Colon and rectal cancer is now the fourth most common cancer diagnosis in the United States. Although scientific advances in recent years have improved overall survival rates for metastatic disease, overall survival remains highly variable. Previous attempts to build models for predicting patient survival, furthermore, have historically amounted to predictions that are only slightly better than a coin flip.
A study led by Beiqun "Mark" Zhao, MD and co-authored by Bryan Clary, MD, recently published in the Journal of Gastrointestinal Surgery, analyzed how machine learning can more accurately predict survival of patients with colorectal cancer.
The researchers used data from National Cancer Database to create a training dataset to build both a machine-learning model and a testing dataset to externally validate the model. They then graphically calculated three-year overall survival and assessed the model using concordance indexes, which estimate the probability of concordance between predicted and observed outcomes, Dr. Zhao and his team were able to create predictive models with superior accuracy to those previously published. This increase in predictive ability is a great assistance to clinicians in understanding treatment of metastatic rectal cancer.