|
Correction
to
Mattson et al.,
|
| |||||||||||||||||||||||||||||
In a prospective, randomized study we found no differences after three years in efficacy between phenobarbital, phenytoin, carbamazepine, and valproate in children and adults with newly diagnosed primary generalized or partial seizures with or without secondary generalization4,5. Table 1 shows that there are important differences in the design of the Veterans Affairs and Medical Research Council studies. Mattson et al. (the Veterans Affairs study) lost 130 patients within the first year for reasons unrelated to the drugs, and their analyses are based on one year of follow-up. The division of their patients into two arbitrary subgroups based on the predominant seizure type, with analysis of only the predominant seizure type in each subgroup, is a source of weakness and bias. They now advise that all patients with symptomatic localization-related epilepsy should be treated with carbamazepine first. Why did they not treat the group as a whole and examine all the data?
|
Our patients were randomly assigned to minimal therapy with each drug, the dose of which was increased only gradually into the range of therapeutic blood levels if necessary. Patients in the Veterans Affairs group were rapidly given maximal ("middle of the target range") therapy, and the dose was then pushed temporarily to the point of clinical toxicity if necessary. The upper range for valproate in the Veterans Affairs study (100 to 150 µg per milliliter) was much higher than is usual in clinical practice and explains the greater toxicity with this drug. In the Medical Research Council study, fewer patients were withdrawn for unacceptable toxicity associated with valproate (5 percent) than for toxicity associated with carbamazepine (11 percent).
The report of Mattson et al. will reinforce underuse of valproate in the United States, in comparison with Europe. This stems in part from the extraordinary failure of the Food and Drug Administration to license a very useful drug, even for tonic-clonic seizures.
E.H. Reynolds, M.D., F.R.C.P.
A.J. Heller, B.Sc., M.R.C.P.
King's College Hospital
London SE5 9RS, United Kingdom
D. Chadwick, D.M., F.R.C.P.
Walton Hospital
Liverpool L9 1AE, United Kingdom
References
Stratification at study entry was not according to the presence or absence of a history of secondarily generalized seizures, which would have provided a clear distinction between the two strata, but according to predominant seizure type. There is no characterization of these two strata, and consequently it is difficult to apply the recommendations for the selection of antiepileptic drugs. Were two patients, each with a history of seven complex partial seizures, placed in separate strata if three seizures were secondarily generalized in one, and four in the other? What was the justification for analyses that ignored seizures (perhaps many) of nondominant types (Table 1 of the article), rather than count all seizures, so as to assess overall seizure control?
The article has many outcome measures and no less than 56 P values; the problem of multiple testing, an issue that has been raised in the Journal, is neither addressed nor discussed1,2.
There are few meaningful summary statistics and no comparative measures that quantify treatment effects with estimates of precision. In Table 1 and Table 4 all the means but one are less than the corresponding standard deviations for data that cannot be negative. Hence, the underlying distributions are skewed, and the mean is not an informative measure.
Some analyses reflect the difficulties involved in using seizure counts and rating scales in clinical trials that require prolonged observation of patients. The only comparison known to be unbiased is one based on all randomized patients, so any analysis that excludes more than a few percent may be suspect. Over half the reported analyses exclude 35 percent or more of the eligible randomized patients; some exclude 63 percent. When actuarial techniques were applied (Figure 1 and Figure 2), follow-up extended only one year from randomization, rather than the three or more years that are achievable when patient accrual has extended over five years.
Unfortunately, recommendations for the selection of drugs are based on analyses within strata, in which the treatment effect is statistically significant within one stratum but not the other. The weakness of such analyses has been pointed out by Pocock3.
This is an important study that continues the trend toward larger cooperative clinical trials of efficacy in epilepsy. It is disappointing that the study did not use better methods of data analysis and that obvious deficiencies were not identified before publication.
Tony Johnson, B.Sc., Ph.D., F.I.S.
MRC Biostatistics Unit
Cambridge CB2 2SR, United Kingdom
References
A. Hernandez-Vidal, M.D.
C/ Bruch, 42
08010 Barcelona, Spain
Mattson et al. conclude that carbamazepine produces fewer long-term side effects than valproate, but this conclusion is not supported by their study. Moreover, many controlled studies and a recent review4 conclude that valproate may be better tolerated and easier to use than carbamazepine. There were side effects described in the text of the article that were not present in the tables (i.e., hyponatremia, granulocytopenia, and cardiac side effects). A table listing all the side effects would have led to different conclusions. On the one hand, some side effects have the same incidence with both drugs: gastrointestinal symptoms, hair change, impotence, hepatic toxicity, and neurologic adverse effects (and of course teratogenicity). On the other hand, the incidence of other side effects depends on the particular medication: weight gain, tremor, thrombocytopenia, and pancreatitis are more frequent with valproate than with carbamazepine, whereas hyponatremia, rash, agranulocytosis, anemia, and heart block are more frequent with carbamazepine than with valproate. The severity of some side effects, such as fatal hepatoxicity with valproate and fatal hyponatremia, Stevens-Johnson syndrome, agranulocytosis, severe arrhythmias, and heart block with carbamazepine, make the choice between them a difficult one.
Jean-Claude Monfort, M.D.
Hopital Albert Chenevier
94010 Creteil, France
References
To the Editor: The letters from Reynolds et al. and Dr. Johnson, of the Medical Research Council Study Group, express doubt about our conclusion that carbamazepine is more effective for the control of complex partial seizures. We believe that this difference was found because our study involved a larger number of patients than the other studies they cite. Indeed, Chadwick and Turnbull,1 in discussing the lack of differentiation among antiepileptic drugs in Medical Research Council studies, stated: "It is possible that the samples of patients studied have been too small to detect reliably significant differences." Despite several inaccurate interpretations of the Veterans Affairs study in Table 1 of Reynolds et al. (e.g., the dose was gradually increased to levels needed for seizure control), it is obvious that the Veterans Affairs study (480 patients in two groups, with moderate-to-large differences detectable with a power of 0.80) has much more statistical power than the Medical Research Council study (410 patients in four groups). In addition, the protocol for the Medical Research Council study is sufficiently dissimilar from that for the Veterans Affairs study that comparisons are not always appropriate (e.g., comparing our double-blind study of adults with localization-related epilepsy to an open trial of children and adults with generalized or localization-related epilepsy).
The Veterans Affairs study was designed to achieve an answer for two groups of seizure types separately (i.e., with separate calculations of sample size, randomizations, and planned analyses), making it two studies performed simultaneously. Secondary efficacy analyses combined the two groups. If we had only reported the results for the combined population, we would have concluded that carbamazepine was superior to valproate for both types of seizures. This would have been erroneous.
All the primary outcome measures, not just selected ones, were presented to allow readers to judge the differences between treatments on the basis of a variety of clinically useful measures; this cannot be considered data dredging. Our conclusion for the group with complex partial seizures was based on statistically significant differences in four of the five planned efficacy analyses. The Wilcoxon rank-sum test was used to account for the skewed distributions. Twelve-month life-table analyses were used because very few new seizures occurred after 12 months. Variables such as life tables and monthly seizure rates allowed us to use data from all patients, including early losses to follow-up. (Note that standard deviations were erroneously given for seizure control in the groups with generalized tonic-clonic and complex partial seizures in our Table 1).
We hope that differences in design and analysis do not obscure the important finding of the controlled Veterans Affairs trial that valproate is a useful medication for the treatment of localization-related epilepsy.
Richard H. Mattson, M.D.
Veterans Affairs Medical Center
West Haven, CT 06516
Joseph F. Collins, Sc.D.
Veterans Affairs Medical Center
Perry Point, MD 21902
Joyce A. Cramer, B.S.
Veterans Affairs Medical Center
West Haven, CT 06516
References
| |||||||||||||||||||||||||||||
This article has been cited by other articles:
HOME | SUBSCRIBE | SEARCH | CURRENT ISSUE | PAST ISSUES | COLLECTIONS | PRIVACY | HELP | beta.nejm.org Comments and questions? Please contact us. The New England Journal of Medicine is owned, published, and copyrighted © 2008 Massachusetts Medical Society. All rights reserved. |
Previous
