Clinical trial protocols
Protocols are written to be complete, not comparable. The same criterion is a bullet in one document and a paragraph in the next, and almost every question worth asking is across the set rather than inside one of them.
Before anything else: this builds a table you check against the protocols, not one you rely on instead of reading them. Ragextract is not a clinical decision support tool, it does not screen or match patients, and it is not regulated as a medical device. What it changes is where the reading goes — not whether it happens.
What a row is
One study. Which document represents it is the decision to settle first: a full protocol and a synopsis answer very differently, and a table mixing the two will report “not stated” for half its cells without that being a finding about the trial.
Amendments belong in a bundle with the protocol they amend, most recent first. Eligibility is the field that moves, and a protocol amended twice has an inclusion list that no longer matches the one in the original document.
The starting columns
This is the Healthcare preset:
| Column | Type | Prompt |
|---|---|---|
| Patient Eligibility Criteria | List | List the key patient inclusion/exclusion eligibility criteria for this trial. |
| Primary Endpoints | List | List this trial's primary endpoints. |
| Dosage Limits | Text | What dosage limits are specified in this protocol? |
| Adverse Event Rules | Text | Summarize the rules for reporting and managing adverse events in this protocol. |
Two of the four are list columns, which are charged at 2–3 credits per cell rather than 1. That is the right shape here — eligibility criteria and endpoints are genuinely enumerations, and flattening them into prose loses the ability to scan a column for one criterion.
What it looks like filled in
| Row | Documents | Key eligibility | Primary endpoints | Dosage limit | AE reporting |
|---|---|---|---|---|---|
| Protocol A — phase II.pdf | ECOG 0–1; ≥1 prior line; no CNS metastases §4.1 | Progression-free survival at 12 months §8.1 | 400 mg daily, 200 mg on grade 3 §6.4 | SAEs within 24 hours to sponsor §9.2 | |
| Protocol B — phase III.pdfAmendment 2.pdf | ECOG 0–2; treatment-naive; CNS mets permitted if treated Am. 2, §4.1 | Overall survival §8.1 | 10 mg/kg every 3 weeks, capped at 1,200 mg §6.2 | SAEs within 24 hours; DSMB review quarterly §9.1 | |
| Protocol D — synopsis.pdf | ECOG 0–1; ≥2 prior lines §2 | Objective response rate §2 | Not stated | Not stated |
Setting it up
- One workspace per review or per therapeutic area. A table can only see documents in its own workspace, and the comparison is the point.
- Bundle each protocol with its amendments, most recent first, before building the table.
- Create the table, take the Healthcare preset, and edit the prompts to name the fields your review actually turns on.
- Run two protocols you already know well and read every citation against the section it points at. Protocols you know are the only honest test of a table you intend to use on protocols you do not.
- Adjust, then run the set.
Columns worth adding
- Phase as a category — I, Ib, II, III, IV, Not stated — so the set can be filtered rather than read.
- Target enrolment as a number, and number of sites as a second. Both are the sort of figure that gets retyped into a feasibility spreadsheet.
- Randomisation as a category and blinding as another, rather than one prose column describing the design.
- Visit schedule burden as text, if the review is about feasibility. It is the question sites actually ask and it is never in the same place twice.
- Secondary endpoints as a separate list. Asking one column for “endpoints” produces a cell that mixes the two, and the distinction is the whole statistical argument.
What to check before relying on it
- Superseded eligibility. If an amendment is not bundled with its protocol, the table will report withdrawn criteria with a citation, which makes a wrong answer more convincing rather than less. This is the failure to guard against above all others here.
- Incomparable units. A flat milligram dose and a weight-based dose with a cap are both correct answers to “what is the dosage limit”, and they cannot be compared down a column. Read the citations before treating the column as a series.
- Inclusion read as exclusion. “No CNS metastases” and “CNS metastases permitted if treated and stable” are close in wording and opposite in effect. If the list column is collapsing them, split inclusion and exclusion into two columns.
- Synopses standing in for protocols. A synopsis will answer the first two columns and blank the rest. That is a fact about which document was uploaded, and it should not be read as a gap in the trial design.
Roughly what it costs
Forty protocols averaging 90 pages is 3,600 pages — 3,600 credits, $36, once. Two typed columns across forty rows is 80 credits; the two list columns are another 240 at the top of their band. About $39.20 at par, and less with a bulk top-up. See how credits work.
Ragextract is not a clinical decision support tool and is not regulated as a medical device. The preset’s prompts are illustrative starting points, not clinical or regulatory definitions, and nothing here is medical, clinical or regulatory advice.