Accurate Prediction of Response to Interferon-based Therapy in Egyptian Patients with Chronic Hepatitis C Using Machine-learning Approaches

Hepatitis C virus' patients with genotypes 1 & 4 have break-even response rates to Pegylated-Interferon (Peg-IFN) and Ribavirin (RBV) treatment. Furthermore, the incompliance to the treatment because of its high cost and related unfavorable effects makes its prediction of paramount importance. By using machine-learning techniques, a significantly accurate predictive model constructed to predict Egyptian patients' response based on their clinical and biochemical data. The model uses Artificial Neural Networks (ANN) and Decision Trees (DT) to achieve this goal. Two-hundred patients treated with Peg-IFN and RBV, 83 responders (41%), and 117 non-responders (59%) retrospectively studied to extract informative features and train the Neural Networks and Decision Trees. Optimization was done by using six different Neural Network architectures, starting with an input layer of 12 neurons, a hidden layer of 70 to 180 neurons and an output layer containing a single neuron. For decision Trees (DTs), the CART classification algorithm was used. Six DTs with two classes, pruning levels of 9, 11, 13, and 17, and nodes from 45 to 61 were investigated. Among the 12 features in the study, the most statistically significant informative features were the patient's Histology activity index, fibrosis, viral-load, Alfa-feta protein and albumin. Validation of the models on a 20% test set was then performed. The best and average accuracy for the ANN and DT models were 0.76 and 0.69, and 0.80 and 0.72 respectively. Sensitivity and specificity were 0.95 and 0.39, and 0.89 and 0.78 respectively. We conclude that decision trees gave a higher accuracy in predicting response, and would help in proper therapy options for patients.