WorkflowManufacturing & Quality

GMP Document Review

GMP document review for API plants and CDMOs: agents read a controlled document against its template, check each citation on the date its results were made, and trace each number to LIMS before the technical reviewer and the QA approver sign. They then find the other approved documents that still cite a superseded revision.

See one stability report checked screen by screen ↓
Watch with transcript →
The problem

A pasted row and a replaced specification both passed review

The technical reviewer prints the LIMS export, ticks numbers against the report, and opens each cited specification and method at the revision the document system shows today. Copied from the row above, a value looks right on paper. A citation that changed halfway through the study gets checked against the wrong date, and a template section that asks for a regression receives one sentence saying the trend is slight.

QA then signs a document whose errors nobody went looking for. When the client lowers a limit, the report under review may get fixed, while the protocol, the test worksheet and the disposition checklist that cite the same specification wait for the next periodic review or an inspector.

Built forQC technical reviewersQA document approversDocument control specialistsStability coordinators

The Problem Today

Reviewers tick numbers against a printout and open citations at today's revision, so a superseded limit or a pasted row gets signed.

243
FY2025 FDA citations: quality procedures unwritten or unfollowed published
18
months at most from a year of data without statistics, ICH Q1E published

Checked by hand, page by page

  • Templates
  • Revisions
  • Results
  • Statistics
  • Replaced specs still citedA client's change reaches one document and misses the rest
  • A sentence where analysis belongsThe template asks for a regression and gets an opinion

The Solution

  1. Read in three layersTemplate sections, cited documents and borrowed numbers, each against its source
  2. Citations judged by dateThe revision that applied when each result was made, not today's
  3. The missing analysis worked outRegression and confidence bound against each market's limit
  4. Two people sign, then the rippleReviewer and QA approver sign; other documents citing the old revision get change requests

Built-in controls

  • No verdict on a draft the agents have not read
  • Authors cannot review or approve their own work
  • A new draft clears each signature already given

What You Get

Documents reach clients and inspectors with each number and citation traced, and a specification change reaches the documents that cite it.

  • Reports clients can fileEach value and citation opens the record behind it
  • Each reviewer checks like the bestApproved review profiles carry what experienced reviewers look for
  • Clients hear before they fileWhat their own limit change means for the data, in fields to answer
Features

GMP document review of one stability report, from first read to approval

The technical reviewer runs the read, the author fixes from each source, the lab head rules on a data question, and the QA approver signs and writes to the client.

demoEverything you see here is sample data: the stability study, its batches, the plant, its client, the people and every figure.

  1. 01Three layers

    Controlled document checked against its template, citations and LIMS at once

    • The document as it prints. Headers, revision and page numbers stay as the inspector will read them, so the reviewer judges the real artifact.
    • A count for each layer. Sections complete, citations in force and numbers matching LIMS sit beside the pages, with each gap opening its mark.
  2. 02Template layer

    Template section filled with a sentence where a regression belongs

    • Each required section outlined. The approved template sits over the draft, so a section that states a conclusion without its analysis stands out amber.
    • Read against the review profile. The profile for stability reports names what each section must hold, so a new reviewer checks it the same way a veteran does.
  3. 03Citation by date

    Specification revision checked against the date each result was made

    • Old and new revision side by side. The limit that changed is highlighted, so the reviewer sees what the citation gets wrong without opening the document system.
    • In force when the test ran. Results made before the change stay judged against the old revision; only later results and the conclusion move to the new one.
  4. 04Number trace

    Stability result traced to its LIMS row, with the row it was copied from

    • Matched at the reported precision. Each number in the tables is compared with LIMS as the report rounds it, so a slip of two hundredths is not lost.
    • Where the wrong value came from. The trace names the neighboring row the value matches, which tells the author it was a paste and not a retest.
  5. 05Evaluation

    ICH Q1E stability evaluation computed when the report left it out

    • Poolability tested, bound per market. Slopes and intercepts are tested before pooling, and the upper confidence bound is drawn against each market's limit.
    • What each market's data support. One market keeps its 24 months; the other reaches its limit at 17.1 months, which the client must know before filing.
    • Assumptions the reviewer can change. Switching to the report's rounded values or another model shows how far the answer moves, and the agents never choose the retest period.
  6. 06Challenger

    The client's regulatory reviewer played back, catching the agents' own fix

    • Questions the client will ask. The challenger reads the report as the person who files it and raises each point before anyone relies on it.
    • Its own drafters challenged too. A proposed fix that rewrote earlier results to the new limit was withdrawn and redrafted by date, and the audit trail keeps both.
    • Points routed to a named role. A question about later batches goes to the QA approver while a data-integrity question is ruled by the laboratory head, never by an agent.
  7. 07Author's marks

    Author fixes each finding only after opening its source record

    • Fix proposed, source opened first. The author reads the LIMS row, both revisions and the evaluation before the fix can be applied, so nothing goes in unread.
    • Fixed is not the same as checked. Applied fixes stay marked not re-checked while the saved draft waits for its next read, so no reviewer signs on a promise.
  8. 08Re-check

    New draft re-read by the agents before any signature counts

    • Saving clears the signatures. A new draft wipes any signature given on the old one and needs a fresh read before sign-off opens.
    • The rebuilt section shown in place. The regression, the bound for each market and the conclusion against the new limit now sit in the document itself.
  9. 09Sign-off

    Technical review and QA approval signed with a meaning, each gate shown

    • Gates each signer must clear. The reviewer signs only after the re-check, the lab head's ruling and a read of the new draft; the approver after reading the evaluation.
    • Signatures with name, meaning and time. Each person signs as themselves with a password and what the signature means, as Part 11 and Annex 11 expect.

The reviewer reads, the author fixes, QA signs the meaning

Four jobs share one controlled document, and nobody holds two of them on the same draft.

QC technical reviewers

Check reports and methods against LIMS, specifications and templates, then sign the review

  • Numbers already traced to their rows
  • Citations judged on the result date
  • A read they can steer mid-run
QA document approvers

Approve controlled documents and own the approval queue and its clocks

  • Only re-checked drafts reach approval
  • Clean minor revisions approved after a random sample
  • The client note sent after approval
Document control specialists

Keep templates and review profiles, and raise document change requests

  • Documents citing a superseded revision, listed
  • Rules learned from rulings, waiting for QA
  • Profiles applied only once QA approves
Stability coordinators

Write stability protocols and reports and fix what review finds

  • Each finding with its source
  • A proposed fix to accept or rewrite
  • One re-check instead of a full re-read

A leaver's account opened the 12-month sequence, and the lab head rules on it

QC technical reviewer
Sees
The re-checked draft, the lab head's ruling and any sentence the author wrote
Signs
The technical review: accurate and complete
QC lab head
Sees
The chromatography data system audit trail for the sequence a leaver's account opened
Signs
The data-integrity ruling while the deviation stays open
QA document approver
Sees
The re-checked draft, the evaluation and the challenger point on later batches
Signs
The approval, and the note sent to the client
Document control specialist
Sees
Each document still citing the superseded revision
Signs
The change requests, and proposed review rules
Plant QA head
Sees
A proposed rule and the two rulings it came from
Signs
Review profiles, learned rules and any looser setting

demoSample data, from the working solution.

More screens

Five more approved documents still carried the limit the client had replaced

Past the one report: where the old revision still sits, what the client answered, the queue and its sampled approval, the profiles and the record of it all.

Protocol, worksheet and checklist still cite itFive approved documents cite the replaced client specification, each with where it cites it and why it matters; document control has raised the first change request.
The client answers the note in fieldsThe note gives 24 months for one market and 17.1 months for the other; the client picks an option, and its comment counts only once QA has opened it.
A queue that counts working daysTwenty-five documents waiting at the plant, one late and one due today, each with its stage, whether the agents found anything and its clock.
Nine minor revisions, two read firstThe agents draw a random sample from the nine clean minor revisions; each opens its was-and-now redline, and the set can be approved only once each sample is read.
Two rulings became one ruleThirty-four of 38 document types run on a QA-approved profile; a rule drawn from two rulings went live once document control proposed it and QA signed it in.
Seventy-eight entries for one reportEach mark, opened record, ruling and signature with who, when, why, the signature's meaning and the session, ready to export for an auditor.

demoSample data, from the working solution.

The 24-month claim rested on a section with no statistics in it

A water result from the row above

The report printed 0.32% where LIMS holds 0.34%: the value of the batch printed above it, pasted one row off.

Two citations wrong on their dates

A client specification and a test method had both changed during the study, and the report cited one revision for every time point.

A retest period the data could not carry

With no regression, ICH Q1E allows 18 months from 12; computed, the tighter market's limit is reached at 17.1 months.

A fix that would have rewritten history

The agents' first fix moved earlier results to the new limit; the challenger caught it before the author applied it.

demoSample data, from the working solution.

The rebuilder computes the bound; people choose the retest period

Six agents read in view, in order, each handing on to the next; reviewers sign.

Profile keeper

Picks the approved review profile for the document type and draws samples for bulk approval.

  • Applies no profile or rule before QA approves it
Template matcher

Lays the template over the document, section by section.

  • Never edits the document
Reference checker

Judges each citation by the revision that applied on its result's date, and reads client change notices when told to.

  • Never reads today's revision into past results
Number tracer

Traces each result to LIMS at the reported precision and recomputes computed values.

  • Never shows a value it did not compare as a match
Evaluation rebuilder

Works out a missing evaluation: regression, poolability and the confidence bound against each market's limit.

  • Never chooses a retest period
Challenger

Reads the document as the client's regulatory reviewer and attacks it, along with the agents' own fixes.

  • Leaves a point that needs a person to that person

Questions approvers and reviewers ask about GMP document review

What is GMP document review?

GMP document review is the check a controlled document gets before QA approves it: does it follow its template, are its citations right, and do its numbers match the records they came from. Here agents do that reading in three layers, and the technical reviewer and the QA approver sign.

Which revision should a stability report cite when a specification changes mid-study?

The one that applied on the day each result was made. Earlier results stay judged against the old limit, while later results and the conclusion use the new one. The reference checker applies that rule date by date, and the challenger tests the agents' own fixes against it.

What does ICH Q1E allow without a statistical evaluation?

When long-term data show change and no statistical analysis is done, extrapolation is limited to one and a half times the period covered and no more than six months beyond it. With a regression and a confidence bound, the period can go further. Where a report leaves the evaluation out, the agents compute it and show what each market's limit supports.

Does it replace our document management system?

No. The agents read the document management system, LIMS, the chromatography data system and the quality system, and write nothing back to them. What this adds is the read before anyone signs, with each mark opening the record behind it.

How do other document types get reviewed?

Each type gets a review profile built from its template and governing procedure, proposed by document control and approved by QA. Procedures, work instructions, protocols, validation reports and specifications each run on their own profile, and a type without one is read for references and numbers only.

Who decides when the data and the client's limit disagree?

People. The QA approver signs the report and the note, and the client's regulatory lead replies in fields on how its market will be handled. The agents compute what the data support and never pick the retest period.

Bring one approved report
and the LIMS export behind it.

Send one stability or validation report with its template, the documents it cites and its LIMS results; we will read it in three layers and show what review missed.