Extended Data Fig. 1: Performance under delayed updates to participant records. | Nature Medicine

Extended Data Fig. 1: Performance under delayed updates to participant records.

From: Screening for idiopathic pulmonary fibrosis using comorbidity signatures in electronic health records

Extended Data Fig. 1

a, b, Out-of-sample ROC curves when the patient data is delayed by 4w vs the no-delay condition, for the UCM and the Truven datasets, respectively. 95% confidence bounds about the mean is shown, computed with n=2,053,277 for Truven and n=68,658 for UCM. Note that there is no significant loss of performance with such delayed data. c, d, ZCoR-IPF performance vs a 87-feature baseline model optimized via logistic regression, where these features denote presence/absence of manually-curated risk factors (Supplemental Table 4) and age (over/under 65 years), for the Truven and the UCM datasets, respectively.

Back to article page