Skip to contents

Structured, validated, traceable data for living systematic reviews.

A metawoRld project is a folder (usually a Git repository) of plain YAML, CSV and JSON files, laid out as defined in SPEC.md:

  • settings, criteria and forms describing what is extracted (_metawoRld.yml, forms/extraction.yml),
  • the search and screening log (searches.yml, records.csv, screening.csv), from which PRISMA counts are computed,
  • one door for data: every input channel (an entry form, CSV files, the LLM pipeline DataFindR, scripts) calls submit_study(), which validates the extraction and stages it in submissions/,
  • accepted studies in data/<study_id>/, with provenance.csv (who or what produced each value, from which source), verification.csv (human checks, tied to a hash of the values) and datapackage.json (a Frictionless Data Package with file hashes),
  • meta-analyses defined in _analyses.yml and a Quarto dashboard website, rebuilt whenever data change.

check_project() checks all of it, locally or in CI, and catches data edited outside metawoRld.

No R code needed to get going: metawoRld::launch() opens an app to start a review, accept and reconcile submissions, verify values, check, publish and share the project with a team (Git, without Git commands):

The metawoRld app
The metawoRld app

Documentation: https://andjar.github.io/metawoRld/, including a worked example with real data from four open-access studies of IL-6 in preeclampsia.

The living meta-analysis dashboard built from the worked example
The living meta-analysis dashboard built from the worked example

Values are verified against the papers in a keyboard-driven review app; every decision is tied to a hash of the values it approved:

Verifying values with the keyboard in review_values()
Verifying values with the keyboard in review_values()

Installation

# install.packages("remotes")
remotes::install_github("andjar/metawoRld")
remotes::install_github("andjar/DataFindR")   # optional, for LLM screening/extraction

Example

Open the app (start a review, or explore the worked example):

metawoRld::launch()

Or try the worked example from R:

library(metawoRld)
p <- example_project(file.path(tempdir(), "il6"))
review_values(p)       # verify values
generate_website(p)    # build the dashboard (needs Quarto)

Or start your own review:

library(metawoRld)

create_metawoRld(
  "my-review",
  project_name = "Biomarker X in pregnancy",
  project_description = "Living review of maternal serum levels of X.",
  inclusion_criteria = c("Human pregnancy", "Serum or plasma X reported"),
  exclusion_criteria = c("Animal study", "Review or case report")
)
add_person("my-review", "ahj", "Anders Hagen Jarmund", roles = c("admin", "extractor", "verifier"))

id <- submit_study("my-review", "hansen2019", metadata = list(...), data = my_data_frame,
                   person = "ahj", channel = "script")
accept_submission("my-review", "hansen2019", id, person = "ahj")
review_values("my-review")          # verify values against the papers
check_project("my-review")          # schemas, hashes, provenance, people
generate_website("my-review")       # living meta-analysis dashboard