“Data and AI in the ICU” – Workshop on what work in the ICU of the future might look like

The social scientist partners from the University of Vienna have organized a virtual workshop to discuss the potentials of “Data Mining and AI in the ICU”. This is a short report written by the team in the University of Vienna about the fascinating workshop that took place on March 17th.

ICU4Covid aims to bring the latest technologies to the Intensive Care Units (ICUs) across Europe: high-quality Talmudical systems, new ways of collecting clinical data, and Artificial Intelligence that senses deteriorations of patients’ health before they occur. This is more than just a question of installing the technologies. To be sustainable and to unfold their full potential these technologies need to become part of the everyday routines of healthcare professionals. Therefore, ICU4Covid aims to co-create the implementation of technologies together with healthcare professionals in the ICU. To facilitate the conversation about this, the social scientist partners from the University of Vienna have organized a virtual workshop to discuss the potentials of “Data Mining and AI in the ICU”. It made available a forum in which data and machine learning scientists and medical professionals could discuss how these technologies could be used in clinical care.

The COVID-19 pandemic is the first pandemic in the digital age. Therefore, it has not only impacted the world in unprecedented ways but has also been the first pandemic about which large volumes of data were collected. Marco Pegoraro (RWTH Aachen) presented insights from the Corona Virus Aachen Study (COVAS). Using data from the first and second waves of the pandemic at the University Hospital of Aachen (UKA), he and his colleagues analyzed event logs, detailed documentations of the sequence of events, for more than 200 patients to derive a model of the treatment process these patients undwernet. Elisabetta Benevento (RWTH Aachen, University of Pisa) concluded how clinical professionals may use such treatment models to monitor and improve healthcare processes in real-time. However, she also cautioned that staff acceptance and transformation of how data is collected – for instance, are all events logged when they occur? – are the pre-conditions for these applications.

The data sets and resulting process model, such as the COVAS can be used for machine learning to predict the next steps in the treatment process. However, the shortcomings of algorithms have been well-publicized. In the ICU, biases and intransparent classifications can be lethal. In her presentation, Barbara Hammer (University of Bielefeld) demonstrated different methods to make algorithms explainable so that their classifications and recommendations are made comprehensible for the human user. A particular challenge in the context of ICU4Covid will be how the algorithms can be transferred from one hospital to the next where other conditions may exist. In other words: There cannot be one-size-fits-all algorithms to account for the differences in the hospitals within the project.

How AI can be and has been used in everyday clinical practice was the topic of the final presentation. Giovanni Paragliola (Consiglio Nazionale delle Ricerche) gave an overview over the CNR’s previous work on algorithms in the healthcare domain: a system to identify motion disorders and changes in gait patterns using wearable sensors; a recommendation system for the prescription of medication and monitoring the adherence to the treatment plan by the patient; an automated prediction system for the occurrence of hypertension to realize the early onset and to be able to respond accordingly and timely. As he demonstrated in his talk such applications of AI also raise questions of data and privacy issues as well as challenges to how healthcare is understood and organized in the future.

This workshop will not be the last of its kind. It has been a rewarding opportunity to connect data and AI scientists with healthcare professionals to discuss where the chance of these technologies may be in healthcare but also how they will impact clinical practice – only through such discussions can ICU4Covid tap into the full potential of what these technologies offer, during the project and beyond.