EvidAIEvidAI
Back to Home
6 min read

Multi-Standards Checklist Engine

Automated checklist evaluation across 13+ reporting standards with bundle emission

Multi-Standards Checklist Engine

PRISMA 2020 is the baseline, but a real review may need to satisfy several reporting standards simultaneously — PRISMA-S for search reproducibility, ROBIS for review-level bias, AMSTAR-2 for methodological quality, and agency-specific checklists on top. EvidAI's checklist engine evaluates every applicable standard automatically and ships each one as a file in the export bundle.

Supported Standards

StandardApplies to
PRISMA 2020 — 27-item reporting checklistAll systematic reviews
PRISMA-S — search reporting extensionSearch phase
PROSPERO — protocol registration checklistProtocol phase
ROBIS — Risk of Bias in Systematic ReviewsOverall review bias
AMSTAR-2 — 16-item critical appraisal of SRsOverall review quality
RAISE — Reviewer AI and Statistical EvaluationAI-assisted reviews
Cochrane MECIR — conduct + reporting standardsCochrane reviews
JBI Critical AppraisalJBI methodology reviews
STROBE — observational study reportingIncluded observational studies
CONSORT 2010 — RCT reportingIncluded RCTs
NICE UK HTA submissionHTA reporting profile
CADTH reimbursement submissionHTA reporting profile
IQWiG benefit assessment dossierHTA reporting profile
ICH M4E(R2) Module 2.5 Clinical OverviewICH reporting profile

The registry is extensible — additional standards can be added without changing the consuming code paths.

How Evaluation Works

Each checklist is a structured definition containing a label, version, and list of items. Each item knows which review-state signals it depends on and emits a status of complete, partial, or missing for the active review.

The evaluator takes the review's phase signals, the artifacts already present in the export bundle, and the active reporting profile, then walks every active checklist deterministically. Missing signals degrade gracefully — a missing input produces a missing status for the affected item, not an error.

Profile-Driven Activation

Which checklists are active is driven by the review's Reporting Standard profile:

Reporting profileActive checklists
PRISMA 2020 (default)prisma-2020, prisma-s, prospero, robis, amstar-2
Cochrane interventionprisma-2020, prisma-s, cochrane-mecir, robis, amstar-2
JBIprisma-2020, jbi-sumari, robis
HTA dossierprisma-2020, prisma-s, nice-hta, cadth-hta, iqwig-hta
ICH M4E + EMA RWEprisma-2020, ich-m4e
AI-reported (RAISE)prisma-2020, raise

The checklist engine iterates over the profile's active standards on every export.

Export Bundle Artifacts

Every bundle emits one Markdown + one JSON + one CSV per active checklist (e.g. PRISMA 2020 ships as a human-readable Markdown report, a machine-readable JSON, and a spreadsheet-friendly CSV — and the same triple is produced for every other active standard). A README in the checklists directory indexes every attached checklist and its coverage score.

Quality Thresholds as Rules

In addition to item-level status, the engine evaluates phase-level quality thresholds that you can set per review:

  • Minimum inter-rater agreement (Cohen's Kappa) for dual screening
  • Minimum proportion of papers dual-extracted
  • Maximum proportion of high-risk-of-bias studies

A threshold fail degrades the containing checklist item (typically to partial) and the reason is surfaced in the bundle's Markdown output.

How to Configure

Set the reporting profile

  1. SLR hub → Reporting Standard selector → pick a profile.
  2. The profile is saved against the review with signer and timestamp, and triggers a Reporting profile selected audit event.

Adjust quality thresholds

  1. SLR hub → Options drawer on any phase card → Quality thresholds section.
  2. Edit the numeric threshold and save.

Preview checklist state

The Reporting Standard selector shows a live coverage score — how many items are complete versus missing across all active checklists. Clicking it opens a per-item breakdown.

Design Principles

  • Pure and deterministic — given the same review state, the engine always returns the same status. This makes the bundle output reproducible.
  • Fail-open — missing signals never throw; they mark the relevant item as missing and keep the rest of the checklist evaluating.
  • Profile-gated — checklists are only evaluated if the review's active profile lists them. Irrelevant standards never clutter the bundle.
Did this article help?
Still stuck?