Application of Natural Language Processing algorithm for Predetermine Asthma Criteria to ascertainment of adult-onset asthma

Background: Natural Language Processing (NLP) for ascertainment of childhood asthma based on Predetermined Asthma Criteria (PAC) using free text in the EMR was successfully developed and validated. However, the approach has not been applied to adult-onset asthma. Objective: To assess whether PAC-based NLP algorithm (NLP-PAC) identifies adult-onset asthma as effective as childhood asthma. Methods: We applied NLP-PAC to EMR of adult cohorts (Olmsted County, Minnesota, residents) who participated in multiple population-based studies. We excluded asthma that occurred before 2000 when EMR was first implemented at our study setting. The study subjects’ asthma status was ascertained by applying PAC through manual chart review by abstractors. The performance of NLP-PAC was assessed by determining concordance of asthma status between NLP-PAC and manual chart review as gold standard. Results: The cohort consists of 1,898 subjects, 56% were female, and the median age at the last follow-up date was 65 years (IQR: 55-76). Manual chart review identified 97 subjects (5.1%) with adult-onset asthma. NLP-PAC ascertained 98 subjects (5.1%) as asthma with 89 of them overlapped with manual chart review. The sensitivity, specificity, positive predictive value, and negative predictive value of NLP-PAC were 91%, 99%, 90%, and 99%, respectively. Conclusion: Leveraging NLP to automated chart review for asthma ascertainment in adults is as effective as that implemented in children. Our NLP-PAC will be a useful tool enabling large-scale clinical studies for research and population management for asthma care.