Bias reduction in repeated-measures observational studies by the use of propensity score : the case of enteral sedation for critically ill patients