Medical prevention is a very important aspect of healthcare informatics research through the prediction of medical events. In this work, we propose a deep learning approach to perform multi-label prediction on acts of medical care. The proposed approach utilizes a time-aware long short-term memory network and extends it with additional information from a fuzzy clus-tering of the same portfolio. The former mechanism (time-aware) is used to handle the tem-poral irregularity between the elements of a medical trajectory whereas the latter mechanism (fuzzy clustering) assists in modeling the heterogeneity among patients and treatments. Using a large portfolio of reimbursed medical records (over 16 million consumed acts of medical care) by a healthcare insurance in France, we show that our approach outperforms traditional and deep learning methods in medical multi-label prediction. Our work has implications for supporting medical prevention and more broadly improving the quality of healthcare service and insurance.