Efficient Modeling of Surrogates to Improve Multi-source High-dimensional Biobank Studies

TL;DR

The paper introduces SASH, a semi-supervised method that leverages unlabeled data and surrogate variables from multiple sites to improve high-dimensional biobank modeling, especially when gold labels are scarce.

Contribution

SASH is a novel semi-supervised approach that combines surrogate-assisted modeling with bias correction and data aggregation to enhance accuracy in high-dimensional biobank studies.

Findings

Outperforms existing methods in simulations.

Effectively integrates multi-site surrogate data.

Successfully applied to diabetes genetic risk modeling.

Abstract

Surrogate variables in electronic health records (EHR) and biobank data play an important role in biomedical studies due to the scarcity or absence of chart-reviewed gold standard labels. We develop a novel approach named SASH for {\bf S}urrogate-{\bf A}ssisted and data-{\bf S}hielding {\bf H}igh-dimensional integrative regression. It is a semi-supervised approach that efficiently leverages sizable unlabeled samples with error-prone EHR surrogate outcomes from multiple local sites, to improve the learning accuracy of the small gold-labeled data. {To facilitate stable and efficient knowledge extraction from the surrogates, our method first obtains a preliminary supervised estimator, and then uses it to assist training a regularized single index model (SIM) for the surrogates. Interestingly, through a chain of convex and properly penalized sparse regressions that approximate the SIM loss with bias-correction, our method avoids the local minima issue of the SIM training, and fully eliminates the impact of the preliminary estimator's large error. In addition, it protects individual-level information through summary-statistics-based data aggregation across the local sites, leveraging a similar idea of bias-corrected approximation for SIM.} Through simulation studies, we demonstrate that our method outperforms existing approaches on finite samples. Finally, we apply our method to develop a high dimensional genetic risk model for type II diabetes using large-scale data sets from UK and Mass General Brigham biobanks, where only a small fraction of subjects in one site has been labeled via chart reviewing.