You are here
NLM_NIH at TREC 2016 Clinical Decision Support Track.
In this paper, we present our approach for TREC 2016 Clinical Decision Support (CDS) track. We combined methods for question analysis, query expansion, document retrieval and result fusion to find relevant documents to a given clinical question. We submitted three automatic runs using the summaries and two automatic runs using the notes, provided for the first time at the CDS track. Our experiments showed that query expansion and rank-based result fusion led to the best performance. Our runs exploring the clinical notes used MeSH for topic analysis and achieved our best P10 score of 0.2533. Using the summaries, we obtained an infNDCG score of 0.1872 and a R-prec score of 0.1465 (score in the top 10 of 107 automatic runs submitted to the 2016 CDS track).