| name | cochrane-search |
| description | Search Cochrane Reviews for systematic-review evidence and effect sizes. Use when the user wants the strongest evidence-base summary for a treatment, screening, or diagnostic question. Produces structured effect-size estimates with confidence intervals. |
Cochrane search
When to use this
- Treatment-effect questions ("does X work for Y?").
- Diagnostic test performance questions.
- Screening-program effectiveness.
- Anywhere you need a defensible effect size with a CI to put in
patient-facing wording.
How to do it
- Search the Cochrane Library:
https://www.cochranelibrary.com/
(or the CDSR JSON API if a key is available).
- For each hit capture: review title, year, conclusion summary,
primary outcome effect (RR / OR / MD with 95% CI), and the
number of studies / participants pooled.
- Note any "low-certainty" / "moderate-certainty" GRADE ratings —
they belong in the recommendation alongside the effect size.
Output
Write the durable source/search/effect-size record to
$FORMTASTIC_RUN_DIR/output/research/bundle.json; include the review as
a kind: "paper" source with DOI/URL, year, citation, relevance, and
effect-size notes plus sourceFile and, when item-specific, extractFile.
Record the search in searches[] with transcriptFile, and write matching
supporting files under output/research/searches/, sources/, and
extracts/. Then write output/findings.json as the compact
claim/evidence summary, and cite with refType=paper plus a note that
summarizes the effect:
{
"refType": "paper",
"refId": "https://doi.org/10.1002/14651858.CDxxxx",
"note": "Pooled RR 0.78 (95% CI 0.65-0.93) across 12 RCTs (n=4,800)."
}
If the review is itself the source you want anchored on the
question (rare), use valueCanonical and refType=guideline only
if the review functions as a recommendation (it usually doesn't).
Pitfalls
- Pooling that mixes interventions or populations the question
doesn't apply to. Read the review's PICO before quoting numbers.
- Old reviews — Cochrane reviews are updated; cite the most recent.
- Heterogeneity (I²>50%): note it. The pooled point estimate may
not be a faithful summary.