Random forests algorithm using basic medical data for predicting the presence of colonic polyps.

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Colorectal cancer (CRC) is often preceded by malignant transformation of colorectal polyps.
• Early detection and removal of polyps significantly reduce CRC mortality and morbidity.

Purpose of the Study:

• To develop a predictive model for colorectal polyp presence using machine learning.
• To utilize basic patient information and common laboratory test results for polyp prediction.

Main Methods:

• A Random Forests algorithm was trained on data from 164 patients, including demographics, medical history, and lab results.
• The model was validated internally on 80% of the data and externally on 42 patients.
• Performance was compared against Generalized Linear Models (GLM) and Support Vector Machines (SVM).

Main Results:

• The Random Forest model achieved an Area Under the Curve (AUC) of 0.820 on the test set and 0.79 on external validation.
• Key predictors included body mass index, platelets, hemoglobin, triglycerides, and transaminase levels.
• Random Forests outperformed GLM and SVM in both internal and external validation.

Conclusions:

• A Random Forest prediction model effectively forecasts the presence of colonic polyps.
• The model leverages demographic data, medical history, and routine blood tests for accurate prediction.
• This tool offers significant potential for early polyp detection and CRC prevention.