Storm disasters are the most common cause of accidents in offshore oil and gas industries. To prevent accidents resulting from storms, it is vital to analyze accident propagation and to learn about accident mechanism from previous accidents. In this paper, a novel risk analysis framework is proposed for systematically identifying and analyzing the evolution of accident causes. First, accident causal factors are identified and coded based on grounded theory (GT). Then, decision making trial and evaluation laboratory (DEMATEL) is integrated with interpretative structural modeling (ISM) to establish accident evolution hierarchy. Finally, complex networks (CN) are developed to analyze the evolution process of accidents. Compared to reported works, the contribution is threefold: (1) the demand for expert knowledge and personnel subjective influence are reduced through the data induction of accident cases; (2) the method of establishing influence matrix and interaction matrix is improved according to the accident frequency analysis; (3) a hybrid algorithm that can calculate multiple shortest paths of accident evolution under the same node pair is proposed. This method provides a new idea for step-by-step assessment of the accident evolution process, which weakens the subjectivity of traditional methods and achieves quantitative assessment of the importance of accident evolution nodes. The proposed method is demonstrated and validated by a case study of major offshore oil and gas industry accidents caused by storm disasters. Results show that there are five key nodes and five critical paths in the process of accident evolution. Through targeted prevention and control of these nodes and paths, the average shortest path length of the accident evolution network is increased by 35.19%, and the maximum global efficiency decreases by 20.12%. This indicates that the proposed method has broad applicability and can effectively reduce operational risk, so that it can guide actual offshore oil and gas operations during storm disasters. © 2022 by the authors. Licensee MDPI, Basel, Switzerland.