Leveraging massive electronic health records (EHR) brings tremendous promises to advance clinical and pre-cision medicine informatics research. However, it is very challenging to directly work with multifaceted patient information encoded in their EHR data. Deriving ef-fective representations of patient EHRs is a crucial step to bridge raw EHR information and the endpoint ana-lytical tasks, such as risk prediction or disease subtyp-ing. In this paper, we propose Health-ATM, a novel and integrated deep architecture to uncover patients' com-prehensive health information from their noisy, longitu-dinal, heterogeneous and irregular EHR data. Health-ATM extracts comprehensive multifaceted patient in-formation patterns with attentive and time-aware mod-ulars (ATM) and a hybrid network structure composed of both Recurrent Neural Network (RNN) and Convolu-tional Neural Network (CNN). The learned features are finally fed into a prediction layer to conduct the risk pre-diction task. We evaluated the Health-ATM on both artificial and real world EHR corpus and demonstrated its promising utility and efficacy on representation learning and disease onset predictions.