Differentiating walled-off pancreatic necrosis with no documented pancreatitis from mucinous cystic neoplasms: a multicenter LASSO-based nomogram.
This study aimed to investigate the clinical characteristics and imaging features of walled-off pancreatic necrosis (WOPNs) involving the pancreatic parenchyma in patients with no documented pancreatitis, and to establish a diagnostic prediction nomogram using Lasso regression to differentiate WOPNs from mucinous cystic neoplasms (MCNs).
A total of 247 cases were retrospectively collected from three independent hospitals, including surgically confirmed post-inflammatory necrotic collections meeting imaging criteria for WOPNs and pathologically confirmed MCNs. Clinical and imaging features were analyzed, and independent predictors were identified using Lasso regression. Clinical, imaging, and combined diagnostic models were constructed and assessed using ROC, calibration, and decision curve analyses (DCA). A diagnostic nomogram was developed from the best model.
Univariate analysis revealed significant differences between WOPNs and MCNs in the training cohort for age, sex, clinical symptoms, lesion location, shape, lesion margin, cyst category, incomplete septation, peripancreatic fat space, density/signal, cyst wall and/or septal thickness, mural nodule, lesion calcification, peripancreatic inflammatory changes, MPD morphology, location of ductal dilation, vascular involvement, and organ involvement (p < 0.05). Lasso regression identified three clinical features, eight imaging features, and five combined clinical and imaging features as independent risk factors, which were subsequently used to construct clinical, imaging, and combined models. The combined model achieved the highest diagnostic performance, with AUC values of 0.880 in the training cohort and 0.858 and 0.856 in the two external validation cohorts, respectively, demonstrating good sensitivity, specificity, and overall accuracy.
We established a reliable, non-invasive diagnostic prediction nomogram based on Lasso regression to differentiate WOPNs with no documented pancreatitis from MCNs.
A total of 247 cases were retrospectively collected from three independent hospitals, including surgically confirmed post-inflammatory necrotic collections meeting imaging criteria for WOPNs and pathologically confirmed MCNs. Clinical and imaging features were analyzed, and independent predictors were identified using Lasso regression. Clinical, imaging, and combined diagnostic models were constructed and assessed using ROC, calibration, and decision curve analyses (DCA). A diagnostic nomogram was developed from the best model.
Univariate analysis revealed significant differences between WOPNs and MCNs in the training cohort for age, sex, clinical symptoms, lesion location, shape, lesion margin, cyst category, incomplete septation, peripancreatic fat space, density/signal, cyst wall and/or septal thickness, mural nodule, lesion calcification, peripancreatic inflammatory changes, MPD morphology, location of ductal dilation, vascular involvement, and organ involvement (p < 0.05). Lasso regression identified three clinical features, eight imaging features, and five combined clinical and imaging features as independent risk factors, which were subsequently used to construct clinical, imaging, and combined models. The combined model achieved the highest diagnostic performance, with AUC values of 0.880 in the training cohort and 0.858 and 0.856 in the two external validation cohorts, respectively, demonstrating good sensitivity, specificity, and overall accuracy.
We established a reliable, non-invasive diagnostic prediction nomogram based on Lasso regression to differentiate WOPNs with no documented pancreatitis from MCNs.
Authors
Xu Xu, Huang Huang, Duan Duan, Chen Chen, Zheng Zheng, Wang Wang, Fan Fan, Wang Wang, Yu Yu
View on Pubmed