DrugEHRQA: A Question Answering Dataset on Structured and Unstructured Electronic Health Records For Medicine Related Queries

This paper develops the first question answering dataset (DrugEHRQA)\ncontaining question-answer pairs from both structured tables and unstructured\nnotes from a publicly available Electronic Health Record (EHR). EHRs contain\npatient records, stored in structured tables and unstructured clinical notes.\nThe information in structured and unstructured EHRs is not strictly disjoint:\ninformation may be duplicated, contradictory, or provide additional context\nbetween these sources. Our dataset has medication-related queries, containing\nover 70,000 question-answer pairs. To provide a baseline model and help analyze\nthe dataset, we have used a simple model (MultimodalEHRQA) which uses the\npredictions of a modality selection network to choose between EHR tables and\nclinical notes to answer the questions. This is used to direct the questions to\nthe table-based or text-based state-of-the-art QA model. In order to address\nthe problem arising from complex, nested queries, this is the first time\nRelation-Aware Schema Encoding and Linking for Text-to-SQL Parsers (RAT-SQL)\nhas been used to test the structure of query templates in EHR data. Our goal is\nto provide a benchmark dataset for multi-modal QA systems, and to open up new\navenues of research in improving question answering over EHR structured data by\nusing context from unstructured clinical data.\n