We have combined two methods to detect anomalies in a time series - in this case in emergency department visit data. The n-gram method applies an existing ICD classifier to a set of emergency department (ED) visits for which both the chief complaint (CC) and ICD code are known. A collection of CC substrings (or n-grams), with associated probabilities, are automatically generated from the training data. This information becomes a CC classifier which is then used to find a classification probability for each patient. The output of this classifier can be used to build volume predictions for a syndromic group or can be combined with a selected threshold to provide syndromic determinations on a per-patient basis. Once the daily volume predictions have been calculated using the n-grams, the HWR anomaly detection algorithm is applied, which alerts both for unusual values and for changes in the overall behavior of the time series in question. The earliest alert was generated by the series of volume predicted by flu n-grams as a proportion of total daily visits.