📖 Core Concepts
PDQC814: Computerized System Validation, Data Governance, and LIMS/ELN
Core Concepts
0. What This Subject Is Really About
Every course in the PDQC sequence so far has improved the measurement. PDQC803 built a stability-indicating HPLC method. PDQC805 qualified the instruments. PDQC811 to PDQC815 added mass spectrometry, counter-ion accounting, and stereochemical purity. Every one of those efforts produces a single thing that leaves the lab: a record. A Certificate of Analysis is a claim. The chromatograms, audit trails, sequence logs, integration events, and signatures behind it are the evidence for that claim.
Computerized system validation (CSV) and data governance answer one question: when Peptide Partners reports "purity 98.7%, largest impurity 0.31%," can anyone (Larissa, a customer auditor, an FDA investigator, a state board of pharmacy) reconstruct, from the records alone, exactly how that number came to exist and confirm that nobody could have quietly made it a different number?
You already know the core logic from PDQC802. Analytical procedure validation shows that a method is fit for its intended use. CSV shows that a computerized system, as configured and operated in this lab, is fit for its intended use. The structure is the same: define intended use (ATP ↔ URS), assess risk, design verification around the risks, generate objective evidence, and maintain the validated state over the lifecycle (Stage 3 continued verification ↔ periodic review and change control). The software is part of the measurement. An integration algorithm is as much a part of your related-substances procedure as the column or the gradient.
McDowall's Data Integrity Model is a useful frame for where this course sits relative to everything you've done [McDowall, 2019]:
- Foundation: Data governance and culture. Management commitment, policies, training, and an environment where reporting a mistake is safer than hiding it.
- Level 1: Right system/instrument for the job. AIQ (USP <1058>, PDQC805) plus CSV (this course).
- Level 2: Right analytical procedure. Validated methods (PDQC802/803).
- Level 3: Right analysis for the right reportable result. Analysts executing correctly, with honest data handling and second-person review.
Each level rests on the one below. A perfectly validated method (Level 2) run on a CDS where analysts can delete injections (Level 1 failure), in a culture that punishes OOS results (Foundation failure), produces untrustworthy numbers. This course builds the Foundation and Level 1, and the controls that make Level 3 observable.
The beautiful idea: data integrity is not a property of a number. It is a property of the system that produced the number, including people, procedures, and software. You cannot inspect integrity into a result after the fact. You can only design it into the process that makes the result, which is the same lesson as Quality by Design.
1. The Record Is the Product: ALCOA+, Original Records, and Dynamic Data
1.1 ALCOA+
The regulatory vocabulary for trustworthy data is ALCOA: Attributable, Legible, Contemporaneous, Original, Accurate. Regulators added "+" attributes: Complete, Consistent, Enduring, Available. MHRA (2018), PIC/S PI 041-1 (2021), WHO TRS 1033 Annex 4 (2021), and FDA's 2018 Q&A converge on these. The definitions differ slightly in wording, so quote the specific guidance you are being audited against.
| Attribute | What it means in a peptide QC lab | Typical failure mode |
|---|---|---|
| Attributable | Every injection, integration, and signature traces to one named individual | Shared "HPLC1" login on an acquisition PC |
| Legible | Readable and interpretable for the full retention period | Raw files only openable by a retired software version |
| Contemporaneous | Recorded when the activity happened, with a trustworthy timestamp | Unsynchronized PC clock 4 min fast; balance printouts transcribed later |
| Original | The first capture (or a verified true copy) is retained | Keeping a PDF of the chromatogram and discarding the .dx raw data |
| Accurate | Correct, with no undocumented edits; calculations verified | Unvalidated Excel template with a wrong cell reference to the TFA % |
| Complete | All data, including failed, aborted, and repeated runs | "Trial" injections deleted or run in an unrecorded project |
| Consistent | Chronology and sequencing make sense; same conventions throughout | Audit-trail timestamps out of order |
| Enduring | Survives for the retention period on durable, backed-up media | Data on a single aging disk in a Windows 7 MS workstation |
| Available | Retrievable for review, audit, and investigation in reasonable time | Archive exists but nobody knows how to restore it |
1.2 Static versus dynamic records: why the printout is not the record
This is one of the most consequential and least intuitive ideas in the field. A static record (a balance printout, a signed paper form) has a fixed form. A dynamic record is one where the user can interact with the content: re-integrate, overlay, zoom, change processing parameters [MHRA, 2018]. A chromatogram is dynamic. The PDF you print shows one rendering of the data under one set of integration events. It does not show that the baseline was dropped at 18.42 min instead of 18.60 min, or that the deamidation shoulder at RRT 1.03 was manually split.
The consequence is that the electronic raw data plus metadata plus audit trail constitute the original record. A paper printout cannot be the GMP record for a CDS-generated result, because it is not a complete and accurate copy of a dynamic record. FDA's 2018 Q&A addresses this directly. Read it alongside 21 CFR 211.194(a), which requires laboratory records to include "complete data derived from all tests." Verify the exact Q&A wording against the current guidance.
A true copy (MHRA's term, akin to "certified copy" in other guidances) must preserve content and meaning, including the dynamic nature where that matters. For chromatography, that means a copy that retains reprocessability, which in practice means the vendor's native data format with its metadata, verified by checksum.
1.3 Metadata is data
"Metadata" sounds like an afterthought. It is the opposite. The number "0.31" is meaningless without: which peak, which sample ID, which processing-method version, which RRF, which disregard limit, who integrated, when. MHRA defines metadata as data that describe the attributes of other data and provide context and meaning. If metadata is lost, the result is no longer reconstructable, even if the number survives. Every backup, migration, and archive test in Weeks 7, 10, and 13 must therefore verify that metadata and audit trails survive, not only that files exist.
2. The Regulatory Architecture: Predicate Rules First, Part 11 Second
2.1 The layering
A common beginner model is "Part 11 is the computer regulation." The correct model is layered:
- Predicate rules say what records must exist and what they must show. For the QC lab these are mainly:
- 21 CFR 211.68: automatic, mechanical, and electronic equipment. Paragraph (b) requires appropriate controls so that changes to records are made only by authorized personnel, checks of input/output accuracy, and a backup file of data entered into the computer system.
- 21 CFR 211.180: general records requirements, including retention for at least 1 year after the batch expiration date (with different provisions for certain OTC products; check the current text).
- 21 CFR 211.194: laboratory records, including complete data from all tests, the record of all calculations, and the signature of the second person who reviewed the original records.
- 21 CFR Part 11 says how electronic records and signatures must be controlled when you choose to keep predicate-rule records electronically, or when you rely on them to perform regulated activities.
- Guidances and international equivalents describe current expectations: FDA Part 11 Scope and Application (2003), FDA Data Integrity Q&A (2018), EU GMP Annex 11 and Chapter 4, PIC/S PI 041-1, MHRA (2018), and WHO TRS 1033 Annex 4.
2.2 The 2003 Scope and Application guidance: what discretion means and what it doesn't
After the 1997 Part 11 rule created a costly compliance burden, FDA narrowed its interpretation in 2003. FDA stated it would exercise enforcement discretion over specific Part 11 requirements:
- validation (§11.10(a))
- audit trails (§11.10(e))
- record retention (§11.10(c))
- record copying (§11.10(b))
- legacy systems
FDA continued to enforce other controls, including:
- limiting system access to authorized individuals (§11.10(d))
- operational, authority, and device checks (§11.10(f)–(h))
- training (§11.10(i))
- written accountability policies (§11.10(j))
- the electronic-signature provisions (Subparts C and the signature/record linking of §11.70)
Check the exact enumerated paragraphs against the 2003 guidance text before citing them in a deliverable.
The trap: enforcement discretion under Part 11 does not remove the requirement where a predicate rule independently imposes it. 211.68(b) still requires controls over changes and backups. 211.194 still requires complete data. FDA explicitly said it would enforce predicate rules, and it expects validation and audit trails wherever they are needed to ensure record integrity under those rules. In practice, for a CDS generating release data, you validate and you keep audit trails, full stop. The 2018 Q&A reinforces this by recommending audit-trail review for CGMP records, and EU Annex 11 §9 requires audit trails based on risk.
2.3 Convergence: one design satisfies all
The guidance documents differ in emphasis but converge on the same control set. Your strategy should be to design once to the strictest common denominator and maintain a cross-reference table rather than separate compliance programs. The Annex 11 headings work well as a checklist skeleton:
- §1 Risk management, §2 Personnel, §3 Suppliers
- §4 Validation, §5 Data, §6 Accuracy checks
- §7 Data storage, §8 Printouts, §9 Audit trails
- §10 Change and configuration management, §11 Periodic evaluation
- §12 Security, §13 Incident management, §14 Electronic signature
- §15 Batch release, §16 Business continuity, §17 Archiving
Flag: the EU has been revising Annex 11, with a draft revision released for consultation. Check the status before finalizing any deliverable, because the revision is expected to be considerably more prescriptive on audit-trail review, identity management, and cloud/service providers.
2.4 Applicability to Peptide Partners specifically
Peptide Partners is a testing laboratory whose customers may include 503A pharmacies and 503B outsourcing facilities. Three layers of expectation reach you:
- Customer-imposed expectations via quality agreements. A 503B outsourcing facility is subject to CGMP expectations (verify current FDA policy for 503B). It will audit you as a contract lab, and its auditors will apply Part 11/211 thinking to your CDS.
- ISO/IEC 17025:2017 §7.11, if you pursue accreditation. It requires that laboratory information management systems be validated for functionality before introduction (including changes), protected from unauthorized access, safeguarded against tampering and loss, operated within supplier/lab-specified environments, maintained to ensure integrity, and that failures and corrective actions are recorded. Note the scope: in 17025, "laboratory information management system(s)" covers all computerized and non-computerized systems for data management, not only a product called "LIMS."
- Direct regulatory exposure. FDA can inspect contract testing labs. A data-integrity finding at your lab becomes a finding against every customer whose release decision relied on your data.
The rational design choice is to build to full CGMP CSV expectations from day one. Retrofitting integrity controls after go-live is far more expensive than designing them in.
3. Compliance Is a System Property: Why "Part 11 Compliant" Software Doesn't Exist
Vendors sell software with "Part 11 compliance features" or a "compliance pack." No software is compliant by itself, because compliance is the product of four independent factors:
Compliance(system) = Technical capability × Configuration × Procedural control × Behavior/culture Technical capability : does the software CAN do it? (audit trail exists, e-sig exists) Configuration : is it TURNED ON and set correctly? (audit trail enforced, roles segregated) Procedural control : do SOPs define use, review, and response? (audit-trail review SOP) Behavior/culture : do people actually follow them, and report problems? Any factor ≈ 0 ⟹ Compliance ≈ 0
The product form, rather than a sum, is the key insight. A system with an excellent audit trail (technical = 1) that analysts can disable (configuration = 0) has zero integrity, however good everything else is.
Recurring patterns in FDA warning letters and inspection findings, as analyzed in the data-integrity literature [Shafiei et al., 2015]; [Rattan, 2017], map directly onto these factors:
- Shared/generic logins (configuration + behavior failure). Every action becomes unattributable. FDA's 2018 Q&A states that shared login credentials are not acceptable for CGMP actions.
- Audit trails disabled or never enabled (configuration failure). A capable product, an incapable installation.
- "Trial" or "test" injections (behavior + procedural failure). Analysts inject the sample, look at the result, and only "officially" run it if it passes. This is testing into compliance. FDA's position (2018 Q&A) is that using actual samples for informal "trial" injections is not acceptable. System-suitability injections using a reference standard, run per a written procedure and retained, are a different and acceptable practice.
- Deletion of failing results; aborting and rerunning sequences (configuration + behavior). Deletion should be technically impossible for analysts. Aborted sequences must remain visible.
- Administrator rights held by analysts (configuration). Whoever can change the system's rules can change the evidence.
Counterintuitive point: the most dangerous person in a CDS is not the dishonest analyst. It is the well-meaning analyst who is also the administrator. They can fix any problem, including the evidence that a problem existed. Segregation of duties protects honest people from being unable to prove they are honest.
4. GAMP 5 (2nd ed.): Risk-Based Lifecycle and the Software Category Continuum
4.1 The lifecycle
GAMP 5 describes a lifecycle of Concept → Project → Operation → Retirement. The project phase follows a specification and verification model: each specification (URS, functional/configuration specification, design specification for custom code) is paired with a verification activity at the appropriate level. This is the familiar "V-model," but GAMP 5 (2nd ed.) emphasizes that the V is a conceptual mapping, not a mandated waterfall. Iterative and agile delivery is acceptable if traceability and verification are maintained [ISPE, 2022].
The second edition's dominant theme is critical thinking: effort proportional to risk to patient safety, product quality, and data integrity, and away from documentation generated for its own sake. It aligns explicitly with ICH Q9(R1), ASTM E2500 principles, and FDA's CSA direction.
4.2 Software categories
| Category | Description | Peptide Partners example | Typical verification emphasis |
|---|---|---|---|
| 1: Infrastructure software | Operating systems, databases, network, security tools | Windows OS on CDS server and acquisition PCs; SQL/database engine; antivirus | Qualify/manage the infrastructure (installation records, configuration baselines, patching under control). No functional "validation" of Windows itself. |
| 3: Non-configured products | Used as supplied. Runtime parameters may be entered but business process is not configured. | Possibly BacT/ALERT embedded software if only standard settings are used. Judgment needed; see Week 8. | Verify intended use. Rely heavily on supplier. |
| 4: Configured products | Configured to the business process using vendor-provided tools | Agilent CDS (roles, audit-trail settings, e-sig meanings, custom fields, report templates); MS acquisition/processing software; COTS LIMS | Verify configuration against the configuration specification. Test critical workflows. Leverage vendor OQ for standard functions. |
| 5: Custom applications | Bespoke code, including macros, scripts, custom calculations, and custom reports | Excel VBA macros; LIMS scripts; custom CDS calculation scripts; result-import parser | Design specification, code review, full functional testing including boundary and error paths |
Correction to a common statement: Category 2 (firmware) was discontinued in GAMP 5 (2008), not newly retired in the 2nd edition (2022). The 2nd edition continues without it. Firmware is classified by what it does, often as Category 3 or 4. Verify phrasing against the guide when writing deliverables.
The categories are a continuum, not boxes. GAMP 5 (2nd ed.) stresses that category is only one input to risk. A real system is usually a composite: the Agilent CDS is Category 4 overall, runs on Category 1 infrastructure, and contains Category 5 elements wherever you write custom calculations (e.g., an RRF-corrected impurity formula built in the custom-calculation editor). Your system inventory must therefore list components, not just products. An Excel spreadsheet that only uses built-in functions with locked cells is often treated as Category 3 or 4 configuration, but one containing a VBA macro is Category 5.
4.3 ASTM E2500 and supplier leverage
ASTM E2500 (written for manufacturing systems and equipment, but the principles transfer) contributes four ideas:
- verification focused on aspects critical to product quality and patient safety
- leadership by subject-matter experts, not document generators
- leveraging supplier documentation and testing when the supplier has been assessed as competent
- good engineering practice as the foundation
For the Agilent CDS this means you do not re-test that the integration algorithm computes areas correctly from first principles. Agilent has tested that across thousands of installations, and you rely on that through a supplier assessment. What you test is what is unique to your installation:
- your role matrix
- your audit-trail settings
- your data paths
- your locked templates
- your custom calculations
- your interfaces
Supplier assessment rigor scales with risk:
- Questionnaire: low-risk Category 3 tools.
- Remote/postal audit of the supplier's QMS documentation: typical for a large, established CDS vendor.
- On-site audit: high-risk, novel, or small-vendor Category 5, or a SaaS LIMS where the supplier operates your GxP data.
Vendor IQ/OQ packages (e.g., Agilent's installation verification tools) can be accepted after a documented review that confirms they cover the functions you rely on. The gap assessment lists what they don't cover, and you test those gaps.
5. Risk-Based Assurance: From ICH Q9(R1) to Test Rigor
5.1 Functional risk assessment
Each URS requirement is assessed for what happens if the function fails. A common FMEA-style scoring scheme is shown below. ICH Q9(R1) does not mandate any particular scale or tool.
Severity (S): impact on patient safety / product quality / data integrity if the function fails
1 = negligible 2 = minor 3 = major (wrong reportable result possible) 4 = critical
Probability (P): likelihood of failure, informed by category, novelty, supplier testing, complexity
1 = remote (std. function, vendor-tested) 2 = possible 3 = likely (custom/novel)
Detectability (D): likelihood the failure escapes detection BEFORE it affects a reported result
1 = will be detected (e.g., SST fails, reviewer sees it) 2 = may be detected 3 = unlikely
RPN = S × P × D (range 1–36 with these scales)
Worked mapping to test rigor (example decision rule; your validation plan defines the real thresholds):
Severity 4 with any P,D → scripted test regardless of RPN (severity override) RPN ≥ 18 → scripted test, positive + negative + boundary cases 8 ≤ RPN < 18 → unscripted (exploratory / scenario) test with recorded outcome RPN < 8, vendor-tested → leverage vendor IQ/OQ; no additional lab test GxP-irrelevant → no test (justify in FRA)
Worked example 1: "The CDS shall prevent users in the Analyst role from deleting injections."
- S = 4: deletion destroys complete-data evidence and enables testing into compliance.
- P = 2: a configured setting, so misconfiguration is plausible.
- D = 3: a deletion leaves nothing for a reviewer to see unless the audit trail catches it.
RPN = 4 × 2 × 3 = 24 → scripted test
The test logs in as an Analyst, attempts deletion through every UI path (and, for file-based components, at the Windows file-system level), and verifies the refusal plus the audit-trail entry of the attempt where the system logs it.
Worked example 2: "The CDS shall display the chromatogram baseline in a user-selectable color."
- S = 1, P = 1, D = 1, so RPN = 1.
Result: no test. Documenting why you didn't test is part of critical thinking.
A mathematical subtlety worth knowing. S, P, and D are ordinal scales. Multiplying ordinal numbers is not strictly meaningful: an RPN of 12 isn't "twice as risky" as 6. Different combinations also produce the same RPN with very different meaning: S=4, P=1, D=3 and S=1, P=4, D=3 both give 12. ICH Q9(R1) added emphasis on reducing subjectivity in QRM. Robust practice is to use the RPN as a sorting aid but write severity override rules (as above) or use a 2-D risk matrix lookup, so a catastrophic-but-rare failure is never waved through by arithmetic.
5.2 Computer Software Assurance (CSA)
FDA's 2022 draft guidance Computer Software Assurance for Production and Quality System Software was issued by CDRH/CBER for medical device production and quality-system software. It does not formally govern drug CGMP. However, its principles are widely adopted in pharma and are consistent with GAMP 5 (2nd ed.). Flag: check whether this guidance has been finalized and whether any drug-side (CDER) statement exists before citing it as an expectation.
CSA's four-step logic:
- Identify the intended use.
- Determine a risk-based approach, distinguishing whether a failure would foreseeably compromise safety ("high process risk") or not.
- Select assurance activities commensurate with risk.
- Establish an appropriate record.
The assurance activities include:
- Scripted testing: robust (step/expected/actual with objective evidence) or limited.
- Unscripted testing:
- ad hoc: no plan
- error guessing: tester uses experience to target likely failures
- exploratory: simultaneous learning, test design, and execution against stated objectives
The surprising insight is that unscripted testing by an expert often finds more defects than scripted testing. A scripted test only confirms what its author thought to check. An experienced reviewer told to "try every way you can think of to change a result without leaving an audit-trail entry" exercises paths no script anticipated, such as editing a processing method mid-sequence or reprocessing under a different project. The record for unscripted testing is lighter (objective, tester, date, issues found, conclusion), but it is not absent.
6. Specification, Verification, and Traceability
6.1 Testable requirements
A URS is the "ATP" of a computerized system. A requirement is useful only if a test can pass or fail it. The rule of thumb is one requirement, one verb (shall), one measurable criterion.
| Bad | Why bad | Good |
|---|---|---|
| "The system shall be Part 11 compliant." | Untestable; bundles ~30 requirements; transfers responsibility to the vendor | "The CDS shall require entry of user ID and password to apply an electronic signature, and shall record the signer's full name, date/time (UTC offset shown), and signature meaning on the signed record." |
| "The system shall have adequate security." | "Adequate" is unmeasurable | "The CDS shall lock a user account after 5 consecutive failed login attempts and require Lab Admin action to unlock." |
| "Audit trail shall be reviewed." | A procedural requirement disguised as a system requirement; no criterion | "The CDS shall provide a filterable audit-trail report per sample set showing all changes to integration events, with user, timestamp, old value, new value, and reason." |
| "The system should be fast and user-friendly." | "Should" is not a requirement; unmeasurable | "Result retrieval for a 100-injection sample set shall complete in ≤ 10 s on the production network." (Only include if it matters.) |
6.2 The traceability matrix as a graph
The requirements traceability matrix (RTM) links:
URS → FRA (risk) → Configuration/Design Spec → Test case(s) → Deviation(s) → Validation Summary Report
Think of it as a directed graph. Its validity checks are graph properties:
Forward coverage = (# URS items with ≥1 linked test or documented "no test" rationale) / (# URS items) → must = 100%
Backward coverage = (# test cases linked to ≥1 URS item) / (# test cases) → should = 100%
(orphan tests waste effort or reveal undocumented requirements)
Open-deviation rule: every failed test step → deviation → resolution or risk-accepted before release-for-use
The RTM's operational value appears later, during change control (Week 11). When Agilent issues a CDS patch, the release notes say "changed: audit-trail export, e-signature dialog." The RTM tells you instantly which URS items and test cases touch those functions. That set is your regression-testing scope. Without an RTM, every change becomes either "retest everything" (unaffordable) or "retest nothing" (indefensible).
6.3 Test evidence and release-for-use
A scripted test step needs four elements:
- prerequisites
- an action
- an expected result written before execution
- an actual result with objective evidence (a screenshot, audit-trail excerpt, or report)
The pass/fail rule must be decided in advance. A tester who writes "as expected" without evidence has produced an assertion, not evidence.
The Validation Summary Report states, against the plan:
- what was done
- deviations and their dispositions
- any restrictions on use
- an explicit release-for-use conclusion signed by the system owner and QA
7. Identity, Segregation of Duties, Electronic Signatures, and Time
7.1 Role design principles
Three principles drive the Agilent CDS role matrix:
- Unique identity. One human, one account, never reassigned. Integration with Active Directory gives central lifecycle control (joiner/mover/leaver), but the CDS must still map AD identities to CDS roles.
- Least privilege. Each role gets exactly the privileges its function requires.
- Segregation of duties. Whoever generates data cannot change the rules governing data; whoever changes the rules cannot generate data.
| Privilege | Analyst | Reviewer | Lab Admin | IT Admin | QA Read-Only |
|---|---|---|---|---|---|
| Acquire data / run sequences | ✔ | ✔ (own work only) | ✘ | ✘ | ✘ |
| Process using approved methods | ✔ | ✔ | ✘ | ✘ | ✘ |
| Manual integration | ✔ (per SOP, reason required) | ✔ | ✘ | ✘ | ✘ |
| Create/edit processing or instrument methods | ✘ (draft only, if permitted) | ✘ | ✔ (versioned, approval required) | ✘ | ✘ |
| Approve methods/templates for use | ✘ | ✔ (or QA) | ✘ | ✘ | ✘ |
| Delete data | ✘ | ✘ | ✘ | ✘ | ✘ |
| Modify audit trail | ✘ (no role) | ✘ | ✘ | ✘ | ✘ |
| User/role management | ✘ | ✘ | ✔ | ✔ (AD level only) | ✘ |
| Server, OS, backup administration | ✘ | ✘ | ✘ | ✔ | ✘ |
| View all data and audit trails | Own projects | ✔ | ✔ | ✘ | ✔ |
| E-sign "Reviewed"/"Approved" | ✘ | ✔ | ✘ | ✘ | ✔ (Approved, if QA releases) |
Small-lab reality. With two people (Zak and Larissa), perfect segregation is impossible. The defensible approach:
- Give each person separate accounts for separate roles (e.g., "larissa.admin" vs "larissa.analyst").
- Prohibit using the admin account for data generation, by procedure and by technical restriction.
- Have the other person periodically review every admin-account action in the system audit trail.
Document this as a risk-accepted compensating control. Regulators accept small-organization constraints when they are recognized, mitigated, and reviewed. They do not accept them when they are ignored.
7.2 Electronic signatures (Part 11 Subpart C and §11.50/§11.70)
- §11.50 manifestation: signed records show the printed name of the signer, date and time of signing, and the meaning of the signature (e.g., "Performed by," "Reviewed by," "Approved by," "Audit trail reviewed"). This must be on human-readable displays and printouts.
- §11.70 linking: signatures must be linked to their records so they cannot be excised, copied, or transferred to falsify another record. In practice, the CDS binds the signature to the specific result-set version. If the result is reprocessed after signing, the signature must be invalidated or flagged.
- §11.200 components: non-biometric signatures use at least two distinct identification components (e.g., ID + password). In a continuous controlled session, the first signing uses all components and subsequent signings may use at least one private component. For non-continuous sessions, all components are required each time. Verify exact wording in §11.200(a)(1).
The signature meaning is not decoration. It is the legal content of the act. "Reviewed" with no definition in an SOP is ambiguous: did the reviewer check integration, or only that the numbers transferred correctly? Define each meaning in the e-signature SOP. Make "audit trail reviewed" a distinct meaning, or an explicit element of "Reviewed," so you can prove audit-trail review happened for every release.
7.3 Time as an integrity control
"Contemporaneous" depends on clocks. Suppose the HPLC acquisition PC is 3 min ahead of the CDS server and the MS workstation is 5 min behind. Then the audit trail's chronology can show a result being processed before it was acquired. This consistency failure is indistinguishable from manipulation on its face. The design requirements are:
- All systems (3 HPLC controllers, CDS server, both MS workstations, BacT/ALERT if network-capable, LIMS) synchronize to a single authoritative NTP source.
- A defined tolerance, e.g., |offset| ≤ 1 s for networked systems (set by the lab; justify it), monitored with alerting.
- Users cannot change system time (no local admin rights).
- Clock changes are recorded in the OS event log and reviewed in the periodic system audit-trail review.
- Isolated legacy PCs that can't reach NTP get a documented manual check (e.g., weekly comparison against a reference clock, logged).
Also define time zone and daylight-saving handling. Store UTC and display local time with offset. Otherwise the October/November DST transition produces a duplicated hour that looks like a timestamp anomaly in every audit trail.
8. Audit Trails: Design, Content, and Review
8.1 Content
A complete audit-trail entry answers who, what, when, why, and old value → new value. For a CDS this exists at three levels:
- System (logins, failed logins, user and privilege changes, configuration changes)
- Project/method (method version creation, edits, approvals)
- Result (integration-event changes, reprocessing, manual integration, signatures)
Configuration requirements:
- always on
- not user-disableable
- not editable by any role
- reason-for-change mandatory, using a controlled pick-list (e.g., "Integration: baseline correction per SOP-xxx §y") plus free text
Pick-lists make audit trails reviewable and trendable. Free text captures specifics. Free text alone produces "fixed," "corrected," and "." entries that defeat the purpose.
8.2 Risk-based review: two cadences
- Result-related audit trails: reviewed with each data package, before the result is reported. These are the trails that could reveal a manipulated reportable result.
- System-level audit trails: reviewed periodically (e.g., monthly) for failed-login clusters, privilege changes, configuration changes, clock changes, and out-of-hours admin activity.
An audit trail nobody reads is a deterrent at best and a liability at worst. When an inspector finds the anomaly you missed, the inspector's conclusion is that your review process doesn't work.
8.3 Quantitative governance metrics
These metrics turn audit-trail review from anecdote into process monitoring. They connect to your MATH201 control-chart thinking.
Manual-integration rate (MIR) = (# injections with ≥1 manual integration event) / (# injections processed) × 100 Reprocessing rate = (# result sets processed more than once) / (# result sets) × 100 Audit-trail findings per 100 runs = (# anomalies raised in review) / (# sample sets reviewed) × 100 Aborted-sequence rate = (# aborted/incomplete sequences) / (# sequences started) × 100
Worked example. In Q1 the related-substances method shows MIR = 4% (12 of 300 injections). In Q2 it shows MIR = 19% (57 of 300).
- Using a p-chart from the Q1 baseline: p̄ = 0.04, n = 300.
- σ = √(0.04 × 0.96 / 300) ≈ 0.0113.
- Upper control limit = 0.04 + 3(0.0113) ≈ 0.074.
- Q2's 0.19 lies far outside the limit.
That is a signal, but not automatically misconduct. Likely causes include column aging degrading resolution of the deamidation pair, a new peptide lot with a new impurity, or an analyst change. The correct response is investigation:
- Is the method's automated integration still fit for purpose?
- Should the processing method be revised and revalidated?
- Is one user responsible for most of the manual integrations?
Metrics turn data integrity into a method-lifecycle signal, linking directly back to PDQC802 Stage 3.
9. Hybrid Systems, File-Based Systems, and Legacy Platforms
9.1 Hybrid systems
A hybrid system combines electronic records with handwritten signatures on printouts. MS workstations often operate this way. Hybrids are permitted but burdensome. You must:
- link each signed printout unambiguously to the specific electronic record (file name, path, checksum, or acquisition date/time and sequence ID on the printout)
- retain the electronic original
- ensure the printout cannot be mistaken for the complete record
The signature attests to the electronic record via the printout, not to the paper alone. The long-term goal is to eliminate hybrids, because every hybrid link is a manual control that can fail silently.
9.2 File-based systems
Many MS acquisition packages store raw data as ordinary files in user-writable folders. In that case anyone with Windows Explorer can delete, rename, or overwrite data, and the application's own audit trail (if it exists) never sees it. Tune and calibration files are a special risk: a changed tune file changes mass accuracy and sensitivity for every subsequent identity result from PDQC811.
Compensating-control stack (defense in depth):
- NTFS permissions: users write to an acquisition folder, but completed data become read-only, or are moved by a service account to a folder users cannot modify.
- Windows object-access auditing on data, tune, and calibration folders as a compensating audit trail. It records who deleted or modified which file and when. It does not record why or old/new scientific values, so it is weaker than an application audit trail. Say so in the risk assessment.
- Automated transfer (scheduled, service-account driven) to a secure server, with SHA-256 checksum verification at source and destination.
- Procedural review of the Windows security log as part of periodic review.
9.3 Legacy operating systems
Instrument PCs often run unsupported Windows versions because the vendor's acquisition software is only qualified on that OS. You cannot patch the OS, so you reduce exposure rather than vulnerability:
- Isolate the PC on a dedicated instrument VLAN behind a deny-by-default firewall that permits only the required ports to the data server.
- Allow no internet or email access.
- Lock down USB and removable media.
- Use application allowlisting.
- Remove local admin rights.
- Set a BIOS/UEFI password.
- Control physical access to the PC.
Then formally accept the residual risk. A residual-risk statement signed by QA and IT carries a review date and explicit triggers for re-assessment, such as:
- a network change
- a new vulnerability class (e.g., a wormable SMB exploit)
- end of hardware support
- a disk SMART warning
ISO/IEC 27001:2022 provides the information-security management vocabulary for these controls. Its Annex A control set was restructured in the 2022 revision, so cite controls by their 2022 numbering, and only after checking.
A useful way to quantify why isolation matters: treat compromise probability as roughly proportional to the attack surface.
Exposure ∝ (# reachable network services) × (# inbound paths) × (# removable-media events) × (time unpatched) Isolation reduces the first three factors by orders of magnitude; you cannot reduce the fourth, so you must drive the others toward zero.
10. Backup, Restore, Archive, and Continuity
10.1 Backup is not archive; and backup is not proven until restored
- Backup protects against loss. It is a copy of current data for recovery, rotated and overwritten on a schedule.
- Archive is the long-term, protected, retention-controlled storage of completed records. It is locked against change, managed by an archivist role, and retrieved by controlled request.
Annex 11 §7.2 expects regular backups whose integrity and ability to be restored are checked during validation and monitored periodically. §17 covers archiving.
The idea to internalize: a backup you have never restored is a hypothesis, not a control. Restore testing is a validated activity. A test restore must:
- restore a defined data set to a non-production location
- verify checksums match
- confirm metadata and audit trails are intact
- confirm the data can be reprocessed in the CDS, not merely opened
10.2 Quantifying resilience
The 3-2-1 principle: 3 copies, on 2 different media types, with 1 off-site. Add an immutable or offline copy for ransomware. The probability argument shows why:
If copies fail independently with annual probability p each: P(all n copies lost) = pⁿ p = 0.05, n = 1 → 0.05 n = 3 → 1.25 × 10⁻⁴ But ransomware creates COMMON-CAUSE failure: every network-writable copy is encrypted together. Effective n = number of copies NOT reachable/writable from the compromised network. → An immutable/offline copy is what keeps n_effective ≥ 1.
The counterintuitive lesson is that three synced copies on the same network can be worth less than one offline copy.
10.3 RPO and RTO
RPO (Recovery Point Objective): maximum tolerable data loss, measured in time.
Data at risk ≈ (data generation rate) × (backup interval)
Example: 3 HPLCs × ~40 injections/day each = 120 injections/day; nightly backup →
worst case ≈ 120 injections of unrecoverable raw data (if acquisition buffers also lost).
Mitigation: acquisition to server with local buffering; shorter interval for the CDS database.
RTO (Recovery Time Objective): maximum tolerable time to restore service.
Driven by business impact: stability pulls with fixed timepoints, customer release deadlines.
For BacT/ALERT, the continuity problem is physical as well as digital. A power interruption during a multi-day incubation threatens both the database and the test itself. The UPS runtime and generator transfer time become data-integrity parameters. The procedure must state what happens to bottles in progress when monitoring is interrupted.
10.4 Checksums
A SHA-256 checksum detects any change to a file. Accidental collision probability is negligible, around 2⁻²⁵⁶ per random pair. A matching hash before and after transfer is strong objective evidence of a complete, unaltered copy. Two limits matter:
- A checksum proves bit-identity, not semantic completeness. If the audit trail lives in a separate database that wasn't copied, every file hash can match while the record is still incomplete.
- Checksum verification must therefore be paired with a record-level completeness check, such as a count of injections per sequence and confirmation that the audit-trail entries are present.
11. Operational Control: Change, Incidents, Periodic Review, and Spreadsheets
11.1 Maintaining the validated state
Validation is a state maintained over time, not a one-time event. Four processes maintain it:
- Change control: impact assessment, then RTM-driven regression scope, then testing, then approval.
- Incident and problem management: crashes, data loss, and failed backups, each with root cause and CAPA.
- Configuration management: the configuration baseline is documented, and any drift from it is detected.
- Periodic review: e.g., annually for high-impact CDS/LIMS, every 2–3 years for lower impact. It covers access recertification, audit-trail trend metrics, deviations and incidents, changes since the last review, supplier support status, and backup/restore performance, and ends with an explicit conclusion: the system remains in a validated state, or doesn't, with actions.
The most common periodic-review failure is the review that concludes "no issues" because nobody looked. Mandatory data inputs, such as the access list exported and compared against current staff, make the review evidence-based.
11.2 End-user computing: spreadsheets
Excel templates for peptide content are high-risk precisely because they are easy to build. A spreadsheet validation package includes:
- a specification of the calculation and its inputs
- cell protection on everything except input cells
- input validation (ranges, data types)
- version control (template ID and version visible on every printout)
- verification of each formula against an independent hand calculation, using normal, boundary, and invalid inputs
- a decision on how the completed workbook is retained as a record
Worked verification case (from PDQC812 accounting). The template computes the peptide content (mass fraction of peptide in the lyophilized material) by mass balance:
Peptide content (%, w/w) = 100 − %water (KF) − %counter-ion (IC or ¹⁹F/HPLC) − %residual solvents − %other non-peptide
Test inputs: water = 6.20% TFA = 11.45% residual solvents = 0.15% other = 0.00%
Hand calc: 100 − 6.20 − 11.45 − 0.15 − 0.00 = 82.20%
Spreadsheet: 82.20% → PASS if |Δ| ≤ predefined tolerance (e.g., 0.005% absolute before rounding)
and displayed value rounds per the lab's rounding rule (verify against USP General Notices).
Boundary tests:
- Water = 0: the template should accept it.
- A negative input: the template should reject it.
- An inputs sum > 100%: the template should flag it rather than report a negative content.
- A blank TFA cell: does the template treat it as 0 silently? This is the most dangerous failure, because the result looks plausible.
Strategic question: should the calculation live in Excel at all? Each spreadsheet is a small, separately validated, separately controlled system with weak native audit trails. Moving the calculation into the CDS (as a custom calculation) or LIMS (as a calculated result) puts it under the platform's audit trail, access control, and change control. The usual answer for a lab at scale is to migrate, keeping validated spreadsheets as a bridge.
12. Data Governance, Master Data, and LIMS/ELN Architecture
12.1 Governance roles
- Process owner: owns the business process, e.g., release testing.
- System owner: responsible for the system's availability, support, and validated state.
- Data owner: accountable for data quality and access decisions.
- Data stewards/users: handle data day to day.
In a two-person lab, one person may hold several roles. The governance document should still name them, so accountability is explicit when the lab grows. PIC/S PI 041-1 emphasizes that data governance is a management responsibility within the PQS (ICH Q10), not an IT function.
12.2 System taxonomy
- LIMS: sample- and result-centric. Covers login, test assignment from specification, results, OOS flagging, CoA, stability scheduling, and inventory.
- ELN: experiment-centric, unstructured or semi-structured. Suited to development work.
- LES (lab execution system): enforces step-by-step procedure execution.
- SDMS: captures and stores instrument raw data centrally.
At launch, a small peptide QC lab primarily needs LIMS functions plus robust CDS data management. ELN and LES functions become valuable at scale or with method development. Buying a platform that covers all four on day one is often an implementation-effort trap.
12.3 Master data: the leverage point
Master data are the reference definitions every transaction uses:
- product/peptide records: sequence, salt form, molecular weight basis, reporting basis
- specification versions (PDQC806)
- test/method definitions
- limits and units
An error in transactional data affects one result. An error in master data affects every result that uses it, silently and consistently. This is why master-data changes go through change control with second-person verification.
Worked consequence. A peptide free base has MW = 3000.0 g/mol and is supplied as a tetra-TFA salt (TFA, C₂HF₃O₂, MW 114.02).
Salt MW = 3000.0 + 4 × 114.02 = 3456.1 g/mol Free-base fraction of salt = 3000.0 / 3456.1 = 0.868 If master data specify the reporting basis as "salt" when the specification is on "free base": reported assay is biased by a factor of 1/0.868 ≈ 1.152 → +15.2% relative error, on every lot, every time, and every individual result looks internally consistent.
No audit trail flags this, because nobody changed anything. The error was born correct. Only master-data verification at creation and change, plus periodic reconciliation against the specification, catches it.
12.4 FAIR alongside ALCOA+
The FAIR principles (Findable, Accessible, Interoperable, Reusable) emphasize machine-actionability of scientific data. Examples include persistent identifiers, rich metadata, and standard vocabularies [Wilkinson et al., 2016]. FAIR is not a GMP standard, and it answers a different question: can this data be reused?, whereas ALCOA+ asks can this data be trusted? The two are complementary, and they converge in LIMS design:
- unique, persistent sample IDs (barcodes)
- controlled vocabularies for test names and units
- structured metadata linking each result to its specification version and raw data
These make data both trustworthy and reusable, for example for trending impurity profiles across lots or building stability models.
12.5 SaaS and the validation responsibility split
With a SaaS LIMS, the supplier operates the infrastructure and controls upgrade timing. You remain accountable for:
- fitness for intended use
- your configuration
- data integrity
The supplier quality agreement must define:
- who validates what
- upgrade notification lead time and access to release notes and test evidence
- backup/restore and DR responsibilities with RTO/RPO
- data residency
- audit rights
- data return in a usable format at contract termination
Data return at termination is the clause most often forgotten. A LIMS you can't exit holds your records hostage.
Connections to Peptide Chemistry
This course can feel like it has left chemistry behind. In fact, it is where your chemistry becomes defensible. Each technical decision from earlier courses has a matching data-integrity control:
- Related substances and integration (PDQC803). Peptide impurity profiles are full of closely eluting species: deamidation (Asn→Asp/isoAsp), Met oxidation, des-amino-acid deletions, and D-isomers (PDQC815). Their resolution often falls near Rs ≈ 1.5. That is exactly where manual integration temptation is highest and where the integrity risk concentrates. The locked processing method should encode the disregard limit, reporting threshold, and RRF corrections. The manual-integration SOP must state when tangent skim versus drop-line is scientifically justified for a shoulder peak, and require that both the automated and the manual results be retained for comparison.
- RRF-corrected impurities. An RRF entered as a custom calculation is Category 5 code. A wrong RRF (e.g., 0.85 entered as 1.85) inflates or deflates an impurity on every lot. Verify custom calculations by hand calculation at validation and after any edit.
- MS identity and tune files (PDQC811). Mass accuracy claims (e.g., ±0.5 Da on a quadrupole, ppm-level on a high-resolution instrument) are only as good as the calibration file in force at acquisition time. Treat the tune/calibration file as a controlled, versioned record linked to each identity result. An undocumented tune change is a data-integrity event, not just an instrument event.
- Peptide content and counter-ions (PDQC812). Mass-balance calculations combine KF water, IC or ¹⁹F counter-ion, residual solvents, and HPLC purity, often from contract labs. Every external value entering your calculation is a transcription risk. LIMS result import with checksums, or at minimum second-person verification of transcribed contract-lab values against the contract lab's CoA, is required.
- Contract-lab bridging. Because ICP-MS, KF, GC-HS, IC, endotoxin, and AAA/qNMR are outsourced for now, your data governance extends to the contract labs' data. Quality agreements must give you the right to review their raw data and audit trails, not just their CoAs. When you later buy an instrument (e.g., a KF titrator), its software enters this course's lifecycle: inventory, GAMP category, URS, FRA, and qualification.
- Sterility and BacT/ALERT (PDQC822 ahead). For 503B-adjacent customers, a positive or negative bottle result can determine whether a sterile preparation is released. The detection algorithm is supplier-validated. Your job is to verify configuration and protect bottle-to-sample traceability (barcode chain of custody), load/unload timestamps, and alarm records.
- Stability programs. LIMS stability scheduling ties pull dates, chamber conditions, and results together. A missed pull, or a result attributed to the wrong timepoint because of a barcode mix-up, corrupts shelf-life regression. Your MATH201 regression is only as good as the timestamped, attributable data feeding it.
Common Misconceptions
- "The vendor says it's Part 11 compliant, so we're compliant." Vendors supply capability. Compliance requires your configuration, procedures, and behavior (Section 3). An auditor will ask to see your configuration specification and your OQ evidence.
- "FDA exercises enforcement discretion on audit trails and validation, so we don't need them." The predicate rules (211.68, 211.194) and the 2018 Q&A independently require controls that amount to validation and audit trails for a CDS producing release data.
- "The signed printout is the official record." For dynamic data, the electronic record with metadata and audit trail is the original. The printout is a representation.
- "More testing is always safer." Testing low-risk functions consumes effort that should go to high-risk ones, and produces documentation that hides what matters. GAMP 5 (2nd ed.) and CSA explicitly favor risk-proportionate assurance.
- "Unscripted testing is undocumented testing." Unscripted testing has a lighter record (objective, tester, date, results, issues, conclusion), but it is still a record and still auditable.
- "An audit trail that exists is an audit trail that works." Unreviewed audit trails don't detect anything. Review must be scheduled, risk-based, evidenced (e-signature meaning), and done by trained reviewers. Seeded-trail exercises are how you prove reviewer competence.
- "Backups mean we're protected." Only a successfully tested restore, with metadata, audit trails, and reprocessability confirmed, proves protection. Networked backups are vulnerable to the same ransomware as production data.
- "Archive and backup are the same thing." Backup is for recovery and is overwritten on rotation. Archive is for retention and is locked, controlled, and retrieved by request.
- "Windows file auditing is equivalent to an application audit trail." It records file-level who/what/when but not scientific old/new values or reasons. It is a compensating control, and its limitations must be stated in the risk assessment.
- "Category 4 software doesn't need code review." True for configuration, false for any custom calculation, script, macro, or report embedded within it. Those elements are Category 5 and need design review and thorough testing.
- "A small lab can't segregate duties, so the requirement doesn't apply." The requirement's intent (no unchecked power over evidence) still applies. Use separate role accounts, mutual review of admin actions, and documented risk acceptance.
- "Data integrity problems are about fraud." Most integrity failures come from poorly designed systems, ambiguous procedures, pressure to meet deadlines, and fear of OOS consequences [Rattan, 2017]. Good design makes the honest path the easy path and makes deviations visible without needing to presume bad intent.
- "An RPN is a precise risk number." RPNs multiply ordinal scores. Use them for prioritization, with severity override rules, not as a quantitative risk measure.
- "Once validated, always validated." The validated state decays with every patch, network change, staff change, and configuration drift. Change control and periodic review are what keep validation true.
The closing thought for this section: a QC laboratory sells trust, and the chromatography, mass spectrometry, and microbiology are how that trust is earned. Computerized system validation and data governance are how it is proven, to anyone, at any time, for as long as the record must be kept. When you can hand an auditor a result and the complete, attributable, contemporaneous, unalterable story of how it came to exist, you have built a lab that can defend itself.
Verification flags for deliverables:
- Confirm the exact Part 11 paragraphs under enforcement discretion against the 2003 guidance.
- Confirm the current status of FDA's CSA guidance (draft vs final) and of the EU Annex 11 revision.
- Confirm the §11.200(a)(1) wording on signing within continuous sessions.
- Confirm ISO/IEC 27001:2022 Annex A control numbering before citing specific controls.
- Confirm the 211.180 retention provisions for products without expiration dates.
- Confirm the lab's rounding conventions against current USP General Notices.
⚙️ Mechanisms & Applications
Mechanisms & Applications — From Analytical Signal to Defensible Decision
My central insight: computerized system validation is not a ceremony performed on software. It is an evidence-based demonstration that a particular configuration, operated by particular people through controlled procedures, can reliably support its intended decisions. For Peptide Partners, those decisions include peptide identity, assay, impurity reporting, and whether a result is suitable for a customer’s quality assessment.
The analogy to protein biochemistry is useful: knowing a protein’s sequence does not establish its activity under every condition. Likewise, knowing a software product’s name—or possessing a vendor’s qualification certificate—does not establish the behavior of the installed laboratory workflow. Configuration, interfaces, environment, and human actions are part of the functional system.
1. Begin with the Record Lifecycle, Not the Software Catalogue
A chromatogram is not simply an image. Its meaning depends on the sample identity, acquisition method, detector settings, processing method, integration history, calculations, and review status. The laboratory must preserve the information needed to reconstruct the analytical decision, including changes that led to the final result. This connects directly to the “complete laboratory records” requirement in 21 CFR 211.194 and FDA’s discussion of dynamic electronic records.[1, 2]
Physical sample
→ accession and custody record
→ preparation and dilution records
→ acquisition worklist and method version
→ detector signal + acquisition metadata
→ processing method + result versions
→ technical review + relevant audit-trail review
→ approval and issued report
→ retained, readable, retrievable record
- Identify the decision. Example: report a specified peptide impurity against an approved limit.
- Identify every record supporting it. Include preparation records, standards, sequence entries, aborted runs, original integrations, and recalculations—not only the passing chromatogram.
- Locate each record. CDS database, acquisition controller, MS workstation, spreadsheet, paper worksheet, or contract-laboratory repository.
- Assign ownership and controls. Specify who creates, changes, reviews, retains, and authorizes disposal.
- Challenge reconstruction. Can an independent reviewer reproduce the reported value and explain its history?
Application at Peptide Partners: first inventory the three Agilent 1260 systems and their actual acquisition architecture. Do not infer that a networked CDS is installed because the instruments are Agilent. Record software name, version, modules, database technology, storage locations, operating system, supplier support status, interfaces, and backup owner. Both MS software configurations and the additional MS model remain identification tasks—not assumptions.
Contract testing belongs in the same lifecycle. Quality agreements should establish access to underlying data and investigations, retention responsibilities, subcontracting controls, and expectations for electronic deliverables. Receiving a CoA does not eliminate the need to evaluate the laboratory and its records.
2. Determine Regulatory Scope Before Designing Controls
Predicate rules determine what records are required; electronic-record requirements determine how applicable electronic records and signatures are controlled. Relevant US drug-CGMP anchors include 21 CFR 211.68 for computerized equipment and backups, 211.180 for retention, and 211.194 for laboratory records. Applicability must be assessed for the laboratory’s actual activities and customer commitments, rather than inferred from the phrase “503A/503B-adjacent.” ISO/IEC 17025:2017 clause 7.11 independently addresses control of data and information management.[1, 3]
A subtle but important correction: FDA’s 2003 Part 11 guidance describes enforcement discretion for certain Part 11 provisions, including validation, audit trails, record retention, and copying. It does not suspend predicate-rule obligations or make integrity controls optional. A defensible laboratory still needs risk-appropriate validation, complete records, controlled access, and traceability of relevant changes. Access and electronic-signature controls cannot be dismissed by treating the guidance as a blanket exemption.[1]
- Technical control: software prevents an analyst from deleting a result.
- Configuration control: that permission is actually denied in the released installation.
- Procedural control: access is reviewed, exceptions are investigated, and privileged actions are supervised.
- Governance control: management funds the controls and does not reward undocumented repeat testing.
“Part 11 capable” is therefore a feature claim, not a transferable compliance status. EU GMP Annex 11, PIC/S PI 041-1, and other data-integrity guidance converge on this lifecycle view, although their legal contexts differ.[4]
3. Build Validation as a Chain of Testable Claims
GAMP 5 organizes validation around intended use, risk, supplier knowledge, and lifecycle control. Software categories describe the nature of the software and configuration—not its importance or a predetermined quantity of testing.[5]
- Category 1: infrastructure software, such as the operating system.
- Category 3: non-configured product software, where applicable to the actual implementation.
- Category 4: configured products, commonly including CDS and LIMS workflows.
- Category 5: custom applications or extensions, such as bespoke macros or interface scripts.
One installation can contain several categories. A configured LIMS may include custom reporting code. BacT/ALERT classification depends on the installed software and how its functions are configured; the instrument name alone does not determine the category.
Intended use
→ user requirement
→ functional risk assessment
→ configuration/design specification
→ verification evidence
→ deviation resolution
→ validation conclusion
→ authorized release
→ continued control
Bad requirement: “The CDS shall ensure data integrity.”
Testable requirement, URS-AT-01: “The system shall record every saved change to a processing method with the user identity, timestamp, changed parameter, previous value, and new value.” A separate requirement should address the reason for change and another should restrict audit-trail administration.
- Assess failure: an analyst changes integration thresholds without a traceable history.
- Identify consequence: an impurity may disappear from a report without the reviewer knowing why.
- Specify controls: versioned methods, restricted permissions, audit trails, and independent review.
- Verify the configured behavior: deliberately change a parameter using an authorized test account.
- Inspect evidence: confirm the saved record and trail against predefined expected values.
- Test prohibited behavior: attempt to alter or disable the trail using an analyst account.
Example traceability row: URS-AT-01 → FRA-07: untraceable method changes → CFG-AT-03: method audit history enabled → OQ-AT-04: change and inspect method parameters → DEV-002, if observed → retest evidence → validation summary disposition
Supplier testing may be leveraged after assessing its relevance to the installed version, configuration, and intended use. It does not replace local verification of users, storage, interfaces, calculations, and end-to-end workflows. ASTM E2500 supplies useful specification-and-verification principles; USP <1058> connects this work to analytical instrument qualification.[5, 6]
A validation summary should identify the released configuration, test coverage, deviations, residual risks, trained users, approved procedures, and operational prerequisites. A missing high-impact audit trail or unverified recovery pathway is not a minor paperwork issue.
4. Use Risk Scores to Structure Thought—not to Manufacture Certainty
ICH Q9(R1) supports systematic quality-risk management but does not mandate a particular FMEA scoring scheme. Severity, occurrence, and detectability scores are usually ordinal judgments, not measured probabilities.[7]
Illustrative local scale: 1–5 for each factor D = 5 means poor detectability. Risk priority number = S × O × D Wrong reporting-basis mapping: S = 5, O = 2, D = 4 RPN = 5 × 2 × 4 = 40
The number 40 does not mean “40% risk.” Nor is an RPN of 40 scientifically established to be twice as dangerous as 20. Different combinations can conceal very different consequences.
- Severity override: loss of original data, unauthorized approval, or incorrect release calculations requires explicit verification even when estimated occurrence is low.
- Scripted testing: use for critical permissions, signatures, calculations, interfaces, and recovery acceptance criteria.
- Exploratory testing: use to discover unexpected workflow behavior, such as cancellation, repeated submission, or interrupted sessions. Record scope, actions, observations, and defects.
- Supplier-leveraged evidence: use for adequately documented standard functions after a justified gap assessment.
The course’s 2022 FDA Computer Software Assurance draft is a historical teaching source. Its medical-device production and quality-system context should not be presented as a blanket drug-QC exemption from validation. Check the current final guidance and applicable scope before using it in controlled documents.
5. Configure the CDS as a Controlled State Machine
A useful model is a sequence of permitted record states. Data are acquired, processed, reviewed, and approved; each transition requires an authorized identity and leaves evidence. Editing an approved result must create a controlled new state, not silently replace the old one.
Acquired → Processed → Reviewed → Approved
↑
└── justified reprocessing → new result version
Approved record changed:
preserve prior approval and record
→ flag changed version as requiring review
→ repeat applicable review/approval
- Analyst: acquire data and use approved methods; perform permitted reprocessing with a reason; no deletion, audit-trail administration, or production-method approval.
- Reviewer: inspect source records and audit trails; return results for correction; do not independently review one’s own work.
- Lab administrator: maintain authorized configuration; no routine production data generation or approval.
- IT administrator: maintain infrastructure; no routine analytical approval. Privileged access remains separately controlled and monitored.
- QA read-only: inspect records and governance evidence without modifying analytical data.
Method development needs a controlled route: separate development permissions or environments, followed by approval and deployment of a versioned production method. A prohibition on analysts changing production methods must not force development work into undocumented workarounds.
Illustrative site parameters—not regulatory constants: centrally managed unique accounts; minimum 14-character passwords where supported; lockout after five failed attempts; screen lock after 10 minutes; prompt account removal at termination. Set password-aging and authentication policies against applicable requirements and current security practice. Part 11 does not prescribe those numeric settings.
Electronic signatures must show the signer’s name, execution date/time, and meaning, and remain linked to the signed record. For non-biometric signatures, §11.200 addresses two identification components and their use during continuous and non-continuous signing sessions; “two components” is not necessarily equivalent to modern multifactor authentication.[1]
Time control: synchronize servers, acquisition controllers, and MS workstations to authorized sources. An example internal acceptance criterion is clock offset ≤60 seconds during qualification and monitoring. Preserve timezone interpretation and detect unauthorized clock changes. NTP synchronization alone does not prove that application audit trails use time correctly.
6. OQ Challenge Testing Is an Experimental Technique
The “experiment” is a deliberate perturbation of a controlled system. As in enzyme kinetics, the independent variable must be explicit and the observed response must discriminate between acceptable and unacceptable behavior.
Test OQ-ACCESS-02 — Analyst deletion challenge Prerequisites: Approved test environment/configuration Analyst test account Disposable chromatographic test dataset Independent evidence capture available Action: Attempt deletion through the application. Attempt deletion through accessible filesystem paths. Expected: Neither route removes protected data. Denied actions produce the configured security evidence. Original data remain readable and processable. Acceptance: All protected records remain intact. No unauthorized privilege or deletion route succeeds.
Do not require an application audit-trail entry for every denied filesystem action unless the design actually supports it. The evidence may reside in operating-system security logs. Specify where each event should appear.
Additional high-value challenges:
- Attempt to approve another analyst’s result using an analyst-only account.
- Change a processing method and verify previous/new values and version linkage.
- Interrupt the acquisition network and verify documented local buffering, alarms, reconciliation, and recovery.
- Submit a duplicate result message to the LIMS and verify that it is not posted twice.
- Attempt report generation using an obsolete specification version.
These tests should use controlled test data. Disrupting a production acquisition or deleting live data is not an acceptable validation technique.
7. Audit-Trail Review Reconstructs Analytical Reasoning
An audit trail is valuable because it exposes the path to the answer. A final impurity value can be numerically correct yet scientifically indefensible if earlier results were concealed or integration changes were selected solely to obtain compliance.[2, 8]
- Reconcile completeness. Compare the worklist, injections, files, sequence status, and issued results. Account for blanks, suitability injections, aborts, and repeats.
- Review result-affecting changes. Inspect method versions, manual integrations, dilution factors, reference-standard assignments, and calculations.
- Evaluate reasons scientifically. “Reprocessed” describes an action, not its justification.
- Check authorization and chronology. Investigate unexpected privilege changes, timestamps, and changes after approval.
- Document review and disposition. Preserve queries and responses; escalate suspected invalidation or concealment through deviation/OOS/data-integrity procedures.
Seeded reviewer exercise:
- 09:10 — impurity result calculated as 0.18%; approved limit is 0.15%.
- 09:16 — analyst changes the integration threshold; reason: “better integration.”
- 09:18 — impurity result becomes 0.12%.
- 09:22 — analyst attempts to delete the first result; access denied.
- 09:25 — only the 0.12% result appears in the proposed report.
Answer key: hold reporting; preserve both result versions and the denied-deletion evidence; investigate the initial potentially OOS result under the applicable procedure; evaluate integration against predefined scientific rules; determine whether similar actions affected other samples. Do not average the values or invalidate the first result merely because a later one passes. The trail establishes events, not intent.
Review relevant result-related trails with each data package before reporting. Review system-level access and configuration events on a justified periodic schedule, and sooner when an event may affect current results. “Monthly” is a possible site choice, not a universal rule.
8. Secure MS Workstations Without Confusing Copying with Integrity
MS systems add tune files, calibration files, spectral libraries, processing applications, and file-based raw datasets to the record lifecycle. These can affect peptide identity and impurity characterization as decisively as chromatographic integration settings.
Instrument workstation on restricted VLAN
├── permitted acquisition communication → instrument
├── approved time/authentication services → designated servers
└── controlled data transfer → hardened repository
Firewall default: deny
Internet/email: blocked
Interactive access: restricted and logged
Backup repository: not writable by ordinary acquisition users
Unsupported operating systems require a documented threat model: removable-media infection, lateral ransomware movement, aging disks, unauthorized administration, and loss of readable software. Controls include segmentation, tested allowlisting, restricted USB use, least privilege, physical access control, and a funded replacement plan.
Automated transfer should copy only completed or properly quiesced datasets, generate a manifest, verify transfer, and alarm on failure. For multi-file MS records, preserve directory structure and dependencies.
For each transferred file i: SHA-256(source_i) = SHA-256(destination_i) Also verify: expected files = received files expected sizes = received sizes sample/run identity matches dataset opens in the authorized application
Important limitation: a matching hash proves matching bytes at the comparison points. It does not establish correct sample identity, trustworthy acquisition, or the absence of modification before hashing. Protect the manifest and the process that creates it.
NTFS permissions and Windows object-access auditing can detect file operations, but usually cannot reconstruct chromatographic reintegration or changes within proprietary records. They are compensating controls, not automatically equivalent substitutes for an application audit trail. If critical risks remain uncontrolled, restrict the system’s GxP use or bridge through a qualified laboratory rather than accepting the gap indefinitely.
9. BacT/ALERT: Database Recovery Cannot Reverse Biology
For BacT/ALERT, map bottle identification, loading and unloading, incubation, instrument readings, positivity determination, alarms, and reports. Verify the installed model’s actual record storage, access control, audit-trail, and export capabilities.
- Establish which records exist electronically and which exports preserve them.
- Verify configuration such as users, protocol durations, and alarm behavior.
- Challenge barcode mismatches and authorized configuration changes using suitable test material or supplier-supported simulation.
- Use supplier-supported database backup methods; do not assume that copying a live database file creates a recoverable backup.
- Restore to a supported test environment and reconcile bottle IDs, events, results, and available histories.
Supplier evidence can support understanding of the detection algorithm, but the laboratory still needs appropriate analytical-method verification or validation for its intended microbiological use. Software validation does not establish product-specific microbial recovery or method suitability.
The surprising distinction: restoring a database restores information, not the incubation conditions experienced during a power failure. UPS runtime, alarm response, exposure assessment, and affected-test disposition therefore belong beside IT recovery procedures. Signed printouts and controlled logbooks may supplement limited native controls, but cannot recreate electronic history that was never captured.
10. Backup, Restore, and Archive Solve Different Problems
- Backup: recover from loss, corruption, or disaster.
- Restore verification: demonstrate that recovered records and functions are usable.
- Archive: retain completed records in a controlled state for their required lifetime.
Use a justified backup architecture, commonly informed by the 3-2-1 principle: three copies, two storage types or independent storage arrangements, and one off-site copy. Add immutability or offline separation appropriate to ransomware risk. A backup account with unrestricted access to every copy can defeat the architecture.
RPO = maximum acceptable loss of recent data RTO = maximum acceptable time to restore required service Illustrative CDS targets: RPO ≤ 1 hour RTO ≤ 8 hours 24 GB restore at effective 20 MB/s: t ≈ 24,000 MB / 20 MB/s = 1,200 s = 20 min Total recovery time also includes: failure diagnosis + infrastructure recovery + database recovery + application startup + reconciliation + authorization to resume
A nightly backup alone does not support a one-hour RPO. Transaction-log backups, protected acquisition buffers, or other suitable mechanisms may be required. Targets must reflect analytical and business consequences; they are not vendor brochure parameters.
Restore-test acceptance criteria: all expected records present; application-supported database consistency checks satisfactory; file hashes match where applicable; audit trails and signature links preserved; selected chromatograms reprocess successfully using retained methods; retrieval permissions work; measured recovery meets the approved target. Bitwise comparison alone is insufficient for a database restored through transaction replay.
For retention, 21 CFR 211.180 includes at least one year after batch expiry for relevant records, with a different provision for certain OTC products lacking expiration dating. Establish the exact schedule using the applicable official text, customer agreements, and other legal obligations. Legal holds override routine disposal. PDF reports alone generally do not preserve the functionality of dynamic chromatographic records.[1, 2]
11. Validate Peptide Calculations at Their Most Dangerous Boundary: Meaning
PDQC812 showed that water, counter-ion, and reporting basis can dominate peptide-content accounting. A spreadsheet can execute flawless arithmetic while answering the wrong chemical question.
Worked example: a qualified reference standard has an assigned peptide content of 0.870 mg peptide per mg material, explicitly on an as-is basis. A 12.00 mg portion is diluted to 25.00 mL, then 1.00 mL is diluted to 10.00 mL.
Initial peptide concentration = 12.00 mg material × 0.870 mg peptide/mg material ÷ 25.00 mL = 0.4176 mg peptide/mL Final peptide concentration = 0.4176 × (1.00/10.00) = 0.04176 mg peptide/mL = 41.76 µg peptide/mL
Do not subtract water or counter-ion again when they are already incorporated into the assigned as-is content. Do not substitute HPLC area purity for mass fraction without a justified measurement model.
Validation tests:
- Independently verify normal cases and dimensional consistency.
- Reject zero or negative volumes, missing content assignments, and inconsistent units.
- Challenge entry of “87” where “0.870” is expected.
- Test basis mismatches: as-is versus dry basis, salt versus peptide equivalent.
- Verify locked formulas, version identity, print/export content, and unauthorized-edit detection.
- Define comparison rules and rounding explicitly; compare unrounded results to limits unless the approved procedure specifies otherwise.
For this test vector, the unrounded computational result must be 41.76 µg/mL within a predefined numerical tolerance. That software-test tolerance is not the analytical uncertainty. Measurement uncertainty still arises from weighing, volumes, and the assigned standard value.
12. LIMS Integration Is a Semantic and Transactional Experiment
A CDS specializes in acquisition and processing; LIMS manages samples, specifications, results, and reporting; an ELN supports experimental narrative; an LES guides execution; an SDMS manages scientific files and metadata. A small laboratory should purchase the functions its controlled workflows require, not all four labels.
Product / peptide definition
→ approved specification version
→ required tests and limits
→ sample accession
→ aliquots and custody events
→ method version and instrument eligibility
→ result version
→ review/approval
→ issued CoA version
Master data must carry sequence or identity definition, salt form, reporting basis, units, method versions, and applicable limits. Historical results must remain linked to the specification version used; updating master data must not silently rewrite their interpretation.
- Export an identified worklist: sample ID, test ID, method reference, and unique transaction ID.
- Acquire and review results: retain source record identifiers and processing versions.
- Import into a staging layer: check message structure, identifiers, units, basis, status, and completeness.
- Commit atomically: either post the complete valid transaction or quarantine it; avoid partly populated official records.
- Acknowledge and reconcile: detect missing, duplicate, rejected, and superseded messages.
Critical failure test: transmit the same message twice. The second transmission must not create a second reportable result. This is idempotency. Also challenge “0.25%” being interpreted as a fraction of 0.25, which would become 25%.
Barcodes reduce transcription but do not establish truth: a perfectly readable label on the wrong vial remains wrong. Verify label issuance, aliquot relationships, scan-at-transfer behavior, duplicate handling, durability, and the handling of unreadable labels.
Illustrative vendor score: Weights: integrity/validation 0.30 workflow fit 0.25 integration 0.20 support/viability 0.15 lifecycle cost 0.10 Scores, each 1–5: 5, 4, 3, 4, 2 Weighted score = 0.30(5) + 0.25(4) + 0.20(3) + 0.15(4) + 0.10(2) = 3.90 / 5
Apply mandatory gates before scoring: unacceptable audit trails, inability to export complete records, or an unworkable SaaS upgrade process cannot be compensated by low price. In SaaS, infrastructure operation may be delegated; intended-use validation, data ownership, and release decisions remain assigned responsibilities.
13. Maintain the Validated State as a Feedback-Control System
Validation is the initial characterization; operational governance is the feedback loop that detects drift. This connects naturally to control systems in physiology: a stable state requires sensing, evaluation, and corrective action—not merely a good starting condition.
Change or incident
→ impact assessment
→ affected requirements and risks
→ targeted regression testing
→ controlled deployment
→ effectiveness monitoring
→ periodic validated-state review
Periodic review should examine access recertification, privileged activity, failed backups, restore evidence, incidents, changes, supplier support, training, audit-trail trends, and residual risks. Annual review may be a suitable starting point for a high-impact CDS; major events can require earlier review.
Audit-trail findings per 100 reviewed runs = 100 × qualifying findings / reviewed runs Example: 6 findings / 240 reviewed runs × 100 = 2.5 findings per 100 runs
Interpret trends with denominators and consistent definitions. A rising finding rate may indicate deteriorating behavior—or better detection. A low manual-integration rate is not inherently good if analysts conceal necessary corrections. Metrics must not reward the disappearance of evidence.
14. Historical Perspective: Why These Ideas Emerged
- 1997 — FDA issued Part 11. The central problem was making electronic records and signatures trustworthy substitutes for paper in regulated activities. The innovation was not simply electronic signing, but enforceable expectations around access, record protection, and linkage.[1]
- 2003 — FDA clarified scope and application. Enforcement discretion and a narrower interpretation addressed implementation concerns without abolishing predicate-rule requirements. This remains a major source of misunderstanding.[1]
- 2015–2017 — regulatory analyses emphasized recurring integrity failures. Shafiei and colleagues, and later Rattan, examined regulatory thinking, historical issues, and remediation. These publications help explain why technical controls must be joined to organizational governance.[9, 10]
- 2016 — Wilkinson and colleagues articulated FAIR. Findable, Accessible, Interoperable, and Reusable principles strengthened scientific data stewardship. FAIR is not a GMP certification and does not require unrestricted public access. A FAIR dataset can still be scientifically wrong; a secure dataset can still be unusable.[11]
- 2022 — GAMP 5 second edition emphasized critical thinking. The trajectory is away from document volume as a proxy for assurance and toward evidence proportionate to intended use and risk—not away from documentation.[5]
My final learning checkpoint: I should be able to take any reported peptide result and answer: What generated it? Which versions transformed it? Who changed it? Why? What evidence was reviewed? Could an unauthorized action alter it? Can we recover and interpret the complete record years later? If those questions cannot be answered, the analytical method may be excellent, but the measurement system is not yet defensible.
Sources and Official-Text Checks
- 21 CFR Part 11, especially §§11.10, 11.50, 11.70, 11.100, and 11.200; 21 CFR §§211.68, 211.180, and 211.194. FDA, Part 11, Electronic Records; Electronic Signatures—Scope and Application (2003). Check current eCFR text for controlled implementation.
- FDA, Data Integrity and Compliance With Drug CGMP: Questions and Answers—Guidance for Industry (2018).
- ISO/IEC 17025:2017, clause 7.11, Control of data and information management.
- EU GMP Annex 11, Computerised Systems, and Chapter 4, Documentation; PIC/S PI 041-1 (2021); MHRA, GXP Data Integrity Guidance and Definitions (2018); WHO TRS 1033, Annex 4 (2021). Verify current effective versions and jurisdictional applicability.
- ISPE, GAMP 5: A Risk-Based Approach to Compliant GxP Computerized Systems, 2nd ed. (2022); ISPE, GAMP Guide: Records and Data Integrity (2017).
- ASTM E2500, Standard Guide for Specification, Design, and Verification of Pharmaceutical and Biopharmaceutical Manufacturing Systems and Equipment; USP <1058>, Analytical Instrument Qualification. Consult current applicable editions.
- ICH Q9(R1), Quality Risk Management; ICH Q10, Pharmaceutical Quality System.
- McDowall RD. Validation of Chromatography Data Systems, 2nd ed. Royal Society of Chemistry (2017); Data Integrity and Data Governance. Royal Society of Chemistry (2019).
- Shafiei N, De Montardy R, Rivera-Martinez E. “Data Integrity—A Study of Current Regulatory Thinking and Action.” PDA Journal of Pharmaceutical Science and Technology (2015). DOI: 10.5731/pdajpst.2015.01082.
- Rattan AK. “Data Integrity: History, Issues, and Remediation of Issues.” PDA Journal of Pharmaceutical Science and Technology (2017). DOI: 10.5731/pdajpst.2017.007765.
- Wilkinson MD, Dumontier M, Aalbersberg IJ, et al. “The FAIR Guiding Principles for scientific data management and stewardship.” Scientific Data 3, 160018 (2016). DOI: 10.1038/sdata.2016.18.
🪬 Skipper's Reflection
Skipper’s Reflection: PDQC814 — Computerized System Validation, Data Governance, and LIMS/ELN
If my earlier chemistry courses taught me that the molecule is the reality, PDQC814 taught me a much more sobering truth for the pharmaceutical industry: the data is the product. Until this week, I viewed a chromatogram of a peptide as a physical truth. Now, I see it as a fragile digital construct that only has value if its chain of custody, from the detector to the database, is unbroken, locked, and fully auditable.
1. What Surprised Me Most
I was genuinely shocked to learn that there is no such thing as "21 CFR Part 11 compliant software" out of the box. I always assumed that if Zak and Larissa bought a modern Agilent CDS or a brand-new LC-MS/MS, the vendor’s compliance claim meant we were automatically shielded from FDA or MHRA data integrity citations. Not true. The software only provides the technical controls (the locks); it is the lab that must configure them, assign the keys, and write the procedural controls (the rules).
The warning letter case studies we reviewed were eye-opening. Seeing companies cited because their analysts shared a generic "Admin" login—allowing anyone to delete a failing injection and run a "trial" injection in its place—made me realize how quickly a multi-million dollar laboratory can be invalidated by a lazy IT policy. The realization that who holds the Windows Administrator password is a Critical Quality Attribute was a profound shift in my thinking.
2. Conceptual Breakthroughs: When It Clicked
My biggest breakthrough was understanding the Requirements Traceability Matrix (RTM) and how it integrates with GAMP 5 (2nd edition) and Computer Software Assurance (CSA). I used to view validation as a mountain of redundant paperwork. But building the RTM in my mind—tracing a User Requirement Specification (URS) to a Functional Risk Assessment (FRA), down to the Configuration Specification, and finally to a specific OQ test script—revealed it as the structural engineering of data integrity.
The other massive "click" was the distinction between static and dynamic records. A printed PDF of a chromatogram is a static, dead record. The FDA (and common sense) dictates that the electronic file—where you can zoom in, check the integration baseline, and see the audit trail—is the true original record. You cannot validate a system by just signing the printouts while leaving the raw data in an unsecured, user-writable Windows folder.
3. Connections to the Peptides I Know
How does IT architecture relate to BPC-157, Tirzepatide, or Retatrutide? It comes down to manual integration and impurity profiles. Peptides are notorious for closely eluting related substances—deamidated forms, oxidized species, or aggregates that trail the main peak.
When running a stability-indicating assay for Retatrutide on our Agilent 1260 HPLCs, the integration baseline is everything. Moving a baseline drop by just 0.05 minutes can shift an impurity from 2.1% (a failing batch) to 1.9% (a passing batch). PDQC814 made me realize that the chemical separation is only half the battle; defending that baseline from unauthorized manual integration is the other half. The audit trail is the immune system of our analytical methods. By locking the related-substances templates and enforcing a "reason for change" prompt whenever manual integration is performed, we aren't just doing IT work—we are defending the chemical truth of the peptide.
4. Questions to Investigate Further
This course raised several urgent questions as we build out Peptide Partners:
- The Mystery MS: We have a second mass spectrometer in the lab with an unidentified model and software version. If this system is running a legacy, unsupported OS (like Windows 7), how exactly will I configure the VLAN and firewall deny-by-default rules to air-gap it while still allowing data egress to a secure server? I need to dig deeper into compensating network controls.
- Training the Reviewers: How do I generate realistic, "seeded" audit trails with planted integrity issues (e.g., deleted injections, out-of-order time stamps) to train Zak and Larissa on the Audit-Trail Review SOP without polluting our validated production CDS environment?
- LIMS Scalability: As we transition from Excel templates (which require locked cells and intense validation) to a full LIMS, how do we handle data residency and vendor audits if we choose a SaaS cloud LIMS? How much of the validation burden really shifts to the vendor?
5. Confidence Rating
Most Confident: I feel highly confident in designing the CDS Configuration Specification and the Role Matrix. I clearly see how to segregate duties (e.g., IT Admin vs. Analyst vs. QA Read-Only) to comply with 21 CFR 211.68 and Part 11. I am also very comfortable drafting the URS-to-test traceability matrices and defining audit trail policies.
Least Confident: I am least confident in the physical IT infrastructure and legacy OS remediation. While I understand the theory of compensating controls, implementing NTFS permissions, Active Directory integration, automated SHA-256 checksum verifications, and backup/restore RTO/RPO targets pushes me out of my chemistry comfort zone and into the realm of a network administrator. I will need to bridge this gap to ensure the raw MS data generated in PDQC811 isn't lost to a crashed hard drive or a ransomware lateral movement.
Ultimately, PDQC814 has completely reshaped my view of a QC lab. A lab is not just a room full of instruments analyzing peptides; it is a secure data factory, and the validation of its digital systems is the ultimate guarantor of its scientific integrity.
🗺️ Visual Aids & Key Takeaways
Diagram 1: GAMP 5 Software Category Decision Tree for Peptide QC Lab Assets
┌─────────────────────────────────────────────────────────────────────────────────┐
│ GAMP 5 (2nd ed.) SOFTWARE CATEGORY DECISION TREE │
│ applied to Peptide Partners QC laboratory systems │
└─────────────────────────────────────────────────────────────────────────────────┘
┌──────────────────────────────┐
│ Is the software GxP-impacting│
│ (used to generate, process, │
│ store, or report GxP data)? │
└──────────────────────────────┘
│ │
YES NO
│ │
↓ └──→ OUT OF SCOPE
(no CSV required;
document rationale)
┌────────────────────────┐
│ Is it infrastructure │
│ software — OS, network,│
│ antivirus, PDF viewer? │
└────────────────────────┘
│ │
YES NO
│ │
↓ ↓
┌─────────────────┐ ┌─────────────────────────────────┐
│ CATEGORY 1 │ │ Is it non-configurable firmware │
│ Infrastructure │ │ or fixed-function instrument │
│ │ │ software with no user options? │
│ Windows 10 OS │ └─────────────────────────────────┘
│ NTP client │ │ │
│ Acrobat Reader │ YES NO
│ │ │ │
│ APPROACH: │ ↓ ↓
│ Keep patched, │ ┌──────────────┐ ┌────────────────────────┐
│ document │ │ CATEGORY 3 │ │ Is it standard product │
│ version, │ │ Non-config │ │ software (COTS), sold │
│ supplier notes │ │ firmware │ │ to many, configurable │
│ sufficient │ │ │ │ but no custom code? │
└─────────────────┘ │ BacT/ALERT │ └────────────────────────┘
│ incubator │ │ │
│ firmware │ YES NO
│ (positivity │ │ │
│ algorithm │ ↓ ↓
│ embedded, │ ┌──────────────────┐ ┌──────────────────┐
│ user cannot │ │ CATEGORY 4 │ │ CATEGORY 5 │
│ alter) │ │ Configured │ │ Custom/bespoke │
│ │ │ COTS │ │ │
│ APPROACH: │ │ │ │ Custom Excel │
│ Supplier │ │ Agilent CDS │ │ macros (VBA) │
│ documentation│ │ (OpenLAB CDS) │ │ │
│ + config │ │ MassHunter │ │ LIMS custom │
│ verification │ │ Agilent MS │ │ calculation │
│ only │ │ data acq. │ │ scripts │
└──────────────┘ │ BacT/ALERT SW │ │ │
│ (user accounts, │ │ APPROACH: │
│ alarm config) │ │ Full code review │
│ │ │ + scripted test │
│ APPROACH: │ │ of every branch │
│ URS → config │ │ + RTM coverage │
│ spec → IQ/OQ/PQ │ │ 100% │
│ (risk-scaled) │ └──────────────────┘
└──────────────────┘
─────────────────────────────────────────────────────────────────────────────────
VALIDATION EFFORT SCALE (least → most)
Cat 1 ──────────→ Cat 3 ──────────→ Cat 4 ──────────→ Cat 5
Document only Supplier docs URS+config spec Full lifecycle
+ config verify IQ/OQ/PQ code review +
(risk-based) scripted tests
─────────────────────────────────────────────────────────────────────────────────
NOTE: Category 2 (spreadsheet-as-product) was retired in GAMP 5 (2nd ed., 2022).
Validated Excel templates now treated as Category 4 (configured) or 5 (custom).
Decision tree mapping every software asset in the Peptide Partners lab to its GAMP 5 category, with the appropriate validation approach and real lab examples at each leaf node; the retired Category 2 is flagged to prevent the common error of citing the old five-category scheme.
---Diagram 2: CDS Data Integrity Architecture — From Injection to Locked Archive
╔══════════════════════════════════════════════════════════════════════════════════════╗
║ CDS DATA FLOW AND INTEGRITY CONTROLS: AGILENT OpenLAB CDS ║
║ (3× Agilent 1260 HPLC + LC-MS/MS at Peptide Partners) ║
╚══════════════════════════════════════════════════════════════════════════════════════╝
ACQUISITION LAYER PROCESSING LAYER REVIEW / RELEASE LAYER
────────────────── ───────────────── ─────────────────────────
┌─────────────────┐ ┌──────────────────┐ ┌────────────────────────┐
│ HPLC-1 (1260) │ │ │ │ REVIEWER │
│ Acquisition PC │ ── raw .D ──→ │ CDS SERVER │ ──────→ │ (role: Reviewer) │
│ [NTP synced] │ (buffered │ (hardened, │ │ │
└─────────────────┘ on loss) │ RAID array, │ │ · Opens sequence │
│ server-side │ │ · Inspects audit trail│
┌─────────────────┐ │ audit trail │ │ · Checks integration │
│ HPLC-2 (1260) │ ── raw .D ──→ │ always on) │ │ · Verifies SS criteria│
│ Acquisition PC │ │ │ │ · e-signs (Reviewed) │
│ [NTP synced] │ │ DATA CONTROLS: │ └────────────────────────┘
└─────────────────┘ │ ┌─────────────┐ │ │
│ │ No delete │ │ ↓
┌─────────────────┐ │ │ No overwrite│ │ ┌────────────────────────┐
│ HPLC-3 (1260) │ ── raw .D ──→ │ │ No local │ │ │ APPROVER (QA role) │
│ Acquisition PC │ │ │ storage │ │ │ │
│ [NTP synced] │ │ │ Versioned │ │ │ · e-signs (Approved) │
└─────────────────┘ │ │ method files│ │ │ · Releases data for │
│ │ Locked │ │ │ reporting │
┌─────────────────┐ │ │ report │ │ └────────────────────────┘
│ LC-MS/MS │ ── raw .D ──→ │ │ templates │ │ │
│ MassHunter PC │ + NTFS │ └─────────────┘ │ ↓
│ [VLAN isolated]│ permissions │ │ ┌────────────────────────┐
│ [NTP synced] │ compensating │ AUDIT TRAIL: │ │ ARCHIVE │
└─────────────────┘ audit trail │ · System log │ │ │
│ · Method log │ │ Project → locked, │
↑ All injections in sequence │ · Result log │ │ read-only archive │
↑ No orphan / aborted data │ WHO/WHAT/ │ │ folder │
↑ NTP timestamps verified │ WHEN/WHY/ │ │ [retention ≥ 1 yr │
(CDS server is NTP master │ OLD VAL/ │ │ past batch expiry, │
for all 4 acquisition PCs) │ NEW VAL │ │ per 21 CFR 211.180] │
└──────────────────┘ └────────────────────────┘
│
↓
┌───────────────────────┐
│ BACKUP SERVER │
│ │
│ · Daily incremental │
│ · Weekly full │
│ · Off-site or │
│ immutable copy │
│ · SHA-256 checksum │
│ verified on restore │
│ · Restore tested │
│ quarterly (scripted)│
└───────────────────────┘
─────────────────────────────────────────────────────────────────────────────────────
KEY INTEGRITY CONTROLS AT EACH BOUNDARY:
[Instrument PC → Server] Network buffer: data queued locally if connection lost,
auto-transferred on reconnect; never written only locally.
[Server storage] WORM-equivalent: no user-level delete; only privileged
archive function can move (not delete) finalized data.
[Method/template changes] Change requires re-approval; old version retained with
timestamp; new version effective date recorded.
[Audit trail] Cannot be disabled by any user role, including Lab Admin.
Reviewed per run before data release (Week 9 SOP).
─────────────────────────────────────────────────────────────────────────────────────
End-to-end data flow for the Agilent OpenLAB CDS at Peptide Partners, annotating every integrity control (buffering, versioning, audit-trail layers, backup chain, archival) at the boundary where data could be corrupted or manipulated, from the moment of injection through to locked archive.
---Diagram 3: Requirements Traceability Matrix (RTM) — Structure and Flow
┌────────────────────────────────────────────────────────────────────────────────────┐
│ REQUIREMENTS TRACEABILITY MATRIX: LIFECYCLE LINKAGE │
│ Demonstrated on CDS Audit-Trail Requirement (representative row) │
└────────────────────────────────────────────────────────────────────────────────────┘
DOCUMENT TIER ARTIFACT EXAMPLE CONTENT
───────────── ──────── ───────────────
┌──────────────┐
│ URS │ URS-CDS-017
│ (User │ "The system SHALL record an audit-trail
│ Requirement │ entry for every result modification,
│ Spec) │ capturing: operator ID, date/time
│ │ (UTC, NTP-synced), old value, new value,
│ │ and a mandatory reason-for-change
│ │ selected from a controlled pick-list."
└──────┬───────┘
│ traces to ↓
┌──────┴───────┐
│ FRA │ Risk ID: R-017
│ (Functional │ Severity: HIGH (GxP record integrity)
│ Risk │ Probability: MEDIUM (user incentive
│ Assessment) │ to alter peak areas in RS methods)
│ │ Detectability: LOW if audit trail absent
│ │ Risk score → HIGH
│ │ Test rigor: SCRIPTED (mandatory)
└──────┬───────┘
│ traces to ↓
┌──────┴───────┐
│ CONFIG │ CS-CDS-017
│ SPEC │ · Audit trail: ENABLED, system-level
│ │ and result-level, non-disableable
│ │ · Pick-list values: "Baseline correction",
│ │ "Peak misidentified", "Reintegration
│ │ after column change", "Other (explain)"
│ │ · Free-text field: mandatory when
│ │ "Other" selected
│ │ · Admin cannot disable via GUI or
│ │ registry edit (tested in OQ)
└──────┬───────┘
│ traces to ↓
┌──────┴───────┐
│ TEST SCRIPTS│ OQ-CDS-017a "Verify audit trail captured
│ IQ/OQ/PQ │ on reintegration"
│ │ Prerequisite: Analyst account logged in,
│ │ test sequence loaded
│ │ Step 1: Modify a peak integration
│ │ boundary in processed result
│ │ Expected: Pick-list dialog appears;
│ │ cannot save without selection
│ │ Step 2: Select reason, save
│ │ Expected: Audit trail entry appears:
│ │ user=testanalyst01,
│ │ time=[±30 s of wall clock],
│ │ old area=[value], new area=[value],
│ │ reason=[selected text]
│ │ Pass/Fail: ALL five fields present AND
│ │ time within ±1 min of NTP
│ │
│ │ OQ-CDS-017b "Verify audit trail cannot
│ │ be disabled by Lab Admin"
│ │ Step 1: Log in as Lab Admin
│ │ Step 2: Navigate all settings menus
│ │ and attempt to find audit-trail
│ │ disable option
│ │ Expected: No such option present
│ │ Pass/Fail: Absence confirmed, screenshot
│ │ attached as objective evidence
└──────┬───────┘
│ traces to ↓
┌──────┴───────┐
│ DEVIATION │ DEV-OQ-003 (if test fails):
│ RECORD │ "Audit trail pick-list not appearing
│ (if needed) │ in CDS v2.6.1 — defect report to Agilent,
│ │ configuration workaround applied,
│ │ re-test OQ-CDS-017a PASS"
└──────┬───────┘
│ traces to ↓
┌──────┴───────┐
│ VALIDATION │ VSR-CDS-2024: "URS-CDS-017 fully met.
│ SUMMARY │ Test OQ-017a/b both PASS.
│ REPORT │ One deviation (DEV-OQ-003) resolved.
│ (VSR) │ System released for GxP use on [date]."
└──────────────┘
─────────────────────────────────────────────────────────────────────────────────
RTM COLUMN STRUCTURE (one row per requirement):
URS ID │ Requirement text │ Risk ID │ Risk level │ Config Spec ID │
Test ID(s) │ Test result │ Deviation ID │ VSR reference │ Status
─────────────────────────────────────────────────────────────────────────────────
The RTM is a living document — every change triggers re-assessment of affected rows.
A gap in any column (e.g., no test for a HIGH-risk requirement) is a critical finding.
Vertical slice through the requirements traceability matrix for a single CDS audit-trail requirement, showing how URS language, risk score, configuration specification, OQ test-script steps with explicit pass/fail criteria, deviation handling, and the validation summary report are linked into one auditable chain — illustrating why every column must be populated for every HIGH-risk requirement before system release.
---Diagram 4: Operational Control Cycle — Change Control, Periodic Review, and Incident Management
╔═══════════════════════════════════════════════════════════════════════════════╗
║ COMPUTERIZED SYSTEM OPERATIONAL CONTROL CYCLE ║
║ (post-release validated-state maintenance) ║
╚═══════════════════════════════════════════════════════════════════════════════╝
┌─────────────────────┐
│ VALIDATED STATE │
│ (system released, │
│ VSR approved, │
│ in routine use) │
└──────────┬──────────┘
│
┌─────────────────────┼────────────────────────┐
│ │ │
↓ ↓ ↓
┌──────────────────┐ ┌──────────────────────┐ ┌──────────────────────┐
│ CHANGE EVENT │ │ INCIDENT / FAILURE │ │ PERIODIC REVIEW │
│ │ │ │ │ (scheduled, annual │
│ · CDS patch │ │ · System crash │ │ for CDS/LIMS; │
│ · OS upgrade │ │ · Failed backup │ │ 2–3 yr lower-impact)│
│ · Network change │ │ · Data loss event │ │ │
│ · New column │ │ · Audit-trail gap │ │ CHECKLIST: │
│ type added │ │ · OOS linked to │ │ · Access recertify │
│ · User role mod │ │ system error │ │ · Audit-trail trends │
└────────┬─────────┘ └──────────┬───────────┘ │ (findings/100 runs)│
│ │ │ · Deviation summary │
↓ ↓ │ · Pending changes │
┌──────────────────┐ ┌──────────────────────┐ │ · Supplier support │
│ CHANGE IMPACT │ │ ROOT CAUSE │ │ status (EOL risk) │
│ ASSESSMENT │ │ ANALYSIS │ │ · VSR conclusion: │
│ │ │ │ │ "System remains │
│ Q1: Does it │ │ · 5-Why or fishbone │ │ in validated state"│
│ affect GxP │ │ · Is data integrity │ └──────────┬───────────┘
│ data or its │ │ compromised? │ │
│ processing? │ │ → Data-integrity │ │
│ │ │ investigation │ │
│ Q2: Does it │ │ (PDQC801 OOS SOP) │ │
│ change tested │ └──────────┬───────────┘ │
│ functionality? │ │ │
└────────┬─────────┘ ↓ │
│ ┌──────────────────────┐ │
│ │ CAPA │ │
│ │ · Corrective action │ │
│ │ · Preventive action │ │
│ │ · Effectiveness │ │
│ │ check (≤90 days) │ │
│ └──────────┬───────────┘ │
↓ │ │
┌──────────────────┐ │ │
│ REGRESSION TEST │ │ │
│ SCOPE (from RTM) │ │ │
│ │ │ │
│ Minor patch: │ │ │
│ → unscripted │ │ │
│ smoke test │ │ │
│ │ │ │
│ Major upgrade: │ │ │
│ → all HIGH-risk │ │ │
│ OQ tests re-run│ │ │
│ → new features: │ │ │
│ full test cycle│ │ │
└────────┬─────────┘ │ │
│ │ │
└───────────────────────┴─────────────────────────┘
│
↓
┌─────────────────────┐
│ VALIDATED STATE │
│ (re-confirmed or │
│ updated VSR) │
└─────────────────────┘
─────────────────────────────────────────────────────────────────────────────
TRIGGER MATRIX: what forces what level of re-validation
Change type │ Impact assessment │ Regression scope
──────────────────────────────── ─────────────────── ─────────────────────
Security patch (no functional │ Required │ Unscripted smoke test
change, vendor confirmed) │ │ + backup verification
──────────────────────────────── ─────────────────── ─────────────────────
CDS minor version (e.g., │ Required │ HIGH-risk OQ subset
2.6.1 → 2.6.2) │ │ per RTM
──────────────────────────────── ─────────────────── ─────────────────────
CDS major version (e.g., │ Required │ Full OQ/PQ re-run
2.x → 3.x) │ │ treat as new system
──────────────────────────────── ─────────────────── ─────────────────────
New user role added │ Required │ Access/privilege tests
──────────────────────────────── ─────────────────── ─────────────────────
Network reconfiguration │ Required │ Data-path and NTP tests
─────────────────────────────────────────────────────────────────────────
The three operational pathways — change control, incident response, and periodic review — that maintain validated state after system release, showing the decision points, regression-test scoping logic derived from the RTM, and the trigger matrix that determines whether a minor patch needs only a smoke test or a full OQ re-run.
---Key Takeaways
- Part 11 compliance is tripartite and non-transferable. A vendor's "Part 11 compliant" claim covers only technical controls built into the software. Compliance additionally requires procedural controls (SOPs, training) and configuration controls (correct settings actually applied in your instance). All three must coexist; a perfectly Part 11–capable CDS running with shared logins is non-compliant regardless of vendor claims. This is the central lesson of every FDA warning letter citing computerized-system failures.
- The 2003 FDA Scope and Application guidance created enforcement discretion, not an exemption. FDA stopped enforcing Part 11 controls for records that exist on paper and are merely scanned (the "predicate rule" records that don't require electronic form). However, it has always enforced Part 11 fully for records that are created, modified, maintained, archived, or transmitted electronically as GxP records — which includes every HPLC chromatogram, integration result, and sequence log in the Agilent CDS. There is no relaxation for audit trails, access controls, or signature linking on those records.
- GAMP 5 (2nd ed., 2022) retired Category 2. The previous five-category scheme (1–5) is now four effective categories (1, 3, 4, 5). Validated Excel calculation templates are Category 4 (configured) if they use only built-in functions, or Category 5 (custom) if they contain VBA macros. Citing the old Category 2 in a validation plan is an immediate credibility failure during audit.
- The audit trail must capture five fields to be compliant: who, what, when (absolute NTP-synced timestamp), old value, and new value — plus a mandatory reason-for-change. An audit trail that logs only "record modified by analyst01" without old and new values is insufficient to reconstruct what happened. The reason-for-change should use a controlled pick-list (preventing free-form rationalization) with an "Other — explain" escape valve. FDA's 2018 Data Integrity Q&A and EU GMP Annex 11 clause 9 both require the audit trail to be retained as long as the record it documents.
- No local-only data storage is the cardinal rule of CDS architecture. Acquisition PCs must write directly to the server, with network-loss buffering that auto-transfers on reconnect. Data that ever exists only on a local instrument PC can be deleted without a server-side audit trail. The compensating control for file-based MS systems (MassHunter) is NTFS permissions (read-only data folder) plus Windows object-access auditing as a surrogate audit trail, combined with automated scheduled copy with SHA-256 checksum verification to the hardened server.
- Time synchronization is a data-integrity control, not an IT nicety. All acquisition PCs, the CDS server, and MS workstations must synchronize to a single NTP source. A clock drift of even a few minutes can make audit-trail timestamps appear out of sequence, creating the appearance of backdating — a critical finding even when the underlying data are legitimate. The OQ must include a test that verifies timestamp agreement to within ±1 minute across all nodes, and the configuration specification must document the NTP server address and clock-change detection policy.
- The requirements traceability matrix (RTM) is the spine of the validation package — not a bureaucratic artifact. Every URS requirement must trace forward to a risk level, then to a configuration specification entry, then to at least one test case with explicit pass/fail criteria, and then to a validation summary report conclusion. A HIGH-risk requirement with no corresponding test case is a critical gap. The RTM also governs regression testing: when a change occurs, the system owner looks up which RTM rows are affected and re-runs only those tests — which is what makes risk-based validation efficient rather than "test everything every time."
- Legacy operating systems (e.g., end-of-life Windows on instrument PCs) require a documented risk acceptance, not just tolerance. The risk-acceptance form must name specific threats (unpatched CVEs, ransomware lateral movement, removable-media malware), list the compensating controls actually implemented (instrument VLAN, firewall deny-by-default, USB port lockdown via Group Policy or physical cover, application allowlisting, local-admin rights removed, BIOS password, scheduled automated data egress with checksum), state the residual risk level, be signed by QA and IT, and carry a review date and explicit triggers for re-assessment. A risk acceptance without compensating controls documented is not defensible.
- Manual integration is the highest-risk single act in HPLC data handling. It must be governed by a written policy specifying: (a) which conditions justify reintegration (defined criteria, not analyst judgment), (b) that both the original automated integration and the manual integration are retained with full audit trail, (c) that a second reviewer must approve before the manual result is used, and (d) that the reintegration rate is tracked as a metric (findings per 100 injections). An anomalously high reintegration rate is a data-integrity signal requiring investigation — exactly the kind of trend that periodic audit-trail review should detect.
- Backup and restore are two different activities; only restore testing validates the backup. A daily backup that has never been restored is an untested assumption. The restore test must be scheduled (quarterly is a defensible frequency for a GxP lab), must compare SHA-256 checksums of the restored files against the originals, must confirm that audit trails survive intact and are not truncated, must verify that data remain reprocessable (i.e., the CDS can open and process the restored raw files), and must be documented as a completed validation activity. The 3-2-1 backup principle (3 copies, 2 different media types, 1 off-site or immutable) is the minimum architecture for ransomware resilience.
- A LIMS and an ELN solve different problems; a small peptide QC lab likely needs LIMS functions first. LIMS (Laboratory Information Management System) manages sample login, chain-of-custody, test assignment from specifications, result entry, OOS flagging, CoA generation, and instrument calibration status — all of which are immediate operational needs in a QC lab. An ELN (Electronic Laboratory Notebook) manages experimental narrative, protocol capture, and research data — valuable for development but not the first priority for a QC testing environment. The selection matrix must weight compliance features, CDS integration capability, and the ability to configure (not code) workflows, because custom code immediately moves the system to GAMP Category 5.
- SaaS/cloud LIMS creates a split validation responsibility that must be explicitly documented. The vendor is responsible for infrastructure qualification (IQ equivalent: server, OS, database) and may provide Customer Qualified Infrastructure documentation. The customer remains responsible for configuring the system to their URS, verifying that configuration, and ensuring that vendor-controlled upgrades do not invalidate the tested state. The supplier quality agreement must address data residency, backup frequency and restore capability, notification period before upgrades, access to audit logs, and the right to audit. Without this agreement, the customer cannot demonstrate control over their GxP data.
- Computer Software Assurance (FDA CSA draft guidance, 2022) shifts emphasis from documentation volume to testing confidence. CSA's key message is that scripted testing should be reserved for high-risk functions, while low-risk functions can be verified by unscripted (exploratory) testing or by leveraging vendor testing with documented rationale. The goal is "thinking, not documenting" — a QC lab that produces 500 pages of scripted test scripts for a spell-checker configuration has misunderstood the risk-based approach. The RTM's risk scores determine which approach applies to each requirement; this is where the intellectual work of validation actually lives.
- The FAIR data principles (Findable, Accessible, Interoperable, Reusable) operationalize data governance beyond regulatory compliance. A raw CDS chromatogram that is locked in a proprietary binary format with no documented metadata schema is technically "backed up" but scientifically inaccessible after a vendor discontinues support. FAIR requires persistent identifiers for datasets, rich metadata in standardized formats, and data in open or well-documented formats where possible. For a peptide QC lab, this means: unique sample IDs traceable through LIMS, method metadata captured with every run, and a documented plan for reading legacy MS raw data (e.g., virtualized MassHunter viewer, vendor-neutral mzML export where scientifically valid) when the acquisition software is eventually retired.
- Audit-trail review is an analytical task, not a checkbox. The reviewer must actively look for specific anomalies: (a) deletions or attempted deletions, (b) reintegration count and justification adequacy, (c) method changes mid-sequence without documented justification, (d) aborted or incomplete sequences followed by a clean re-run, (e) out-of-hours activity without explanation, (f) timestamps out of chronological order, and (g) repeated sample-set injections that might represent "test injections" not reported. The FDA 2018 Data Integrity Q&A specifically asks whether labs review audit trails as part of the data-review process — it is an inspectable element, not an internal best practice. Reviewer training should use seeded audit trails containing planted anomalies, with a documented answer key to establish competency before live use.
References
- U.S. Food and Drug Administration. 21 CFR Part 11: Electronic Records; Electronic Signatures. Code of Federal Regulations, Title 21, Part 11. U.S. Government Publishing Office. Available at: https://www.ecfr.gov/current/title-21/chapter-I/subchapter-A/part-11
- U.S. Food and Drug Administration. 21 CFR Part 211: Current Good Manufacturing Practice for Finished Pharmaceuticals (Sections 211.68, 211.180, 211.194). Code of Federal Regulations, Title 21. U.S. Government Publishing Office. Available at: https://www.ecfr.gov/current/title-21/chapter-I/subchapter-C/part-211
- U.S. Food and Drug Administration. Guidance for Industry: Part 11, Electronic Records; Electronic Signatures — Scope and Application. FDA, August 2003. Available at: https://www.fda.gov/media/75414/download
- U.S. Food and Drug Administration. Guidance for Industry: Data Integrity and Compliance With Drug CGMP — Questions and Answers. FDA, December 2018. Available at: https://www.fda.gov/media/119267/download
- U.S. Food and Drug Administration. Draft Guidance for Industry and FDA Staff: Computer Software Assurance for Production and Quality System Software. FDA, September 2022. Available at: https://www.fda.gov/media/161521/download
- European Commission. EU Guidelines for Good Manufacturing Practice for Medicinal Products for Human and Veterinary Use, Annex 11: Computerised Systems. European Commission, 2011. Available at: https://health.ec.europa.eu/system/files/2016-11/annex11_01-2011_en_0.pdf
- European Commission. EU Guidelines for Good Manufacturing Practice, Chapter 4: Documentation. European Commission. Available at: https://health.ec.europa.eu/system/files/2016-11/2014-02_gmp_chapter-4_en.pdf
- Pharmaceutical Inspection Co-operation Scheme (PIC/S). PI 041-1: Good Practices for Data Management and Integrity in Regulated GMP/GDP Environments. PIC/S, July 2021. Available at: https://picscheme.org/layout/document.php?id=1507
- Medicines and Healthcare products Regulatory Agency (MHRA). GXP Data Integrity Guidance and Definitions. MHRA, March 2018. Available at: https://www.gov.uk/government/publications/guidance-on-gxp-data-integrity
- World Health Organization. WHO Technical Report Series No. 1033, Annex 4: Guideline on Data Integrity. WHO, 2021. Available at: https://www.who.int/publications/m/item/trs1033-annex4
- International Society for Pharmaceutical Engineering (ISPE). GAMP 5: A Risk-Based Approach to Compliant GxP Computerized Systems, 2nd edition. ISPE, 2022.
- International Society for Pharmaceutical Engineering (ISPE). GAMP Guide: Records and Data Integrity, 1st edition. ISPE, 2017.
- McDowall, R. D. Validation of Chromatography Data Systems: Ensuring Data Integrity, Meeting Business and Regulatory Requirements, 2nd edition. Royal Society of Chemistry, 2017.
- McDowall, R. D. Data Integrity and Data Governance: Practical Implementation in Regulated Laboratories, 1st edition. Royal Society of Chemistry, 2019.
- Wilkinson, M. D., Dumontier, M., Aalbersberg, I. J., et al. 'The FAIR Guiding Principles for scientific data management and stewardship.' Scientific Data, 2016, 3, 160018. DOI: 10.1038/sdata.2016.18
- Shafiei, N., De Montardy, R., Rivera-Martinez, E. 'Data Integrity — A Study of Current Regulatory Thinking and Action.' PDA Journal of Pharmaceutical Science and Technology, 2015, 69(1), 17–29. DOI: 10.5731/pdajpst.2015.01082
- Rattan, A. K. 'Data Integrity: History, Issues, and Remediation of Issues.' PDA Journal of Pharmaceutical Science and Technology, 2017, 71(4), 287–313. DOI: 10.5731/pdajpst.2017.007765
- ASTM International. ASTM E2500-13(2018): Standard Guide for Specification, Design, and Verification of Pharmaceutical and Biopharmaceutical Manufacturing Systems and Equipment. ASTM International, 2018. DOI: 10.1520/E2500-13R18
- International Council for Harmonisation. ICH Q9(R1): Quality Risk Management. ICH, January 2023. Available at: https://www.ich.org/page/quality-guidelines
- International Council for Harmonisation. ICH Q10: Pharmaceutical Quality System. ICH, June 2008. Available at: https://www.ich.org/page/quality-guidelines
- International Organization for Standardization. ISO/IEC 17025:2017: General Requirements for the Competence of Testing and Calibration Laboratories (Clause 7.11: Control of data and information management). ISO, 2017.
- International Organization for Standardization. ISO/IEC 27001:2022: Information Security Management Systems — Requirements. ISO, 2022.
- U.S. Pharmacopeial Convention. USP General Chapter <1058>: Analytical Instrument Qualification. United States Pharmacopeia. [Verify current edition in official USP online catalog.]
- U.S. Pharmacopeial Convention. USP General Chapter <1029>: Good Documentation Guidelines. United States Pharmacopeia. [Verify current edition in official USP online catalog.]