The ascertainment and availability of phenotype data varied widely among the 43 studies (see Supplementary Tables 1 and 3). For example, birth weight was measured by trained personnel in some studies, but in others was self-reported in adulthood. Gestational age was not available as a covariable in all studies. We therefore performed further analyses to verify data quality and check that the effect size estimates in our meta-analyses were not greatly influenced by poor quality data or lack of adjustment for gestational age.
