Aim:To determine if the critical periods of growth and asthma/wheeze phenotypes identified in the BiB cohort be replicated in other populations. (ii) To determine if there are links between growth in early life and asthma.

Hypotheses:That the relationships observed between asthma/wheeze and growth in the BiB and other cohorts will be replicated in the ALSPAC cohort.

Exposure variables: Asthma/Wheeze symptoms and diagnoses (examined as latent class phenotypes)

Confounding variables: maternal/paternal history of asthma, maternal/paternal height, maternal/paternal smoking, maternal BMI, gestational length, mode of birth.

Outcome variable: Trajectories of growth (length and weight).

Statistical analyses: Firstly Latent class analyses will be undertaken to determine discrete phenotypes of asthma to be established that are specific to the ALSPAC cohort. Latent class analysis (LCA) allows classification of individuals into groups based on conditional probabilities as within each class individuals will have a similar pattern of response. As these models work with probabilities rather than absolute values this allows children to potentially be fractional members of all classes and use probability to assign class membership. The first stage of LCA will be to determine the optimal number of classes to the data by evaluating the best fitting model using multiple indices of model fit including Akaikes Information Criterion, Bayesian Information Criterion, entropy and likelihood ratio test using bootstrapping as well as assessing the face validity and meaningfulness of the resultant classes. The second stage of the modelling process, to determine the relationship between co-variates including ethnicity, gender and the latent class groups of phenotypes, will be conducted using multinomial logistic regression adjusted for probability weighted class assignment. For the regression analysis we will include co-variates that are significant at the 5% level in the univariate analysis. Any variables with greater than 3% missing data will be tested at the end of the modelling procedure using imputation to determine the resultant sensitivity of results to missingness.

Separate models of growth, for height and weight, will then be created for each latent class grouping. Growth models will be undertaken using heirachical linear spline models using knot points for growth identified in previous studies undertaken using ALSPAC data.In line with previous ALSPAC growth analysis the source of measurement (height and weight) will be included as a covariate in the models. Appropriate adjustment for other known confounders such as gestational length, maternal height and childs gender.