Development and External Validation of a Prediction Model for Colorectal Cancer Among Patients Awaiting Surveillance...

Area of Science:

• Gastroenterology
• Oncology
• Epidemiology

Background:

• Surveillance colonoscopy demand can exceed healthcare capacity, particularly post-pandemic.
• No existing tools prioritize patients for surveillance colonoscopy based on colorectal cancer (CRC) risk.
• This study addresses the need for a predictive model to identify high-risk individuals for CRC during surveillance.

Purpose of the Study:

• To develop and validate a multivariable prediction model for CRC detection during surveillance colonoscopy.
• To compare the performance of the multivariable model against a guideline-based model using only polyp characteristics.
• To assess the model's performance across different time periods and adapt it through updating.

Main Methods:

• Logistic regression was employed for model development using data from 2014-2019.
• Predictors included index colonoscopy findings, adenoma detection rates, and patient clinical characteristics.
• Internal validation used a separate cohort (n=15,854), and external validation used 2020-2022 data (n=30,015), followed by model updating.

Main Results:

• The multivariable model identified polyp size ≥10 mm, low adenoma detection rate, older age, and smoking history as significant CRC predictors.
• The multivariable model demonstrated superior performance compared to the guideline-based model (AUC 0.73 vs 0.52 in internal validation).
• Initial performance decline at external validation was recovered after model updating (AUC 0.72).

Conclusions:

• A multivariable prediction model utilizing common clinical factors can effectively prioritize patients at high risk for CRC during surveillance.
• Regular updates are crucial to maintain model performance and address potential drift over time.
• This tool can aid in resource allocation when surveillance colonoscopy capacity is limited.