Designing and Evaluating Explanations for a Predictive Health Dashboard: A User-Centred Case Study

Abstract

As predictive health technologies become increasingly prevalent, the need for effective explanations that aid health experts and practitioners in understanding the underlying factors driving predictions is paramount. While many different explanation methods have been elaborated, recent research suggests that these explanations are often too complex for AI novices. In addition, little work has been done to evaluate whether the proposed methods indeed enhance human interpretability. In this study, we develop and evaluate explanations tailored to the needs of health experts, following an iterative user-centred design process to ensure understanding and usefulness of our explanation designs. Our findings underscore the importance of data-centric explanations, prioritising an emphasis on the underlying data rather than solely focusing on the model's internal workings. Additionally, explanations should not only highlight points of congestion or areas for improvement but also emphasise positive aspects to promote a holistic understanding of the predictive health dashboard.