Latent Variable Models for Development and Validation of an Infant Morbidity Index

• Activity Number: 515
• Type: Contributed
• Date/Time: Thursday, August 2, 2007 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Xuefeng Liu*+ and Jeffrey Roth
• Companies: Wayne State University and
• Keywords: infant morbidity outcomes ; latent variable model ; modified Gauss-newton algorithm ; validation ; infant morbidity index

Abstract

Birth Defect (BD), Abnormal Conditions (AC), Developmental Delay or Disability (DDD) and Low Birth Weight (LBW) are four major pregnancy outcomes which are associated with infant morbidity. Little attention has been paid to development of a composite index which is a summary construct of infant morbidity outcomes. In this paper, a new single variable, called infant morbidity index ($IMI$) which is a summary of these four infant morbidity outcomes, was developed by using latent variable models (LVM) and modeling conditional probability as a function of $IMI$ in LVM. The validity of this index was then assessed in detail. It was shown that $IMI$ was correlated with each of the individual outcome, with infant mortality and with a face-valid index of morbidity outcomes, and could be used in future research as a measure of infants propensity for morbidity.