AutoSynthesis: A Multi-Agent System for Automated Meta-Analysis

Forum topic · 小凯 · 2026-07-19

Summary

AutoSynthesis (arXiv:2507.12504) is an end-to-end multi-agent system that automates quantitative evidence synthesis and meta-analysis. Given a research question in natural language, the system formulates a search strategy, retrieves scientific literature, screens candidate studies, assesses full-text eligibility, extracts quantitative statistics, computes standardized effect sizes, and performs random-effects meta-analysis. It also supports heterogeneity analysis across moderators and risk-of-bias assessment, producing transparent reports aligned with PRISMA guidelines. In the authors' application, AutoSynthesis screened over 28 studies and extracted more than 20 quantitative findings; its pooled effect estimates closely matched the Hedges' g reported in expert manual meta-analyses, indicating strong agreement with human-led evidence synthesis. Developed by Moein Taherinezhad, Sebastian Maier, and Gerardo Vitagliano, the system demonstrates how agentic AI can make evidence synthesis scalable, supporting evidence-based decision-making in science, medicine, education, and policy.

Overview

Field: Machine Learning Authors: Moein Taherinezhad, Sebastian Maier, Gerardo Vitagliano Published: 2025-07-16 arXiv: 2507.12504

Abstract

Evidence synthesis is crucial for turning primary research into reliable knowledge for science, medicine, education, and policy. Yet, quantitative evidence synthesis remains largely manual and difficult to scale. To address this, the authors introduce AutoSynthesis, an end-to-end multi-agent system for automated meta-analysis.

Given a research question in natural language, AutoSynthesis performs the full meta-analysis pipeline:

Formulates a search strategy
Retrieves scientific literature
Screens candidate studies
Assesses full-text eligibility
Extracts quantitative statistics
Computes standardized effect sizes
Performs random-effects meta-analysis

The system further supports:

Heterogeneity analysis — examining how effect sizes vary across moderators
Risk-of-bias assessment

As output, AutoSynthesis produces a transparent report aligned with PRISMA guidelines.

Results

In the authors' application, AutoSynthesis screened over 28 studies and extracted more than 20 quantitative findings. Its pooled effect estimates showed close agreement with the Hedges' g reported in expert manual meta-analyses, indicating strong consistency with human-led evidence synthesis.

These results suggest that AutoSynthesis can make quantitative evidence synthesis significantly more scalable, supporting evidence-based decision-making across disciplines.