Repository files navigation

The Semi-supervised Calibration of Risk with Noisy Event Times (SCORNET)is a consistent, semi-supervised, non-parametric survival curveestimator optimized for efficient use of Electronic Health Record (EHR)data with a limited number of current status labels. Derived from vander Laan and Robins’ Inverse Probability of Censoring Weighted (IPCW)estimator, it achieves locally efficient survival curve estimation usingcurrent status labels – binary indicators of phenotype status atcensoring time – rather than more expensive event time labels. SCORNETboosts efficiency over IPCW in the typical EHR setting by (1) utilizingunlabeled patients in a semi-supervised fashion, and (2) leveraginginformation-dense engineered EHR features to maximize imputationprecision in the unlabeled set.

See Ahuja et al. (2020) for details.

Install stable version from CRAN:

install.packages("SCORNET")

Install development version from GitHub:

# install.packages("remotes")remotes::install_github("celehs/SCORNET")

• Ahuja Y, Liang L, Huang S, Liao K, Cai T (2020). Semi-supervisedCalibration of Risk with Noisy Event Times (SCORNET) UsingElectronic Health Record Data. BioArxiv.

• Mark J. van der Laan & James M. Robins (1998) Locally EfficientEstimation with Current Status Data and Time-Dependent Covariates,Journal of the American Statistical Association, 93:442, 693-701,DOI: 10.1080/01621459.1998.10473721