Tensor decomposition has recently been gaining attention in the machine learning community due to its versatility in processing large-scale data. In particular, it has become popular for the analysis of Electronic Health Records (EHR). However, this task becomes signifi- cantly more difficult when the data follows complex temporal patterns. This paper introduces the notion of a temporal phenotype as an arrange- ment of features over time and it proposes SWoTTeD (Sliding Window for Temporal Tensor Decomposition), a novel method to discover hidden temporal patterns. SWoTTeD integrates several constraints and regu- larizations to enhance the interpretability of the extracted phenotypes. We validate our proposal using both synthetic and real-world patient data from MIMIC-IV and the Greater Paris University Hospital. The results show that SWoTTeD outperforms the recent state-of-the-art tensor decomposition models.