Towards Semantic Modeling of Contradictions and Disagreements: A Case Study of Medical Guidelines

Computing Conceptual Distances between Breast Cancer Screening Guidelines: An Implementation of a Near-Peer Epistemic Model of Medical Disagreement

“…Notice that unlike our earlier work Zadrozny et al (2017), in this article we have not performed any logical analysis of the guidelines. Thus the similarities and differences in distances are conceptual, they reflect the conceptual knowledge of the writers (through the word and concept embeddings of their texts),…”

“…To switch to more recent and more technical work, Bowles et al (2019) focuses on finding and resolving conflicting recommendations using a formal model and automated proof systems -it relies on a manual translation into a formal language, Labelled Event Structure. This is a very interesting work, somewhat in the spirit of our own attempts to do it, Zadrozny et al (2017), using a combination of NLP and information retrieval tools. In another article dealing with contradictory recommendations, Tsopra et al (2018) focus on the semi-automatic detection of inconsistencies in guidelines and apply their tools to antibiotherapy in primary care.…”

“…We are sure this accuracy can be improved, and we are actively working towards that goal. In particular, we (or someone else) should be able to use an architecture modeled after our previous system, Zadrozny et al (2017), to add explicit detection of contradictions to the model. Please note such effort would be highly non-trivial, since the search space of the CDC summary (as in Fig.…”

“…Since medical guidelines are developed by different medical associations which count on experts with different specialties and sub-specialties, there is a high possibility that there may be disagreement in the guidelines. And indeed, as noted by CDC (2017) and discussed in Zadrozny & Garbayo (2018); Zadrozny et al (2017), breast cancer screening guidelines contradict each other.…”

“…Zadrozny & Garbayo (2018) Zadrozny et al (2017) Garbayo (2019) we have not made a distinction between contradictions and disagreements. Thus, some of our ongoing work is about incorporating logical structures of documents and finding metrics reflecting them.Description of individual contributions to this research: HH performed the majority of experiments discussed in this article, and substantially contributed to writing.…”