A1245
Title: Imputation of fetal growth trajectories: A comparison between GLS models and the non-homogeneous lognormal process
Authors: Ana Garcia-Burgos - University of Granada (Spain) [presenting]
Beatriz Gonzalez-Alzaga - Andalusian School of Public Health (Spain)
Marina Lacasana - Andalusian School of Public Health (Spain)
Paola Paraggio - University of Salerno (Italy)
Nuria Rico-Castro - University of Granada (Spain)
Desiree Romero-Molina - University of Granada (Spain)
Abstract: The reconstruction and imputation of fetal growth trajectories from longitudinal data collected in a prospective cohort of pregnant women in Andalusia is examined. Five fetal biometric measurements obtained by ultrasound are analyzed: biparietal diameter, abdominal circumference, head circumference, femur length, and estimated fetal weight, recorded at different gestational times, which makes direct comparison across individuals difficult. After a clinical and temporal data-cleaning process, the final sample consists of 782 fetuses. The objective is to obtain comparable values at three gestational ages of clinical interest: weeks 12, 20, and 34. Two methodological approaches are compared. The first is based on multiple regression models estimated by feasible generalized least squares, incorporating heteroscedasticity and temporal correlation. The second uses the non-homogeneous lognormal process as a stochastic framework for growth curves. The main methodological contribution is the development of an imputation strategy based on the nearest observed value, combining classical conditioning on past observations with a new formulation that also allows conditioning on future observations. The results show similar performance for both approaches, although the non-homogeneous lognormal process exhibits greater structural stability and more homogeneous implementation.
