[Application research and evolutionary history of risk prediction models for postoperative pancreatic fistula after pancreaticoduodenectomy].

Postoperative pancreatic fistula (POPF) is a severe complication following pancreaticoduodenectomy,among which clinically relevant postoperative pancreatic fistula has garnered substantial attention due to its high incidence (10% to 25%) and fatal consequences. This article reviews the evolutionary trajectory of risk prediction models for POPF after pancreaticoduodenectomy,progressing from single-center risk prediction models dominated by intraoperative factors, to weighted risk prediction models and their modified versions,and further to novel models integrating radiomics with machine learning. This evolutionary process reflects the tendency of the prediction system toward preoperativeness,simplification,and objectivity. Early models were limited by their dependence on intraoperative factors and insufficient external validation. Modified models enhanced preoperative applicability by streamlining variables and underwent adequate multicenter external validation;however,they remain constrained by the lack of effective preoperative methods for assessing pancreatic texture. In recent years,machine learning models have demonstrated superior predictive efficacy by leveraging multimodal data;however, they still rely on manual delineation,and challenges persist in terms of interpretability and clinical translation. Future studies should achieve in-depth integration of preoperative imaging and artificial intelligence,focus on multicenter validation,and develop fully automated prediction systems to provide a basis for individualized risk stratification and precise intervention more efficiently.