AI evidence extraction

L-PRISMA: An Extension of PRISMA in the Era of Generative Artificial Intelligence (GenAI)

Samar Shailendra, R. W. Kadel, Aakanksha Sharma, Islam Mohammad Tahidul, Urvashi Rahul Saxena · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

0.00

FWCI

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.36227/techrxiv.177040570.06710783/v1

Methodology & findings

Study design

Conceptual framework development and methodological extension; the paper presents a proposed approach integrating human oversight with GenAI-assisted statistical screening rather than conducting an empirical experiment or systematic review itself..

Main result

The study proposes that "human-led synthesis with a GenAI-assisted statistical pre-screening step" can address reproducibility and transparency challenges in systematic reviews. The authors argue that "human oversight ensures scientific validity and transparency, while the deterministic nature of the statistical layer enhances reproducibility," suggesting a hybrid approach that combines manual human validation with automated computational screening to improve efficiency while maintaining PRISMA principles.

Reports effect sizes.

Research paradigm

Interpretive/argumentative - proposing a methodological extension rather than conducting empirical testing

Author conclusions

The authors conclude that "the proposed approach systematically enhances PRISMA guidelines, providing a responsible pathway for incorporating GenAI into systematic review workflows." However, this conclusion is theoretical rather than empirically demonstrated.

Risk of bias

Hallucination risk from LLM-based screening; Bias amplification through automated processing; Potential selection bias in how GenAI flags relevant studies; Transparency and auditability concerns with black-box LLM decisions; Lack of empirical validation - no experimental testing of the proposed L-PRISMA framework; No comparison against standard PRISMA workflow; Potential confirmation bias in framework design favoring LLM integration; Unquantified assumptions about 'human oversight' effectiveness; No assessment of how much hallucination or bias amplification actually occurs in the proposed workflow; LLM non-determinism; Hallucination risk from language models; Bias amplification from generative AI systems; Hallucination in LLMs; Bias amplification; Non-determinism of generative models; hallucination and bias amplification in large language models; non-determinism of LLMs affecting reproducibility; potential for loss of scientific validity if human oversight is insufficient; Hallucination in large language models; Bias amplification from GenAI systems; Non-deterministic nature of LLMs affecting reproducibility

Limitations

The proposed approach acknowledges that "the manual processes of data extraction and literature screening remain time-consuming and restrictive" and that "reproducibility, transparency, and auditability, the core PRISMA principles, are being challenged by the inherent non-determinism of LLMs and the risks of hallucination and bias amplification." The framework relies on human oversight, which may not fully eliminate the risks associated with LLM non-determinism.

Open questions raised

The paper identifies the need for responsible integration of GenAI into systematic review workflows while preserving PRISMA's core principles of reproducibility, transparency, and auditability. Future work should address validation of the proposed L-PRISMA extension and empirical testing across different domains.
The paper identifies the need for responsible integration of GenAI into systematic reviews while maintaining PRISMA principles of reproducibility, transparency, and auditability. Future work should include empirical validation of the L-PRISMA framework against traditional PRISMA workflows.
The paper identifies the need for responsible integration of GenAI into systematic review processes while maintaining PRISMA principles of reproducibility, transparency, and auditability.
The paper identifies the need for frameworks that balance the efficiency benefits of GenAI with the reproducibility, transparency, and auditability requirements of systematic reviews. It addresses the gap between emerging AI capabilities and the need for responsible implementation in evidence synthesis.
The paper identifies the need for responsible incorporation of GenAI into systematic review workflows while maintaining PRISMA principles of reproducibility, transparency, and auditability. Future research should address validation of the L-PRISMA framework in real-world systematic review contexts and refinement of the balance between automation and human oversight.
The paper identifies gaps in current PRISMA frameworks regarding automation of manual data extraction and literature screening processes, and the need for responsible integration of GenAI technologies while maintaining reproducibility and transparency standards.

Data: not_statedCode: not_statedExtracted from: pdfAgreement 69%

Explore related topics

Systematic Review Automation Literature Discovery Conceptual