Thinking outside the curve, part I: modeling birthweight distribution

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Understanding fetal-infant mortality requires analyzing birthweight distributions and prognostic factors.
• Existing frameworks for birthweight analysis may not capture complex population heterogeneity.
• A realistic and tractable analytical framework is needed for public health implications.

Purpose of the Study:

• To introduce a novel framework for analyzing birthweight distributions.
• To improve the understanding of relationships between birthweight and fetal-infant mortality.
• To develop a method that can reveal previously undetectable heterogeneity in birthweight.

Main Methods:

• Describing birthweight distributions using a normal mixture model.
• Determining the number of model components from data using model selection criteria.
• Addressing methodological issues such as sample size dependency and criterion influence.

Main Results:

• A 4-component normal mixture model effectively describes birthweight for white singleton infants of smoking mothers.
• A 6-component normal mixture model may be more suitable for general black singleton populations.
• The proposed model was compared to a contaminated normal model and a 2-component normal mixture model.

Conclusions:

• The developed framework does not assume specific birthweight intervals for compromised pregnancies.
• It avoids constraining birthweights within compromised pregnancies to be normally distributed.
• This approach can detect birthweight heterogeneity missed by other models like the contaminated normal model.