On the Use of Population Attributable Fraction to Determine Sample Size for Case-Control Studies of Gene-Environment Interaction