Hierarchical Semi-Bayes Methods for Misclassification in Perinatal Epidemiology

Area of Science:

• Epidemiology
• Biostatistics
• Maternal and Child Health

Background:

• Prepregnancy body mass index (BMI) is prone to misclassification in epidemiological studies.
• Validation data from the Penn MOMS cohort were used to quantify BMI misclassification.
• Accurate BMI assessment is crucial for understanding its impact on pregnancy outcomes.

Purpose of the Study:

• To estimate the association between maternal prepregnancy BMI and early preterm birth (<32 weeks).
• To develop and apply a semi-Bayesian hierarchical model for adjusting exposure misclassification.
• To provide more flexible adjustment for misclassification in epidemiological analyses.

Main Methods:

• A two-stage semi-Bayesian hierarchical model was employed.
• The first stage modeled bias parameters within a Bayesian framework, shrinking estimates for precision.
• The second stage used probabilistic bias analysis to adjust a frequentist outcome model for misclassification.

Main Results:

• Bias parameters derived from the hierarchical model showed improved reasonableness and reduced variance.
• Adjusting for misclassification generally attenuated unadjusted associations.
• Severely obese mothers demonstrated an increased risk of early preterm birth compared to normal weight mothers.

Conclusions:

• The two-stage semi-Bayesian model effectively adjusted for exposure misclassification by borrowing strength across group-specific bias parameters.
• Findings support an elevated risk of early preterm birth for severely obese mothers relative to normal weight mothers.