A Mixture Model to Correct Misclassification of Gestational Age

Abstract

Misclassification of gestational age based on the last menstrual period (LMP) in routinely collected data creates bias in newborn birthweight and gestational age-related indicators. Common correction methods have not been evaluated. We developed a normal mixture model for use with SAS software to correct misclassification of gestational age and compare its performance with other available correction methods and estimates of gestational age. Using the 2007 United States natality file from the National Center for Health Statistics, we compared LMP preterm and postterm birth rates and gestational age-specific birthweight percentiles against a reference subset of births, where the likelihood of misclassification in gestational age was minimized, before and after correction by a normal mixture model, two truncation methods, and the clinical/obstetric estimate of gestational age. The mixture model corrected preterm and postterm birth rates by 90% and 41% respectively, but previous methods performed poorly. The mixture model was also superior in correcting birthweight percentiles 50 and 90 with error reductions in the range of 68% to 85% between 28 and 36 weeks of gestation, where most misclassification occurred. The mixture model behaved consistently better than truncation methods, particularly between weeks 28 and 36 of gestation.