Title : Early risk prediction of non-alcoholic fatty pancreas disease: A machine learning–based nomogram from the UK Biobank cohort
Abstract:
Background: Non-alcoholic fatty pancreas disease (NAFPD) is an under-recognized metabolic disorder with no standardized diagnostic criteria, making early detection challenging. This study aimed to identify easily accessible clinical indicators strongly associated with NAFPD, and to develop a simple, practical, and non-invasive diagnostic prediction tool for early screening and risk assessment.
Method: A cohort of 8,746 adults from the UK Biobank was analyzed to identify indicators associated with NAFPD. Candidate predictors from lifestyle factors, routine blood tests, biochemical parameters, and composite metabolic and inflammatory indices were first screened using LASSO regression, followed by machine learning modeling to refine and optimize the selection. The final set of indicators was then used to construct a diagnostic nomogram, providing a visual and practical tool for NAFPD prediction. Its performance, robustness, and clinical utility were assessed using ROC curve analysis and decision curve analysis (DCA).
Results: Following LASSO and machine learning-based screening, a diagnostic nomogram for NAFPD was constructed using seven indicators: sex, age, body mass index (BMI), red blood cell count, red cell distribution width to high-density lipoprotein cholesterol ratio (RDW/HDL), cystatin C, and C-reactive protein to albumin ratio (CAR). The nomogram showed good discrimination, with an AUC of 0.769 in the test set. Its overall accuracy was 0.725, with a sensitivity of 0.855, specificity of 0.488, precision of 0.753, and an F1 score of 0.801.
Conclusion: The seven-indicator nomogram offers a simple, accessible, and non-invasive tool for accurately predicting NAFPD risk. It demonstrates robust discrimination and generalizability across populations, supporting early clinical screening and individualized risk assessment. This tool also provides a practical framework for future studies on pancreatic steatosis and may guide the management and prognosis of patients at risk of NAFPD.
Keywords: Non-alcoholic Fatty Pancreas Disease; UK Biobank; Machine Learning; Cohort Study; Nomogram; Risk Prediction
