A Method for Computing Conceptual Distances between Medical Recommendations: Experiments in Modeling Medical Disagreement

Using natural language processing tools, we investigate the semantic differences in medical guidelines for three decision problems: breast cancer screening, lower back pain and hypertension management. The recommendation may cause undue variability patient treatments outcomes. Therefore, having a better understanding of their causes can contribute to discussion on possible remedies. We show that these recommendations are highly correlated with knowledge brought problem by different societies, as reflected conceptual vocabularies used groups authors. While this article is case study using sets guidelines, proposed methodology broadly applicable. Technically, our method combines word embeddings novel graph-based similarity model comparing collections documents. For main study, use CDC summaries (very short documents) full (long) texts represented bags concepts. other studies, compare text abstracts tables, summarizing between recommendations. approach evaluated models distance measures. In all experiments, results statistically significant. discuss significance results, extensions, connections domains knowledge. conclude automated methods, although not perfect, be applicable comparisons enable analysis at scale.