Methods: A random sample of a birth cohort (2002-2006) was enrolled for a retrospective study. Performance of NLP-based asthma ascertaining using predetermined asthma criteria was assessed for both criterion (manual chart review as a gold standard) and construct validity (association with known risk factors for asthma). We collected and compared time taken for ascertaining asthma status between NLP vs. manual chart review to assess efficiency.
Results: Among 480 children, 51% were female, 75% Whites, and median age at last follow-up date was 5.3 years (range:0.2–8.8). Sensitivity, specificity, positive predictive value and negative predictive value for NLP algorithm in predicting asthma status were 94%, 97%, 89%, and 99%. Children with asthma determined by NLP compared to those without had higher risks of having a family history of asthma, a history of allergic rhinitis, smoking during pregnancy (p<0.05 in each), and childcare attendance before 3 years (p=0.079), except eczema (29% vs. 24%, p=0.350). NLP-based asthma ascertaining for this cohort was more time-efficient than manual chart review (1.3 hours vs. 384 hours for all subjects).
Conclusions: NLP approach for asthma ascertainment should be considered in the era of big data and EMR as it enables large-scale clinical studies and care and significantly improves cost- and time-efficiency.