We thank Arthritis & Rheumatism for the opportunity to respond to Dr. Bredemeier. We agree that it is important to understand the various approaches to handling missing data; these approaches were a key consideration in the design of our study. However, we would like to emphasize a few main points in response to Dr. Bredemeier.
First, we believe that the important consideration here is not the method used to account for missing data per se, but the assumptions underlying such methods. Dr. Bredemeier does not acknowledge that our study was powered for the assessment of continuous end points, using a predefined testing strategy to adjust for multiplicity. Our primary analyses of all continuous end points utilized a mixed-effects model repeated-measures (MMRM) analysis. As described by the National Research Council (NRC) Panel on Handling Missing Data in Clinical Trials, in MMRM analysis it is assumed that data are missing at random and all observed values from individual patients are used to compensate for missing values without explicitly imputing any data (Panel on Handling Missing Data in Clinical Trials: National Research Council. The prevention and treatment of missing data in clinical trials. Washington, DC: The National Academies Press; 2010). To support these primary MMRM analyses, secondary sensitivity analyses with last observation carried forward (LOCF) and baseline observation carried forward (BOCF) approaches were used to impute missing data.
Given that there is no ideal or established method to account for missing data, we believed this multifaceted approach demonstrated the robustness of our results and was consistent with the NRC Panel on Handling Missing Data in Clinical Trials recommendations. With respect to our secondary sensitivity analyses, the LOCF method was used to make our results comparable to those reported in previous studies of esreboxetine treatment, and the BOCF method was used because the Food and Drug Administration has requested that it be used in fibromyalgia trials.
Second, Dr. Bredemeier's letter focused on the use of the LOCF approach to missing data in our analysis of pain responders. As discussed above, this particular analysis was only used to support the clinical relevance of our primary pain analysis. Additionally, it is important to stress that a BOCF approach would also have had limitations, given that not all patients withdrew from the trial because of adverse events or lack of efficacy and, therefore, would not necessarily be classified as nonresponders.
In summary, we assumed that data were missing at random for the primary analysis of the primary end points in this study (one of which was the pain end point). Sensitivity analyses, such as the one discussed in Dr. Bredemeier's letter, were only used to support our primary analyses. Given that there is no ideal method to account for missing data, we believe this was a pragmatic approach that is consistent with current recommendations regarding the handling of missing data in clinical trials.