Assemble the ICH E3 CSR
without the blank page
BuildCSR assembles the ICH E3 Clinical Study Report from your TLFs, with AI-drafted section narratives from metadata only, a medical-writer review loop, and cited provenance to every supporting table.
The last mile to submission is the slowest
The CSR is the final document before filing, yet every section starts from a blank page, and linking prose to the right tables is manual. The work that gates the submission has the least leverage, so the most senior writers spend their hours restating what the outputs already show.
× Blank-page drafting is slow
Writers spend hours restating what the tables already say before they can add the judgment that only a person can.
× Citations are stitched by hand
Tying each statement to its supporting TLF is manual and easy to get wrong, and a broken link is a finding waiting to happen.
× Status is hard to see
Which sections are drafted, in review, or approved lives in a tracker kept separate from the document itself.
Sections as first-class units, drafted from metadata, approved by writers
BuildCSR treats each ICH E3 section as a workflow unit with its own state. An agent drafts the narrative from section metadata and TLF titles, never patient data, and the medical writer edits and approves. Every section cites its tables.
Sections carry their E3 numbers and structure, so the document is built the way reviewers expect to read it.
The agent proposes prose, the writer refines and signs, and the human owns the text that ships.
Every section links to the TLF titles that support it, captured as you write rather than reconstructed later.
The CSR as a tree of E3 sections
- Sections such as 11.4.1 Demographics are first-class units, each one named and numbered the way ICH E3 lays them out.
- Each section moves through draft, in review, and approved, and you see its state next to its title.
Why it mattersProgress is visible at the section level, and nothing ships unapproved, so the document and its status are the same thing.
Draft, in review, approved, on the record
- Each section follows a state machine with a human gate into approved, and the gate cannot be skipped.
- Moving a section forward writes an event, so the transition itself is part of the trail.
Why it mattersThe status of the document is unambiguous and audited, captured in the section itself rather than tracked on the side.
A first draft from the section's metadata
- The agent drafts from the section number, its title, and the titles of its TLFs, and it never sees patient rows.
- The proposed prose lands in the editor as a starting point the writer can keep, rewrite, or discard.
Why it mattersWriters start from a structured draft and spend their time on judgment, not on restating what the tables already report.
Every statement tied to its table
- Each narrative statement links to the supporting TLF titles, captured as the section is built.
- The citation map travels with the section, so the link is never re-derived from memory.
Why it mattersA reviewer can confirm any claim against the exact table behind it, with no manual cross-checking of the prose against the outputs.
| Statement | Cited TLF | Link |
|---|---|---|
| 312 subjects randomized across two arms | Table 14.1.1 Demographics | linked |
| Groups comparable at baseline | Table 14.1.1 Demographics | linked |
| Disposition by treatment arm | Table 14.1.2 Disposition | linked |
| Exposure summary | Table 14.1.3 Exposure | pending |
The writer edits and signs
- The writer accepts, rewrites, or rejects the draft, then approves the section, and the human owns the final text.
- Every edit and the approval are recorded against the section as the writer works.
Why it mattersAI removes the blank page while the writer keeps authorship and accountability, so the prose that ships is the writer's own.
Section templates you can reuse
- Start sections from a library of E3 templates and from your own approved patterns.
- A template seeds the structure and the expected citations, so each new study begins further along.
Why it mattersConsistent structure across studies, with less setup each time, so house style is built in rather than re-typed.
Every draft and sign-off, on the record
- Proposals, edits, and approvals are append-only with actor, server time, and a content hash.
- The ledger is scoped to your tenant, and updates or deletes to it are rejected even for an admin.
Why it mattersAn inspector sees how each section reached approved, and by whom, captured as the work happens rather than reconstructed under deadline.
| When | Actor | Event | Hash |
|---|---|---|---|
| 14:22:07 | a.duval | 11.4.1 approved | 9f2c…d1 |
| 14:10:33 | a.duval | writer edit applied | 3b71…8e |
| 13:58:46 | system | narrative agent proposed | c40a…55 |
| 13:41:12 | a.duval | section opened | 7d18…02 |
Built for the document that gates your filing
| Capability | Word + copy/paste | Generic AI writing | Document management | BuildCSR |
|---|---|---|---|---|
| ICH E3 structure enforced | ||||
| AI draft from metadata only, no patient data to the model | ||||
| Cited provenance to supporting TLFs | ||||
| Section-level state machine and sign-off | ||||
| Narrative and supporting tables in one workflow | ||||
| Append-only Part 11 audit trail |
One assembly, six points of view
Medical writer
Start from a structured draft instead of a blank page, edit and own the prose, and sign each section into approved.
Regulatory affairs
Read a document built to ICH E3 structure, with every claim cited to the table behind it before it leaves for the agency.
Lead statistical programmer
See your TLF titles flow straight into the narratives that cite them, with no re-keying of table names into prose.
Biostatistician
Confirm that each narrative statement is tied to the correct supporting output, so the science is described accurately.
Submission lead
Watch sections move to approved on one board, and ship a CSR whose status and citations are part of the document.
QA / auditor
A read-only, tamper-evident trail of every proposal, edit, and sign-off, ready for inspection on demand.
Regulatory rigor, by construction
The controls auditors look for are inherent in the system, not bolted on after the fact.
21 CFR Part 11
Gated sections with human sign-off, server-time stamping, and access control on every move into approved.
AI audit trail
Every narrative proposal is hashed and linked to the user action that accepted, edited, or rejected it.
Metadata-only AI access
The agent sees section numbers, titles, and TLF lists, never patient rows, so no patient data reaches the model.
Tenant isolation
Row-level security separates every tenant's sections and audit at the database, verified through the request path.
Assemble one of your CSR sections in a 30-day pilot
See BuildCSR draft narratives from your TLF metadata, cite every table, and move sections to approved in days. Writers own the text, and the audit trail is yours to keep.
- We load a study's TLFs into a private workspace
- Your writers draft, cite, and approve a set of sections
- You keep the narratives and the citation map
- A readout on drafting time saved
info@the-bdkm.com · BDKM LLC · Back to Life Sciences