Semiparametric analysis of a generalized linear model with multiple covariates subject to detection limits

Area of Science:

• Environmental Health Sciences
• Biostatistics
• Epidemiology

Background:

• Analyzing health effects of environmental mixtures is challenging due to multiple correlated exposures below the limit of detection (LOD).
• Conventional methods like complete-case analysis or imputation can lead to biased results or loss of efficiency.
• Existing models for single exposure below LOD do not extend well to multiple correlated exposures.

Purpose of the Study:

• To develop a robust statistical method for estimating health outcomes associated with multiple correlated environmental exposures below LOD.
• To generalize a semiparametric accelerated failure time (AFT) model for multivariate censored data.
• To provide a more accurate and efficient approach than conventional methods for environmental health studies.

Main Methods:

• A multivariate AFT model was developed for multiple correlated covariates subject to LOD, coupled with a generalized linear model for the health outcome.
• A two-stage estimation procedure using semiparametric pseudo-likelihood was proposed.
• The consistency and asymptotic normality of the proposed estimators were theoretically derived for a fixed dimension of covariates.

Main Results:

• Simulation studies showed the proposed method performs well in realistic scenarios with large sample sizes.
• The new method demonstrated superior performance compared to conventional approaches in terms of efficiency and bias.
• The method was successfully applied to real-world data from the LIFECODES birth cohort study.

Conclusions:

• The proposed semiparametric pseudo-likelihood approach effectively handles multiple correlated environmental exposures below LOD in health studies.
• This method offers a statistically sound and practical alternative to conventional techniques, improving the analysis of complex environmental mixtures.
• The findings have significant implications for epidemiological research on environmental exposures and health outcomes, particularly in vulnerable populations.