Systematic Review (PRISMA)
Real literature search, screening, meta-analysis, and a PRISMA-grade write-up.
On the Generate tab, pick "Systematic Review (PRISMA)" to run an end-to-end review. You can paste a study list, or run a real database search — the search retrieves and screens actual records and reports genuine PRISMA counts.
Workflow:
- Search — query Europe PMC + Crossref (merged and de-duplicated). Enter optional inclusion criteria.
- Screen — the AI screens each title/abstract against your criteria (include/exclude with reasons). "Screen all" extends beyond the first page (up to 200 records).
- Dual screening (optional) — a second independent AI reviewer screens the same set; you get Cohen’s κ, % agreement, and a list of conflicts to adjudicate.
- Integrity — included DOIs are checked against Crossref for retractions/withdrawals and flagged.
- Open access — each study is flagged OA with a direct full-text link (Unpaywall / Europe PMC).
- Generate — produces Abstract → Introduction → Search Strategy → Inclusion Criteria → Quality Assessment → Findings → PRISMA Flow (real counts) → Confidence & Limitations → References → PRISMA 2020 Checklist → Provenance.
Analysis aids (Generate tab, below the search):
- Meta-analysis + forest plot — enter per-study effect data (OR/RR/MD/SMD or effect+CI); computes fixed and random-effects pooled estimates with heterogeneity (I², τ²).
- Risk-of-bias appraisal — an AI traffic-light matrix across five domains, conservative from abstracts (confirm with full text).
- GRADE + Summary of Findings — set the GRADE factors per outcome; certainty (High→Very low) is computed and assembled into a SoF table.
Reproducibility: save a search to re-run it later (it stores the query, criteria, and results). Re-running a saved search flags studies that are NEW since the last run — a living review. The included studies export to RIS (EndNote/Zotero/Mendeley) and BibTeX (LaTeX/JabRef).
Honesty note: a real search retrieves and screens the top-ranked records (not the entire hit set), and the report states this caveat. Effect sizes are entered by you because abstracts rarely expose them in machine-readable form.