Clinical study document search that answers from the newest approved version — and flags the older ones
Any study question answered in seconds, from the newest approved version.
See one case, screen by screen ↓Search over a study's own files — protocol and amendments, plans, manuals, decision logs, minutes, trackers — and its CTMS, EDC and blinded IRT data, answered in plain words. Clinical Study Document Search returns each answer as a fact card: the value, the date it is true as of, and the document, version and section it came from, with the supporting passages.
How to choose AI solutions for life sciences →
Updated
A Phase 3 study runs on more than a thousand documents in many versions — protocol and amendments, statistical analysis plan, monitoring, pharmacy and laboratory plans, decision logs, minutes, trackers — plus daily CTMS and EDC extracts. A question from a site or a CRA usually crosses them: the rule is in one document, the number in EDC, the status in a tracker. Answering it by hand takes 30 to 60 minutes, and the study manager is the study's human index.
The expensive failure is the copy found first. Three in four protocols are amended, and every amendment leaves superseded numbers behind in synopses, site guides, checklists and trackers: a sample size of 600 after the analysis plan moved it to 640, a ± 5-day window printed on visit cards after the protocol narrowed it to ± 3, a 60-minute temperature rule on a pharmacy card after the manual said 30. Nobody owns the question “is any other document now out of date?” — it is found by accident, as a deviation or an inspection finding.
Estimated split of one study manager's week spent answering cross-document study questions, by hand and with the solution — based on the study team's own estimate of 30–60 minutes per question.
Six specialist agents route, read, query, date every version, answer and plan corrections; the study manager decides.
VLT-302 is a Phase 3 study of veltrimab in plaque psoriasis: 88 sites in 11 countries, Protocol v4.0 after Amendment 3. On Wednesday 7 October, Lena Ortiz, the study manager, has questions waiting. Here is what happens, screen by screen, in the working solution.
The search reads 1,284 study documents, the CTMS extract from 06:00, EDC from 05:30, blinded IRT and 4 trackers and logs — blinded data only, within her site access. Her first answer is already on screen: 67 participants still to re-consent for Amendment 3, at 7 sites, 14 overdue at 3 sites, as of today 05:30. The CRAs' tracker says 64; the gap is explained.
“The gap is Site 214: three participants signed v5.0 on paper on 2 Oct. The tracker has them; EDC does not yet. The answer uses EDC until they are entered.”
Lena types the question. The question router sends it to the documents of VLT-302, the document reader finds 9 passages and 4 statements of the fact, the data analyst reads the enrolment extract, and the version checker dates every statement before the answer writer starts. Each step shows as it runs.
“Finding every statement of this in VLT-302's files…”
The answer leads with the number: 640 participants to randomise, 3:1 — 480 veltrimab, 160 placebo — as of 18 Sep 2026, from Statistical analysis plan v3.0, section 9.2, page 21. 512 are randomised so far, 80% of the target. Under the card, every approved statement of the fact sits on one line, oldest to newest, with the passage behind each.
Timeline: Analysis plan v2.0 · 600 · superseded → Protocol v4.0 synopsis · 600 · out of date → Decision D-024 · 640 · agrees → Analysis plan v3.0 · 640 · used.
Protocol v4.0 was approved on 14 Aug, 35 days before analysis plan v3.0. The blinded sample-size review of 11 Sep found pooled discontinuation of 11.8% — 57 of 483 randomised by 31 Aug — and decision D-024 on 16 Sep raised the target to 640, with the synopsis correction put in Amendment 4. One click opens the older passage at page 3, marked out of date.
The version checker never answers from a superseded version, never hides a disagreement and never edits a source document — it flags it for its owner.
From the card, Lena clicks “Add to the Amendment 4 list”. A follow-up is created for her — add the synopsis sample size (600 → 640) to the Amendment 4 change list, all sites, due Oct 21 — and traced back to the question that raised it. The statement moves to “Correction asked”.
“Follow-up created for Lena Ortiz — due Oct 21”
Each morning, after the TMF index and the extracts arrive, every document is re-checked: today 1,284 documents, 6 new since yesterday. Five statements are listed with the fact, the out-of-date wording, the newest approved one, how much older it is, its owner and its status — from the CRA monitoring visit checklist still saying “verify 100% of pages” to a pharmacy card already corrected and sent to all 88 sites.
Sam asked about the Week 16 window at 07:31. The protocol says ± 3 days around Day 113; the site visit scheduling guide v1.0, issued in January before Amendment 2 narrowed it, still says ± 5 — and sites still print visit cards from it. Opened side by side, the two passages can be marked as explained or sent to Lena Ortiz to correct.
“Every answer about this fact uses the newest approved statement until the older one is corrected.”
Every question asked about VLT-302 is listed with who asked, whether it was answered from documents, data or both, and its confidence. Below 0.90, or where sources disagree without an explanation, the answer waits for her check: consent deviations this quarter came back at 0.71 — 17 in the CTMS log, 2 more in signed monitoring reports.
Opening the re-consent answer shows the reasoning in full: Amendment 3 requires re-consent at the next scheduled visit after ethics approval; 81 of 88 sites have it; 371 of 438 participants there have signed v5.0. The 14 overdue are at Site 214 Warsaw, 131 Leeds and 305 Valencia. Sources, the plain-language steps and the read-only query sit beside it.
Answers are working notes until a person files them. Lena files this one to TMF 01.01 Trial oversight as a study note, with the data as of CTMS 06:00 and EDC 05:30 on 7 Oct 2026, and signs with its meaning: “I confirm this answer is accurate for the data stated”.
“Signed and filed to TMF 01.01 — the answer is now a study record”
The same study, seen by the people who ask and answer its questions every day — what that looked like, and what it looks like now.
Six specialist agents route, read, query, date every version, answer and plan corrections; the study manager decides.
Reads the question, decides whether it needs documents, data or both, and picks the smallest set of study sources that can answer it, within the asker's site access.
Searches the protocol, amendments, plans, manuals, minutes and decisions, current versions first, and returns every statement of the fact with document, version, page and section.
Writes one read-only query per question over this morning's CTMS, EDC and blinded IRT extracts, explains each step in plain words and returns the rows used so they can be cited.
Dates every statement behind an answer — documents by approval and version, data by extract time — uses the newest approved one, flags any still-current document that disagrees and explains gaps between data sources.
Writes the fact card and the plain-words answer: the number first, numbered citations, and a table or chart when the answer has more than three rows.
Turns a gap in an answer into proposed actions for the right owner — the CRA for the site — with due dates from the next visit or the plan's deadline.
Checks, files and decides. The agents propose; a named person decides.
Ask about consent, visits and windows, queries, enrolment, dosing, drug supply, lab samples or site start-up — or ask it to filter a list or create a follow-up. Every answer points to its source.
Which sites still need re-consent for Amendment 3?
67 participants at 7 sites still need to sign ICF v5.0; 14 are overdue at three sites — 214 Warsaw (6), 131 Leeds (5) and 305 Valencia (3). Seven sites are still waiting for ethics approval.
Which study documents are out of date?
Two statements are open and one correction is under way: the target sample size (protocol synopsis 600, analysis plan v3.0 640), the Week 16 window (visit guide v1.0 ± 5 days, protocol v4.0 ± 3 days) and verification of other pages (CRA checklist v2.4 100%, monitoring plan v3.1 20%). Every answer uses the newest approved statement.
Show me the answers that need my check
Three answers are waiting: the consent-deviation count where sources disagree (Q-1043), the Week 16 window (Q-1041) and the temperature excursions (Q-1036). The Questions list is filtered to them.
Create a follow-up for the paper signatures at Site 214
Created: “Enter the 3 paper signatures from 2 Oct in EDC” for Priya Raman, Site 214, due Oct 8 — linked to Q-1042. Priya has been told by email.
13 screens from the working solution, on its sample data. Pick one to see it large.
What may be read, how answers are given, who can file and how out-of-date statements are handled are settings, per study.
| Setting | Default | Choose from |
|---|---|---|
| Never read unblinded data | On | Treatment arm, kit-to-arm and closed-session sources are excluded |
| Use superseded document versions | Only to compare | Only to compare · Never |
| Answer below this confidence goes to a person | 0.90 | 0.80 · 0.85 · 0.90 |
| CRAs see only their own sites' data | On | Data answers filtered to the asker's site access in CTMS |
| Filing an answer to the TMF | Study manager | Study manager · Study manager or medical monitor |
| Re-check every document each morning at | 06:30 | 06:30 · 07:00 · 08:00 |
| When two approved documents disagree, use | The newest approved statement | The newest approved statement · The protocol, then the newest plan |
| Send follow-ups to | Follow-ups board and email | Follow-ups board and email · Follow-ups board only |
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, based on the study team's own estimate of 30–60 minutes per cross-document question · “typical” = published figures (Tufts CSDD, Getz et al., Therapeutic Innovation & Regulatory Science, 2024 — protocol amendments). People, studies, sites and companies named on this page are fictional — characters and sample data in the working solution.
Search over a study's own files — protocol and amendments, plans, manuals, decision logs, minutes, trackers — and its CTMS, EDC and blinded IRT data, answered in plain words. Clinical Study Document Search returns each answer as a fact card: the value, the date it is true as of, and the document, version and section it came from, with the supporting passages.
The version checker finds every approved statement of the fact, dates each one by approval and version — or by extract time for data — and uses the newest approved statement. Superseded versions are read only when a question asks to compare. A still-current document that disagrees is flagged with how much older it is.
The answer uses the newest approved statement and flags the other on the card and on the out-of-date list, with both passages side by side. The study manager decides whether it is corrected, explained or reopened, and one click turns it into a follow-up for its owner. Where data sources differ, the answer gives both numbers and explains the gap or sends it for a person's check.
Yes. It reads blinded views only. Treatment allocation lists, unblinded pharmacist dispensing logs and closed data monitoring committee reports are never read, and data answers are filtered to the asker's site access, so CRAs see only their own sites.
Yes. The study data analyst writes one read-only query over this morning's CTMS, EDC and blinded IRT extracts, explains each step in plain words and cites the rows used. The rule from the document and the number from the data come back in one answer.
No. Every source is read in place and read-only. The agents never edit a source document; they flag it for its owner, and a person creates every follow-up.
Yes. Answers are working notes until a person files them. Below 0.90 confidence, or where sources disagree without an explanation, an answer waits for a person's check, and filing to the TMF needs the study manager's e-signature with its meaning.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your study files, TMF index and CTMS and EDC extracts — and it goes live once every quality gate has passed. We will walk you through it on your own study 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 demoClinical Study Report AuthoringResults drafted within days of database lock, linked to tables.
1-min demoWe'll run Clinical Study Document Search on a sample of your own study documents and extracts.