Hostname: page-component-7c8c6479df-5xszh Total loading time: 0 Render date: 2024-03-28T16:22:31.549Z Has data issue: false hasContentIssue false

Multivariate analyses of the Lifeways cross-generation study of BMI

Published online by Cambridge University Press:  06 September 2018

G.E. Kelly
Affiliation:
School of Mathematics and Statistics, University College Dublin, Dublin 4, Ireland.
Q. Zhang
Affiliation:
School of Mathematics and Statistics, University College Dublin, Dublin 4, Ireland.
C. Murrin
Affiliation:
School of Public Health, Physiotherapy and Sports Science, University College Dublin, Dublin 4, Ireland.
C. Kelleher
Affiliation:
School of Public Health, Physiotherapy and Sports Science, University College Dublin, Dublin 4, Ireland.
Rights & Permissions [Opens in a new window]

Abstract

Type
Abstract
Copyright
Copyright © The Authors 2018 

Increasing levels of body mass index (BMI) present a problem, particularly in children. Determining possible pathways of familial transmission is important, both for prevention and management intervention(Reference Murrin, Kelly and Tremblay1, Reference Davey Smith, Steer and Leary2, Reference Kelly, Murrin and Viljoen3). The aim of the present analysis was to demonstrate the use of multivariate statistical models to examine familial patterns in the Lifeways cross-generation cohort study.

Expectant mothers were recruited initially in 2001–2003(Reference O'Mahony, Fallon and Hannon4). In addition to their index infant, mothers recruited if possible their partner and at least one living grandparent. There were 1094 children born to 1082 mothers. Height and weight at age 5 were recorded for 570 children. Where available from either baseline or follow-up, BMI measurements of their parents, maternal and paternal grandparents were included. The data are multivariate in that there are up to seven variables corresponding to family members' BMI. We used principal component analysis (PCA) to reduce the dimensionality of the data(Reference Everitt and Duun5). We then used latent variable models in the form of factor analysis and structural equation models (SEM) to describe the structure of the covariance matrix of the seven variables(Reference Everitt and Duun5). All models were fitted using the lavaan package in R(Reference Rosseel6). Missing data were imputed(Reference Josse and Husson7).

With PCA, two components accounted for 68% of variation in the data. These were firstly overall level of BMI and then variation between individuals in the maternal family (including the child) and paternal family. With factor analysis, there were again two factors interpreted as maternal family and paternal family (including the child). SEM was then used to test a specific pathway- the intrauterine effect/ mtDNA effect on BMI. Maternal grandmother, mother, child comprised one factor and paternal grandmother, father comprised the second factor (Fig. 1). The loadings of the second factor were not significant but when the father was omitted from the second factor, the fit was better, providing confirmation of the model.

Fig. 1. Path diagram for intrauterine/mtDNA effect. f1 and f2 are latent factors. The numbers denote factor loadings. mgm/pgm=maternal/paternal grandmother.

In conclusion, these analyses confirmed particularly maternal-line patterns associated with child's BMI, consistent with intrauterine exposure, as well as shared familial or environmental exposures.

References

1.Murrin, CM, Kelly, GE, Tremblay, RE et al. (2012) BMC Public Health 12, 111.Google Scholar
2.Davey Smith, G, Steer, C, Leary, S et al. (2007) Arch Dis Child 92, 876–80.Google Scholar
3.Kelly, GE, Murrin, C, Viljoen, K et al. (2014) BMJ Open 4, e005732.Google Scholar
4.O'Mahony, D, Fallon, UB, Hannon, F et al. (2007) Ir Med J 100, 36.Google Scholar
5.Everitt, B & Duun, G (1991) Applied Multivariate Data Analysis. London: Edward Arnold.Google Scholar
6.Rosseel, Y (2012) J Stat Softw 48, 136.Google Scholar
7.Josse, J & Husson, F (2016) J Stat Softw 70, 131.Google Scholar
Figure 0

Fig. 1. Path diagram for intrauterine/mtDNA effect. f1 and f2 are latent factors. The numbers denote factor loadings. mgm/pgm=maternal/paternal grandmother.