Development and validation of a clinical prediction model for fatty pancreas disease: from CT-based indicators to an easy-to-use prediction application

作者：Yuan Zhou, Cong Chen, Fangsheng Chen, Weiwei Xue, Hanting Dai, Haoteng Luo, Tianhong Teng, Heguang Huang, Fengchun Lu · 发表于：Nutrition Journal · 年份：2026 · DOI：10.1186/s12937-026-01299-6 · 研究领域：Pancreatitis Pathology and Treatment、Pancreatic and Hepatic Oncology Research、Liver Disease Diagnosis and Treatment

BACKGROUND: Fatty pancreas disease (FPD) is closely associated with the pathogenesis of pancreatic cancer (PC) and diabetes mellitus (DM). However, current imaging modalities for assessing intra-pancreatic fat deposition (IPFD) have limitations and are not widely used in clinical practice. Developing a practical prediction tool for FPD would facilitate the evaluation of pancreatic health. METHODS: A total of 852 subjects were included in the model construction cohort, and 202 PC patients were included to investigate the association between fat deposition and PC. Quantitative assessment of IPFD and diagnosis of FPD were based on the pancreas-to-spleen attenuation ratio (P/S ratio) from CT scans. Multivariate logistic regression was used to identify risk factors associated with FPD, and a web-based application was deployed based on these factors. The performance of the model was validated in an internal cohort with respect to discrimination, calibration and clinical benefit. RESULTS: The incidence of PC and DM significantly increased as the P/S ratio decreased, particularly when P/S ratio fell below 0.8. Multivariate analysis identified seven independent risk factors: age > 65 years, abnormal waist circumference, abnormal gamma-glutamyl transferase, fasting plasma glucose > 6.1mmol/L, NLR > 1.97, fatty liver index > 24.7, and mFIB-4 > 3.05. A web application was deployed based on these risk factors. The model demonstrated good discrimination, with an AUC of 0.750 (95% CI: 0.707...