EOCRCPred: an AI Model to Predict Survival in EOCRC Patients After Surgery

Overview

The goal of this observational study is to develop a predictive model for overall survival in patients under the age of 50 who have undergone surgery for early-onset colorectal cancer (EOCRC). The main question it aims to answer is: Can machine learning models accurately predict the long-term survival of EOCRC patients after surgical treatment? Participants who have already undergone surgery for EOCRC as part of their regular medical care will have their clinical data analyzed, with survival outcomes tracked through follow-up assessments. An online survival calculator will also be developed to aid clinicians and patients in predicting personalized survival outcomes.

To avoid duplicating information that will be entered or uploaded elsewhere in the record, here is a concise summary of the key components of the study: * Study Title\*\*: \*EOCRCPred: An AI Model to Predict Survival in Early-onset Colorectal Cancer Patients After Surgery\* * Introduction\*\*: This study addresses the increasing incidence and mortality of early-onset colorectal cancer (EOCRC) in patients under 50. EOCRC exhibits distinct clinical and pathological features compared to late-onset CRC, including higher recurrence rates and advanced disease stages at diagnosis. Current predictive models for postoperative outcomes in EOCRC are limited, highlighting the need for specialized tools to guide treatment decisions. * Objectives\*\*: 1. Develop AI models for predicting overall survival (OS) in postoperative M0 EOCRC patients. 2. Propose a new survival risk stratification system. 3. Deploy an online survival calculator to assist clinical decision-making. * Methods\*\*: * \*\*Data Source\*\*: SEER database (2010-2019) for training/testing; two Chinese hospitals for external validation (2014-2024). * \*\*Inclusion Criteria\*\*: Pathologically confirmed primary EOCRC, radical surgery (stage I-III), and complete follow-up. * \*\*Models\*\*: Six predictive models, including CoxPH, RSF, S-SVM, XGBSE, GBSA, and DeepSurv. * \*\*Evaluation Metrics\*\*: Discrimination (C-index, time-dependent AUC), calibration (Brier score, calibration curves), and clinical utility (Decision Curve Analysis). * Statistical Analysis\*\*: Comparisons were made using t-tests, Mann-Whitney U tests, and chi-square tests, with P \< 0.05 indicating significance. \*\*Risk Stratification\*\*: Risk groups were classified based on RSF-derived scores (low, intermediate, high), and survival differences were assessed via Kaplan-Meier curves and log-rank tests. This streamlined summary covers the primary goals, methodology, and analysis without repeating specifics that will be detailed in other sections of the record.