Toward Automated Meta-Analysis: Leveraging Large Language Model-Driven Outcome Alignment and Data Extraction

Meta-analyses stand as a cornerstone in consolidating research findings to assess treatment efficacy within evidence-based medicine. Yet, they are often laborintensive and quickly outdated due to the rapid publication of new studies. On the other hand, harmonization of research outcomes reported in papers following clinical trials, is essential for ensuring clarity and consistency within the scientific community. These outcomes, often described using disparate terminologies by different research teams, necessitate a process of meticulous extraction and standardization to achieve uniformity. This standardization is critical for facilitating data harmonization, which in turn enhances transparency, reproducibility, and comparability across various studies. This study proposes a novel automated approach for the extraction and analysis of data from open-source research articles for meta-analysis. Utilizing large language models, our proposed system identifies and aligns semantically similar outcomes from various papers published subsequent to the clinical trials. The system further extracts pertinent statistical data, facilitating a more efficient and robust meta-analytic process. By automating the assimilation of emerging research, the system ensures that meta-analyses remain current and reflective of the latest scientific evidence. This method reduces the manual effort traditionally required and offers a continually updated, time-efficient solution for meta-analytic research.