Predicting Adverse Drug Events by Analyzing Electronic Patient Records

摘要: Diagnosis codes for adverse drug events (ADEs) are sometimes missing from electronic patient records (EPRs). This may not only affect safety in the worst case, but also number of reported ADEs, resulting incorrect risk estimates prescribed drugs. Large databases (EPRs) potentially valuable sources information to support identification ADEs. study investigates use machine learning predicting one specific ADE based on extracted EPRs, including age, gender, diagnoses and Several predictive models developed evaluated using different algorithms feature sets. The highest observed AUC is 0.87, obtained by random forest algorithm. model can be used screening EPRs that not, possibly should be, assigned a diagnosis code under consideration. Preliminary results presented.