Generative models are powerful deep learning methods that have demonstrated great potential for applications in the medical domain. Their performance has continuously improved over the last few years. They address several challenges that are due to the lack of sufficient publicly available medical training data, by providing approaches for synthetic dataset creation, unsupervised pre-training, transfer learning, or generation of missing image modalities. However, generative models often generalize poorly, e.g., to data from different domains and are criticized as black boxes due to a lack of interpretability and controllability. Some of these concerns can be handled by analyzing and disentangling the latent space representation of generative models, encouraging a comprehensive, human interpretable, and compressed representation of the data. The goal of this MICCAI workshop is to analyze the different proposed definitions of disentangled representations, review state-of-the-art methods to achieve disentanglement, evaluate and compare existing quality metrics, as well as to discuss new ideas and methods. By considering the mathematical background alongside applied methods, the workshop will combine theory and practice. Furthermore, these foundations will be discussed in the context of present and future medical applications. Results of the workshop and future directions of the disentanglement approach for medical applications are planned to be summarized in a workshop paper.