Philip J. Hashkes, MD, MSc; Steven J. Spalding, MD; Edward H. Giannini, DrPH, MSc; Bin Huang, PhD; Anne Johnson, CCRP; Grace Park, MD, MPH; Karyl S. Barron, MD; Michael H. Weisman, MD; Noune Pashinian, MD; Andreas O. Reiff, MD; Jonathan Samuels, MD; Dowain A. Wright, MD, PhD; Daniel L. Kastner, MD, PhD; Daniel J. Lovell, MD, MPH
Acknowledgment: The authors thank the following site study coordinators, nurses, and pharmacists: Tara Barker, RN, and John Petrich, RPh, from the Cleveland Clinic; Anne Jones, RN, Beverly Barham, RN, and Judith Starling, RPh, from the National Institutes of Health; Ana Cabrera, CCRC, and John Pech, MS, from the Children's Hospital of Los Angeles; and Claudine Davis, RN, from New York University Medical Center. They also thank Professor Philip Hawkins from the National Amyloidosis Centre at the Royal Free Hospital for measurement of serum amyloid A levels.
Grant Support: By the U.S. Food and Drug Administration, Office of Orphan Products Development (R01 FD003435). This trial was also supported by the Intramural Research Programs of the National Institute of Arthritis and Musculoskeletal and Skin Diseases, the National Human Genome Research Institute, and the Cleveland Clinic Foundation. Rilonacept and placebo were supplied by Regeneron Pharmaceuticals. Target Health provided and partially subsidized the electronic data capture system and electronic database.
Potential Conflicts of Interest: Disclosures can be viewed at www.acponline.org/authors/icmje/ConflictOfInterestForms.do?msNum=M11-3106.
Reproducible Research Statement:Study protocol: Available from Dr. Hashkes (e-mail, firstname.lastname@example.org). Statistical code: Available in Appendix 1. Data set: Available to approved individuals through written agreements with the trial steering committee (contact Dr. Hashkes).
Requests for Single Reprints: Philip J. Hashkes, MD, MSc, Pediatric Rheumatology Unit, Shaare Zedek Medical Center, POB 3235, Jerusalem, 91031 Israel; e-mail, email@example.com.
Current Author Addresses: Dr. Hashkes: Pediatric Rheumatology Unit, Shaare Zedek Medical Center, POB 3235, Jerusalem, 91031 Israel.
Dr. Spalding: Pediatric Rheumatology-A111, Cleveland Clinic Foundation, 9500 Euclid Avenue, Cleveland, OH 44195.
Drs. Giannini, Huang, and Lovell and Ms. Johnson: Pediatric Rheumatology, Cincinnati Children's Medical Center, 3333 Burnet Avenue, MLC 4010, Cincinnati, OH 45229-3039.
Dr. Park: National Human Genome Research Institute, MSC 1517, 10 Center Drive, Bethesda, MD 20892-1517.
Dr. Barron: National Institute of Allergy and Infectious Diseases, MSC 3207, 33 North Drive, Bethesda, MD 20892-3207.
Drs. Weisman and Pashinian: Division of Rheumatology, Cedars-Sinai Medical Center, 8700 Beverly Boulevard, B131, Los Angeles, CA 90048.
Dr. Reiff: Children's Hospital Los Angeles, 4650 Sunset Boulevard, MS#60, Los Angeles, CA 90027.
Dr. Samuels: New York University Langone Hospital for Joint Disease, 246 East 20th Street, New York, NY 10003.
Dr. Wright: Children's Hospital Central California, 9300 Valley Children's Place, MB08, Madera, CA 93636.
Dr. Kastner: National Human Genome Research Institute, MSC 8002, 50 South Drive, Bethesda, MD 20892-8002.
Author Contributions: Conception and design: P.J. Hashkes, E.H. Giannini, B. Huang, K.S. Barron, M.H. Weisman, D.L. Kastner, D.J. Lovell.
Analysis and interpretation of the data: P.J. Hashkes, S.J. Spalding, E.H. Giannini, B. Huang, M.H. Weisman, N. Pashinian, A.O. Reiff, D.L. Kastner, D.J. Lovell.
Drafting of the article: P.J. Hashkes, E.H. Giannini, B. Huang, J. Samuels, D.L. Kastner, D.J. Lovell.
Critical revision of the article for important intellectual content: P.J. Hashkes, E.H. Giannini, B. Huang, K.S. Barron, M.H. Weisman, A.O. Reiff, J. Samuels, D.A. Wright, D.L. Kastner, D.J. Lovell.
Final approval of the article: P.J. Hashkes, S.J. Spalding, E.H. Giannini, B. Huang, A. Johnson, G. Park, K.S. Barron, M.H. Weisman, A.O. Reiff, J. Samuels, D.A. Wright, D.L. Kastner, D.J. Lovell.
Provision of study materials or patients: P.J. Hashkes, S.J. Spalding, K.S. Barron, M.H. Weisman, N. Pashinian, D.A. Wright, D.L. Kastner.
Statistical expertise: E.H. Giannini, B. Huang.
Obtaining of funding: P.J. Hashkes, E.H. Giannini, D.L. Kastner, D.J. Lovell.
Administrative, technical, or logistic support: S.J. Spalding, A. Johnson, M.H. Weisman, D.L. Kastner, D.J. Lovell.
Collection and assembly of data: P.J. Hashkes, S.J. Spalding, E.H. Giannini, A. Johnson, G. Park, K.S. Barron, N. Pashinian, A.O. Reiff, J. Samuels, D.A. Wright, D.L. Kastner.
Hashkes P., Spalding S., Giannini E., Huang B., Johnson A., Park G., Barron K., Weisman M., Pashinian N., Reiff A., Samuels J., Wright D., Kastner D., Lovell D.; Rilonacept for Colchicine-Resistant or -Intolerant Familial Mediterranean Fever: A Randomized Trial. Ann Intern Med. 2012;157:533-541. doi: 10.7326/0003-4819-157-8-201210160-00003
Download citation file:
Published: Ann Intern Med. 2012;157(8):533-541.
This article has been corrected. The original version (PDF) is appended to this article as a Supplement.
Currently, there is no proven alternative therapy for patients with familial Mediterranean fever (FMF) that is resistant to or intolerant of colchicine. Interleukin-1 is a key proinflammatory cytokine in FMF.
To assess the efficacy and safety of rilonacept, an interleukin-1 decoy receptor, in treating patients with colchicine-resistant or -intolerant FMF.
Randomized, double-blind, single-participant alternating treatment study. (ClinicalTrials.gov number: NCT00582907).
6 U.S. sites.
Patients with FMF aged 4 years or older with 1 or more attacks per month.
One of 4 treatment sequences that each included two 3-month courses of rilonacept, 2.2 mg/kg (maximum, 160 mg) by weekly subcutaneous injection, and two 3-month courses of placebo.
Differences in the frequency of FMF attacks and adverse events between rilonacept and placebo.
8 males and 6 females with a mean age of 24.4 years (SD, 11.8) were randomly assigned. Among 12 participants who completed 2 or more treatment courses, the rilonacept–placebo attack risk ratio was 0.45 (SD, 0.13) (equal-tail 95% credible interval, 0.26 to 0.77). The median number of attacks per month was 0.77 (0.18 and 1.20 attacks in the first and third quartiles, respectively) with rilonacept versus 2.00 (0.90 and 2.40, respectively) with placebo (median difference, −1.74 [95% CI, −3.4 to −0.1]; P = 0.027). There were more treatment courses of rilonacept without attacks (29% vs. 0%; P = 0.004) and with a decrease in attacks of greater than 50% compared with the baseline rate during screening (75% vs. 35%; P = 0.006) than with placebo. However, the duration of attacks did not differ between placebo and rilonacept (median difference, 1.2 days [−0.5 and 2.4 days in the first and third quartiles, respectively]; P = 0.32). Injection site reactions were more frequent with rilonacept (median difference, 0 events per patient treatment month [medians of −4 and 0 in the first and third quartiles, respectively]; P = 0.047), but no differences were seen in other adverse events.
Small sample size, heterogeneity of FMF mutations, age, and participant indication (colchicine resistance or intolerance) were study limitations.
Rilonacept reduces the frequency of FMF attacks and seems to be a treatment option for patients with colchicine-resistant or -intolerant FMF.
U.S. Food and Drug Administration, Office of Orphan Products Development.
November 1, 2013
Comments on the Implementation of Bayesian Model
TO THE EDITORS:
The authors of this article choose to use non-informative, skeptical and enthusiastic prior densities for the attack rate risk ratio (RR) (Events on rilonacept - Events on placebo). These are modeled via a log-normal(mean, standard deviation (SD)) density. We are not given the exact parameterisation of the log-normal densities but it is assumed that these are the mean and SD of the variables natural logarithm. The skeptical prior is log-normal(0,0.125) implying a median (exp(mean)) RR of 1 which seems sensible. However, the ‘enthusiastic’ prior is log-normal(0.5, 0.125) implying a median RR of 1.65 i.e. that placebo is the better treatment option.
A further concern here is that both the skeptical and enthusiastic prior probability density functions at RR = 0.42 (a measure of centrality reported in the posterior) is effectively zero. By Bayes’ theorem if the prior density is zero the posterior density should be zero.
The reported posterior values could be explained by the way the WinBUGS code has been implemented in Appendix 1. The authors suggest the function dnorm and dlnorm within WinBUGS are parameterised using the mean and SD, they are not. Referring to the WinBUGS manual referenced by the authors these functions use the mean and the precision as input. Had the SD been used instead of the precision the actual priors for the Bayesian model would have been very different then suggested in the article.
Phillip J. Hashkes, MD, MSc
Shaare Zedek Medical Center
January 23, 2014
We thank Dr. Bell for the thoughtful comments. After revisiting all of our Bayesian analyses, we agree that we made errors to the WinBUGS code, and these errors in fact led us to report rather conservative results of this trial. The paper has been corrected at annals.org.The correct results under the non-informative prior are a rilonacept-placebo attack risk ratio of 0.45 (SD, 0.13) as opposed to the previously reported 0.59 (SD 0.11). Similarly, the corrected risk ratios are 0.47 (SD, 0.11) and 0.60 (SD, 0.13) under the enthusiastic and skeptical priors respectively, as oppose to 0.42 and 0.40 reported previously. Given the observed median attack rate of 0.77 during treatment with rilonacept and 2 during placebo use (ratio of 0.385), it is not surprising that the Bayesian posterior estimate of risk ratio using the enthusiastic prior of LN(ln(0.5), .125), i. e. log-normal with mean of ln(0.5) and precision parameter of 0.125 is slightest more conservative. This is because the study pre-specified enthusiastic prior were concentrated on the relative risk to ½, thus pulling the posterior estimate away from potential stronger treatment effect. For the skeptical and enthusiastic priors, we purposely chose smaller variances, or wider precision (in WinBUGS syntax). The choice of variances was set in such way that it represented a reasonable range of treatment effect under two opposite opinions. Our pre-specified priors corresponded to risk ratio ranges between (0.78 = exp(-1.96*.125), 1.28= exp(1.96*.125)) under the skeptical prior and (0. 40 = exp(-0.68-1.96*.125), 0.65= exp(-0.68+1.96*.125)) under the enthusiastic prior. In other words, in the enthusiastic model we believed the FMF attack rate during rilonacept treatment would be between 40 and 65% of the attack rate when using placebo. In the skeptical model, we believed the FMF attack rate during rilonacept treatment would be between 78% and 128% of the attack rate when using the placebo. As for the non-informative prior, we used a risk ratio range between 0.3E-8(= exp(-1.96* 10)) to 3.3E8(= exp(1.96* 10)). We conducted sensitivity analyses considering several different prior specifications: 1) using a more enthusiastic prior of LN(ln(0.33), .125), and obtained the posterior estimate of 0.41 (SD, 0.09), which is a more enthusiastic estimate than the non-informative posterior estimate. 2) using the precision of sqrt(.125) instead of .125 , which corresponds to wider range of risk-ratio, i.e. 2.5%-tile to 97.5%-tile of 50-100% under skeptical prior, and 25-100% under enthusiastic prior. The corresponding posterior under the enthusiastic prior was 0.47 (SD, 0.12), nearly the same compared to results using the precision of .125. The result under the pessimistic prior was 0.53 (SD, 0.13) using sqrt(.125) compared to 0.60 (SD, 0.13) using .125. 3) We considered a non-informative prior of LN(0, sqrt(10)) and obtained an identical result to the primary study results. In fact the estimate of 0.45 is nearly the same as the frequentist median estimate of 0.44 ( = exp(-0.833)) based on the log-transformed risk ratio data (last column of Table 2), log(risk ratio+0.1). Overall our findings suggest that rilonacept has in fact greater efficacy than the initial enthusiastic estimates.In summary, we feel these investigations ensure the accuracy of the results of our corrected paper.
to gain full access to the content and tools.
Learn more about subscription options.
Register Now for a free account.
Results provided by:
Copyright © 2016 American College of Physicians. All Rights Reserved.
Print ISSN: 0003-4819 | Online ISSN: 1539-3704
Conditions of Use
This PDF is available to Subscribers Only