Abstract
OBJECTIVE: Accurate preoperative identification of benign or malignant pancreatic cystic neoplasms (PCN) may help clinicians make better intervention choices and will be essential for individualized treatment. METHODS: Preoperative ultrasound and laboratory examination findings, and demographic characteristics were collected from patients. Multiple logistic regression was used to identify independent risk factors associated with malignant PCN, which were then included in the nomogram and validated with an external cohort. The Net Reclassification Index (NRI) and Integrated Discrimination Improvement (IDI) were calculated to evaluate the improvement in the predictive power of the new model with respect to that of a combined imaging and tumor marker prediction model. RESULTS: Malignant PCN were found in 83 (40.7%) and 33 (38.7%) of the model and validation cohorts, respectively. Multivariate analysis identified age, tumor location, imaging of tumor boundary, blood type, mean hemoglobin concentration, neutrophil-to-lymphocyte ratio, carbohydrate antigen 19-9, and carcinoembryonic antigen as independent risk factors for malignant PCN. The calibration curve indicated that the predictions based on the nomogram were in excellent agreement with the actual observations. A nomogram score cutoff of 192.5 classified patients as having low vs. high risk of malignant PCN. The model achieved good C-statistics of 0.929 (95% CI 0.890-0.968, P < 0.05) and 0.951 (95% CI 0.903-0.998, P < 0.05) in predicting malignancy in the development and validation cohorts, respectively. NRI = 0.268; IDI = 0.271 (P < 0.001 for improvement). The DCA curve indicated that our model yielded greater clinical benefits than the comparator model. CONCLUSIONS: The nomogram showed excellent performance in predicting malignant PCN and may help surgeons select patients for detailed examination and surgery. The nomogram is freely available at https://wangjunjinnomogram.shinyapps.io/DynNomapp/.