Preprint · 2026 · Open access

A Configurable LLM System for Automated Data Extraction in Meta-Analysis Using a Minimal Atomic Unit Framework

Yijia Yin, Haoxin Feng, Mengqi Shao, Yujia Pan, Chuyu Zhao, Chenxin Zhu, You Wan, Tiejun Tong, Xiaoyu Tang

Research Square · Published 9 September 2026 · DOI 10.21203/rs.3.rs-10662331/v1

Open PDFView at the publisherCite this paper

In brief

“A Configurable LLM System for Automated Data Extraction in Meta-Analysis Using a Minimal Atomic Unit Framework” is a 2026 preprint by Yijia Yin and 8 co-authors, published in Research Square. OpenAlex counts 0 citations as of 18 November 2026. It is open access under a CC BY licence: the full text is free to read at Research Square.

Citations
0
References
26
Authors
9

Abstract

Background: Extracting numerical data from randomized controlled trial reports is a major bottleneck in systematic reviews. Large language models (LLMs) can assist with this task, but complex combinations of study arms, time points, subgroups, and outcomes create risks of omission, misalignment, and unsupported values. We developed a configurable LLM extraction system based on a Minimal Atomic Unit (MAU) representation. Methods: The system uses GPT-4.1 and a three-level pipeline that produces study-level, arm-level, and outcome-level records, with source text attached to every extracted field. During prompt development, an exploratory comparison used the same three continuous positive airway pressure trial reports for prompt refinement and for comparing MAU and non-MAU prompts. The two settings used the same model, extraction scope, output fields, and source-tracing requirements; only the MAU setting explicitly decomposed the task. The finalized system was independently evaluated on 88 full-text study reports from five clinical domains. Performance was assessed at the aligned-cell level using precision, recall, F1 score, and hallucination rate. Results: In the exploratory prompt-development comparison (1,160 cells), precision, recall, and F1 were 1.0000, 0.9595, and 0.9793 with MAU and 0.9288, 0.2138, and 0.3476 without MAU; neither setting produced a hallucinated response. In the independent validation dataset (25,887 evaluated cells), precision was 0.997 (95% confidence interval [CI], 0.996–0.997), recall was 0.984 (95% CI, 0.983–0.985), F1 was 0.990 (95% CI, 0.989–0.991), and the hallucination rate was 0.2% (95% CI, 0.2%–0.3%). Conclusions: The MAU framework produced structured, source-traceable extraction with high cell-level performance across five domains. The MAU versus non-MAU finding is exploratory because the comparison reused the prompt-development studies. External validation with other models and review teams remains necessary before unsupervised use.

Cite this paper

Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.

APA 7

Yin, Y., Feng, H., Shao, M., Pan, Y., Zhao, C., Zhu, C., Wan, Y., Tong, T., & Tang, X. (2026). A Configurable LLM System for Automated Data Extraction in Meta-Analysis Using a Minimal Atomic Unit Framework. Research Square. https://doi.org/10.21203/rs.3.rs-10662331/v1

MLA 9

Yin, Yijia, et al. “A Configurable LLM System for Automated Data Extraction in Meta-Analysis Using a Minimal Atomic Unit Framework.” Research Square, 2026. https://doi.org/10.21203/rs.3.rs-10662331/v1.

Chicago

Yin, Yijia, Haoxin Feng, Mengqi Shao, Yujia Pan, Chuyu Zhao, Chenxin Zhu, You Wan, Tiejun Tong, and Xiaoyu Tang. 2026. “A Configurable LLM System for Automated Data Extraction in Meta-Analysis Using a Minimal Atomic Unit Framework.” Research Square. https://doi.org/10.21203/rs.3.rs-10662331/v1.

Harvard

Yin, Y. et al. (2026) ‘A Configurable LLM System for Automated Data Extraction in Meta-Analysis Using a Minimal Atomic Unit Framework’, Research Square. doi:10.21203/rs.3.rs-10662331/v1.

IEEE

Y. Yin et al., “A Configurable LLM System for Automated Data Extraction in Meta-Analysis Using a Minimal Atomic Unit Framework,” Research Square, 2026, doi: 10.21203/rs.3.rs-10662331/v1.

BibTeX

@misc{yin2026configurable,
  title     = {{A Configurable LLM System for Automated Data Extraction in Meta-Analysis Using a Minimal Atomic Unit Framework}},
  author    = {Yin, Yijia and Feng, Haoxin and Shao, Mengqi and Pan, Yujia and Zhao, Chuyu and Zhu, Chenxin and Wan, You and Tong, Tiejun and Tang, Xiaoyu},
  howpublished = {Research Square},
  year      = {2026},
  doi       = {10.21203/rs.3.rs-10662331/v1},
  url       = {https://doi.org/10.21203/rs.3.rs-10662331/v1}
}

Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)

Topic and keywords

Social Sciences › Decision Sciences › Statistics, Probability and Uncertainty › Meta-analysis and systematic reviews

  • automated data extraction
  • large language models
  • meta-analysis
  • randomized controlled trials
  • prompt engineering
  • GPT-4
  • source tracking
  • hallucinations

Work with this paper

Most-cited papers in Decision Sciences

All highly cited Decision Sciences papers

Questions about this paper

How do I cite this paper in APA style?

Yin, Y., Feng, H., Shao, M., Pan, Y., Zhao, C., Zhu, C., Wan, Y., Tong, T., & Tang, X. (2026). A Configurable LLM System for Automated Data Extraction in Meta-Analysis Using a Minimal Atomic Unit Framework. Research Square. https://doi.org/10.21203/rs.3.rs-10662331/v1

How many times has this paper been cited?

0 times, according to OpenAlex on 18 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.

Is this paper free to read?

Yes. It is open access at Research Square: sign in and open the PDF from this page.

Who wrote this paper?

Yijia Yin, Haoxin Feng, Mengqi Shao, Yujia Pan, Chuyu Zhao, Chenxin Zhu, You Wan, Tiejun Tong, Xiaoyu Tang.

What is the DOI of this paper?

10.21203/rs.3.rs-10662331/v1, permanently at https://doi.org/10.21203/rs.3.rs-10662331/v1. Sign in and open it from this page.

Bibliographic data from OpenAlex (CC0), last checked 18 November 2026. ResearchGuru does not host this paper; once you are signed in, the links open it at the publisher or the repository that does.