Abstract
Cancer development is influenced by a complex interplay of diverse risk factors, but synthesizing this fragmented research remains challenging. To address this, we developed CanRisk-DB, an Artificial intelligence (AI)-driven database that systematically aggregates and standardizes published evidence on cancer-associated risk factors. Using a multi-stage AI pipeline based on the PICOS framework, we analyzed 435,975 publications from PubMed, Embase, and Cochrane (2000-2024), employing a Graph-based Retrieval-Augmented Generation framework to extract cancer types, risk factors, and quantitative estimates (e.g., relative risk [RR], hazard ratio [HR], standardized incidence ratio [SIR]). In the literature screening stage, the system demonstrated high accuracy and efficiency. From 9550 relevant articles, CanRisk-DB compiled 445,646 standardized records covering 76 risk factor groups and 42 cancer types across 80 countries over 50 years. Validated against benchmark datasets, this publicly accessible resource constitutes a comprehensive knowledge base of cancer risk factors, supporting etiological research, risk analyses, and the development of evidence-informed prevention strategies. The CanRisk-DB is available at http://www.canrisk-ai.com .