A0192
Title: Efficient inference on high-dimensional logistic regression
Authors: Alexander Giessing - National University of Singapore (Singapore) [presenting]
Wenjie Guan - Cornell University (United States)
Yikun Zhang - University of Washington (United States)
Doudou Zhou - National University of Singapore (Singapore)
Abstract: A novel procedure is proposed for inference on high-dimensional logistic regression models when the number of covariates is exponentially larger than the sample size. The procedure yields semi-parametrically efficient estimates of the conditional log odds ratio and conditional case probabilities for arbitrary future high-dimensional observations. Moreover, the procedure does not require positivity of case and control probabilities and is thus more robust than existing methods. Finite sample simulation studies are provided that support the theoretical, large sample semiparametric efficiency and robustness properties of the procedure. The efficacy of the procedure is further illustrated by proposing a risk prediction model for T2D against demographic variables and genetic variants in a genome-wide association study from the Mass General Brigham (MGB) biobank dataset.
