Presentation: Composite Responder Endpoints for Fibromyalgia Trials - Experience with Milnacipran (2007)

1525 Composite Responder Endpoints for Fibromyalgia Trials - Experience with Milnacipran

Purpose: Composite responder analyses are known to sacrifice power when compared to continuous analyses; however, they have the advantage of inherent face validity. If responder definitions are based on minimal clinical important differences (MCID), then statistically significant differences in response rates are clinically meaningful by definition. In a large, placebo-controlled study (N=888) of milnacipran for the treatment of fibromyalgia, we were able to compare and contrast composite responder with continuous endpoints.
Methods: The primary endpoint of the study involved a composite “pain” responder analysis based on the proportion of patients who (a) achieved a 30% improvement in pain (recorded on an electronic diary) and (b) were also “improved,” “much improved,” or “very much improved” on a 7-point Patient Global Impression of Change (PGIC) scale. An additional analysis was conducted to assess whether the patient’s entire “syndrome” improved. Syndrome improvement required the above criteria plus a 30% improvement in functional status, as measured by the Physical Function subscale of the Fibromyalgia Impact Questionnaire (FIQ).
Results: The composite “pain” responder analysis at the 3-month primary endpoint showed that milnacipran 200 mg daily was superior to placebo with P values of .048/.058 (BOCF/LOCF). In contrast, using the pain intensity score mean change from baseline, milnacipran was superior to placebo with P=.01. Similarly, the mean difference in PGIC between groups was even more robust, reaching a P value of <.001. Interestingly, minor but scientifically grounded alterations in the definitions of responder (eg, requiring “much” or “very much improved” on the PGIC) led to large differences in the discriminative abilities of the responder indices. Because 47% of placebo-randomized patients reported themselves to be at least “improved” on this outcome measure, stricter criteria improved discrimination. The composite responder endpoint for “syndrome” did not show significant differences between treatment groups, in large part due to the poor psychometric performance of the FIQ; changes in the FIQ did not correlate well with changes in pain or global status or with other functional status measures such as the SF-36 PCS. However, substitution of the SF-36 PCS (MCID of 6) resulted in significant differences between milnacipran and placebo on the syndrome analysis above.
Conclusions: Use of responder analyses can be less powerful than analyses of continuous variables, but they have the advantage of ensuring clinical relevance and face validity. Nuances in how these measures are constructed and applied can lead to significant differences in the results obtained.

  R. Gendreau, Cypress Bioscience, Inc., 3; D.A. Williams, Cypress Bioscience, Inc., 5; R.H. Palmer, Forest Laboratories, Inc., 1; Forest Research Institute, 3; P. Mease, Cypress Bioscience, Inc., 2; Forest Laboratories, Inc., 2; Forest Laboratories, Inc., 5; Cypress Bioscience, Inc., 5; D.J. Clauw, Cypress Bioscience, Inc., 1; Cypress Bioscience, Inc., 5; Pfizer Inc., 5; Wyeth Pharmaceuticals, 5; Eli Lilly and Company, 5.