The Use of Multiple Emergency Department Reports per Visit for Improving the Accuracy of Influenza Case Detection
We present an assessment of the diagnostic accuracy of probabilistic Bayesian models that detect influenza from one or multiple emergency department (ED) reports per visit. We extracted clinical findings from ED reports using two natural language processing (NLP) tools. Then, we measured and compared the area under the ROC (AUROC) curve of existing models applied to single and multiple ED reports per visit respectively, and found that using multiple reports per visit increased the detection performance.
Recent research studies demonstrated that ED reports improve influenza detection performance over chief complaints. Those studies used one ED report per visit; however, in ED practice, multiple ED reports exist for one visit, which is due to multiple clinicians (e.g., residents, attending physicians) seeing the same patient and/or one clinician dictating/writing multiple (supplemental) reports. To the best of our knowledge, no previous research studied the impact of influenza detection performance using single vs multiple ED reports for each ED visit. We hypothesize that the use of multiple ED reports per visit can improve the accuracy of influenza detection through the collection of more complete clinical findings across multiple reports over time.