EP591 - Development of an Artificial Intelligence-Based Predictive Model for 3-Year and 5-Year Survival in Rectal Cancer

Sunday, May 10, 2026

2:35 PM - 2:40 PM ET

ePoster Presenter(s)

SA

Seyma N. Avci, MD

General Surgery Resident
Cleveland Clinic Florida
Cleveland Clinic Florida
Sunrise, FL, United States

Disclosure information not submitted.

Author(s)

Sameh H. Emile, MBBCh, MSc, MD, FACS

Project Scientist
Cleveland Clinic Florida
Cleveland Clinic Florida
Weston, FL, United States

Disclosure information not submitted.
Nir Horesh, MD

Resident Physician
Cleveland Clinic Florida
Tel Aviv University
Weston, FL, United States

All of the relevant financial relationships listed below have been mitigated.
This individual has no financial relationships with ineligible companies.
Jingyuan Zhang, MD (she/her/hers)

Resident
Cleveland Clinic Florida
Cleveland Clinic Florida
Weston, FL, United States

Disclosure information not submitted.

Senior Author(s)

Steven D. Wexner, MD, PhD (Hon)

Surgeon
Cleveland Clinic Florida
Cleveland Clinic Lerner College of Medicine at Case Western Reserve University
Weston, Florida, United States

Disclosure information not submitted.

All of the relevant financial relationships listed below have been mitigated.

Nir Horesh, MD: This individual has no financial relationships with ineligible companies.

Purpose/Background: Colorectal cancer is the second most common cause of cancer-related deaths. Artificial Intelligence (AI) applications using diverse data types have been introduced to generate individualized survival predictions that often outperform traditional statistical methods. This study aimed to develop and validate AI models to predict 3-year and 5-year survival in rectal cancer patients and to assess the relative importance of predictive features.

Methods/Interventions: This retrospective cohort study used data from the National Cancer Database (NCDB) from 2004 to 2020 to develop two separate models to predict 3-year and 5-year survival in rectal adenocarcinoma patients. The dataset was randomly divided into training (70%), testing (15%), and validation (15%) dataset. Gradient-boosted decision tree models were developed using the XGBoost library. Model performance was assessed using accuracy, precision, recall, and the area under the receiver operating characteristic curve (ROC AUC). Predictors included demographics, stage, operation type and approach, perioperative variables, and systemic therapies.

Results/Outcomes: A total of 285,868 patients was included in the database and analyzed. The 3-year survival model achieved an accuracy of 82.0%, precision of 84.7%, recall of 90.2%, and ROC AUC of 0.87 (Figure 1). The 5-year survival model achieved accuracy of 80.3%, precision of 79.6%, recall of 85.4%, and ROC AUC of 0.88 (Figure 2). Feature importance analysis revealed that the 3-year model was more influenced by short-term oncologic and surgical quality factors (e.g., surgical margins, surgery type, staging, biomarkers, early metastases, readmission), whereas the 5-year model reflected long-term disease biology, comorbidities, systemic therapies, and treatment era (e.g., year of diagnosis, SARS-CoV-2 status, radiation dose, age, distant metastases).

Conclusion/Discussion: AI-based predictive models demonstrated strong accuracy and discrimination in predicting 3-year and 5-year OS in rectal cancer patients. These models may serve as valuable tools to support personalized treatment planning, counseling, and follow-up strategies. Future work should integrate additional clinical and molecular data and validate performance across external datasets.

Figure 1

3-year survival model
Figure 2

5-year survival model