**5. Statistical analysis**

The sample size was chosen to assure a power of at least 90% in detecting at least a 2 mm Hg difference between groups with a standard deviation of 2 mm Hg with a twosided α error of 5%.

Evaluation of continuous variables was achieved using the Student's *t*-test. To evaluate the difference in intraocular pressures between follow-up intervals, the paired *t*-test was used. All *t*-tests were two-tailed. Categoric variables were evaluated with the chi-square test, the Fisher exact test, or the Spearman rank correlation as appropriate. A level of P < 0.05 was accepted as statistically significant. Each potential confounding variable was screened for association with the outcome. Only those confounding variables that were statistically associated were eligible to be incorporated into the potential final multivariate model.

For the pairing of groups, age, sex, best-corrected visual acuity, and intraocular pressure at baseline were used for matching. We studied a correlation between the paired observations. If observations were correlated, the F-test was used to study two population variances.

Because a representation of mean intraocular pressure over time could be misleading because of exclusion of cases after failure, the mean intraocular pressure was recalculated by carrying forward the last intraocular pressure reading before repeat surgery. The proportion of surgical failures and adverse events in each treatment group was compared. Success was evaluated on the basis of Kaplan-Meier cumulative probability (log rank test).
