6ER-020 Large language model: methodology proposal for an assisted meta-analysis to catalogue medical devices used in preoperative progressive pneumoperitoneum for the management of complex ventral hernias

<h3>Background and Importance</h3> With the rise of artificial intelligence (AI) tools, especially Deep-Learning models, meta-analyses are likely to evolve. However, it is necessary to provide a framework for this purpose. This study focused on the Goñi-Moreno technique, which increases abdominal cavity volume prior to complex ventral hernia surgery in order to reduce postoperative respiratory issues. Subcutaneous emphysema, caused by multi-perforated catheters, is a common complication. Medical devices used depend on practitioners. Here we conducted a meta-analysis to explore medical devices mentioned in literature related to this technique using AI. <h3>Aim and Objectives</h3> The objective of this study was to propose a state-of-the-art meta-analysis technique, assisted by a large language model (LLM), to identify an exhaustive list of the medical devices and gases employed in the Goñi-Moreno technique worldwide. And compared these findings with our practices. <h3>Material and Methods</h3> Articles with the keywords ‘goñi-Moreno’ and ‘preoperative progressive pneumoperitoneum’ from PubMed (1948–2024) where selected. Full-text articles from the PMC repository were retrieved using a web crawler, while others were manually downloaded. We then used the GPT-4 LLM API to extract sentences referring to medical device and gases. Prior to this, a pre-processing step was carried out to extract text from the PDFs using Python’s PyPDF2 library. The extracted output was subsequently integrated, along with the corresponding PDFs, into an SQLite3 database. Each extracted sentence was meticulously reviewed to accurately identify the mentioned medical devices and gas injected. <h3>Results</h3> Out of 200 PubMed articles (1948–2024), 63 were analysed after excluding 105 inaccessible paper articles and 32 irrelevant articles. Most cited devices were intra-abdominal implantable chamber and central venous catheter, followed by multi-perforated drainage catheter (MPDC) and lumbar needle. Air was the most used gas, with some mentions of NO and CO2. At our hospital centre, filtered air is used with MPDCs. <h3>Conclusion and Relevance</h3> Some studies, like Takehiko et al. (2024), show promising results using LLMs in meta-analysis. Here we propose a methodology to help practitioners quickly answer surgical questions, complementing rigorous methods that require re-evaluation. Fine-tuning small language models for medical device research is the next step. <h3>References and/or Acknowledgements</h3> Oami T, Okada Y, Nakada T. Performance of a large language model in screening citations. <i>JAMA Netw Open.</i> (2024) <h3>Conflict of Interest</h3> No conflict of interest