Using machine learning to construct nomograms for patients with metastatic colon cancer

Area of Science:

• Oncology
• Data Science
• Biostatistics

Background:

• Predicting overall survival (OS) for patients with synchronous colon cancer metastases is challenging due to high variability.
• Accurate predictive models are needed to guide clinical decision-making and patient care.

Purpose of the Study:

• To develop and validate machine learning-based nomograms for predicting 3-year OS in patients with right colon (RC) and left colon (LC) cancer and synchronous metastases.
• To improve the accuracy of OS prediction in this patient population.

Main Methods:

• Utilized the National Cancer Database (2010-2014) including 11,018 RC and 8,346 LC patients with synchronous metastases.
• Developed nomograms using Cox proportional hazard regression with lasso regression, splitting data into training and testing sets.
• Evaluated model calibration and validation using concordance index (c-index).

Main Results:

• Nomograms demonstrated good predictive accuracy, with predicted OS within 95% confidence intervals of observed OS in four out of five risk groups for both RC and LC models.
• Externally validated 3-year c-indexes were 0.794 for RC models and 0.761 for LC models.
• Machine learning models showed improved accuracy in predicting OS compared to traditional methods.

Conclusions:

• Machine learning-driven nomograms provide more accurate OS predictions for patients with metastatic colon cancer.
• These nomograms can serve as valuable tools for shared decision-making between clinicians and patients regarding cancer treatment.
• The study highlights the potential of leveraging big data and machine learning in precision oncology.