Growth and infancy in three contemporary cohorts : selection bias and other methodological issues

There is broad recognition that early life growth trajectories are important predictors for the onset of several diseases. This thesis addresses two methodological challenges that arise in life-course studies of infant growth: (i) the bias that may derive from participants’ selection in cohort studies, and (ii) the modelling of individual growth trajectories. Data from socio-economically diverse populations were used to address them: the Italian NINFEA web-based birth cohort, the Portuguese GXXI birth cohort, and the Chilean GOCS cohort. Participants’ selection affects all cohorts, but web-based designs are thought to be more affected than traditional ones. The thesis first examines possible selection mechanisms by Monte Carlo simulations and then uses population registry data to assess evidence of selection bias among NINFEA participants. The simulations show that under sensible scenarios there is only weak bias in the effects estimated from a selected sample. Comparisons of NINFEA participants with their source population (via registry data) show that the confounding patterns present in NINFEA differ from those in the source population, revealing that participants’ restriction may either increase or decrease the confounding bias in an association of interest. Studying individual early life growth data requires dealing with the quality of the growth measurements and the nonlinearity of the trajectories. Alternative models are compared in terms of their ability to address these problems while extracting salient features of weight growth. SITAR results to be the most useful model for life-course enquiries. An extension of this model, that includes explanatory variables, is fitted on the three cohorts to study the effect of prenatal exposures on different biologically defined dimensions of the growth process. This reveals some interesting mechanisms. This thesis contributes to the interpretation of results obtained from cohort studies with restricted participation, and to the implementation of advanced growth models useful for life-course research.