Uncertainty-Aware Text-to-Program for Question Answering on Structured Electronic Health Records

Question Answering on Electronic Health Records (EHR-QA) has a significant\nimpact on the healthcare domain, and it is being actively studied. Previous\nresearch on structured EHR-QA focuses on converting natural language queries\ninto query language such as SQL or SPARQL (NLQ2Query), so the problem scope is\nlimited to pre-defined data types by the specific query language. In order to\nexpand the EHR-QA task beyond this limitation to handle multi-modal medical\ndata and solve complex inference in the future, more primitive systemic\nlanguage is needed. In this paper, we design the program-based model\n(NLQ2Program) for EHR-QA as the first step towards the future direction. We\ntackle MIMICSPARQL*, the graph-based EHR-QA dataset, via a program-based\napproach in a semi-supervised manner in order to overcome the absence of gold\nprograms. Without the gold program, our proposed model shows comparable\nperformance to the previous state-of-the-art model, which is an NLQ2Query model\n(0.9% gain). In addition, for a reliable EHR-QA model, we apply the uncertainty\ndecomposition method to measure the ambiguity in the input question. We\nempirically confirmed data uncertainty is most indicative of the ambiguity in\nthe input question.\n