Clinical study report authoring with every number linked to its table cell — and re-checked on every re-cut
Results drafted within days of database lock, linked to tables.
See one case, screen by screen ↓The database locks, biostatistics delivers hundreds of tables, listings and figures, and the clock on the clinical study report starts. The medical writer turns them into ICH E3 results sections — disposition, baseline, efficacy, safety, conclusions — and writes a narrative for every death, serious adverse event, discontinuation and event of special interest. Much of that time is not writing: it is copying numbers from RTF tables into the draft by hand, then a QC pass that checks every number against the tables again.
Then programming QC re-runs an output. A response flag was derived on a rounded value, one participant moves from responder to non-responder, and four efficacy outputs come back with new numbers. Nobody can say which sentences quoted the changed cells, so the number QC starts over — and a stale number that slips into the submitted report becomes a regulatory question and a correction.
Estimated weeks of writing per report, by hand and with the solution, as the working solution's dashboard shows them on its sample data.
Five specialist agents draft, link and re-check every number; a biostatistician and a medical writer sign.
VLT-302 is a Phase 3 trial of Veltrimab 150 mg against placebo (2:1, N=642) in moderate-to-severe plaque psoriasis, with co-primary endpoints PASI 75 and sPGA 0/1 at Week 16. Here is its report, screen by screen, in the working solution.
Biostatistics delivers 412 outputs from the locked database — 186 tables, 142 listings and 84 figures. The output reader reads them in 49 minutes and indexes 9,840 cells, each with its row, column, population and program run. Every output shows the sections that quote it and how many of its numbers they use.
“Tables, listings and figures from the VLT-302 locked database — the only place a number in the report may come from.”
The narrative writer drafts 53 narratives in 47 minutes — serious adverse events including one death, 9 participants who stopped because of an adverse event, and 6 events of special interest. Every fact is checked against the listings. Two facts are missing at the site, so it raises two queries instead of guessing: a discharge date for 302-0207-0119 and a CT pulmonary angiogram report for 302-0618-0026.
Its rule: “Missing facts become queries, never guesses.”
Maya's screen is the report: the ICH E3 section list on the left with each section's status, the draft in the centre set like a report page, and the source tables it quotes on the right. 5 of 11 sections are approved. Section 11.4.1 Primary efficacy endpoints, drafted by the results writer, is in review with Omar Haddad.
“19 numbers · all 19 match their table cell — click any number to open it beside the text.”
Maya clicks “62.4%”. Table 14.2.1.1 opens beside the text at row PASI 75 responders, column Veltrimab 150 mg (N=428), and a dashed line joins the number to the cell: 267 (62.4). The footer names the source — ADEFF, program t_eff_prim.sas, run 02OCT2026 18:42 — and the chips below list all 10 numbers the section quotes from this table.
“Exactly as delivered: 267 (62.4) · the text says 62.4%.”
Independent programming QC found that the PASI 75 response flag was set on a rounded improvement. Re-derived on the unrounded value per SAP Section 9.2, participant 302-0915-0137 (74.6% improvement) is now a non-responder, and Tables 14.2.1.1–14.2.1.3 and Figure 14.2.1.4 are re-run. A banner appears on the section; Maya clicks “Check the text” and the number verifier works through it step by step: 8 of 38 cells changed, all in the Veltrimab column.
Steps on screen: read the re-cut outputs · compare cell by cell with delivery 1 · find every sentence that quotes a changed cell · re-check the other numbers · flag for the writer.
Every sentence that quotes a changed cell is shaded, the old value struck and the new one beside it: PASI 75 62.4% → 62.1% (266/428), the difference 57.3 → 57.0, the 95% CI 51.9, 62.6 → 51.7, 62.3, multiple imputation 63.8% → 63.6%, per-protocol 65.0% → 64.8%. The table shows the struck values in the changed cells. The other 17 numbers that quote the re-cut outputs still match.
“Old and new value side by side · nothing changes until you accept.”
Section 11.4.3 Subgroup analyses was signed by Omar Haddad at 09:40 that morning. Two of its six numbers now quote changed cells: biologic-naive 64.5% → 64.2%, and weight 100 kg or less 65.0% → 64.7%. Section 13 Discussion and overall conclusions quotes 62.4% too, and is flagged with them.
The number check lists all 79 numbers in the VLT-302 results with the section, the text, the source cell, the cell value, the delivery and the check. Filtered to “Changed by re-cut”, it shows the 8 flagged numbers; 25 come from the re-cut outputs and 71 still match. Approval and export stay held while anything is flagged.
“8 numbers flagged by the 10:44 re-cut.”
Maya clicks “Update 5 sentences”. Before anything changes, the update lists each section with its status and every replacement: 11.4.1 goes from In review to To review, 11.4.3 from Approved to To review, and 13 stays in To review. She confirms, and the results writer replaces each flagged number with the re-cut value.
“Sections that were in review or approved go back to review and are signed again.”
The text now quotes the re-cut, all 6 numbers in 11.4.3 match their cells, and the section is marked “signature withdrawn by the re-cut”. The report shows 4 of 11 approved until Omar signs 11.4.3 again. No stale number can be approved or exported.
“Text updated to the re-cut · 8 numbers changed · Omar Haddad signs 11.4.3 again · the change is version 0.3.”
The same report, seen by the five people who write, check and sign it — what their weeks looked like, and what they look like now.
Five specialist agents draft, link and re-check every number; a biostatistician and a medical writer sign.
Reads every table, listing and figure from the locked database and turns each one into addressable cells — row, column, population and program run — and detects re-issues and re-cuts.
Drafts the ICH E3 results sections — disposition, baseline, efficacy and safety — in the report template, following the SAP for endpoint names, populations and methods, and proposes new wording after a re-cut.
Writes patient narratives for deaths, serious events, discontinuations and events of special interest from the participant listings and SAE eCRF pages, with the ICH E3 12.3.2 elements.
Checks every number in every section and narrative against its linked cell — value, rounding, population and column — compares each new delivery cell by cell and flags the sentences that quote a changed value.
Confirms every required ICH E3 heading is present, every “Table x” and “Section y” cross-reference resolves, and the house style is followed.
Sign efficacy and narratives. The agents propose; a named person decides.
Ask about any number, section, narrative or re-cut in the report — or say what to change. Every answer points to its source.
Where does 62.4% come from?
62.4% is the PASI 75 response at Week 16 with Veltrimab 150 mg — 267 of 428 participants. It comes from Table 14.2.1.1, row “PASI 75 responders”, column “Veltrimab 150 mg (N=428)”, full analysis set with missing data counted as non-response. Program t_eff_prim.sas, run Oct 2 at 18:42. It is quoted in sections 11.4.1 and 13.
Which narratives are waiting on a site?
Two: 302-0207-0119 (acute cholecystitis — discharge date missing) and 302-0618-0026 (pulmonary embolism — CT report not yet received). Both queries went out on Oct 3; the narrative writer will re-draft as soon as the site replies.
Add a rule: flag any narrative without a hospital discharge date
Done. Every serious-event narrative that involves a hospital stay is now checked: if the discharge date is not in the listings or the SAE eCRF page, the narrative is flagged and a query is drafted for the site. It applies to all 38 serious-event narratives — 1 is flagged today (302-0207-0119).
Summarize safety for the medical monitor
To Week 16, adverse events were reported in 57.9% of Veltrimab and 52.3% of placebo participants. Serious events: 6.1% versus 5.6%; serious infections 1.4% versus 0.9%. One death — a fatal myocardial infarction on Day 97, adjudicated as cardiovascular death and assessed as not related. Injection site erythema was the only event clearly more frequent with Veltrimab (3.3% versus 0.9%).
13 screens from the working solution, on its sample data. Pick one to see it large.
The template the agents write into, how numbers are quoted, who signs each part and the rules every draft is checked against are settings.
| Setting | Default | Choose from |
|---|---|---|
| Report template | ICH E3 house template v4.2 | ICH E3 house template v4.2 · ICH E3 abbreviated report v2.0 · Phase 1 report template v1.3 |
| Percentages | As in the table (one decimal) | As in the table (one decimal) · Whole numbers in the summary only |
| Writing style | House style guide 2026 | House style guide 2026 · Journal style (AMA 11th) |
| Efficacy sections signed by | Omar Haddad · Biostatistics | Omar Haddad or Dana Okafor (Biostatistics) |
| Safety sections signed by | Dr. Sam Patel · Medical monitor | Dr. Sam Patel (medical monitor) or Dr. Maya Chen (medical writing) |
| Final report signed by | Lena Ortiz · Clinical lead | Lena Ortiz (clinical lead) or Dr. Maya Chen (medical writing) |
| Block approval when a number has no source cell | On | On · Off |
| Send safety sections to the medical monitor before medical writing | Off | On · Off |
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 study data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · by hand = the same team's last six reports, in the solution's sample data · report structure follows ICH E3. People, studies and products named on this page are fictional — characters and sample data in the working solution.
Software that helps medical writers turn a trial's locked tables, listings and figures into the ICH E3 clinical study report. Clinical Study Report Authoring drafts the results sections and patient narratives with five agents, links every number to its table cell and every narrative fact to its listing row, and leaves the decisions and signatures to people.
Every delivery is kept and compared cell by cell with the last. When an output is re-run, the number verifier finds every sentence that quotes a changed cell and flags it with the old and new value side by side. Approval and export are held until the writer updates the text, and an approved section that changes goes back to its signer.
Yes. The narrative writer drafts narratives for deaths, serious adverse events, discontinuations and events of special interest from the participant listings and SAE eCRF pages, with the ICH E3 12.3.2 elements, and cites every fact to its source. When a fact is missing it drafts a query to the site instead of guessing.
It is built not to. The results writer writes only numbers it can link to a cell in the locked outputs, and the output reader stores every value exactly as printed. The number verifier checks each number's value, rounding, population and column, and a section cannot be approved while any number lacks a source cell or does not match it.
Yes. The agents draft, check and propose; they never approve or sign. Efficacy sections are signed by biostatistics, safety sections by the medical monitor, narratives by medical writing and the final report by the clinical lead, each with an electronic signature and its meaning.
Yes. The report template, how percentages are rounded, the writing style, who signs efficacy, safety, narratives and the final report, and the rules every draft is checked against are settings.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your CSR template, style guide, SAP and a past study's outputs — and it goes live once every quality gate has passed. We will walk you through it on one of your own studies first.
A checked first-draft protocol in days, in your approved language.
1-min demoeTMF Quality CheckAn inspection-ready trial master file every day, each site scored on what is due today.
1-min demoProtocol Design OptimizationAvoidable amendments fixed before the committee meets, each flag tied to a past amendment.
1-min demoWe'll run Clinical Study Report Authoring on the outputs and template of one of your own completed studies.