A prediction model for colon cancer surveillance data

Area of Science:

• Oncology
• Biostatistics
• Preventive Medicine

Background:

• Colorectal cancer (CRC) surveillance in high-risk individuals requires accurate risk prediction.
• Dynamic models can improve individualized risk assessment using longitudinal patient data.

Purpose of the Study:

• To develop and validate a dynamic prediction model for advanced adenoma or colorectal cancer (AAC) risk.
• To incorporate time-dependent colonoscopy (COL) and fecal occult blood test (FOBT) results into risk prediction.

Main Methods:

• A generalized nonlinear model with a complementary log-log link, motivated by a Poisson process, was developed.
• Time-dependent covariates representing COL and FOBT results were used to update risk predictions.
• Model selection utilized Akaike information criterion, and goodness-of-fit was assessed using calibration plots.

Main Results:

• Abnormal COL results significantly increased AAC risk for 1 year post-test.
• Positive FOBTs significantly increased AAC risk for 3 months post-result.
• Updated test results as covariates demonstrated greater significance and impact on risk than baseline variables.

Conclusions:

• The proposed dynamic model effectively predicts individualized AAC risk.
• Incorporating time-dependent COL and FOBT results enhances risk prediction accuracy.
• This approach offers a valuable tool for personalized CRC surveillance strategies.