Program Information
Biomarkers Evaluated During Radiotherapy Improve Outcome Predictions in Liver Cancer Patients
C Grassberger*, D Craft , D Duda , T Hong , T Bortfeld , Massachusetts General Hospital, Boston, MA
Presentations
WE-G-FS1-3 (Wednesday, August 2, 2017) 4:30 PM - 6:00 PM Room: Four Seasons 1
Purpose: After curative radiotherapy to the liver, patients suffer from high mortality despite low local recurrence rates, highlighting the potential of personalized treatments for this indication. In this study we develop the best predictive model from clinical factors pre-treatment, and then update it with additional blood biomarkers taken before and during treatment to investigate their value for outcome prediction.
Methods: We analyzed 83 liver cancer patients who received proton radiotherapy at our institution. The standard clinically available information consisted of all demographic data, tumor and liver volume, underlying liver diseases, chemotherapy and CLIP/CTP scores. We consecutively added two sets of predictors: 1) other standard liver biomarkers, 2) non-standard liver biomarkers and circulating lymphocyte counts during treatment.Based on these features we employed 4 commonly used classification techniques (k nearest neighbours, support vector machines (SVM), random forests and naïve Bayes) to predict survival 2 years after treatment (binary classification). Principal Component Analysis was used for feature-reduction, 5-fold cross validation to protect against overfitting and area under the receiving operator characteristic curve (AUC) and sensitivity were used to score the models.
Results: The best models based on all standard clinical features predicted survival very poorly, with AUC<0.6 for all models and poor sensitivity. Adding the liver biomarkers markedly improved model performance to (AUC=0.68), but still showed very poor sensitivity to classify patients with high confidence. Adding the additional (in-treatment) biomarkers not only improved general model performance (AUC=0.79), but especially boosted sensitivity. Based on cross-validation we estimate that we can detect ~50% of the patients who will be alive at 2 years with a very high confidence (>95%).
Conclusion: Biomarkers evaluated during radiotherapy significantly improve outcome predictions for liver patients and could enable optimal stopping strategies that balance damage to the liver with probability of tumor control.
Contact Email: