Statistical inferences for data from studies conducted with an aggregated multivariate outcome-dependent sample...

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Sampling

Background:

• Outcome-dependent sampling (ODS) is cost-effective but underdeveloped for multivariate responses.
• Existing ODS methods are limited to univariate responses (e.g., case-control, case-cohort).
• Biological studies often involve clustered data with multivariate responses, necessitating advanced sampling techniques.

Purpose of the Study:

• To propose a novel multivariate outcome-dependent sampling (multivariate-ODS) design for clustered continuous responses.
• To develop a semiparametric inference procedure for the proposed multivariate-ODS design.
• To enhance statistical efficiency in biological studies with complex response structures.

Main Methods:

• Developed a multivariate-ODS design based on general selection of continuous responses within clusters.
• Employed a semiparametric inference procedure using empirical likelihood methods for nonparametric covariate distribution modeling.
• Established consistency and asymptotic normality properties for the proposed estimator.

Main Results:

• The proposed multivariate-ODS estimator demonstrated higher efficiency compared to simple random sampling or using only the SRS portion of multivariate-ODS.
• Simulation studies confirmed the superior performance of the new estimator.
• The multivariate-ODS design offers improved study efficiency within a fixed budget.

Conclusions:

• The multivariate-ODS design and associated semiparametric estimator provide a valuable tool for analyzing complex biological data.
• This approach enhances statistical power and efficiency for studies with multivariate, clustered outcomes.
• The method was illustrated using data on polychlorinated biphenyl exposure and childhood hearing loss.