A predictive model of longitudinal, patient-specific colonoscopy results

Area of Science:

• Gastroenterology and Medical Informatics
• Computational Biology and Bioinformatics
• Epidemiology and Public Health

Background:

• Individual risk for colonic neoplasia (abnormal growths in the colon) is influenced by previous colonoscopy findings.
• Existing models may not fully capture the dynamic nature of neoplasia development over time.

Purpose of the Study:

• To develop and validate a model framework for predicting longitudinal colonoscopy results.
• To personalize colorectal cancer (CRC) risk assessment based on individual patient history and characteristics.

Main Methods:

• A neoplasia natural history model was developed, incorporating patient age and four distinct neoplasia development states.
• Bayesian adjustments were applied to update parameter set combination likelihoods and model predictions after each colonoscopy.
• Longitudinal colonoscopy data from 4084 patients, including procedure details and characteristics, were used for model identification and validation.

Main Results:

• At least two sex-specific parameter sets with model adjustments were necessary to accurately capture longitudinal colonoscopy data.
• Including multiple parameter set combinations was crucial for predicting second-time colonoscopy findings in patients with a history of advanced adenomas.
• The model revealed significant variations in CRC risk based on patient age, gender, and preparation quality, highlighting the need for personalized screening intervals.

Conclusions:

• The developed model framework effectively predicts longitudinal colonoscopy results and individual patient risk for colonic neoplasia.
• This approach allows for dynamic risk assessment, incorporating new colonoscopy findings to refine predictions.
• The findings underscore the need for personalized CRC screening strategies and further investigation into current guideline recommendations.