Development and Internal Validation of a Machine Learning-Based Colorectal Cancer Risk Prediction Model

Background: Colorectal cancer (CRC) remains a leading cause of cancer-related mortality worldwide. While screening tools such as the fecal immunochemical test (FIT) aid in early detection, they do not provide insights into individual risk factors or strategies for primary prevention. This study aimed to develop and internally validate an interpretable machine learning-based model that estimates an individual’s probability of developing CRC using readily available clinical and lifestyle factors. Methods: We analyzed data from 154,887 adults, aged 55–74 years, who participated in the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial. A risk prediction model was built using the Light Gradient Boosting Machine (LightGBM) algorithm. To translate these findings into clinical practice, we implemented the model into a risk estimator that categorizes individuals as average, increased, or high risk, highlighting modifiable risk factors to support patient–clinician discussions on lifestyle changes. Results: The LightGBM model incorporated 12 predictive variables, with age, weight, and smoking history identified as the strongest CRC risk factors, while heart medication use appeared to have a potentially protective effect. The model achieved an area under the receiver operating characteristic curve (AUROC) of 0.726 (95% confidence interval [CI]: 0.698–0.753), correctly distinguishing high-risk from average-risk individuals 73 out of 100 times. Conclusions: Our findings suggest that this model could support clinicians and individuals considering screening by guiding informed decision making and facilitating patient–clinician discussions on CRC prevention through personalized lifestyle modifications. However, before clinical implementation, external validation is needed to ensure its reliability across diverse populations and confirm its effectiveness in real-world healthcare settings.