Todays’ conventional monitoring techniques of patients during intensive chemotherapy treatment protocols for aggressive hematologic malignancies are cumbersome and often fail to detect life threatening complications early. Continuous monitoring of vital signs by means of medical wearables will potentially lead to an earlier diagnosis and better treatment. The major challenge is to extract clinically relevant information from the large volume of data provided by wearables. The dataset from the ASSISTO trial contains 31,315.4 hours of records obtained by a wearable from 79 patients during intensive chemotherapy treatment. According to the clinical documentation, there is a marked imbalance in the data in favor of episodes with absence of complications. Those episodes in the data set were manually annotated as “normal”. In ASSISTO, we take an Out-of-distribution (OOD) detection method to identify whether defined episodes of vital signs are “normal” or not in order to detect clinical complications. We use self-supervised contrastive representation learning to distinguish “normal” (in-distribution) and OOD episodes. We randomly split the “normal” episodes into two subsets, the training (90%) and test (10%) datasets. The training dataset is then used to train the model and the test dataset is used for the OOD detection. For a given sample x, we find the cosine similarity to the nearest training sample and x is classified as OOD if the cosine similarity falls below a threshold.