An XGBoost-Based Fitted Q Iteration for Finding the Optimal STI Strategies for HIV Patients

The computational algorithm proposed in this article is an important step toward the development of computational tools that could help guide clinicians to personalize the management of human immunodeficiency virus (HIV) infection. In this article, an XGBoost-based fitted Q iteration algorithm is proposed for finding the optimal structured treatment interruption (STI) strategies for HIV patients. Using the XGBoost-based fitted Q iteration algorithm, we can obtain acceptable and optimal STI strategies with fewer training data, when compared with the extra-tree-based fitted Q iteration algorithm, deep Q-networks (DQNs), and proximal policy optimization (PPO) algorithm. In addition, the XGBoost-based fitted Q iteration algorithm is computationally more efficient than the extra-tree-based fitted Q iteration algorithm.