Comparing dynamic treatment regimes using repeated-measures outcomes: modeling considerations in SMART studies