EvidenceFlow Documentation
Overview
EvidenceFlow is a platform for conducting systematic reviews and evidence synthesis. It combines literature import, screening, data extraction, meta-analysis, and reporting in a single workspace, with AI-assisted screening, extraction, and manuscript drafting available as a paid upgrade.
Core workflow
1
Create project
2
Import literature
3
Screen studies
4
Extract data
5
Meta-analysis
6
Export report
Projects
Each systematic review lives inside a Project. A project holds your research question, PICO criteria, inclusion/exclusion criteria, all imported studies, screening decisions, extraction records, and reports.
Creating a project
Go to Projects → New Project
Fill in your review details
- Name: Short title for the review
- Research question: Your primary PICO question
- Population, Intervention, Comparator, Outcome: Used by AI to screen studies
- Inclusion / Exclusion criteria. Plain text, one criterion per line; AI uses these automatically
Save
Project dashboard
The project page shows a summary of study counts by screening stage, quick links to each workflow stage, and your team members.
Literature Import
Import references into your project from PubMed or by uploading a file.
Supported sources
PubMed searchRecommended
Enter a search query. EvidenceFlow fetches results directly via PubMed API
RIS file
Export from Endnote, Zotero, Mendeley, or Web of Science and upload here
BibTeX (.bib)
Standard bibliography format used by most reference managers
CSV
Spreadsheet with columns: title, abstract, authors, year, doi
Duplicate detection
After import, click Detect Duplicates. EvidenceFlow compares titles and DOIs to flag duplicates, which are excluded from screening counts automatically.
Full-text retrieval
EvidenceFlow can automatically download open-access PDFs via Unpaywall. On the Import page, click Auto-retrieve PDFs for included studies. PDFs that cannot be retrieved automatically must be uploaded manually.
Screening
Screening happens in two stages: Title/Abstract and Full-text. Both stages use the same interface.
Screening interface
Open a project → Screening
Review the study
Record your decision
- Include: Meets all inclusion criteria
- Exclude: Does not meet criteria
- Maybe: Uncertain, review later
AI screening
Use the AI buttons to get a suggested decision based on your PICO and criteria:
✦ Ask AI Fast
Uses the fast AI model. Returns include/exclude/maybe with a confidence score and one-sentence reason. Responds in ~5 seconds.
✦ Deep AI Thorough
Uses the deep reasoning AI model with chain-of-thought analysis. More accurate for borderline cases. Takes ~20–40 seconds.
Progress tracking
The progress bar at the top shows how many studies have been reviewed vs. total eligible. The summary panel shows include / exclude / maybe counts for the current stage.
Data Extraction
Data extraction captures the quantitative and qualitative information needed for meta-analysis from each included study.
Required fields for meta-analysis
| Field | Description | Required for |
|---|---|---|
| Effect size | SMD, OR, RR, HR, MD, etc. | Clinical meta-analysis |
| Standard error | SE of the effect estimate | Clinical meta-analysis |
| Beta | Regression coefficient from GWAS | Omics meta-analysis |
| p-value | Statistical significance | Omics meta-analysis |
| Sample size | Total N in the study | Both |
| Study label | First author + year | Both |
AI extraction
Click AI Extract on any study with a full-text PDF. EvidenceFlow reads the PDF and pre-fills effect sizes, sample sizes, and other fields. Always verify AI-extracted values against the source paper.
Meta-Analysis
Meta-analysis pools quantitative data from included studies to produce a single summary effect estimate with confidence intervals and heterogeneity statistics.
Prerequisites
Configuration
Analysis type
Clinical: for interventional/observational studies with SMD/OR/RR. Omics: for GWAS/genetic association studies.
Model
Random Effects (DerSimonian-Laird) is the default and appropriate when studies differ in population or methods. Fixed Effects assumes a single true effect size.
Effect measure
Choose the measure that matches your extracted data: SMD, MD, OR, RR, RD, HR, or correlation.
Results
After running, EvidenceFlow displays:
- Pooled effect with 95% confidence interval
- I² statistic: 0-25% low, 25-75% moderate, >75% high heterogeneity
- Cochran's Q: formal heterogeneity test
- Forest plot: individual study effects with pooled diamond
- Funnel plot: asymmetry suggests publication bias
Reports & Export
Generate publication-ready outputs from the Reporting page inside any project.
Export formats
Full report with PRISMA diagram, study table, and meta-analysis results
Word (.docx)
Editable document with methods, PRISMA table, and included studies list
Excel
Multi-sheet workbook: PRISMA counts, screening decisions, extraction data
CSV
Flat file of all screening decisions, one row per decision
JSON
Machine-readable export of all project data for programmatic use
PRISMA 2020 diagram
The interactive PRISMA flow diagram lives on the PRISMA Workflow page (sidebar). It's generated automatically from your screening decisions — records identified, after deduplication, screened, excluded at each stage, and finally included — and every export format above includes the current PRISMA counts.
AI manuscript draft
Click Generate manuscript draft to have EvidenceFlow write Methods, Results, and Discussion sections based on your PRISMA counts and meta-analysis results. Requires Ollama to be running. The draft is a starting point. Always review and edit before submission.
AI Features
EvidenceFlow has AI built in and ready to use. No setup is required on your end — just open a project and start using the AI buttons directly from the interface.
AI Screening
While reviewing a study in the Screening page, you will see two AI buttons on the right panel:
✦ Ask AI — Fast
Click this button to get an instant include/exclude/maybe suggestion with a confidence score and a one-line reason. Best for quickly moving through large study lists.
✦ Deep AI — Thorough
Click this for a detailed chain-of-thought analysis. The AI reads the full abstract against your PICO criteria and explains its reasoning step by step. Best for borderline studies.
AI Data Extraction
On the Data Extraction page, click AI Extract next to any study that has a full-text PDF attached. EvidenceFlow will read the PDF and auto-fill effect sizes, sample sizes, and other quantitative fields for you. Review each value before saving.
AI Manuscript Draft
Once your meta-analysis is complete, go to the Reporting page and click Generate manuscript draft. EvidenceFlow will write a Methods, Results, and Discussion section based on your PRISMA counts and pooled results. Use this as a starting point and edit before submission.
FAQ
Can I use EvidenceFlow without the AI features?+
Yes. All core features — import, manual screening, data extraction, meta-analysis, and reporting — work without using any AI. The AI buttons are optional and can be ignored if you prefer to work manually.
Where is my data stored?+
All project data is stored in a PostgreSQL database on the server running the EvidenceFlow backend. No data is sent to third-party AI services. PDFs are stored locally on the backend server.
How many studies can I import?+
There is no hard limit. Performance is good up to ~10,000 studies per project. For larger datasets, use PubMed filters to narrow your import before uploading.
Why is the Run Analysis button disabled?+
Meta-analysis requires at least 3 extraction records with effect size + standard error (for clinical) or beta + p-value (for omics). Go to Data Extraction and fill in these fields for your included studies.
What is the difference between clinical and omics analysis?+
Clinical meta-analysis pools effect sizes (SMD, OR, RR, etc.) from intervention or observational studies. Omics meta-analysis is designed for genetic association studies and uses beta + p-value from GWAS or candidate gene studies.
Can I collaborate with a team?+
Yes. Open your project settings and invite team members by email. Each member can make independent screening decisions, enabling dual-reviewer conflict resolution.
Is EvidenceFlow free?+
The core platform — import, screening, extraction, meta-analysis, PRISMA reporting, and export — is free with no subscription. AI-assisted screening, extraction auto-fill, and manuscript drafting are a paid upgrade. EvidenceFlow is proprietary software; the source code is not publicly available.