An Annotated Corpus of Textual Explanations for Clinical Decision Support

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 2317–2326 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 2317 An Annotated Corpus of Textual Explanations for Clinical Decision Support Roland Roller1, Aljoscha Burchardt1, Nils Feldhus1, Laura Seiffe1, Klemens Budde2, Simon Ronicke2∗, Bilgin Osmanodja2∗ 1German Research Center for Artificial Intelligence (DFKI), Berlin, Germany 2Charit´e – Universit¨atsmedizin Berlin, Germany firstname.secondname@{dfki/charite}.de Abstract In recent years, machine learning for clinical decision support has gained more and more attention. In order to introduce such applications into clinical practice, a good performance might be essential, however, the aspect of trust should not be underestimated. For the treating physician using such a system and being (legally) responsible for the decision made, it is particularly important to understand the system’s recommendation. To provide insights into a model’s decision, various techniques from the field of explainability (XAI) have been proposed whose output is often enough not targeted to the domain experts that want to use the model. To close this gap, in this work, we explore how explanations could possibly look like in future. To this end, this work presents a dataset of textual explanations in context of decision support. Within a reader study, human physicians estimated the likelihood of possible negative patient outcomes in the near future and justified each decision with a few sentences. Using those sentences, we created a novel corpus, annotated with different semantic layers. Moreover, we provide an analysis of how those explanations are constructed, and how they change depending on physician, on the estimated risk and also in comparison to an automatic clinical decision support system with feature importance. Keywords: Natural Language Explanations, XAI, Clinical Decision Support 1. Introduction Machine-learning-based recommendation or decision support systems have become a commodity in online shopping, entertainment platforms and other consumer apps and services. In many professional domains, the uptake has been comparably slower, and especially in “high-risk” domains such as medical services, a number of requirements and concerns still need to be addressed before applications from research will find broad usage, e.g., in hospitals and clinical care. In the context of automatic clinical decision support, research has shown that machine learning can outper- form physicians on very particular, narrow tasks, or can help physicians to work more efficiently (see, e.g., Gul- shan et al. (2016) or Rajpurkar et al. (2017)). Still, good performance is only one building block towards the final goal of trustworthy AI. Among the many is- sues that need to still be figured out in these complex socio-technical systems, explainability is a prominent one (Markus et al., 2021). Trying to ‘explain’ a decision made by a machine learn- ing model (or the model itself) is currently a research topic with an increasing popularity - not only in the medical context, but overall in the machine learning community. In recent years, a large variety of novel techniques were presented in the context of making ma- chine learning models – in particular neural networks – and …

Content truncated to 500 words