Medical Entity Corpus with PICO elements and Sentiment Analysis

Medical Entity Corpus with PICO Elements and Sentiment Analysis Markus Zlabinger1, Linda Andersson1, Allan Hanbury1, Michael Andersson2, Vanessa Quasnik3, Jon Brassey4 1TU Wien, 2Stockholm University, 3Uppsala University, 4Trip Database Austria, Sweden, United Kingdom 1forename.surname@tuwien.ac.at, 2micche@gmail.com, 3vanessa@quasnik.de, 4jon.brassey@tripdatabase.com Abstract In this paper, we present our process to establish a PICO and a sentiment annotated corpus of clinical trial publications. PICO stands for Population, Intervention, Comparison and Outcome — these four classes can be used for more advanced and specific search queries. For example, a physician can determine how well a drug works only in the subgroup of children. Additionally to the PICO extraction, we conducted a sentiment annotation, where the sentiment refers to whether the conclusion of a trial was positive, negative or neutral. We created both corpora with the help of medical experts and non-experts as annotators. Keywords: sentiment analysis, PICO, medical corpus, annotation 1. Introduction Text mining, like data mining or knowledge discovery, is a process to discover implicit knowledge and potentially use- ful patterns from large text collections. For machine learn- ing (ML) algorithms, it is essential that the labeled data is designed in such a way that the optimal learning for an al- gorithm is achieved. We extracted PICO (Figure 1) elements from the text of publications of clinical trial results in order to improve a medical search mechanism of clinical questions. In this paper, we present the process of creating an annotated, phrase-level PICO (Population, Intervention, Comparison, Outcome) corpus and a sentence-level sentiment analysis corpus. In contrast, in other PICO annotation approaches (Boudin et al., 2010; Kim et al., 2011; Wallace et al., 2016), the PICO elements were only labeled on a sentence-level or abstract-level. The PICO elements are associated with different aspects of noun phrases and domain terminology. The Population (P) elements generally consist of a patient description (e.g. children, men) with one or more post modifications; e.g., patient over forty with type 2 diabetes. The Intervention (I) respectively Comparison (C) describes a treatment method (e.g. drug treatment, surgery) and the Outcome (O) de- scribes the aim of a conducted study (e.g. reduce pain). Our contributions can be summarized in three main points: i) creation of a corpus dataset labeled with PICO elements on a phrase-level, ii) a sentiment analysis of the Outcome, i.e. if an Intervention had a positive effect on a target population or not, and iii) description of a new annotation methodology for community annotation of medical data, which requires less domain-specific knowledge. The rest of the paper is organized as follows: In Section 2, existing PICO corpora and annotation schemata are pre- sented. Section 3 contains a description of the data and the developed annotation interfaces. The conclusion and future work are described in Section 4. Figure 1: Example of PICO elements in a sentence 2. Related Work As mentioned, there are existing datasets based on PICO elements. For instance, (Huang et al., 2006) presented PICO frames as a knowledge representation by analysing 59 real-world, primary-care clinical questions. They ob- tained …

Content truncated to 500 words