Marginal regression model using inverse probability weights for handling dropouts missing at random in longitudinal data