Analysis of repeated pregnancy outcomes

Area of Science:

• Reproductive epidemiology
• Biostatistics
• Perinatal research

Background:

• Reproductive outcomes exhibit significant intra-woman correlation, meaning past events influence future ones.
• Adverse reproductive history approximately doubles the risk of subsequent adverse outcomes.
• Ignoring this clustering can lead to biased estimates and inaccurate variance calculations.

Purpose of the Study:

• To review and evaluate statistical approaches for analyzing reproductive outcomes.
• To highlight the impact of statistical modeling on inferences regarding birth weight determinants.
• To contrast traditional methods with advanced techniques like GEE and mixed models.

Main Methods:

• Review of basic analytic approaches (ignoring history, covariate, single pregnancy analysis).
• Evaluation of modern statistical methods: Generalized Estimating Equations (GEE) and mixed models.
• Application and comparison of methods using the Collaborative Perinatal Project dataset.

Main Results:

• Basic methods can mask effects or yield inaccurate estimates.
• Advanced methods like GEE and mixed models appropriately handle correlated reproductive events.
• Statistical model choice significantly impacts summary statistics and conclusions in birth weight etiology.

Conclusions:

• Accurate analysis of reproductive outcomes requires methods that address intra-woman correlation.
• Generalized estimating equations and mixed models offer robust solutions for clustered reproductive data.
• Proper statistical modeling is crucial for understanding factors influencing birth weight and other reproductive health outcomes.