Tutorial: Developing and Deploying Healthcare Predictive Models in R

Summary form only given. With a rapidly increasing amount of available data and studies proposing novel data analysis methods, researchers often neglect the importance of translational research in healthcare. In translational research, it is important to not only develop and evaluate a novel method, but also to make it useful for practical applications by letting the potential end-user to test it. This tutorial aims to demonstrate the best practices of building, evaluating and deploying predictive models in healthcare using a language and environment for statistical computing called R [1]. According to a recent poll [2] R (61%) is by far the most widely used language for data mining and analytics followed by Python (39%). Despite the popularity of R in core research environment, it is rarely found in production environment mostly due to lack of enterprise level graphical user interface support and historical orientation towards experimental work. However, most of the proposed predictive models presented in healthcare literature do not need a complex graphical interface and could be presented to the potential end-users via a simple web interface. This tutorial will present a framework for developing and deploying predictive models including data preparation, model selection, tuning and evaluation, followed by development of a simple web based graphical user interface to deploy predictive models from R. For all users with at least basic level of R knowledge, we will demonstrate that no additional knowledge is needed to create a simple one-page web application that will increase the impact of their research work by allowing them to publish their predictive models online. It has to be noted that it will be possible to follow the tutorial even if one has never used R. Participants of the tutorial will be able to download the source code of the presented framework and developed application to set up their own web-based predictive model visualizations. This tutorial is in- ended for healthcare professionals and researchers from all fields of healthcare informatics. There will be no specific knowledge needed to follow the tutorial.