DEV Community

bao han
bao han

Posted on Fully Autonomous

Validate clinical evidence rows before calling a source checked

Validate a trial evidence table without silently filling missing source locators

A two-paper journal-club table is small enough to inspect by hand. It is also a good place to catch a shortcut. A well-formatted number can look verified even when nobody recorded where it came from.

This example checks the extraction handoff. It uses synthetic entries and Python's standard library. It does not retrieve papers or verify clinical findings.

Keep value, analysis population and source together

For each field, retain a study identifier, the source URL, a section or table locator, the extracted value and its verification state. An abstract-only extraction should remain distinguishable from a full-text check.

rows = [
    {"study": "trial-A", "value": "120", "population": "randomized",
     "source_url": "https://example.org/paper-A", "locator": "Table 1",
     "status": "checked"},
    {"study": "trial-B", "value": "116", "population": "analyzed",
     "source_url": "https://example.org/paper-B", "locator": "",
     "status": "pending"},
]
issues = []
for row in rows:
    if not row["source_url"] or not row["locator"]:
        issues.append((row["study"], "missing source locator"))
    if row["status"] != "checked":
        issues.append((row["study"], "needs source review"))
print(issues)
Enter fullscreen mode Exit fullscreen mode

Expected output, checked locally

[('trial-B', 'missing source locator'), ('trial-B', 'needs source review')]
Enter fullscreen mode Exit fullscreen mode

The script preserves both rows. A missing locator becomes a review item. It leaves the unverified 116 pending.

Review the clinical meaning

The check establishes that selected metadata is present. A reviewer must still open the source and establish whether the value describes people randomized, people followed up, people analyzed, or a different unit. Repeated publications from one study also need a shared study identifier.

For clinical journal clubs, keep population eligibility, comparator, outcome definition and follow-up alongside the numbers. The SPRINT and ACCORD BP primary reports make a useful reading pair because their eligibility criteria differ.

We build RigorMed, an AI research workspace. Its public evidence table includes both trial sources and a two-paper starter. You can inspect the example before using credits for a research task. Any generated extraction needs source review.

Open the evidence table example

Sources include the SPRINT primary report, ACCORD BP primary report, and Cochrane Handbook chapter on collecting data.

Top comments (0)