Bonferroni corrections were used to account for multiple comparisons between groups at the different follow-ups. The level of statistical significance was set at p < 0.05. The under- lying test assumptions for all tests were ful lled. The statisti- cian was blinded to the group allocation.