Dr Garudadri and associates raise several issues. As described in our baseline and outcomes paper, we anticipated greater dropouts in the brimonidine arm, thus randomizing more patients to brimonidine than timolol. The rate of subjects unable to continue brimonidine 0.2% likely resembles real-world clinical practice.

We disagree that the rate of brimonidine dropouts “… skewed the results in favor of brimonidine.” The proportion of patients progressing was much higher in the timolol group, regardless of dropout rates, which was confirmed by 3 different progression criteria, all being highly statistically significant. We consider it unlikely that the higher dropout in the brimonidine group alone explains the differences in observed progression rates.

Simulation modeling was performed to explore the possibility that higher dropout in the brimonidine group could nullify or reverse the primary results. Simulations varied the rate of progression among dropouts in the brimonidine and timolol groups. Modeling used different rates of progression (0.10 to 1.00) for dropouts. The observed rate of progression among participants completing the study was 5% to 9.1% for brimonidine and 26.6% to 44.3% for the timolol group. Simulations suggested that the primary hypothesis would remain statistically significantly different with a progression rate of 35% of the brimonidine dropouts, 4 times the observed rate in the brimonidine group who completed the study. We examined the possibility that timolol dropouts did better than those who completed the study. Simulations indicated that the progression rate among timolol dropouts would have to be 10% in order for the primary hypothesis to become statistically nonsignificant. Finally, we explored a worst-case scenario in which brimonidine dropouts fared worse than those completing the study and timolol dropouts did much better. Simulations suggest that a statistically nonsignificant test of the protective effect of brimonidine would require the conjunction of a progression rate among the brimonidine dropouts of ≥30% and a progression rate ≤15% in the timolol group.

We do not agree that an intent-to-treat approach is inappropriate for our study. The authors did not propose any explanation for this statement and their cited reference does not state that substantial loss of participants during follow-up would be incompatible with an intent-to-treat approach. This publication suggests that studies that claim to use an intent-to-treat approach should always report how “… any deviations from randomised allocation, false inclusions, or missing outcomes were handled.” This was rigorously described in our methods and results sections, flow chart, and tables.

Garudadri and associates advocate the χ ² test to compare absolute numbers of nonprogressing patients in each group. The use of the χ ² test in this fashion is inappropriate. It is better used to compare cross-sectional data from independent groups. Applying this method to the group with stable fields at last follow-up visit does not take into account the differences between the survival curves throughout the study and factors such as progression endpoints and drop-out rates. The ideal statistical approach in these cases is a survival analysis, as we employed.