Performance of the four models of the EHR algorithm for identifying multiple sclerosis patients (at 95% specificity).

aThe ICD model uses the number of ICD-9 code for MS as the only variable. The Codified (COD) model includes codified variables in addition to the number of ICD-9 code for MS. The NLP model includes narrative variables extracted from clinical texts. The combined (ALL) model uses both codified and narrative variables. Performance parameters are calculated using 0.632 bootstrap cross-validation in the training set. The standard errors are estimated based on 1,000 bootstrap replications.Abbreviations: AUC, area under the curve; NPV, negative predictive value; PPV, positive predictive value; SE, standard error of the estimates.