Ò
PAIN 153 (2012) 1541–1542
www.elsevier.com/locate/pain
Letters to the Editor Estimate at your peril: Imputation methods for patient withdrawal can bias efficacy outcomes in chronic pain trials using responder analyses
To the Editor: Moore et al. [5] correctly emphasize the importance of imputation methodology in assessing the efficacy of pain drugs. This problem has frequently been ignored in reviews and meta-analyses comparing the efficacy of different drugs. There are 2 important considerations that prevent unalloyed enthusiasm for considering all dropouts as nonresponders (ie, using BOCF [baseline observation carried forward] imputation). First, while classifying dropouts as nonresponders may decrease the bias associated with other methods of imputation and reduce the apparent treatment effect in a traditional parallel study, the consequences may be just the opposite in randomized withdrawal studies in which dropouts contribute to the endpoint (loss of therapeutic effect). Second, and more to the point of the review, the appropriateness of using imputation or not (and which kind of imputation) depends on the question being asked. If the question is how many patients starting treatment will be responders, BOCF imputation answers the question. However, if the question is how many patients who can tolerate a drug will have a response, then no imputation (observed cases (OC) analysis) may be preferable. The problem with BOCF imputation is that it conflates 2 important but completely separate concepts (efficacy and tolerability) that should be evaluated independently rather than combined. This is particularly true when there may be substantial differences in dropout rate with the same drug in the same condition, depending on the precautions taken to minimize tolerability issues and the particular population being studied [6]. In 3 very similar trials of milnacipran for the treatment of fibromyalgia (in similar populations), the numbers needed to treat (NNTs) (BOCF) for pain responders (P30% improvement in pain) were 16.3, 17.0, and 9.5, and the difference probably reflects a different up-titration schedule to prevent dropouts in the latter study [1,2,4]. However, the OC analyses, which did not depend on dropout rates, were more similar (5.9, 7.2, and 6.2), reflecting the common treatment effect. The importance of the particular population evaluated was demonstrated by a study in fibromyalgia patients who had shown an insufficient response to pregabalin [3]. In this population, the addition of milnacipran to pregabalin gave an NNT (BOCF) of 4.1, perhaps because patients who were able to tolerate pregabalin may have been less sensitive to side effects in general and, therefore, were less likely to drop out in this study. The OC NNT in this study (3.4) was closer to the OC rates cited above (5.9, 7.2, and 6.2), which may better reflect the pharmacological effect of the drug on the pain of fibromyalgia. The data cited suggest that, ideally, several methods of imputation should be presented in study reports in order to better understand the underlying pharmacological effect of the drug and the contributions of withdrawals to its estimation.
References [1] Arnold LM, Gendreau RM, Palmer RH, Gendreau JF, Wang Y. Efficacy and safety of milnacipran 100 mg/day in patients with fibromyalgia: results of a randomized, double-blind, placebo-controlled trial. Arthritis Rheum 2010;62:2745–56. [2] Clauw DJ, Mease P, Palmer RH, Gendreau RM, Wang Y. Milnacipran for the treatment of fibromyalgia in adults: a 15-week, multicenter, randomized, double-blind, placebo-controlled, multiple-dose clinical trial. Clin Ther 2008;30:1988–2004. [3] Mease P, Farmer MV, Trugman JM, Wang Y, Gendreau RM, Graham SM. Effect of adding milnacipran to pregabalin for the management of fibromyalgia: a randomized, open-label, controlled study. Honolulu, Hawaii: Presented at the American Academy of Neurology Annual Meeting; 2011. [4] Mease PJ, Clauw DJ, Gendreau RM, Rao SG, Kranzler J, Chen W, Palmer RH. The efficacy and safety of milnacipran for treatment of fibromyalgia. A randomized, double-blind, placebo-controlled trial. J Rheumatol 2009;36:398–409. [5] Moore RA, Straube S, Eccleston C, Derry S, Aldington D, Wiffen P, Bell RF, Hamunen K, Phillips C, McQuay H. Imputation methods for patient withdrawal can bias efficacy outcomes in chronic pain trials using responder analyses. Pain 2012;153:265–8. [6] Palmer RH, Escher T, Wang Y. Estimating effect size with number needed to treat in milnacipran trials for fibromyalgia. J Pain 2011;12(Suppl. 4):22. Presented at the American Pain Society Annual Meeting, Austin, TX, May 19–21.
Robert H. Palmer Clinical Development, Forest Research Institute, 185 Hudson Street, 9th floor, Jersey City, NJ 07311, USA Tel.: +1 201 427 8218. E-mail address:
[email protected]
0304-3959/$36.00 Ó 2012 International Association for the Study of Pain. Published by Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.pain.2012.04.024
Response to Palmer To the Editor: In broad terms, we agree with Palmer’s comments, particularly in regard to enriched enrolment randomised withdrawal (EERW) studies, which are quite different from ‘‘classic’’ parallel group studies [1]. While there are few high-quality, large, EERW studies of sensible duration, that study design is arguably more appropriate for chronic pain because the outcome is often a combination of initial efficacy and long-term tolerability with sustained efficacy. For example, the FREEDOM trial published in these pages [2] demanded an initial response of at least 50% pain intensity reduction over the baseline 6 weeks previously AND self-rating of overall improvement on the Patient Global Impression of Change scale of ‘‘much improved’’ or ‘‘very much improved.’’ Subsequent failure in the 6-month double blind randomisation between active drug or placebo was either
