Neurology -- Correspondence for Meininger et al., 66 (1) 88-92

Correspondence published:

Pentoxifylline in ALS: A double-blind, randomized, multicenter, placebo-controlled trial: Bruce Levin, John L.P. Thompson, Gilberto Levy, Hiroshi Mitsumoto, Petra Kaufmann (8 March 2006)

Reply from the Authors: Vincent Meininger, Bernard Asselain, P. Nigel Leigh, Albert Ludolph, Lucette Lacomblez and Philippe Guillet (8 March 2006)

Pentoxifylline in ALS: A double-blind, randomized, multicenter, placebo-controlled trial 8 March 2006

Bruce Levin,
Columbia University
Department of Biostatistics, 722 West 168th Street, New York, NY 10032,
John L.P. Thompson, Gilberto Levy, Hiroshi Mitsumoto, Petra Kaufmann
Send Correspondence to journal:
Re: Pentoxifylline in ALS: A double-blind, randomized, multicenter, placebo-controlled trial

Bruce.Levin{at}Columbia.edu Bruce Levin, et al.

Meininger et al [1] report discrepant trial results for survival and functional outcome measures for Pentoxifylline in ALS. Their discussion of functional measures, and particularly the Amyotrophic Lateral Sclerosis Functional Rating Scale – Revised (ALSFRS-R), [2] deserves attention.
The authors report a negative effect of Pentoxifylline on survival. However, their a priori design was for a one-sided test of the hypothesis that Pentoxifylline is beneficial. The primary conclusion should therefore be that the drug is not beneficial (p > 0.90), [3] although this is not stated. The authors’ interpretation of the harmful outcome not allowed for in the original design as statistically significant is post hoc. While their analysis stresses adjusted rather than unadjusted results, the unadjusted result may be preferable. Adjustment for prognostic factors is unnecessary, since randomization balanced prognostic factors, and the stepwise adjustment procedure does not yield a replicable analysis. The unadjusted result is not significant (p=0.107 on the required two-sided test).
Regarding the ALSFRS-R, in secondary analyses the authors confirm that it strongly predicts survival time [4] (hazard ratio for the dichotomized ALSFRS-R, adjusted for prognostic variables, is 0.46, p<0.001, Table 3). As with survival, the impact of Pentoxifylline on the ALSFRS-R is also in the harmful direction, and not significant (p=0.35). By conventional criteria, the results suggest compatible rather than discrepant outcomes for survival and the ALSFRS-R.
Taking qualitatively different results using two correlated outcome measures, such as survival and the ALSFRS-R, at face value would require equivalent statistical power to detect an effect of corresponding magnitude for each outcome, and equally reliable measurement. For example, what magnitude of effects for the secondary functional outcomes would correspond to the 15% survival rate difference for which the Pentoxyfilline trial had more than 90% power (Appendix (E) A-1), and what was the power to detect these corresponding effects?
The authors state that “All studies had the statistical power to detect a difference in functional or survival endpoints”. However, since analyses for secondary, functional outcome measures are commonly underpowered, susceptible to missing data, and often less reliably measured than mortality, we doubt that most meet the requirement we describe. These problems may explain the apparent lack of qualitative agreement with survival which some of them report.
The current question of whether, despite being strongly associated, survival and the ALSFRS-R as outcomes may produce qualitatively discrepant results is unsettled.
References
1. Meininger V, Asselain B, Guillet P, et al for the Pentoxifylline European Group. Pentoxifylline in ALS: A double-blind, randomized, multicenter, placebo-controlled trial. Neurology 2006; 66: 88-92.
2. Cedarbaum JM, Stambler N, Malta E, et al. The ALSFRS-R: a revised ALS functional rating scale that incorporates assessments of respiratory function. Journal of the Neurological Sciences 1999;169:13-21.
3. Fleiss JL, Levin B, Paik MC. Statistical Methods for Rates and Proportions. Third Edition. Hoboken: John Wiley & Sons, 2003.
4. Kaufmann P, Levy G, Thompson JLP, et al. The ALSFRSr predicts survival time in an ALS clinic population. Neurology 2005;64:38-43.
Disclosure: The authors report no conflicts of interest.

Reply from the Authors 8 March 2006

Vincent Meininger,
Fédération des maladies du système nerveux
APHP. Hôpital de la Salpêtrière, 47 boulevard de l'Hôpital 75013 PARIS France,
Bernard Asselain, P. Nigel Leigh, Albert Ludolph, Lucette Lacomblez and Philippe Guillet
Send Correspondence to journal:
Re: Reply from the Authors

vincent.meininger{at}psl.aphp.fr Vincent Meininger, et al.

We thank Levin et al for their interesting Correspondence. We apologize for not clearly stating that the drug was not beneficial. However, the drug had a negative rather than a positive effect, so we assumed that it was obvious that the drug was not beneficial.
We agree that the original design was for a one-sided test of the hypothesis that pentoxyfilline was beneficial. However, even if the trial was designed primarily to detect a beneficial effect of the drug, we considered it important to inform neurologists and physicians of the possible detrimental effect of the drug on survival.
We find it difficult to believe that Levin et al would consider statistical fine points more important than clinical common sense and safety. Levin et al consider that adjustment is unnecessary, even if they also used such an adjustment in their recent publication in Neurology. [5] In our trial, adjustment was planned a priori. Adjusted and unadjusted results are consistent demonstrating a 28% increased risk of death in the unadjusted analysis (HR=1.28, CI 0.94-1.71) and 43% in the adjusted analysis (HR=1.43, CI 1.05-1.96), even if the unadjusted confidence interval includes 1.00.
The principle of using adjustment or not in the statistical analyses in randomized trials led to a large debate in the statistical community, but the majority of statisticians are now in favor of adjustment, if adjustment factors and the statistical procedures are of course a priori defined as in our trial.
Levin et al suggest that in order to compare survival and ALSFRS-R, we have to consider the equivalence of statistical power. However, as discussed in the paper, the ALSFRS-R analysis in most trials (including our trial) is equally powered compared to the survival analysis. Figures 1 and 2 demonstrate that the treatment effect on the ALSFRS is less than the effect on survival.
Finally Levin et al state that “functional outcome measures are commonly underpowered, susceptible to missing data, and often less reliably measured than mortality”. It seems that they provide a strong argument for survival analysis in ALS trials.
References
5. Kaufmann P. Levy G, Thompson JLP et al. The ALSFRS-R predicts survival time in an ALS clinic population. Neurology 2005; 64: 38-43.
Disclosure: ExonHit Pharma funded the study and participated in the design and conduct of the study being discussed in this Correspondence; collection, management, analysis, and interpretation of the data; and preparation, review, and approval of the discussed manuscript. P.G. was an employee of ExonHit Pharma.