The presence of responder and survivor biases is unsurprising in longitudinal studies. However, if there is a bias towards less depression among our respondents, this would only be a matter of concern – and have implications for generalizability – if the relationships between risk factors and depression differed systematically between those who took part and those who did not ; this seems unlikely. Third, our analyses were constrained by the availability of comparable data in the cohorts. We were unable to examine some potentially important factors, such as physical activity. Fourth, the multivariable-adjusted estimates need to be interpreted with caution. Although we identified potential confounding or mediating factors for each risk factor from among the variables included here, lack of data on some factors in some cohorts – such as cognition and neuroticism in the HCS – means that estimates from each cohort could not always be adjusted for the same variables. Finally, we were not able in these meta-analyses to explore whether the simultaneous presence of different risk factors has an additive or interactive effect on risk of anxiety or depression.