Reliability of attachment loss measurements in a longitudinal clinical trial.