The deep learning revolution of the past years promised to deliver hitherto-unprecedented improvements in understanding massive amounts of data. Given the prevalence of such data sets in the clinical practice, ranging from time series of vital parameters of patients to irregularly-sampled information about drugs that are administered, this domain constitutes a prime target for machine learning research. The hope is that such clinical machine learning approaches are capable of (foremost) improving patient welfare, detecting novel biomarkers for complex syndromes such as sepsis or circulatory failure, and may assist doctors in their daily routine.

Clinical data, however, is also fraught with idiosyncratic challenges that need to be overcome in order for machine learning models to perform well. One of these challenges, for example, is that some measurements are sampled at irregular time intervals. This necessitates special choices for the models. Other hurdles include differences in measurement modalities—impeding the transfer of models between different hospital sites, for example—and differences in prevalence (for classification tasks), exacerbating model comparison.

In this track, we will bring together practitioners and researchers to showcase state-of-the-art machine learning models for the clinical practice. Particular emphasis will be placed on discussions about the use of machine learning for prospective studies. Which additional aspects (concerning ethics, legal discussions, and many more) have to be considered? What success stories are already out there? What can we learn from successful, ongoing, or failed implementations? We aim to provide a track with stimulating discussions about all of these aspects, culminating (ideally) in participants authoring a white paper detailing the future of this field.