Development and validation of a machine learning model for colorectal cancer status classification using NHANES data:...

Area of Science:

• Oncology
• Biostatistics
• Machine Learning

Background:

• Colorectal cancer (CRC) presents a significant global health challenge.
• Existing methods for CRC identification often lack comprehensiveness.
• There is a need for tools utilizing routinely collected data to aid in CRC status identification.

Purpose of the Study:

• To develop and validate an interpretable machine learning model for classifying colorectal cancer (CRC) status.
• To compare the performance of various machine learning algorithms for CRC prediction.
• To identify key predictors of CRC status from a multidomain dataset.

Main Methods:

• Utilized data from the National Health and Nutrition Examination Survey (1999-2018).
• Employed a case-control design with logistic regression, random forest, SVM, k-NN, and extreme gradient boosting (XGBoost).
• Assessed model performance using discrimination metrics (AUC) and post hoc calibration techniques (Platt scaling, SHAP).

Main Results:

• XGBoost demonstrated superior discrimination in the validation cohort with an AUC of 0.787.
• Key predictors identified included alcohol use, hypertension, age, triglycerides, and serum cotinine.
• The developed web-based classifier showed potential for clinical evaluation and triage.

Conclusions:

• An interpretable machine learning framework effectively classified CRC status in a large population-based cohort.
• The XGBoost model offers strong discriminatory power for CRC status identification.
• Further validation is recommended for probability-based outputs to support clinical decision-making.