Intention-to-treat and per-protocol analysis

We thank Dr. Shah for his thoughtful comments on our article.1 As he pointed out, 50 patients were initially randomized to each treatment arm. The four patients randomized to the splint group for whom there were diagnostic errors had to be excluded immediately for safety reasons because the fractures required a different treatment strategy. Some experts would advocate continuation of such patients in the trial, while others would agree that patients enrolled in error should be excluded. We adopted the latter approach because all errors in enrolment were related to diagnostic mistakes that were revealed within 24 hours after randomization.

Among the remaining 46 patients in the splint group and 50 in the cast group, we did not have any primary outcome data for 4 (3 splint, 1 cast) because they were lost to follow up for this outcome. We chose not to account for missing data beause it was such a small number of patients and unlikely to affect the outcome. However, Dr. Shah raises a valid point. We conducted the analysis again, giving the missing cast patient the highest possible score of 100 and the three missing splint patients the lowest observed score in their group (73.28). The lower limit of 90% confidence interval was then −3.37 and the p value < 0.0001, thereby rejecting the null hypothesis that the splint is worse than the cast by more than 7 points. These results support the original findings in our article.