Mining the Epidemiological Literature for Risk Factors and Incidence Data of Kawasaki Disease

Abstract Determining the efficacy of semantic based searching in the analysis of the epidemiology of Kawasaki Disease as a case study, the MetaMap tool processed 159 MEDLINE abstracts into Unified Medical Language System Semantic Types. Based on a random selection of 79 citations, we identified semantic patterns representing risk factors and incidence data. These patterns were tested on the remainder. Results, compared to a gold standard, were 0.19 sensitivity, 0.47 positive predictive value, and 0.99 specificity for incidence data; and 0.55 sensitivity, 0.55 positive predictive value, and 0.86 specificity for risk factors. Thus, a simple algorithm using semantics produced modest results.