Identification and validation of respiratory virus immunization using natural language processing

We demonstrated the effectiveness of NLP methods in identifying immunizations from EHR. High precision and recall for COVID-19 and influenza immunizations suggest that the algorithm can effectively identify immunization references when they are present in the text; however, low recall when compared to the structured data suggests that there are many more immunizations in the structured data not present in the text. Thus, the algorithm has specialized utility for augmenting immunization records using text data from individual notes; however, the algorithm's extensibility and generalizability can serve as a framework for future EHR-based research.