Previously we developed an “Ngram” classifier for syndromic surveillance of emergency department (ED) chief complaints (CC) in Turkish for bioterrorism. The classifier is developed from a set of ED visits for which both the ICD diagnosis code and CC are available. A computer program calculates the associations of text fragments within the CC (e.g. 3 characters for a “3-gram”) with a syndromic group of ICD codes. The program then generates an algorithm which can be deployed to evaluate chief complaint data in real-time. However, the N-gram method differs from most other classifiers in that it assigns a probability that each visit falls within the syndrome rather than ruling the visit “in” or “out” of the syndrome. It is possible to dichotomize visits “in” or “out” using N-grams by choosing a cut-off sensitivity for the n-grams used, but this affects the specificity of the method. The effect of this trade-off is best measured by a receiveroperator curve.
Objective
Our objective was to determine the sensitivity and specificity of the Ngram CC classifier for individual ED visits. We also wish to compare these results to those obtained when we substituted anglicized characters for 6 problematic Turkish characters.