Clinical decision making is a highly complex task. It frequently involves multiple experts from different domains. Large amounts of fragmented patient data, a multitude of clinical guidelines and a vast number of clinical studies need to be considered. Furthermore, the information integration into a model is currently performed mentally by each expert in the course of clinical decision making. Altogether, this leads to a series of problems such as incomplete models due to only partial access to data and limited temporal as well as cognitive facilities, biased models due to the different background and interest of each expert, and obviously, multiple different models that need to be communicated and interrelated in expert meetings. The most critical problem in therapeutic decision making is the weighting of therapy options under consideration of the respective hypothetical patient outcome and the mental patient model.
In this tutorial, we address the abovementioned problems and present two complementary approaches to building a predictive disease- and patient-specific therapeutic decision model supporting medical experts.
- First, we describe techniques for building a probabilistic therapy model which represents weighted causalities between patient information aggregated from medical health records and knowledge derived from medical textbooks, clinical studies, and therapeutic guidelines. The model is implemented as a Bayesian network and predicts the most appropriate therapy options as well as patient outcome by means of Bayesian inferencing. We demonstrate this approach for laryngeal cancer treatment.
- Second, we detail techniques for building a physiological patient model which represents the function of an anatomical structure as well as therapy-induced functional variations, both derived from medical images and image-based, patient-specific simulations. The latter play a crucial role in predictive medicine since they can forecast changes caused by possible therapeutic interventions. We demonstrate this approach for endovascular cerebral aneurysm treatment.
- Finally, we elaborate on an integration of both models and discuss our thoughts with the audience.
The tutorial attendees will learn about probabilistic therapy models, physiological patient models, and their integration. They will acquire knowledge on model generation as well as on model main- tenance, storage, communication, and exchange. Furthermore, they will become acquainted with ways to validate the models and to visualize and explore them for therapeutic decision making. Sufficient room for questions and discussions will allow them to consolidate the acquired knowledge, present own ideas, and jointly with the speakers, elaborate on open challenges and future steps.