Semiparametric Efficient Estimation Of Treatment Effect In A Pretest-Posttest Study With Missing Data