Efficient Semiparametric Inference Under Two-Phase Sampling, With Applications to Genetic Association Studies.

Area of Science:

• Epidemiology
• Biostatistics
• Genomics

Background:

• Large-scale epidemiological studies often involve expensive covariates (e.g., genome sequencing, biomarker assays, medical imaging).
• Measuring these expensive covariates on all subjects is often prohibitively costly.
• Two-phase study designs offer a cost-effective solution by selecting a subset of subjects for expensive measurements.

Purpose of the Study:

• To propose a semiparametric regression approach for general two-phase designs.
• To accommodate continuous or discrete outcomes and continuous covariates.
• To address situations where inexpensive covariates are correlated with expensive ones.

Main Methods:

• Utilizing a semiparametric regression framework.
• Approximating conditional density functions of expensive covariates using B-spline sieves.
• Employing a computationally efficient and numerically stable EM-algorithm for likelihood maximization.

Main Results:

• Established consistency, asymptotic normality, and asymptotic efficiency of the proposed estimators.
• Demonstrated superior performance of the new methods compared to existing approaches via extensive simulations.
• Successfully applied the methods to the National Heart, Lung, and Blood Institute (NHLBI) Exome Sequencing Project (ESP).

Conclusions:

• The proposed semiparametric approach provides a statistically sound and computationally efficient method for analyzing data from two-phase studies.
• This methodology enhances the feasibility of incorporating high-dimensional and expensive covariates in epidemiological research.
• The findings have significant implications for optimizing resource allocation in large-scale biomedical studies.