Evaluating response to treatment and predicting outcome in patients with metastatic colorectal carcinoma using statistical learning theory

Statistical Learning Theory (SLT), combined with new methods of measuring change in lesion burden, can generate a significant improvement over the current methods of evaluating response to treatment using CT; the Response Evaluation Criteria in Solid Tumors (RECIST) and World Health Organization (WHO) standards. Furthermore, SLT techniques can be used to predict patient outcome, compare the efficacy of a particular method of measuring change in lesion burden, and analyze the variability between observers. Two SLT techniques, Logistic Regression (LR) and Support Vector Machines (SVMs) were utilized to this end. The SVM technique performed significantly better than the LR technique.;Predictions of survival after 8 months in 38 patients with metastatic colorectal carcinoma using the SVM technique improved over 30% when using additional information (Visual Anatomic Scoring) with WHO or RECIST compared to WHO or RECIST size measurements alone. The best combination of features resulted in a .84 Area under the Curve (AUC) of a ROC (Receiver Operating Characteristic) curve value, which is a strong performance for an outcome predictor. When using both LR and SVMs, it was discovered that there is no statistically significant difference in performance between WHO and RECIST. The SVM and LR techniques also quantifiably demonstrated that one radiologist consistently outperformed another radiologist. This research effort shows the potential of SLT to assess new methods of measuring change in tumor lesions for evaluating response to treatment, to provide more information to patients making treatment decisions, and to deal with the issue observer variability.