Details
Anaphylaxis is a severe, life-threatening reaction to foods, insect venom, and medications. This study uses natural language processing (NLP) and machine learning methods, the positive predictive values (PPV) of algorithms for identifying validated events in visits with an anaphylaxis diagnosis code improved to 79%. However, little is known about the sensitivity of diagnosis codes for anaphylaxis. This study explored the extent to which anaphylaxis events are missed by algorithms that require the presence of an anaphylaxis diagnosis code. This study used a 10% random sample of a complete population of visits in the Emergency Department(ED) setting represented by complete electronic health record (EHR) data. The six potential predictors of anaphylaxis were explored for identifying ED encounters likely to contain genuine uncoded anaphylaxis events. The study estimated the sensitivity of anaphylaxis diagnosis codes in the ED setting by extrapolating from the event rate for uncoded events estimated in this study and the event rate for coded events estimated in our previous study to the full population of ED visits.