Using time-varying quantile regression approaches to model the influence of prenatal and infant exposures on childhood growth. Issue 1 (1st January 2017)

• Record Type: Journal Article
• Authors: Wei, Ying; Ma, Xinran; Liu, Xinhua; Terry, Mary Beth
• Is Part Of: Biostatistics & epidemiology. Volume 1:Issue 1(2017)
• Page Start: 133
• Page End: 147
• Publication Date: 2017-01-01
• Subjects: Quantile regression -- time-varying coefficient model -- longitudinal methods -- childhood growth; Epidemiology -- Periodicals; Biometry -- Periodicals; 614.42
• DOI: 10.1080/24709360.2017.1358137
• Languages: English
• ISSNs: 2470-9360

Abstract

ABSTRACT: For many applications, it is valuable to assess whether the effects of exposures over time vary by quantiles of the outcome. We have previously shown that quantile methods complement the traditional mean-based analyses, and are useful for studies of body size. Here, we extended previous work to time-varying quantile associations. Using data from over 18, 000 children in the U.S. Collaborative Perinatal Project, we investigated the impact of maternal pre-pregnancy body mass index (BMI), maternal pregnancy weight gain, placental weight, and birth weight on childhood body size measured 4 times between 3 months and 7 years, using both parametric and non-parametric time-varying quantile regressions. Using our proposed model assessment tool, we found that non-parametric models fit the childhood growth data better than the parametric approaches. We also observed that quantile analysis resulted in difference inferences than the conditional mean models in three of the four constructs (maternal per-pregancy BMI, maternal weight gain, and placental weight). Overall, these results suggest the utility of applying time-varying quantile models for longitudinal outcome data. They also suggest that in the studies of body size, merely modelling the conditional mean may lead to incomplete summary of the data.