On the use of population attributable fraction to estimate sample size required in case-control study of gene-environment interaction