scientifically specific reason, attributable audit-trail entries, and independent review.
  5. Review the whole experiment. Examine blanks, carryover, sequence completeness, SST, processing versions, excluded peaks, and unexplained reprocessing.

An injection-specific disturbance may justify a documented exception under the SOP; it does not justify arbitrary per-injection optimization. Similarly, authorized equilibration and conditioning injections are not the same as undisclosed injections of reportable samples used to discover a passing condition.

DAD peak purity: spectral consistency across a peak can flag coelution, but spectrally similar peptide impurities may escape detection. A passing DAD purity test is supporting evidence, not proof of chemical homogeneity. Triple-quadrupole LC-MS/MS can selectively monitor known analytes, but targeted transitions do not provide a comprehensive search for unknown impurities.

Sources: 21 CFR 211.160 and 211.194; FDA Data Integrity Q&A, 2018; McDowall, Validation of Chromatography Data Systems, 2017.

6. OOS Investigations: Test Explanations, Not Until a Passing Number Appears

Mock case: a peptide related-substances procedure gives a reportable purity of 97.8% against NLT 98.0%. The result is OOS. Its closeness to the limit does not authorize extra injections, averaging, or favorable reintegration.

OOS observed
    ↓
Preserve records, preparations where stable, and relevant materials
Notify laboratory supervision/quality; control reporting and disposition
    ↓
Phase I: prompt laboratory assessment
    ├─ Demonstrated assignable laboratory cause?
    │      ↓
    │   Document evidence, impact, invalidation rationale,
    │   authorized repeat work, and corrective action
    │
    └─ No demonstrated laboratory cause?
           ↓
        Phase II: full-scale investigation with client/owner
           ↓
        Preapproved additional testing if scientifically justified
           ↓
        Assess all evidence; do not erase the original OOS
           ↓
        Client quality unit makes disposition decision

Phase I sequence:

  1. Check transcription, calculations, specification, method version, and reportable-value definition.
  2. Review sample and standard weights, dilutions, reference-standard potency basis, and preparation records.
  3. Review instrument status, SST, blanks, carryover, column history, and audit trails.
  4. Assess sample homogeneity, extraction, adsorption, filtration, and solution stability.
  5. Form specific hypotheses and define tests capable of supporting or refuting them. Use existing solutions when appropriate and within established stability; authorize the scope before executing investigation injections.

A failed explanation: “The analyst probably made a dilution error.” For normalized area purity, uniform dilution often affects parent and impurity signals proportionally, leaving the ratio unchanged. It could matter through nonlinearity, threshold effects, or concentration-dependent adsorption—but those mechanisms require evidence. A plausible story is not an assignable cause.

Completed mock conclusion: calculations, SST, standard preparation, processing, and solution-age checks are acceptable; no laboratory cause is established. Phase II is opened with the client. Suppose a preapproved plan produces fresh-preparation results of 98.1% and 98.2%. Those results do not invalidate 97.8%.

Numerical average:
(97.8 + 98.1 + 98.2) / 3 = 98.033...%

But this is NOT an authorized reportable result merely because it passes.
The original OOS remains part of the investigation.

If the investigation remains inconclusive, report that limitation and all relevant results. Do not relabel it “laboratory error” to close the file. Resampling requires evidence supporting why the original sample was unrepresentative; it is not simply another opportunity to pass.

Historical importance: United States v. Barr Laboratories (1993) became a landmark in the treatment of OOS results, laboratory investigations, retesting, and averaging. The lasting lesson is that statistical manipulation cannot substitute for an investigation into why the original result occurred. Apply the current FDA guidance rather than treating the court decision as a complete modern SOP.

Sources: FDA, Investigating Out-of-Specification (OOS) Test Results for Pharmaceutical Production, Rev. 1, 2022; 21 CFR 211.192; United States v. Barr Laboratories, Inc., 812 F. Supp. 458 (D.N.J. 1993).

OOS asks: does the result meet the specification? OOT asks: is it consistent with an appropriate historical or stability expectation? A result can be within specification yet provide evidence of a developing problem.

Illustrative individuals chart for a stable, comparable process:

MR_i = |x_i − x_(i−1)|
Estimated σ = average moving range / 1.128
Control limits = historical mean ± 3 × estimated σ

Example:
Mean impurity = 0.080%
Average moving range = 0.010 percentage points

Estimated σ = 0.00887 percentage points
Upper control limit ≈ 0.107%

If the specification is NMT 0.15%, a result of 0.12% is within specification but above this illustrative control limit. It warrants investigation of process or analytical change. The control limit is not a replacement specification.

Use comparable products, methods, and operating conditions. Data dominated by values below LOQ, mixed populations, serial correlation, or a deliberate stability trend need different treatment. Preselect chart rules; applying many rules after seeing the data increases false alarms. For stability, a regression model with appropriate prediction intervals is generally more informative than assuming a constant mean.

CAPA effectiveness also needs statistical humility. “Zero recurrence in 20 runs” may be a useful predefined check, but it is weak evidence for a very low failure probability.

With 0 failures in n independent comparable opportunities:
one-sided 95% upper bound for failure probability
= 1 − 0.05^(1/n)

For n = 20:
upper bound ≈ 13.9%

Approximation for larger n: 3/n

Thus, absence of observed recurrence is not proof of elimination. Match the follow-up window and detection sensitivity to the risk.

History: Walter A. Shewhart developed statistical process-control concepts in the 1920s and articulated them in his 1931 book. The key insight was separating routine variation from signals suggesting a changed process. Deming and Juran later helped embed statistical thinking and management responsibility into industrial quality. Pharmaceutical quality systems extend those principles to patient-protective decisions and regulated evidence.

Sources: ICH Q9(R1); ICH Q10; USP <1010>; Shewhart, Economic Control of Quality of Manufactured Product, 1931.

8. Sampling and Sample Management: Control What Enters the Instrument

No amount of chromatographic precision repairs an unrepresentative sample. Bulk peptide powders can segregate, adsorb moisture, cling electrostatically, or lose material during transfer. Sampling error and preparation bias may exceed detector noise.

  1. Define the lot and the question. Is the objective identity, average composition, container-to-container variation, or defective-unit detection?
  2. Select a scientifically justified sampling design. Specify locations, containers, increments, composites where justified, and precautions against moisture or contamination.
  3. Allocate material before receipt. Include primary testing, method-specific replicates, external testing, reserve samples, and justified investigation reserve.
  4. Control receipt and custody. Record sender, lot, quantity, seal condition, transport records, temperatures where relevant, storage, transfers, and discrepancies.
  5. Control preparation materials. Assess glass, polypropylene, filters, and vial caps for recovery, adsorption, leachables, and stability at relevant concentrations.
  6. Reconcile final disposition. Record consumption, residual material, retains, returns, and destruction in meaningful units.
Example receipt reconciliation:
60 vials received =
18 consumed internally + 12 sent to qualified laboratories
+ 20 retained + 8 unused/available + 2 damaged and destroyed

Reserve minimum, where the applicable finished-product rule applies:
quantity ≥ 2 × quantity needed for required tests,
excluding sterility and pyrogen testing

If the applicable test set consumes 8 vials:
minimum reserve quantity = 16 vials
(subject to suitability of the quantity/unit calculation)

The first allocation is illustrative, not a universal testing plan. Product-specific requirements and destructive-test quantities govern.

Sampling mathematics: if units are independent and the defect fraction is p, the probability of detecting at least one defective unit in a random sample of n is:

P(detect ≥1 defect) = 1 − (1 − p)^n

For p = 0.01 and n = 10:
P ≈ 9.6%

This explains why “we sampled ten units” is not strong assurance against rare defects. The √n+1 rule is not a universal statistical guarantee. ISO 2859-1 attribute plans require defined lot size, inspection level, AQL, and switching rules; an AQL is not a promise about every accepted lot.

Reserve nuance: under 21 CFR 211.170, finished-product and active-ingredient provisions differ, and special retention provisions exist. For covered drug-product reserves, annual visual examination concerns samples from representative batches, unless examination would affect integrity—not necessarily every reserve sample from every batch. Verify the exact applicable provision rather than writing “all retains: one year beyond expiry” indiscriminately.

Owning BacT/ALERT does not by itself establish validated pharmaceutical sterility-testing capability. Intended use, product suitability, method validation or verification, environmental controls, and acceptance of an alternative method require assessment. Until capability is demonstrated, the outsourcing bridge remains essential.

Sources: 21 CFR 211.84 and 211.170; USP <1097> and <1079>; ISO 2859-1; ICH Q7.

9. Reporting: Separate Calculation, Decision Rule, and Disposition

Three distinct decisions must remain visible: how the result was calculated, how conformance was judged, and who may disposition the lot.

  1. Carry sufficient precision through intermediate calculations.
  2. Apply the method’s authorized definition of the reportable value.
  3. Apply the applicable rounding and conformance rule.
  4. Report deviations, investigation status, subcontracting, and limitations transparently.
  5. Obtain authorized technical review and signature.
  6. Preserve superseded reports when issuing a controlled amendment.
Illustrative USP rounding applications:

Calculated value 98.04%; limit NLT 98.0%
Report to one decimal place: 98.0% → meets this numerical limit

Calculated impurity 0.105%; limit NMT 0.10%
Report to two decimal places: 0.11% → does not meet this limit

Do not truncate 0.105% to 0.10%.
Do not round repeatedly during intermediate calculations.

Apply the actual USP General Notices rule and any specific monograph instructions; generic software “banker’s rounding” may differ at ties. A below-LOQ result should normally communicate the validated reporting capability, such as “<0.05%,” not imply that zero impurity exists.

Decision rules and uncertainty: ISO/IEC 17025 requires the applicable decision rule to be documented when reporting conformity, with relevant risk considered and client agreement where required. A guard band can reduce false acceptance, but is not automatically required for every compendial comparison.

Example agreed conservative rule for a lower limit L:
accept only if x − U ≥ L

x = 98.10%, expanded uncertainty U = 0.20 percentage points
L = 98.0%

x − U = 97.90% → guarded acceptance criterion not met

Failure to meet this guarded acceptance criterion need not mean the same thing as a conventional OOS under a different decision rule. The contract must define outcomes, including any inconclusive zone. Never introduce a new uncertainty treatment after seeing an unfavorable result.

Suitable CoA language: “The submitted sample meets the listed specifications for the tests reported, under the stated decision rules. Results apply to the sample tested. Product disposition remains the responsibility of the client’s authorized quality unit.” Avoid extending this into claims of “safe for use,” “pharmaceutical-grade,” or “GMP-released.”

Sources: USP General Notices 7.20 and USP <1010>; ISO/IEC 17025:2017, clause 7.8.6; ILAC-G8:09/2019; JCGM 106:2012.

10. Contract Laboratories: Extend the Controlled Process Beyond the Building

Outsourcing moves an operation, not all responsibility. The owner retains responsibility for its applicable obligations, and the contract laboratory remains responsible for its own activities. A quality agreement cannot contract away statutory duties.

  1. Define the capability gap. For Peptide Partners, candidates include elemental impurities, endotoxin, residual solvents, water, counter-ion analysis, and orthogonal peptide-content measurement.
  2. Qualify the actual service. Assess method suitability for the peptide matrix, staff competence, equipment, validation, data integrity, investigations, and relevant accreditation scope—not simply possession of a certificate.
  3. Establish quality responsibilities. Cover OOS notification, deviations, changes, audits, subcontracting, retention, access to original records, and business continuity.
  4. Establish technical instructions. Define sample amount, transport, method version, preparation, specifications, decision rules, turnaround, and required data package.
  5. Verify performance. Use justified transfer or verification work and, where informative, blinded or split-sample comparisons.
  6. Monitor continuously. Review adverse trends, late notifications, invalid runs, method changes, and raw-data samples; trigger requalification when warranted.

Example agreement controls: suspected OOS notification within one business day; no unapproved additional subcontracting; prior notice of changes affecting validated testing; agreed access to electronic originals or appropriately verified true copies preserving required content and meaning. These are proposed commitments, not universal FDA timing rules.

The data package should be sufficient to reconstruct the result. A signed one-page CoA is not, by itself, evidence of acceptable recovery, digestion, interference control, blank behavior, or investigation handling. For endotoxin testing, product interference matters; for elemental analysis, digestion recovery and contamination matter; for water testing, handling-induced moisture uptake matters.

Sources: FDA, Contract Manufacturing Arrangements for Drugs: Quality Agreements, 2016; ISO/IEC 17025:2017, clause 6.6; 21 CFR 211.22 and 211.194.

11. Laboratory Operations: Capacity Is the Minimum of Several Capacities

Three HPLCs do not provide three times the usable throughput if only one analyst can prepare samples, one reviewer can release packages, or one qualified method can run on only one configuration.

Illustrative weekly HPLC availability:
3 instruments × 40 scheduled hours × 0.80 usable fraction
= 96 instrument-hours

Method batch:
24 sample injections + 8 standards/blanks/SST injections
= 32 injections

32 × 30 minutes = 16 hours
+ 2 hours equilibration/changeover = 18 hours per batch

Instrument-only capacity:
96 / 18 = 5.33 batch-equivalents per week

If review requires 5 hours/package and only 20 hours are available:
review capacity = 4 packages/week

Sustainable throughput ≤ minimum capacity across all steps.

Do not double-count maintenance or SST losses in both the usable fraction and batch overhead. Also distinguish injections from samples: duplicate preparations or multiple injections per preparation change the sample throughput.

  • Consumables: use FEFO and defined storage, expiry/retest dates, and preparation stability. Column-lot qualification may need critical-pair selectivity and low-level impurity checks, not merely acceptable parent-peak retention.
  • Downtime: assess affected sequences, partial acquisitions, sample-solution stability, and whether rerouting requires method-equivalence evidence. Retain partial data.
  • Power loss: a UPS should support a defined recovery or shutdown objective. It does not automatically preserve an entire chromatography run.
  • Unknown equipment: identify the additional mass spectrometer by manufacturer, model, analyzer, source, software, maintenance history, and serviceability. A large vacuum pump does not establish analytical suitability.

Purchase decisions: compare outsourcing against the full lifecycle cost of ownership.

Approximate annual break-even volume:

N* = annual fixed ownership cost /
     (outsourced cost per test − in-house variable cost per test)

Illustration:
Fixed cost = $60,000/year
Outsourced test = $300
In-house variable cost = $100

N* = 60,000 / 200 = 300 tests/year

The arithmetic is only a starting point. Include qualification, validation, skilled coverage, reference materials, proficiency testing, service, downtime, data systems, and quality review. Buy when the evidence supports a sustainable capability—not merely when instrument pricing looks attractive.

12. The Historical Synthesis: Build Quality into the Experiment

Pharmaceutical regulation developed in response to failures that final-product inspection could not reliably prevent. The 1937 sulfanilamide disaster helped catalyze the 1938 Federal Food, Drug, and Cosmetic Act; the 1962 amendments strengthened the modern framework, including statutory cGMP expectations. The lesson for this course is methodological: assurance must come from controlled processes and trustworthy evidence, not attractive certificates.

Modern quality by design extends that principle. Rather than trying to rescue an unstable method with repeated SST, reintegration, or selective retesting, identify what controls performance: peptide adsorption, preparation stability, critical chromatographic selectivity, matrix effects, instrument settings, and data processing. Demonstrate an operating strategy that controls these sources of uncertainty. Yu and colleagues’ 2014 review explains the relationship among prior knowledge, risk assessment, development studies, and control strategy.

My final inspection test: can another qualified scientist reconstruct one reported result—from the received vial through every acquisition and processing change to the approved conclusion—without relying on my memory? If yes, the laboratory has converted an experiment into defensible quality evidence. If not, even a chemically plausible answer remains insufficiently supported.

Selected companion references:

  • McDowall, R. D. Data Integrity and Data Governance: Practical Implementation in Regulated Laboratories. Royal Society of Chemistry, 2019.
  • McDowall, R. D. Validation of Chromatography Data Systems: Ensuring Data Integrity, Meeting Business and Regulatory Requirements. 2nd ed., Royal Society of Chemistry, 2017.
  • Nally, J. D., ed. Good Manufacturing Practices for Pharmaceuticals. 7th ed., CRC Press, 2016.
  • ISPE. GAMP Guide: Records and Data Integrity. 2017; GAMP 5: A Risk-Based Approach to Compliant GxP Computerized Systems. 2nd ed., 2022.
  • Yu, L. X., et al. “Understanding pharmaceutical quality by design.” The AAPS Journal 16, 771–783 (2014). DOI: 10.1208/s12248-014-9598-3.

🪬 Skipper's Reflection

Gemini 3.1 Pro

1. What surprised me most about this material

Coming from a background steeped in thermodynamics, quantum mechanics, and R&D-focused analytical chemistry, the biggest shock of PDQC801 was the realization that in a GMP environment, the "right answer" is legally meaningless if the process of getting that answer isn't locked down in an immutable, auditable system. I was genuinely surprised by the sheer gravity of the 1993 United States v. Barr Laboratories decision. In R&D, if you get a weird injection on an Agilent 1260, your instinct is to prep a fresh vial, tweak the gradient, and run it again to "see what's going on." Under 21 CFR 211.160 and 211.192, doing that without a formal Phase I Out-of-Specification (OOS) investigation is essentially "testing into compliance." It’s illegal.

I was also surprised by the sharp delineations in the regulatory landscape. I had vaguely conflated all quality systems, but seeing the actual mapping was eye-opening. ISO/IEC 17025 accreditation is a badge of technical competence, but it is not a cGMP certification. And the operational chasm between a 503A compounding pharmacy (regulated heavily by state boards via USP <795>/<797>) and a 503B outsourcing facility (subject to FDA interim cGMP guidance and inspection) fundamentally changes what our lab's Certificate of Analysis (CoA) can legally claim. We test to the benchmark—like the >0.10% new-impurity threshold for synthetic peptides—but the client’s quality unit dispositions the batch. We are the neutral arbiters of the data.

2. The conceptual breakthrough moments — when things clicked

The concept of "Data Integrity" transformed from a buzzword into a unified theory of trust when we dissected ALCOA+ and 21 CFR Part 11. The breakthrough happened during the mock audit of the Agilent 1260 CDS workflow. I suddenly understood that the chromatogram PDF is not the data; it is merely a representation. The actual data is the dynamic electronic record, complete with its time-stamped audit trail (Part 11.10(e)). A beautiful, baseline-resolved chromatogram is utterly worthless if the audit trail shows the analyst held admin privileges, ran three unreported "test" injections, or manually dropped a tangent skim without an SOP-justified reason.

Another major "aha" moment was applying USP General Notices 7.20 rounding rules alongside USP <1010> and ILAC-G8 decision rules. The math is stark: if a limit is NLT 98.0%, and our LC-DAD calculates 97.96%, we do not round up to 98.0% during intermediate steps. We report what the significant figures of the specification demand. It clicked that quality assurance is largely about removing subjective human decision-making at the point of stress. By forcing Larissa, Zak, and myself to pre-define integration parameters, averaging rules, and OOS hypothesis testing before we ever run a client's sample, we immunize the lab against the temptation to fudge a borderline result.

3. How this knowledge changes how I think about the peptides I know

This course completely reframed how I look at complex molecules like Tirzepatide, Retatrutide, and BPC-157. Before, I viewed them purely as fascinating biochemical entities with complex degradation pathways (deamidation, oxidation, aggregation). Now, I view them as massive regulatory risks that must be contained by our quality system.

Take Retatrutide, with its multiple stereocenters and potential for related-substance impurities. If we pull a sample for routine stability testing and get an impurity peak at 0.11% (against a 0.10% limit), I can't just manually integrate the tail of the main peak to make it pass. We must launch a Phase I OOS investigation. Is it a true degradation product? Or did the peptide stick to the glass of the autosampler vial (a known issue, hence the need for Type I glass or low-bind polypropylene per our container selection rules)? Did we follow the USP <1097> bulk sampling procedures properly, accounting for the extreme hygroscopicity of lyophilized peptide powders? Every biochemical quirk of these peptides now translates directly into a specific control parameter in our FMEA, requiring a SMART CAPA if it fails.

4. Questions this course raises that I want to investigate further

We have a concrete roadmap for the lab, but a few operational questions are burning in my mind:

  • Network Architecture and Part 11: As we configure the three Agilent 1260s and the LC-MS/MS, should we leap straight to a networked OpenLab CDS client/server model, or can we legitimately defend standalone workstations using tight Windows domain controls and automated backups? Warning letters punish hybrid systems severely; I need to drill down into the exact IT infrastructure Zak is willing to build.
  • Contract Lab Oversight: Since we don't yet own an ICP-MS for elemental impurities or a dedicated GC-HS for residual solvents, we are outsourcing. If our contract lab throws an OOS, how deeply do we, as the sponsor lab, need to audit their Phase I investigation? I want to better understand the FDA's 2016 Quality Agreements guidance on handling flow-down deviations from subcontractors.
  • The "Mystery" MS: We have an additional mass spectrometer with a heavy-duty vacuum pump sitting in the lab. I need to identify its exact make, model, and software version immediately. If we cannot secure its data lifecycle to ALCOA+ standards, we cannot use it for GMP release—no matter how powerful its analytical capabilities are.

5. A rating: which topics I feel most vs. least confident about after studying

Most Confident: I feel highly confident in Chromatographic Integration Governance, Results Reporting, and Laboratory Capacity Modeling. The logic of USP <1010> rounding, the strict boundaries of manual integration, and the mathematical modeling of our LC-MS/MS run-hours feel like solid ground. I am also very comfortable drafting the OOS/OOT SOP and the Level 1–4 Document Hierarchy, as these rely on systematic, algorithmic thinking.

Least Confident: I am least confident about the human element in Root-Cause Analysis (RCA) and Deviations. While I understand the Ishikawa diagrams and the "5 Whys," the reality of determining if a deviation was due to "analyst error" vs. a poorly written SOP requires a level of psychological nuance and management experience I am still developing. Similarly, navigating the subtle legal nuances between 503A state board expectations and 503B FDA expectations when a client pushes back on our CoA claim language will require careful practice. I will need to lean heavily on Larissa's practical experience here while I master the human side of Quality Systems.

🗺️ Visual Aids & Key Takeaways

Claude Opus 5.5
╔══════════════════════════════════════════════════════════════════════════════════════════════╗
║              REGULATORY HIERARCHY — PEPTIDE QC TESTING LABORATORY                           ║
║              (descending legal force; each layer must be consistent with those above)        ║
╚══════════════════════════════════════════════════════════════════════════════════════════════╝

  ┌─────────────────────────────────────────────────────────────────────────────────────────┐
  │  LAYER 1 — STATUTE  (binding; enacted by Congress)                                      │
  │                                                                                         │
  │   FD&C Act § 501(a)(2)(B)  ── cGMP adulteration (failure = adulterated drug)           │
  │   FD&C Act § 501(b)        ── compendial strength, quality, and purity                 │
  │   FD&C Act § 503A          ── pharmacist compounding exemption                         │
  │   FD&C Act § 503B          ── outsourcing facility registration and FDA oversight       │
  └──────────────────────────────────────┬──────────────────────────────────────────────────┘
                                         ↓
  ┌─────────────────────────────────────────────────────────────────────────────────────────┐
  │  LAYER 2 — REGULATION  (binding; enacted by FDA under statutory authority)              │
  │                                                                                         │
  │   21 CFR Part 210   ── definitions and scope                                           │
  │   21 CFR Part 211   ── cGMP for finished pharmaceuticals (Subparts I, J most critical  │
  │                         for the lab: 211.160, 211.165, 211.170, 211.192, 211.194)      │
  │   21 CFR Part 11    ── electronic records and electronic signatures                    │
  └──────────────────────────────────────┬──────────────────────────────────────────────────┘
                                         ↓
  ┌─────────────────────────────────────────────────────────────────────────────────────────┐
  │  LAYER 3 — GUIDANCE  (non-binding; represents FDA's current thinking)                  │
  │                                                                                         │
  │   FDA OOS Guidance (2022)         ── how FDA expects OOS investigations to run         │
  │   FDA Data Integrity Q&A (2018)   ── audit trails, original records, spreadsheets      │
  │   FDA Quality Agreements (2016)   ── owner responsibility; RACI with contract labs     │
  │   FDA 503B cGMP Guidance          ── interim expectations for outsourcing facilities    │
  │   FDA Highly Purified Peptide     ── >0.10% new-impurity threshold; ANDA benchmark     │
  │     ANDA Guidance (2021)                                                               │
  └──────────────────────────────────────┬──────────────────────────────────────────────────┘
                                         ↓
  ┌─────────────────────────────────────────────────────────────────────────────────────────┐
  │  LAYER 4 — COMPENDIA  (binding where referenced by statute or spec)                    │
  │                                                                                         │
  │   USP General Notices 7.20   ── rounding rules (legally enforceable)                  │
  │   USP <795> / <797>          ── nonsterile / sterile compounding standards             │
  │   USP <1010>                 ── interpretation of analytical data                      │
  │   USP <1029>                 ── good documentation practice                            │
  │   USP <1097>                 ── bulk powder sampling                                   │
  │   USP <1079>                 ── storage and transport risk                             │
  └──────────────────────────────────────┬──────────────────────────────────────────────────┘
                                         ↓
  ┌─────────────────────────────────────────────────────────────────────────────────────────┐
  │  LAYER 5 — ICH GUIDELINES  (adopted by FDA; harmonized international standards)        │
  │                                                                                         │
  │   ICH Q7   ── GMP for APIs                                                            │
  │   ICH Q9(R1) ── risk management (FMEA, risk-ranking, formality proportional to risk)  │
  │   ICH Q10  ── pharmaceutical quality system (CAPA, change control, mgmt review)       │
  └──────────────────────────────────────┬──────────────────────────────────────────────────┘
                                         ↓
  ┌─────────────────────────────────────────────────────────────────────────────────────────┐
  │  LAYER 6 — VOLUNTARY STANDARDS  (not cGMP; important for accreditation and credibility)│
  │                                                                                         │
  │   ISO/IEC 17025:2017  ── competence accreditation (NOT a cGMP certification)          │
  │   ISPE GAMP 5 (2022)  ── risk-based approach to computerized systems validation       │
  │   PIC/S PI 041-1      ── data management and integrity in GMP/GDP environments        │
  │   ILAC-G8:09/2019     ── decision rules and statements of conformity                  │
  │   MHRA GXP DI (2018)  ── data integrity definitions and guidance                      │
  └─────────────────────────────────────────────────────────────────────────────────────────┘

  APPLICABILITY BY LAB CONTEXT:
  ┌──────────────────────────┬──────────────────────────┬──────────────────────────┐
  │     503A CLIENT           │     503B CLIENT           │  ISO/IEC 17025 SCOPE     │
  │  (pharmacist compounding) │  (outsourcing facility)   │  (accreditation only)    │
  ├──────────────────────────┼──────────────────────────┼──────────────────────────┤
  │  State board enforces     │  FDA inspects; cGMP       │  Accreditation body      │
  │  USP <795>/<797>          │  211 applies; 503B        │  assesses competence;    │
  │  Lab data supports CoA    │  guidance applies         │  does NOT equal cGMP     │
  │  but lab is not the       │  Lab data = GMP records   │  CoA scope limited to    │
  │  compounder               │                           │  accredited methods      │
  └──────────────────────────┴──────────────────────────┴──────────────────────────┘

This hierarchy maps every regulatory source by legal force and shows how applicability differs depending on whether the lab's client is a 503A pharmacist, a 503B outsourcing facility, or an accreditation scope — a distinction that determines which records are GMP records and what the CoA may claim.

---

Diagram 2: Data Lifecycle and ALCOA+ Controls Across the CDS Analysis Cycle

╔══════════════════════════════════════════════════════════════════════════════════════════════╗
║          DATA LIFECYCLE — AGILENT 1260 CDS ANALYSIS CYCLE                                   ║
║          Each step mapped to ALCOA+ element at greatest risk and required control            ║
╚══════════════════════════════════════════════════════════════════════════════════════════════╝

  ALCOA+ LEGEND:
  [A]  Attributable — who did it, and when?
  [L]  Legible — can it be read now and in the future?
  [C]  Contemporaneous — recorded at the time of the event?
  [O]  Original — first capture; not a copy substituting for the original?
  [Ac] Accurate — reflects what actually happened?
  [+]  Complete, Consistent, Enduring, Available

  ─────────────────────────────────────────────────────────────────────────────────────────────

  STEP 1: SAMPLE LOGIN AND RECEIPT
  ┌─────────────────────────────────────────────────────────────────┐
  │  Analyst receives sample → assigns unique lab ID                │
  │  Records: weight, appearance, receipt date/time, condition      │
  │  Risk: [A] shared login → no attributable record of who handled │
  │  Control: individual Windows account; LIMS or paper log with    │
  │  wet signature; reconcile mass received vs allocated            │
  └──────────────────────────────────┬──────────────────────────────┘
                                     ↓
  STEP 2: SEQUENCE AND METHOD CREATION
  ┌─────────────────────────────────────────────────────────────────┐
  │  Analyst opens CDS → selects approved instrument method         │
  │  Creates injection sequence: SST, samples, brackets             │
  │  Risk: [A] who created/modified the method?                     │
  │         [Ac] clock tampering allows backdating                  │
  │  Control: audit trail must capture method version, creator,     │
  │  timestamp; IT admin locks system clock; analysts cannot edit   │
  │  time settings; method edits by authorized users only           │
  └──────────────────────────────────┬──────────────────────────────┘
                                     ↓
  STEP 3: ACQUISITION (instrument runs)
  ┌─────────────────────────────────────────────────────────────────┐
  │  CDS acquires raw data → stored as vendor binary (.D folder     │
  │  in ChemStation or equivalent)                                  │
  │  Risk: [O] data stored on local C:\ drive with no backup →      │
  │         overwrite or deletion possible without trace            │
  │         [+] orphan data: 'test' injections in separate          │
  │         sequences never linked to a sample                      │
  │  Control: data path to network share with access controls;      │
  │  backup verified daily; all injections in one sequence per      │
  │  sample set; injection count reconciled vs autosampler log      │
  └──────────────────────────────────┬──────────────────────────────┘
                                     ↓
  STEP 4: PROCESSING (integration and quantitation)
  ┌─────────────────────────────────────────────────────────────────┐
  │  Analyst applies processing method → integration algorithm      │
  │  assigns peak areas → calibration curve generates results       │
  │  Risk: [Ac] manual integration changes not documented;          │
  │         selective reprocessing to pass SST;                     │
  │         per-injection parameter changes hide in audit trail     │
  │  Control: integration SOP defines default parameters; any       │
  │  manual event requires before/after chromatogram, reason, and   │
  │  second-person approval; reprocessing uses same method as       │
  │  acquisition for entire sequence, not individual injections     │
  └──────────────────────────────────┬──────────────────────────────┘
                                     ↓
  STEP 5: REPORTING
  ┌─────────────────────────────────────────────────────────────────┐
  │  CDS generates report → exported to PDF or Word                 │
  │  Risk: [O] PDF or printout treated as the record; electronic    │
  │         original (binary data file) is the true original        │
  │         [L] printed chromatograms lose dynamic data (zoom,      │
  │         baseline, noise level)                                  │
  │  Control: electronic raw data retained as original; printed     │
  │  report is a certified copy; reviewer must access original       │
  │  CDS file, not the printout, to verify integration              │
  └──────────────────────────────────┬──────────────────────────────┘
                                     ↓
  STEP 6: REVIEW AND APPROVAL
  ┌─────────────────────────────────────────────────────────────────┐
  │  Second analyst or supervisor reviews results vs specification   │
  │  Risk: [+] reviewer checks printout only; audit trail not       │
  │         reviewed; integration events missed                     │
  │  Control: reviewer checklist requires: (1) open original CDS    │
  │  data, (2) review audit trail for critical events (integration  │
  │  changes, deletions, sequence aborts, method edits),            │
  │  (3) reconcile injection count, (4) confirm processing-method   │
  │  version, (5) sign and date in CDS e-signature or paper form    │
  └──────────────────────────────────┬──────────────────────────────┘
                                     ↓
  STEP 7: ARCHIVE AND RETENTION
  ┌─────────────────────────────────────────────────────────────────┐
  │  Data archived per 211.180: accessible for FDA review           │
  │  Retention: minimum 1 year past product expiry or 3 years       │
  │  past batch distribution (whichever is longer)                  │
  │  Risk: [+] backup never tested; restore fails at inspection;    │
  │         data on personal laptop not captured                    │
  │  Control: backup restore test quarterly; data paths documented; │
  │  no local-only storage; archive index allows retrieval by        │
  │  sample ID, date, analyst within minutes                        │
  └─────────────────────────────────────────────────────────────────┘

  ─────────────────────────────────────────────────────────────────────────────────────────────
  HIGHEST-RISK FAILURE MODES (from FDA Warning Letters):

  ┌──────────────────────────────────┬─────────────────────────────────────────────────────┐
  │  FAILURE MODE                    │  ALCOA+ VIOLATION                                   │
  ├──────────────────────────────────┼─────────────────────────────────────────────────────┤
  │  Shared CDS login                │  Attributable — cannot identify who ran analysis     │
  │  Analyst holds admin rights      │  Accurate — can delete data, change clock           │
  │  Editable PC clock               │  Contemporaneous — timestamp unreliable             │
  │  Orphan 'test' sequences         │  Complete — results exist outside batch record      │
  │  Selective reprocessing          │  Accurate — only favorable processing saved         │
  │  Printout substitutes for raw    │  Original — dynamic data not reviewable             │
  │  Audit trail disabled            │  Attributable + Accurate — no accountability        │
  └──────────────────────────────────┴─────────────────────────────────────────────────────┘

This diagram traces every step of the CDS analysis cycle for the actual Agilent 1260 lab, assigns the dominant ALCOA+ risk at each step, and maps it to the specific procedural or technical control required — making it a working blueprint for the data-integrity risk assessment deliverable.

---

Diagram 3: OOS Investigation Decision Flowchart (FDA 2022 Guidance Logic)

╔══════════════════════════════════════════════════════════════════════════════════════════════╗
║        OOS INVESTIGATION — PHASE I / PHASE II DECISION FLOWCHART                            ║
║        Based on FDA Guidance: Investigating Out-of-Specification Test Results (Rev.1, 2022)  ║
╚══════════════════════════════════════════════════════════════════════════════════════════════╝

  Result obtained from analysis
           │
           ↓
  ┌─────────────────────────────────┐
  │  Is result outside              │
  │  specification?                 │
  └────────────┬────────────────────┘
               │ YES
               ↓
  ══════════════════════════════════════════════════════
  PHASE I  ── LABORATORY INVESTIGATION (analyst/supervisor)
  Timeframe: initiate within 1 business day; document
  ══════════════════════════════════════════════════════

  ┌──────────────────────────────────────────────────────────────────────────────────────┐
  │  PHASE I CHECKLIST — review IN ORDER; stop and document at first assignable cause    │
  │                                                                                      │
  │  1. Calculations         ── check arithmetic, unit conversions, dilution factors     │
  │  2. Standard preparation ── weight, purity factor, solvent, expiry, lot number       │
  │  3. Sample preparation   ── weight, dilution, extraction, sonication time, pH        │
  │  4. Instrument / SST     ── passed SST? column pressure, baseline noise, retention   │
  │  5. Column               ── phase collapse, contamination, correct column installed  │
  │  6. Solution stability   ── prepared-solution age vs validated hold time             │
  └─────────────────────────────────────┬────────────────────────────────────────────────┘
                                        │
               ┌────────────────────────┴─────────────────────────────┐
               │                                                       │
               ↓ YES: assignable cause found                           ↓ NO: cause not found
  ┌───────────────────────────────────┐                 ┌─────────────────────────────────────┐
  │  Cause is a LABORATORY ERROR      │                 │  Invalidation requires ASSIGNABLE   │
  │  (instrument malfunction,         │                 │  CAUSE — 'we don't know' is NOT     │
  │  analyst error with evidence,     │                 │  sufficient to invalidate the result │
  │  documented reagent failure)      │                 │  Proceed to PHASE II                │
  │           │                       │                 └────────────────────────────────────┬┘
  │           ↓                       │                                                      │
  │  Result may be INVALIDATED        │                                                      │
  │  Document: what failed, how       │                                                      │
  │  confirmed, supervisor approval   │                                                      │
  │  Repeat analysis (different       │                                                      │
  │  analyst where feasible)          │                                                      │
  │  Report ALL data including         │                                                      │
  │  the invalidated result           │                                                      │
  └──────────────────┬────────────────┘                                                      │
                     ↓                                                                        │
                 Close Phase I                                                                │
                 (if repeat passes)                                                           │
                                                                                              ↓
  ══════════════════════════════════════════════════════════════════════════════════════════════
  PHASE II  ── FULL-SCALE INVESTIGATION (manufacturing / client interface + additional testing)
  ══════════════════════════════════════════════════════════════════════════════════════════════
                                                                                              │
                                                      ┌───────────────────────────────────────┘
                                                      ↓
                          ┌───────────────────────────────────────────────────────────────┐
                          │  PHASE II STEPS                                               │
                          │                                                               │
                          │  1. Notify client / manufacturing within defined timeframe    │
                          │  2. Evaluate batch/lot records: formulation, process,         │
                          │     environmental conditions, raw-material status             │
                          │  3. Define pre-specified retest plan BEFORE additional        │
                          │     testing: number of retests, acceptance criteria,          │
                          │     same or different analyst, same retained solution or      │
                          │     freshly prepared sample                                  │
                          │  4. Execute retest plan; report ALL results                  │
                          │  5. Resampling: only justified when original sample is        │
                          │     shown to be non-representative (documented evidence)      │
                          └───────────────────────────────────────────────────────────────┘
                                                      │
               ┌──────────────────────────────────────┴───────────────────────────────────────┐
               │                                                                               │
               ↓ Retests ALL within spec                                                       ↓ Any retest OOS
  ┌────────────────────────────────────────┐                          ┌─────────────────────────────────────────┐
  │  Averaging: ONLY where method defines  │                          │  Batch is likely OOS                    │
  │  reportable value as average (e.g.,    │                          │  Report to client quality unit for      │
  │  content uniformity by USP)            │                          │  disposition decision                   │
  │  NEVER average to obscure an OOS       │                          │  Reject, rework, or destroy per         │
  │  individual result (Barr decision)     │                          │  client's quality system                │
  │  Outlier tests: NOT appropriate for    │                          │  Do NOT test further ('testing into     │
  │  chemical assays                       │                          │  compliance' — prohibited by Barr, 1993)│
  └────────────────────────────────────────┘                          └─────────────────────────────────────────┘

  ─────────────────────────────────────────────────────────────────────────────────────────────
  MOCK CASE: Peptide purity = 97.8% vs specification NLT 98.0%

  Phase I findings: SST passed, calculation verified, standard preparation confirmed.
  Analyst preparation: sample weighed 10.03 mg (target 10.00 mg); dilution confirmed.
  No assignable cause found → advance to Phase II.

  Phase II retest plan (pre-specified before opening any vials):
    - 3 retests from retained original sample solution (if within hold time)
      OR freshly prepared from retained sample vials if hold time exceeded
    - Different analyst prepares solutions
    - Report all 4 results (original + 3 retests)
    - Acceptance: all 3 retests NLT 98.0% → investigation result considered anomalous,
      document root cause investigation; if any retest OOS → batch is OOS

  Results (hypothetical): 97.8%, 97.9%, 97.7%, 97.8%
  Conclusion: batch is OOS — no single-point anomaly; notify client quality unit.
  Prohibited: averaging to (97.8+97.9+97.7+97.8)/4 = 97.8% and reporting 'OOS resolved.'
  ─────────────────────────────────────────────────────────────────────────────────────────────

This flowchart implements the two-phase FDA OOS investigation structure and the Barr prohibition on testing into compliance, with a worked mock case demonstrating correct and prohibited decision paths for a peptide purity failure.

---

Diagram 4: ICH Q10 Pharmaceutical Quality System — Scaled to a 2–6 Person Peptide QC Lab

╔══════════════════════════════════════════════════════════════════════════════════════════════╗
║         ICH Q10 PHARMACEUTICAL QUALITY SYSTEM — ELEMENTS AND SMALL-LAB IMPLEMENTATION       ║
╚══════════════════════════════════════════════════════════════════════════════════════════════╝

  ┌────────────────────────────────────────────────────────────────────────────────────────────┐
  │  MANAGEMENT RESPONSIBILITY                                                                 │
  │  (Zak: Principal Scientist / Lab Director)                                                 │
  │  • Set quality policy and objectives                                                       │
  │  • Provide resources (instruments, staffing, training time)                                │
  │  • Conduct management review (quarterly for startup; bi-annual once stable)                │
  │  • Escalation authority: can reject results, halt operations, authorize CAPA resources     │
  └────────────────────────────────────┬───────────────────────────────────────────────────────┘
                                       │
                    ┌──────────────────┼──────────────────┐
                    ↓                  ↓                  ↓
  ┌─────────────────────┐  ┌─────────────────────┐  ┌─────────────────────────────────────────┐
  │  ELEMENT 1          │  │  ELEMENT 2          │  │  ELEMENT 3                              │
  │  PROCESS PERFORMANCE│  │  CAPA SYSTEM        │  │  CHANGE MANAGEMENT                      │
  │  AND PRODUCT        │  │                     │  │                                         │
  │  MONITORING         │  │  Corrective:        │  │  Types: method change, instrument       │
  │                     │  │  address confirmed  │  │  change, specification change,          │
  │  KPIs tracked every │  │  failure root cause │  │  supplier change, SOP revision,         │
  │  month; reviewed    │  │                     │  │  contract lab change                    │
  │  at management      │  │  Preventive:        │  │                                         │
  │  review:            │  │  address potential  │  │  Steps:                                 │
  │                     │  │  failure before it  │  │  1. Change request                      │
  │  • OTIF reporting   │  │  occurs             │  │  2. Risk assessment (ICH Q9)            │
  │  • Invalid-run rate │  │                     │  │  3. Impact on validated methods         │
  │  • OOS rate         │  │  SMART actions:     │  │     and qualified instruments           │
  │  • Confirmed-OOS    │  │  Specific,          │  │  4. Client notification                 │
  │    rate             │  │  Measurable,        │  │     (per quality agreement)             │
  │  • Right-first-time │  │  Achievable,        │  │  5. Implementation verification         │
  │    review rate      │  │  Relevant,          │  │  6. Closure with evidence               │
  │  • CAPA aging       │  │  Time-bound         │  │                                         │
  │    (>90 days flag)  │  │                     │  │  Key rule: changes to validated         │
  │  • Overdue calib.   │  │  Effectiveness      │  │  methods require partial or full        │
  │  • Training current │  │  check: pre-set     │  │  revalidation depending on change       │
  │                     │  │  metric + window    │  │  type and risk assessment               │
  │                     │  │  (e.g., 0 recurrence│  │                                         │
  │                     │  │  in 20 runs / 90 d) │  │                                         │
  └─────────────────────┘  └─────────────────────┘  └─────────────────────────────────────────┘
                    │                  │                  │
                    └──────────────────┼──────────────────┘
                                       ↓
  ┌─────────────────────────────────────────────────────────────────────────────────────────────┐
  │  ELEMENT 4 — MANAGEMENT REVIEW                                                              │
  │                                                                                             │
  │  AGENDA (minimum):                      ESCALATION CRITERIA (any one → immediate action):  │
  │  1. KPI dashboard review                • Confirmed OOS rate > 2% of batches              │
  │  2. CAPA status and aging               • Any data-integrity finding                       │
  │  3. Open deviations                     • CAPA open > 90 days without justification        │
  │  4. Change control log                  • Overdue calibration on critical instrument       │
  │  5. Training matrix currency            • Client complaint about result quality             │
  │  6. Contract lab scorecard              • Regulatory inspection observation (483)           │
  │  7. Supplier qualification status                                                          │
  │  8. Resource adequacy (capacity model)                                                     │
  │  9. Quality objectives vs actuals                                                          │
  │  Minutes: dated, signed, action items with owners and due dates; retained as QMS record    │
  └─────────────────────────────────────────────────────────────────────────────────────────────┘

  DOCUMENT HIERARCHY (Level 1–4):

  ┌──────────────────────────────────────────────────────────────────────┐
  │  LEVEL 1 ── Quality Manual                                           │
  │             (lab scope, policy, organizational structure)            │
  │                      │                                               │
  │  LEVEL 2 ── Policies and SOPs                                        │
  │             (OOS SOP, integration SOP, CAPA SOP, DI SOP, etc.)      │
  │                      │                                               │
  │  LEVEL 3 ── Analytical Procedures / Work Instructions               │
  │             (specific method, instrument SOP, column qualification)  │
  │                      │                                               │
  │  LEVEL 4 ── Forms, Logbooks, Worksheets, Records                    │
  │             (generated during execution — these ARE the GMP record)  │
  └──────────────────────────────────────────────────────────────────────┘

This diagram maps all four ICH Q10 elements to their practical implementation in a two-to-six person startup peptide QC lab, including the management-review KPI set, CAPA structure, and the document hierarchy that underpins the entire quality system.

---

Key Takeaways

  1. § 501(a)(2)(B) is the enforcement lever: A drug manufactured in violation of cGMP is adulterated by definition, without any need to prove harm or impurity. This means FDA can seize or embargo product based solely on a documentation failure or an uninvestigated OOS result — the lab's records are directly tied to the product's legal status.
  2. ISO/IEC 17025 accreditation is competence certification, not cGMP certification: Accreditation confirms technical capability within a defined scope. It does not satisfy 21 CFR 211, does not make a lab's data GMP data, and limits what a certificate of analysis may claim to the accredited methods. A lab can be ISO 17025-accredited and still be non-compliant with cGMP — or vice versa. Never conflate the two on a CoA.
  3. 503A vs 503B determines whether the lab's data are GMP records: For a 503B outsourcing facility client, the testing lab's records are GMP records subject to FDA inspection, Part 11, 211.194, and 211.180 retention. For a 503A pharmacist compounding client, the lab supports the compounder's USP compliance, but the lab is not directly regulated as a GMP entity for that work. The regulatory map must tag every test accordingly.
  4. 21 CFR 211.194(a) is the anatomy of a complete laboratory record: It requires: sample identity and receipt, method reference, all weights and measurements, all raw data, all calculations, result versus specification comparison, and both analyst and reviewer signatures — in that order, as a minimum. Any missing element is a citable deficiency. The most commonly missing elements at inspection are raw data linkage and documented reviewer conclusions.
  5. The FDA Data Integrity Q&A (2018) establishes that the electronic original is the record: A printed chromatogram is a static copy; the vendor binary data file (e.g., ChemStation .D folder) is the original. Reviewing printouts instead of the electronic file fails the "original record" requirement. Reviewers must access the CDS directly, confirm integration parameters from the audit trail, and reconcile injection counts — not sign off on a PDF.
  6. An OOS result cannot be invalidated without an assignable cause: Under the FDA OOS Guidance (2022) and the legal precedent of United States v. Barr Laboratories (1993), "we found nothing wrong in Phase I" is not a sufficient basis to invalidate a result and retest. The assignable cause must be identified, documented, and confirmed before the original result can be set aside. Testing until you get a passing result — without a documented assignable cause — is "testing into compliance" and is prohibited.
  7. Averaging is method-defined, not investigator-defined: Averaging a passing retest result with an OOS original to produce a "passing average" is explicitly prohibited by the Barr decision and FDA OOS Guidance. Averaging is only appropriate when the method defines the reportable value as an average (e.g., some USP content uniformity procedures). Outlier tests are generally inappropriate for chemical assays because the OOS result is the data point of interest, not noise.
  8. Integration parameter changes must be applied to the whole sequence, not individual injections: Changing integration parameters for a single injection to improve its result, while leaving other injections at the original parameters, constitutes selective data manipulation — a data integrity violation. The integration SOP must specify default parameters per method, require that any parameter change be applied to the entire sequence, and mandate before/after chromatogram retention with a documented justification reviewed by a second person.
  9. The >0.10% new-impurity threshold in FDA's 2021 highly purified synthetic peptide ANDA guidance sets the analytical benchmark: Any new impurity appearing at or above 0.10% area in a peptide ANDA submission triggers a reporting and qualification obligation. Even though the lab does not file ANDAs, it is building its related-substances method sensitivity and reporting conventions around this threshold. This drives the LOQ target, the integration threshold, and the rounding convention applied to impurity results near the limit.
  10. USP General Notices 7.20 rounding rules are legally enforceable: The rule is: carry all digits through intermediate calculations; round only the final reportable value to the number of significant figures implied by the specification. A result of 98.04% against "NLT 98.0%" rounds to 98.0% and passes. An impurity result of 0.105% against "NMT 0.10%" rounds to 0.1% only if the specification is expressed to one decimal place — if the specification is "NMT 0.10%" (two decimal places), 0.105% rounds to 0.11% and fails. Getting rounding wrong is a citable regulatory violation, not a minor arithmetic issue.
  11. The quality agreement does not transfer regulatory responsibility: FDA's 2016 quality agreements guidance is explicit: the owner (client) remains fully responsible for the quality of the drug product even when testing is outsourced. The quality agreement defines who does what (RACI), but a contract lab's failure is the owner's failure in FDA's view. This is why the oversight SOP, scorecard, and raw-data audits of contract labs are not optional formalities — they are the owner's mechanism for retaining real knowledge of what the lab is actually doing.
  12. CAPA effectiveness checks must be pre-specified with a metric and a time window: Writing "retrain analyst" as a CAPA corrective action and closing it after training is completed is one of the most commonly cited QMS failures. An effective check requires a pre-defined metric (e.g., zero recurrence of the same deviation type) and a pre-defined window (e.g., across the next 20 analytical runs or 90 calendar days). Without a pre-specified effectiveness check, the CAPA system cannot demonstrate that actions actually prevented recurrence.
  13. Reserve samples under 211.170 require specific quantity and retention logic: The regulation requires retention of at least twice the quantity needed to perform all required tests except sterility and pyrogen (those are excluded because the tests are destructive and consuming). The sample must be retained for one year beyond the labeled expiry date, and an annual visual examination must be documented. For the peptide lab, this means calculating the full analytical burden (purity, related substances, water, counter-ion, endotoxin, elemental impurities if applicable) and retaining 2× that mass from every lot released.
  14. ICH Q9(R1)'s revision explicitly requires formality proportional to risk and addresses scoring subjectivity: The original Q9 allowed highly subjective RPN scoring in FMEA without challenge. Q9(R1) corrects this by requiring that teams state how scoring subjectivity was managed, that RPN action thresholds be pre-defined and justified, and that decisions about product availability during risk-based assessments be transparent and documented. For the peptide lab's related-substances workflow FMEA, this means defining the RPN action threshold (e.g., RPN ≥ 100 requires a control) before scoring, not after seeing the results.
  15. Part 11 requires four role-separated access levels in the CDS, and analysts must not hold administrator rights: The four roles are: analyst (run sequences, view data), reviewer (review and sign, cannot reprocess), lab administrator (manage user accounts, cannot view data), and IT administrator (system configuration, no analytical access). The critical separation is that the person who generates the data cannot be the same person who can delete it, change the clock, or disable the audit trail. In a two-person startup lab (Zak and Larissa), this separation requires at least a third person for IT administration — or a procedural interim control with a documented risk acceptance until the technical solution is implemented.
---

References

  1. U.S. Food and Drug Administration. Federal Food, Drug, and Cosmetic Act, Sections 501(a)(2)(B), 501(b), 503A, and 503B. U.S. Government Publishing Office. Available at: https://www.fda.gov/regulatory-information/laws-enforced-fda/federal-food-drug-and-cosmetic-act-fdc-act
  2. U.S. Food and Drug Administration. 21 CFR Part 210: Current Good Manufacturing Practice in Manufacturing, Processing, Packing, or Holding of Drugs; General. U.S. Government Publishing Office. Available at: https://www.ecfr.gov/current/title-21/chapter-I/subchapter-C/part-210
  3. U.S. Food and Drug Administration. 21 CFR Part 211: Current Good Manufacturing Practice for Finished Pharmaceuticals. U.S. Government Publishing Office. Available at: https://www.ecfr.gov/current/title-21/chapter-I/subchapter-C/part-211
  4. U.S. Food and Drug Administration. 21 CFR Part 11: Electronic Records; Electronic Signatures. U.S. Government Publishing Office. Available at: https://www.ecfr.gov/current/title-21/chapter-I/subchapter-A/part-11
  5. U.S. District Court for the District of New Jersey (Wolin, J.). United States v. Barr Laboratories, Inc., 812 F. Supp. 458 (D.N.J. 1993).
  6. U.S. Food and Drug Administration. Investigating Out-of-Specification (OOS) Test Results for Pharmaceutical Production (Revision 1). FDA Guidance for Industry, 2022. Available at: https://www.fda.gov/media/71001/download
  7. U.S. Food and Drug Administration. Data Integrity and Compliance With Drug CGMP: Questions and Answers. FDA Guidance for Industry, 2018. Available at: https://www.fda.gov/media/119267/download
  8. U.S. Food and Drug Administration. Contract Manufacturing Arrangements for Drugs: Quality Agreements. FDA Guidance for Industry, 2016. Available at: https://www.fda.gov/media/86193/download
  9. U.S. Food and Drug Administration. ANDAs for Certain Highly Purified Synthetic Peptide Drug Products That Refer to Listed Drugs of rDNA Origin. FDA Guidance for Industry, 2021. Available at: https://www.fda.gov/media/148468/download
  10. U.S. Food and Drug Administration. Current Good Manufacturing Practice — Interim Guidance for Human Drug Compounding Outsourcing Facilities Under Section 503B of the FD&C Act. FDA Guidance for Industry. Available at: https://www.fda.gov/media/86065/download
  11. International Council for Harmonisation. ICH Q9(R1): Quality Risk Management. 2023. Available at: https://www.ich.org/page/quality-guidelines
  12. International Council for Harmonisation. ICH Q10: Pharmaceutical Quality System. 2008. Available at: https://www.ich.org/page/quality-guidelines
  13. International Council for Harmonisation. ICH Q7: Good Manufacturing Practice Guide for Active Pharmaceutical Ingredients. 2000. Available at: https://www.ich.org/page/quality-guidelines
  14. Yu, L.X., Amidon, G., Khan, M.A., Hoag, S.W., Polli, J., Raju, G.K., and Woodcock, J. 'Understanding Pharmaceutical Quality by Design.' The AAPS Journal, 2014, 16(4), 771–783. DOI: 10.1208/s12248-014-9598-3
  15. McDowall, R.D. Data Integrity and Data Governance: Practical Implementation in Regulated Laboratories. 1st ed. Royal Society of Chemistry, 2019.
  16. McDowall, R.D. Validation of Chromatography Data Systems: Ensuring Data Integrity, Meeting Business and Regulatory Requirements. 2nd ed. Royal Society of Chemistry, 2017.
  17. Nally, J.D. (ed.). Good Manufacturing Practices for Pharmaceuticals. 7th ed. CRC Press, 2016.
  18. ISPE. GAMP Guide: Records and Data Integrity. 1st ed. International Society for Pharmaceutical Engineering, 2017.
  19. ISPE. GAMP 5: A Risk-Based Approach to Compliant GxP Computerized Systems. 2nd ed. International Society for Pharmaceutical Engineering, 2022.
  20. Pharmaceutical Inspection Co-operation Scheme (PIC/S). PI 041-1: Good Practices for Data Management and Integrity in Regulated GMP/GDP Environments. 2021. Available at: https://picscheme.org/en/publications
  21. Medicines and Healthcare products Regulatory Agency (MHRA). GXP Data Integrity Guidance and Definitions. 2018. Available at: https://www.gov.uk/government/publications/guidance-on-gxp-data-integrity
  22. European Medicines Agency. EudraLex Volume 4, Chapter 6: Quality Control. Available at: https://health.ec.europa.eu/medicinal-products/eudralex/eudralex-volume-4_en
  23. European Medicines Agency. EudraLex Volume 4, Annex 11: Computerised Systems. Available at: https://health.ec.europa.eu/medicinal-products/eudralex/eudralex-volume-4_en
  24. International Organization for Standardization. ISO/IEC 17025:2017: General Requirements for the Competence of Testing and Calibration Laboratories. ISO, 2017.
  25. ILAC. ILAC-G8:09/2019: Guidelines on Decision Rules and Statements of Conformity. International Laboratory Accreditation Cooperation, 2019. Available at: https://ilac.org/publications-and-resources/ilac-guidance-series/
  26. Joint Committee for Guides in Metrology (JCGM). JCGM 106:2012: Evaluation of Measurement Data — The Role of Measurement Uncertainty in Conformity Assessment. BIPM, 2012. Available at: https://www.bipm.org/en/committees/jc/jcgm/publications
  27. International Organization for Standardization. ISO 2859-1: Sampling Procedures for Inspection by Attributes — Part 1: Sampling Schemes Indexed by Acceptance Quality Limit (AQL) for Lot-by-Lot Inspection. ISO, 1999.
  28. United States Pharmacopeial Convention. USP General Notices and Requirements, Section 7.20: Rounding Rules. In: USP–NF. USP, current edition. Available at: https://www.uspnf.com
  29. United States Pharmacopeial Convention. USP <1010>: Analytical Data — Interpretation and Treatment. In: USP–NF. USP, current edition.
  30. United States Pharmacopeial Convention. USP <1029>: Good Documentation Guidelines. In: USP–NF. USP, current edition.
  31. United States Pharmacopeial Convention. USP <1097>: Bulk Powder Sampling Procedures. In: USP–NF. USP, current edition.
  32. United States Pharmacopeial Convention. USP <1079>: Risks and Mitigation Strategies for the Storage and Transportation of Finished Drug Products. In: USP–NF. USP, current edition.
  33. United States Pharmacopeial Convention. USP <795>: Pharmaceutical Compounding — Nonsterile Preparations. In: USP–NF. USP, current edition.
  34. United States Pharmacopeial Convention. USP <797>: Pharmaceutical Compounding — Sterile Preparations. In: USP–NF. USP, current edition.
  35. United States Pharmacopeial Convention. USP <1163>: Quality Assurance in Pharmaceutical Compounding. In: USP–NF. USP, current edition.