Modelling Birthweight in the Presence of Gestational Age Measurement Error – A Semi-parametric Multiple Imputation Model

by cp2530 on February 23, 2011

Title: Modelling Birthweight in the Presence of Gestational Age Measurement Error – A Semi-parametric Multiple Imputation Model
Location: New York University – 3rd floor conference room of The Research Alliance for New York City Schools (RANYCS), at 285 Mercer St., just south of Waverly
Description: Speaker: Russ Steele, McGill University

Sponsor: PRIISM (Center for the Promotion of Research Involving Innovative Statistical Methodology)

Abstract: Gestational age is an important variable in perinatal research, as it is a strong predictor of mortality and other adverse outcomes, and is also a component of measures of fetal growth. However, gestational ages measured using the date of the last menstrual period (LMP) are prone to substantial errors. These errors are apparent in most population-based data sources, which often show such implausible features as a bimodal distribution of birth weight at early preterm gestational ages (≤ 34 weeks) and constant or declining mean birth weight at postterm gestational ages (≥ 42 weeks). These features are likely consequences of errors in gestational age. Gestational age plays a critical role in measurement of outcome (preterm birth, small for gestational age) and is an important predictor of subsequent outcomes. It is important in the development of fetal growth standards. Therefore, accurate measurement of gestational age, or, failing that, a reasonable understanding of the structure of measurement error in the gestational age variable, is critical for perinatal research.

In this talk, I will discuss the challenges in adjusting for gestational age measurement error via multiple imputation. In particular, I will emphasize the tension between flexibly modelling the distribution of birthweights within a gestational age and allowing for gestational age measurement error. I will also discuss strategies for incorporating prior information about the measurement error distribution and averaging over uncertainty in the distribution of the birthweights conditional on the true gestational age.

Prof. Steele’s research interests include statistical computing, mixture models, multiple imputation, Bayesian modeling and inference, model-based clustering, and large datasets, and his areas of application are quite broad, with an emphasis on health policy analysis. He will be visiting NYU the week of February 28th. Please contact Marc Scott ( if you would like to arrange a meeting.
Start Time: 10:45
Date: 2011-03-02
End Time: 11:45

Previous post:

Next post: