Analyzing the Performance of Information Extraction System for Annotation of Patient Discharge Summary

Abstract

In large tertiary hospitals where patient turnover is exorbitantly high, physicians have a tough time understanding the patients discharge summary particularly if it is very voluminous. Discharge summaries and other medical reports have an unstructured format that includes details on various topics, including illnesses, treatments, and medications. Because of their narrative structure, it is difficult to extract useful and crucial information from them within a short meeting. Since the physician has to spent much time in deciphering the voluminous medical reports of patients before any treatment plans, the quality time of the physicians is wasted proving detrimental in the delivery of quality healthcare. This research intends to overcome the said problem by exploring the various techniques for automated extraction of vital information from discharge summary. Though there are several Information Extraction (IE) techniques for capturing clinical information from medical documents, this paper aims to compare the five most popular and open-source tools: MedTagger, GATE, cTAKES, NCBO Annotator and CLAMP. Experiments were carried out on 108 discharge summaries obtained from MTsamples, and the results indicate that CLAMP outperforms other tools with regard to recall, precision, and F-score value; proving that it could be used as an effective summarization during doctor-patient dialogues.