BMC Cancer (Jun 2023)

Two machine learning-based nomogram to predict risk and prognostic factors for liver metastasis from pancreatic neuroendocrine tumors: a multicenter study

  • Jianbo Li,
  • Long Huang,
  • Chengyu Liao,
  • Guozhong Liu,
  • Yifeng Tian,
  • Shi Chen

DOI
https://doi.org/10.1186/s12885-023-10893-4
Journal volume & issue
Vol. 23, no. 1
pp. 1 – 18

Abstract

Read online

Abstract Background Pancreatic neuroendocrine tumors (PNETs) are one of the most common endocrine tumors, and liver metastasis (LMs) are the most common location of metastasis from PNETS; However, there is no valid nomogram to predict the diagnosis and prognosis of liver metastasis (LMs) from PNETs. Therefore, we aimed to develop a valid predictive model to aid physicians in making better clinical decisions. Methods We screened patients in the Surveillance, Epidemiology, and End Results (SEER) database from 2010–2016. Feature selection was performed by machine learning algorithms and then models were constructed. Two nomograms were constructed based on the feature selection algorithm to predict the prognosis and risk of LMs from PNETs. We then used the area under the curve (AUC), receiver operating characteristic (ROC) curve, calibration plot and consistency index (C-index) to evaluate the discrimination and accuracy of the nomograms. Kaplan-Meier (K-M) survival curves and decision curve analysis (DCA) were also used further to validate the clinical efficacy of the nomograms. In the external validation set, the same validation is performed. Results Of the 1998 patients screened from the SEER database with a pathological diagnosis of PNET, 343 (17.2%) had LMs at the time of diagnosis. The independent risk factors for the occurrence of LMs in PNET patients included histological grade, N stage, surgery, chemotherapy, tumor size and bone metastasis. According to Cox regression analysis, we found that histological subtype, histological grade, surgery, age, and brain metastasis were independent prognostic factors for PNET patients with LMs. Based on these factors, the two nomograms demonstrated good performance in model evaluation. Conclusion We developed two clinically significant predictive models to aid physicians in personalized clinical decision-makings.

Keywords