Development of an artificial intelligence–based model to predict early recurrence of neuroendocrine liver metastasis after resection

Purpose: We sought to develop an artificial intelligence (AI)–based model to predict early recurrence (ER) after curative-intent resection of neuroendocrine liver metastases (NELMs). Methods: Patients with NELM who underwent resection were identified from a multi-institutional database. ER was defined as recurrence within 12 months of surgery. Different AI-based models were developed to predict ER using 10 clinicopathologic factors. Results: Overall, 473 patients with NELM were included. Among 284 patients with recurrence (60.0%), 118 patients (41.5%) developed an ER. An ensemble AI model demonstrated the highest area under receiver operating characteristic curves of 0.763 and 0.716 in the training and testing cohorts, respectively. Maximum diameter of the primary neuroendocrine tumor, NELM radiologic tumor burden score, and bilateral liver involvement were the factors most strongly associated with risk of NELM ER. Patients predicted to develop ER had worse 5-year recurrence-free survival and overall survival (21.4% vs 37.1% [P = .002] and 61.6% vs 90.3% [P = .03], respectively) than patients not predicted to recur. An easy-to-use tool was made available online: (https://altaf-pawlik-nelm-earlyrecurrence-calculator.streamlit.app/). Conclusion: An AI-based model demonstrated excellent discrimination to predict ER of NELM after resection. The model may help identify patients who can benefit the most from curative-intent resection, risk stratify patients according to prognosis, as well as guide tailored surveillance and treatment decisions including consideration of nonsurgical treatment options.