Artificial intelligence-based analysis of thrombotic and infectious risk factors in peripherally inserted central...

Peripherally inserted central catheters (PICCs) are increasingly used in France for prolonged intravenous therapies such as chemotherapy, parenteral nutrition, or antibiotics. They are easier to place than traditional central venous catheters and carry fewer immediate risks, but remain associated with delayed complications, mainly infections, thromboembolic events, and mechanical issues.

This literature review aimed to identify risk factors for PICC-related infections and thrombosis. The study used BIBOT, an artificial intelligence program for natural language processing, already validated in prior reviews and the IA language model LLaMA3. PubMed was searched for studies published between 2013 and 2023.

Using the AI-based tool BIBOT, 1896 PubMed abstracts on PICCs were automatically screened. After filtering and AI-assisted content analysis, 343 original articles focusing on PICC complications were identified, enabling a targeted selection of 113 articles on infectious and 281 on thrombotic complications. In total, 20 infectious and 59 thrombotic risk factors were manually identified. Among these, the most frequently reported in the reviewed articles for infections were number of lumens, chemotherapy and catheter dwell time. For thrombosis, the most commonly cited factors included cancer or hematologic disease, chemotherapy, PICC diameter and number of lumens (>2).

The study highlights the value of AI-based tools to accelerate article selection and data extraction in medical literature. However, human validation remains essential to avoid errors and misinterpretations, particularly with acronyms or heterogeneous definitions. Hybrid approaches combining AI with expert review save considerable time and are expected to improve further with multimodal models and multi-agent strategies.