Identifying implausible gestational ages in preterm babies with Bayesian mixture models.

Area of Science:

• Obstetrics and Gynecology
• Biostatistics
• Epidemiology

Background:

• Infant birth weight and gestational age are critical in obstetric research.
• Current US birth data relies on last menstrual period recall, introducing potential errors.
• Previous studies used mixture models assuming Gaussian birth weight distributions to address misreporting.

Purpose of the Study:

• To develop advanced Bayesian mixture models for more accurate gestational age estimation.
• To address uncertainty in the number of components within mixture models.
• To identify and correct implausible gestational age records in US birth data.

Main Methods:

• Developed a Bayesian mixture model building on previous work.
• Extended methods using reversible jump Markov chain Monte Carlo to handle model component uncertainty.
• Applied methods to US singleton birth data (2001-2008) for gestational ages 23-32 weeks.

Main Results:

• A three-component mixture model best fit data for gestational ages ≤25 weeks.
• A two-component mixture model best fit data for gestational ages >25 weeks.
• The models provide statistical tools for identifying implausible gestational ages.

Conclusions:

• Bayesian mixture models offer improved statistical tools for analyzing gestational age data.
• The techniques can aid in multiple imputation for missing or inaccurate gestational age records.
• Accurate gestational age data is crucial for obstetric research and public health.