Genetic and pharmacogenomic data on smoking: the bigger sample size, the less reliable phenotype? A critical review.