Automated machine learning predicts liver metastases in patients with early-onset gastroenteropancreatic neuroendocrine tumors

Abstract

Background: The incidence of early-onset gastroenteropancreatic neuroendocrine tumors (GEP-NETs) is increasing, with liver metastases often occurring early and adversely affecting prognosis. This study aimed to develop a predictive model for liver metastases detection in patients with early-onset GEP-NETs (<50 years) using an automated machine learning (AutoML) approach.

Methods: A retrospective analysis was conducted on patients diagnosed with early-onset GEP-NETs [2000-2021] using data from the Surveillance, Epidemiology, and End Results (SEER) database. Patients were randomly divided into a training set (n=8,983) and a validation set (n=3,819) in a 7:3 ratio. A nomogram-based scoring system was constructed using least absolute shrinkage and selection operator (LASSO) and logistic regression. AutoML was applied to build predictive models using gradient boosting machine (GBM), generalized linear model (GLM), deep learning (DL), and distributed random forest (DRF) algorithms. Model performance was assessed using receiver operating characteristic (ROC), calibration, decision curve analysis (DCA), and interpretability tools including SHapley Additive exPlanations (SHAP), partial dependence plots (PDPs), and locally interpretable model-agnostic explanations (LIME) plots.

Results: A total of 12,802 patients were included, of whom 1,187 (9.3%) developed liver metastases, comprising 851 (9.5%) and 336 (8.8%) cases in the training and validation sets, respectively. Comparative analyses demonstrated that the AutoML models outperformed traditional logistic regression models, with the GBM algorithm achieving the highest performance. The GBM model achieved an area under the curve (AUC) of 0.961 in the training set and 0.953 in the validation set. Tumor location was identified as the most important predictor in the GBM model, followed by surgery, tumor size, chemotherapy, and T-staging.

Conclusions: The AutoML model leveraging the GBM algorithm provides a robust and clinically valuable tool for the early prediction of liver metastases in patients with early-onset GEP-NETs.

Keywords: Automated machine learning (AutoML); age; gastroenteropancreatic neuroendocrine tumors (GEP-NETs); liver metastasis; predictive modeling.

Figure 1

The flowchart of this study. DL, deep learning; DRF, distributed random forest; GBM, gradient boosting machine; GEP-NENs, gastroenteropancreatic neuroendocrine neoplasms; GLM, generalized linear model; LASSO, least absolute shrinkage and selection operator.

Figure 2

Nomogram of logistic regression model for predicting liver metastasis in patients with early-onset gastroenteropancreatic neuroendocrine tumors. N, node.

Figure 3

The ROC curves (A,D), calibration curves (B,E), and decision curves (C,F) of the logistic regression model in the training and validation sets. AUC, area under the curve; CI, confidence interval; ROC, receiver operating characteristic

Figure 4

Variable importance and SHAP of the GBM model in the training cohort. Panel (A) shows that tumor location is the most important feature. As illustrated in (B), when the variable value approaches 1, the likelihood of liver metastasis increases for the patient. GBM, gradient boosting machine; N, node; SHAP, SHapley Additive exPlanations; T, tumor.

Figure 5

LIME visualization of variable importance in a random sample from the validation cohort (demonstrating the impact of key variables on individual predictions; p0 represents no liver metastasis, and p1 represents liver metastasis). DX, diagnosis; LIME, locally interpretable model-agnostic explanations; N, node; T, tumor.