Using mixture models with linear predictors to identify incorrect gestational age in state birth records.

Area of Science:

• Perinatal epidemiology
• Biostatistics
• Public Health

Background:

• Birthweight distributions show bimodality at early gestational ages.
• Some birthweights are implausible for the stated gestational age.
• Mixture models can identify these implausible birthweights.

Purpose of the Study:

• Identify maternal and infant factors for optimal mixture models.
• Improve accuracy of gestational age reporting in New Jersey birth records.

Main Methods:

• Utilized mixture models with covariates as linear predictors.
• Modeled means and proportions of component distributions.
• Allowed means and proportions to vary with covariates.

Main Results:

• Maternal age and prenatal care entry timing were key covariates.
• Older mothers with early care had fewer implausible births.
• Teen mothers with late care had the highest proportion of implausible births.
• Over 44% of births were classified with incorrect gestational age.

Conclusions:

• Maternal age and care timing improve models for identifying incorrect gestational age.
• Late prenatal care contributes to erroneously short gestational ages.
• Including linear predictors enhances classification validity.