In order to account for censoring of the data, the association between expected allelic dosage and ‘age at first tooth’ was analysed using parametric survival analysis, with the Gaussian distribution used to model event time. The ALSPAC data were modelled as ‘right censored’, whereas the data in NFBC1966 were modelled as ‘interval censored’. The association between expected allelic dosage and the number of teeth was analysed using proportional odds logistic regression (ordinal regression). Teeth are known to erupt in pairs; hence Poisson regression (which assumes that the events of interest are independent) was not appropriate. Analyses were adjusted for sex (ALSPAC and NFBC1966), gestational age (ALSPAC and NFBC1966) and age of completion (ALSPAC only, all NFBC1966 measurements were recorded at 12 months). In addition, in NFBC1966, the top 10 ancestry-derived principal components were tested for association with the phenotypes and were included in the GWAS of that phenotype if they associated at P < 0.05. This resulted in the inclusion of the second principal component in the NFBC1966 analysis of tooth eruption, and no principal components in the analysis of ‘number of teeth’. Data were analysed using the R software package 2.9.1.
