Figure 2. Mean Roland Disability Scale scores at baseline, 6, 12, and 26 weeks by treatment group. Classes ended at week 12. The x-axis signifies week after starting treatment. Higher scores signify greater disability. The P values for the omnibus F tests of any differences in mean Roland scores among the 3 treatment groups (derived from a linear regression model that was fitted to all follow-up time points, adjusting for baseline Roland score by using generalized estimating equations) are 0.046 at 6 weeks, 0.002 at 12 weeks, and 0.002 at 26 weeks.