Global value dossier authoring where every sentence traces to its evidence
Every GVD sentence linked to its source and checked against the label.
See one case, screen by screen ↓Software that helps a market access and HEOR team write the global value dossier — disease burden, unmet need, clinical, comparative and economic value — from the evidence library. Global Value Dossier Authoring drafts each section with a citation on every sentence, checks every claim against its source and the draft label, and keeps an evidence matrix of every value claim against every source.
How to choose AI solutions for life sciences →
Updated
A global value dossier is written once at headquarters by several writers from an evidence library of 100+ sources — clinical study reports, publications, SLR extraction tables, a network meta-analysis, burden-of-illness studies, guidelines, the draft core data sheet and the cost-effectiveness and budget impact model reports. Over months of drafting, claims drift from their evidence: a week-2 timepoint outside the testing hierarchy, an indirect comparison written as “more effective”, a superseded data cut, a cost offset that rests on a model assumption.
Nobody can see, claim by claim, how strong the evidence is across trials, reviews, real-world studies and models. Affiliates then adapt the GVD for NICE, the German benefit dossier, a US payer dossier and the rest, so every unsupported sentence is copied into each market — and the gap surfaces only when an HTA body or a payer asks.
Estimated split for one section of a global value dossier, by hand and with the solution. The totals follow the team's plan of about 9 days by hand and the 2.1-day target.
Six specialist agents index the evidence, draft the sections, check every claim and rate each claim against each source; the dossier lead decides and reviewers sign.
Talmirex is a monthly self-injected antibody for moderate-to-severe hidradenitis suppurativa, with two Phase 3 trials, LUMEN-1 and LUMEN-2. Its GVD v0.6 has 8 sections and is due as v1.0 on Fri Nov 20. Here is the day the LUMEN-X week-52 report arrives, screen by screen, in the working solution.
Dana's home screen is the evidence matrix: 15 value claims under five pillars down the side, 15 current sources across — clinical trials, reviews, real-world studies, guidance and models — and 31 supported cells. Each row shows the claim's strength and how many sentences, tables and figures in the dossier trace to it. Three challenges from markets are waiting on the right.
“Every sentence, table and figure in the dossier traces to a cell.”
The LUMEN-X week-52 interim report arrived this morning. The Evidence Librarian reads it and marks the week-36 interim superseded, leaving 15 current sources. The Evidence Matrix Mapper then rates the 31 claim-source cells by study design.
A source is marked superseded only when a newer report of the same study exists.
The checker applies 3 claim rules to 15 claims, and 1 fails. M5.3 — “Offsets part of its cost through fewer flare-related emergency visits” — has one clinical trial, one real-world study and one model behind it. The rule asks for two real-world studies, so the row is rated thin, the claim goes to Dana and gap G-1 is updated.
“A cost-offset claim needs at least two real-world studies · 1 real-world study found · the trial shows fewer flares; the model assumes fewer visits.”
Dana opens the US claims cell. The passage is Table 3 of the US claims burden study: 29% of patients had at least one HS-related emergency department visit a year. It shows how often patients need emergency care — not that Talmirex lowers it. The cell also lists what in the dossier is built on it: a sentence in §3 and Table 3-1.
“This is the only real-world study behind the claim, and it measures how often patients need emergency care — not a fall with Talmirex.”
Hannah Price in the UK had asked: “Section 8 says Talmirex offsets part of its cost through fewer emergency visits. What is that based on?” Dana drafts the answer from the row. It cites the US claims study, the LUMEN-1 flare result (24.3% to 9.6%) and the cost-effectiveness model, says surgery is not modelled, and presents the offset as a model assumption until a second real-world study exists. She sends it.
“Answer sent to Hannah Price · every sentence cites its source.”
The section board shows 61 of about 74 pages drafted, 386 of 390 claims checked and clear against the source and the draft label, 4 findings to decide and 3 of 8 sections signed. Each section has its writing agent, its owner and reviewer, its share of writer edits and its due date; section 8, economic value, is ready to draft from the model reports.
The Clinical Value Writer's draft of section 6 has three findings. The draft says skin-pain relief within 2 weeks, but week 2 (24.1% versus 17.9%, nominal p = 0.08) sat outside the testing hierarchy; the source supports week 4 (38.6% versus 21.0%, p < 0.001). Adolescent data (12–17, n = 38) is outside the expected label, which covers adults only, so Omar Haddad decides whether it moves to the payer-only appendix or leaves the dossier. And one sentence cites the superseded week-36 cut (74.0%) where the week-52 report says 71.3%.
“The source supports pain relief from week 4, not week 2.”
Section 7 says Talmirex was more effective than anti-TNF therapy. The network meta-analysis gives an odds ratio of 1.21 with a 95% credible interval of 0.88–1.67, which crosses 1. The checker proposes “comparable” to anti-IL-17A and “numerically higher” than anti-TNF, a difference that was not conclusive — and links the sentence to its matrix row, M4.1.
The checker cannot change text — it proposes, and a person decides.
Five gaps are open across four sections. Sam owns the two high-severity ones in section 8: G-1, the cost-offset claim resting on one real-world study, due Oct 23, and G-4, utility values mapped from DLQI that are not validated for HS. Germany's Hurley III subgroup missing from the indirect comparison is G-5, with Katrin Vogel.
Section 4, current management and unmet need, is ready to sign: 64 of 64 claims clear and no findings open. Dr. Maya Chen signs it for scientific accuracy and balance; her name, the date and time and the meaning of the signature are bound to the section version. A section reaches this point only once its findings are decided.
“Your name, the date and time, and the meaning below are bound to section version 0.6 and recorded in the audit trail.”
The same dossier, seen by the five people who carry it — what their weeks looked like, and what they look like now.
Six specialist agents index the evidence, draft the sections, check every claim and rate each claim against each source; the dossier lead decides and reviewers sign.
Reads every source in the evidence library, classifies it by study, design and data cut, and indexes each passage with page and table number so a sentence can cite it. When a newer report of the same study arrives, it marks the older one superseded and lists every sentence that cites it.
Drafts sections 2–4 — disease background, burden of illness, current management and unmet need — against the GVD template headings, preferring systematic reviews and meta-analyses over single studies.
Drafts sections 5–7 — product profile, clinical value and comparative effectiveness — from the study reports, publications and the network meta-analysis, with endpoint, timepoint, arm sizes, effect and uncertainty as the reports state them.
Drafts section 8 — cost-effectiveness and budget impact — from the model technical reports: structure, perspective, time horizon, comparators, base case, sensitivity analyses and key assumptions.
Checks every sentence against the passage it cites and the draft core data sheet — numbers, timepoint and testing hierarchy, comparative wording, label, data cut — applies the claim rules to the matrix, and raises a finding with the evidence and a suggested rewording.
Keeps the evidence matrix and value message map current: rates every value claim against every source — strong, moderate, low or unsupported by study design — and flags a claim whose evidence is too thin for its kind.
Decides findings; section owners e-sign. The agents propose; a named person decides.
Ask in plain words about a claim, a source, a section or a market. Every answer cites the passage it came from.
What does the evidence say about pain relief?
Skin-pain relief (NRS30) is supported from week 4: 38.6% versus 21.0% on placebo, p < 0.001, in the testing hierarchy. Week 2 (24.1% versus 17.9%, p = 0.08) sat outside the hierarchy and did not reach significance. Section 6 still says “within 2 weeks” — the finding is waiting for you.
Which claims are still unsupported?
Two claims lack the evidence they need: G-1, the cost-offset claim, and G-4, utility values mapped from DLQI that are not validated for HS — both with Sam Patel. G-1 is the one to watch: it rests on one real-world study, and the model only assumes fewer emergency visits.
What does Germany need from the dossier?
Katrin Vogel adapts sections 2, 3, 6 and 7 into the benefit dossier due at launch, in Sep 2027. She asked for patient-relevant endpoints and the comparison in the Hurley III subgroup, which the network meta-analysis does not cover yet (gap G-5).
How do we compare with other biologics?
Against placebo the odds ratio is 2.79 (2.17–3.58). Against anti-TNF it is 1.21 (0.88–1.67) and against anti-IL-17A 0.97 (0.71–1.33); both intervals cross 1, so the dossier can say comparable, and numerically higher than anti-TNF — not “more effective”.
13 screens from the working solution, on its sample data. Pick one to see it large.
The template, styles, claim rules and signers are settings. Rules that protect compliance are locked.
| Setting | Default | Choose from |
|---|---|---|
| Template | GVD template v4 | GVD template v4 · GVD template v3 · Payer dossier (AMCP format) |
| Citation style | Vancouver | Vancouver · AMA · Harvard |
| Writing style | Scientific — HTA reviewers | Scientific — HTA reviewers · Plain — affiliate summaries |
| Every sentence cites a passage in the evidence library | On · locked | Turning it off needs the legal & regulatory reviewer's approval |
| Numbers must match the cited passage exactly | On · locked | Effect sizes, intervals, timepoints and arm sizes compared character by character |
| Comparative wording must fit the evidence | On | On · Off — an indirect comparison whose interval crosses 1 supports “comparable”, never “better” |
| A cost-offset claim needs at least two real-world studies | On | On · Off — fewer than two and the claim is rated thin; the dossier lead decides |
| Declare AI assistance in every export | On | On · Off — adds a statement to the title page |
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 dossier · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (an ISPOR Europe 2025 poster on generative AI for global value dossiers — 70–80% time saving with a human-plus-AI workflow versus human-only). Talmirex, the LUMEN trials, the studies and the people named on this page are fictional — characters and sample data in the working solution.
Software that helps a market access and HEOR team write the global value dossier — disease burden, unmet need, clinical, comparative and economic value — from the evidence library. Global Value Dossier Authoring drafts each section with a citation on every sentence, checks every claim against its source and the draft label, and keeps an evidence matrix of every value claim against every source.
The Evidence Matrix Mapper rates every value claim against every current source by study design: controlled trials and the label rate strong; extensions, reviews, indirect comparisons and real-world studies moderate; models low. The claim rules run on the matrix, so a cost-offset claim with fewer than two real-world studies is rated thin, turns amber and raises a gap for the dossier lead to decide.
The Evidence Librarian reads the new report, marks the older report of the same study superseded and lists every sentence that cites it. The matrix is re-mapped, and the Claim & Label Checker flags each citing sentence with the new figure for the writer to decide.
Yes. Comparative wording must fit the evidence — an indirect comparison whose interval crosses 1 supports “comparable”, never “better”. Data outside the expected label, such as an adolescent sub-study when the draft core data sheet covers adults only, goes to the payer-only appendix with its statement, and the legal & regulatory reviewer decides.
A challenge from a market is linked to the claim it questions. Its row opens the passages behind the claim and every sentence, table and figure built on them, and drafts an answer in which every sentence is cited, for the dossier lead to send.
Yes. The agents draft, check and rate: the checker only proposes changes, and no agent edits a source or rewrites a value claim. The medical writer decides wording, the dossier lead decides thin claims, the legal & regulatory reviewer decides outside-label content, and sections are e-signed with meaning once their findings are decided.
Yes. The template (GVD v4, v3 or the AMCP payer dossier format), citation style, writing style, claim rules and signers per section are settings. Rules that protect compliance — every sentence cited, numbers matching exactly, outside-label content in the payer-only appendix — are locked.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your GVD template, value message map and evidence library — and it goes live once every quality gate has passed. We will walk you through it on a section of your own dossier first.
Every rebate invoice line checked, every dispute backed by its clause.
1-min demoPayer Contract ManagementEvery payer contract turned into cited terms, every deadline tracked
1-min demoPayer Coverage Policy TrackingEvery payer policy change caught within a day of posting, checked against the contract.
1-min demoWe'll run Global Value Dossier Authoring on a section of your own GVD and its evidence library.