1: \begin{abstract}

2: Dynamic treatment regimes assign personalized treatments to patients sequentially over time based on their baseline information and time-varying covariates. %in a multi-stage decision process. 

3: In mobile health applications, %a mobile-health setting, such regimes will be developed based on collected patient information over a long period of study. The covariates of patients involved in decision making often include data 

4: these covariates are typically collected at different frequencies over a long time horizon. In this paper, we propose a deep spectral Q-learning algorithm, which integrates principal component analysis (PCA) with deep Q-learning to handle the mixed frequency data. In theory, we prove that the mean return under the estimated optimal policy converges to that under the optimal one and establish its rate of convergence. The usefulness of our proposal is further illustrated via simulations and an application to a diabetes dataset.

5: \end{abstract}

6: