Age bias in survey sampling and implications for estimating HIV prevalence in men who have sex with men: insights from mathematical modelling.