CTD Module 2 summaries drafted in weeks, every statement cited to its source page
Module 2 summaries drafted in weeks, every statement cited to its source page.
See one case, screen by screen ↓Module 2 condenses thousands of pages — the CSRs, the integrated summaries of efficacy and safety, the Module 3 specifications and stability data, the Module 4 toxicology reports — into a clinical overview, a clinical summary and a quality overall summary. Every statement has to trace to a page in Modules 3 to 5, and writing it by hand takes months.
Most of that time is not writing. It goes on pulling the right figure from the right table, checking each citation against its page, and reconciling the same number stated in 2.5, 2.7.3 and 2.7.6. One slip — a full-analysis-set count called “randomized”, an exposure figure carried over from an earlier safety update, a development limit quoted instead of the registered one — and the reviewer finds a number that disagrees with itself.
Estimated split for one section such as 2.7.3, by hand and with the solution. The totals follow the 14 → 3 weeks shown for 2.7.3 on the working solution’s dashboard.
Six specialist agents read the sources, draft each section, verify every citation and check the numbers agree; the section owner signs.
The Veltrimab BLA is due on December 14. 2.7.3 Summary of Clinical Efficacy is on Dr. Maya Chen’s desk — with one number that does not agree. Here is what happens next, screen by screen, in the working solution.
Maya opens the solution: 6 of 14 Module 2 sections approved, 69 days to the December 14 submission. 2.7.3 is ready for her signature with one consistency flag. Dr. Omar Haddad has asked her to review 2.5.4, and a new version of CSR VLT-303 — the Week 104 interim — touches 14 statements in 2.7.3.5 and 2.5.4.
“VLT-302 sample size disagrees — 2.5.4 says 824 randomized · 2.7.3.2 says 818 · CSR Table 14.1.1.1”
On the dossier, the clinical summary writer finishes 2.7.3.5 Persistence of efficacy. The citation verifier then opens every one of the 214 cited pages, and the consistency agent compares 2.7.3 with 2.5.4 and 2.7.6. Every section of the BLA sits in its CTD place, with its owner, stage, statement count and share verified.
“2.7.3 drafted — 214 statements, every one cited · 213 verified · 1 consistency flag”
The draft follows the 2.7.3 headings, from 2.7.3.1 Background and overview of clinical efficacy to 2.7.3.5 Persistence of efficacy and tolerance effects. Each statement carries numbered citation chips and a tick once verified. One in 2.7.3.2 does not: “In VLT-302, 818 patients were randomized 2:1 to veltrimab (n=546) or placebo (n=272).”
“Says 818 randomized. The cited table shows 824 randomized and 818 in the full analysis set — and 2.5.4 states 824.”
Maya clicks the citation. CSR VLT-302 opens at page 112 of 2,140 — Table 14.1.1.1 Subject Disposition. Randomized: 549 and 275, total 824. Received at least one dose (full analysis set): 546 and 272, total 818. The draft gave the full-analysis-set count and called it randomized.
She presses Approve section anyway. The solution stops her: 2.7.3 cannot be signed while a statement disagrees with its source. It offers to open the flag.
“Agents can propose the fix; you decide.”
312 facts tracked, 309 matching everywhere, 3 open flags. The VLT-302 count sits next to two others: patients exposed to veltrimab (1,942 in 2.5.5, 1,947 in 2.7.4.1) and the drug product protein concentration (100 ± 10 mg/mL in 2.3.P.5, 95–105 mg/mL in 3.2.P.5.1). For each, the board shows where it is stated and the source of truth.
“2.7.3.2 uses the full-analysis-set count (818) but calls it ‘randomized’. The final CSR and 2.5.4 both give 824 randomized.”
The agent opens Table 14.1.1.1, compares the populations, and proposes a tracked change to 2.7.3.2: 824 randomized (549 and 275), with 818 receiving at least one dose and forming the full analysis set. It re-checks 2.5.4 and 2.7.6 — both already say 824, so no change there.
“You decide — nothing changes until you apply it.”
Maya applies the change to 2.7.3. The statement goes green, marked “Edited by the consistency agent · applied by Dr. Maya Chen · re-verified against the cited page”. The section moves to version 0.7, with 214 of 214 statements verified.
“Verified. The citation verifier opened the cited page; every figure in this statement matches.”
The signing screen sums up the section: 214 statements, every one cited; 214 of 214 verified against source pages; all figures match 2.5.4, 2.7.6 and the sources; one tracked change, in 2.7.3.2. She signs as section owner — meaning “Approved as section owner”, with her password.
“Your signature is recorded with date, time and meaning. The section is locked as version 1.0 and goes to the eCTD build.”
2.7.3 is approved and locked as version 1.0. The history shows every step in order — each agent’s draft and verification, the flag, Maya’s accepted correction and her signature. Any change from here opens version 1.1, with a reason.
The same submission, seen by four of the people who write and sign it — what their weeks looked like, and what they look like now.
Six specialist agents read the sources, draft each section, verify every citation and check the numbers agree; the section owner signs.
Reads study reports and Module 3 documents page by page, and records every key fact — sample sizes, endpoint results, exposure, doses, specification limits, shelf life — with its population and exact page, table or figure.
Drafts 2.5 Clinical Overview and the 2.7 Clinical Summary from the CSRs and integrated summaries, in the order set by ICH M4E(R2) and the company template.
Drafts 2.4 Nonclinical Overview and the 2.6 written and tabulated summaries from the Module 4 study reports, building the tabulated summaries from the report tables.
Drafts the 2.3 Quality Overall Summary — drug substance, drug product and appendices — summarising each 3.2.S and 3.2.P section under its matching 2.3 heading.
Opens every cited page and checks the statement says what the page says — every number, the population and the direction of the claim — then marks it verified, to check, or disagrees.
Finds every figure stated in more than one section and checks they agree with each other and with the source of truth; when they do not, it opens a flag with the values, the pages and a proposed correction.
Approves and signs. The agents propose; a named person decides.
Section owners and writers can ask in plain words — or tell it what to change. Every answer comes with its page.
Where else do we state the VLT-302 sample size?
Three places. 2.5.4 and 2.7.6 say 824 randomized, which matches the final CSR. 2.7.3.2 says 818 randomized — that is the full analysis set, not the randomized count. The flag is open on the consistency board.
Which statements still need attention?
Four across the dossier: three numbers that disagree with their source — the 2.7.3.2 sample size, the 2.5.5 exposure and the 2.3.P.5 protein concentration — and one wording check in 2.5.4 on body-weight subgroups.
Does response differ by body weight?
Yes. PASI 75 response was seen in every subgroup but was lower above 100 kg: 61.0% versus 75.9%. Population PK explains it — steady-state exposure is 32% lower above 100 kg. 2.5.4 calls the response “consistent” across subgroups; the verifier suggests “seen in all subgroups, with a lower rate above 100 kg”.
Add a rule: cite final study reports, never topline memos
Added to the citation rules. The citation verifier now rejects any statement that cites a topline memo or draft report. Every statement in the dossier already cites a final or signed version.
13 screens from the working solution, on its sample data. Pick one to see it large.
Owners, citation rules and house style are settings. Headings and order follow ICH M4 and your company style guide.
| Setting | Default | Choose from |
|---|---|---|
| Who signs each section | Its named section owner | Any person on the team, per section |
| Every statement needs a source page | On | On · Off |
| Numbers must match the key facts register | On — before owner review | On · Off |
| Flag wording that overstates a result | On | On · Off |
| Re-check when a source gets a new version | On — as tracked changes | On · Off |
| How small p-values are written | p<0.001 | p<0.001 · p<.001 · exact |
| Decimal places in results text | 1 | 0 · 1 · 2 |
| Your own citation rules | Your rules | Add in plain words |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample Veltrimab BLA data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · Module numbers follow the ICH Common Technical Document (M4). People, products and studies named on this page are characters and sample data in the working solution.
It drafts the CTD Module 2 summaries — 2.3 Quality Overall Summary, 2.4 and 2.6 nonclinical, 2.5 Clinical Overview and the 2.7 Clinical Summary — from your Module 3, 4 and 5 documents. Every statement is cited to its source page, every citation is checked, and figures are kept the same across sections. The section owner reviews and signs.
The citation verifier opens every cited page and checks the numbers, the population and the direction of the claim. A statement is marked verified, to check if the wording overstates or understates the result, or disagrees if a number or population differs.
The source reader records every key fact with its page in a key facts register, and the writers take numbers only from it. The consistency agent tracks each figure everywhere it is stated and opens a flag, with a proposed correction, when two places disagree or one disagrees with the final report.
The new version is read and indexed page by page, the changed facts are found, and every statement that cites them gets a tracked change proposed for its owner. In the working solution, the Week 104 interim of CSR VLT-303 touched 14 statements in 2.7.3.5 and 2.5.4.
Yes. Agents draft, cite, verify and propose — they never sign. Nothing changes until a person applies it, only the named section owner can approve, and approval is an electronic signature with its meaning, recorded with date and time.
Yes. Headings and order follow ICH M4 and your company style guide, and settings cover how p-values are written, decimal places in results text, who owns each section and your own citation rules.
Yes. The quality summary writer drafts 2.3 from Module 3, quoting specification limits and shelf life exactly as registered. The nonclinical summary writer drafts 2.4 and 2.6 from the Module 4 reports, quoting doses and NOAELs exactly as printed.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your Module 2 template, style guide and a set of study reports — and it goes live once every quality gate has passed. We will walk you through it on one of your own sections first.
Every agency change reaches the right team within hours, translated and cited.
1-min filmAgency ResponsesEvery agency question gets a cited first draft within hours, consistent with every past answer.
1-min filmLabel Change CascadeEvery core data sheet change finds each local label it touches and puts every market on a clock from day one.
1-min filmWe’ll draft a Module 2 section from your own study reports and template.