LETTERS TO THE EDITORS
The value of publishing negative results from a randomized controlled trial: the Rosenheck's study
Maurício Silva de LimaI; Bernardo Garcia de Oliveira SoaresII
IFederal University of Pelotas and Catolic University of Pelotas, Brazil; Institute of Psychiatry, London, United Kingdon; and Eli Lilly Brasil
IICochrane Center of Brasil and Federal University of São Paulo, Brazil
Mr Editor,
In the last few years, attention has been givendrawn tofor the problem ofwith publications bias: itit is well established that papers with negative results (when the null hypothesis is not refused) are less likely to be published in scientific journals than those with results favoring a given intervention.1
The paper published in November 23rd, 2003 in JAMA by Rosenheck et al.2 reports negative results (no differences) in randomized clinical outcomes when comparing for the comparison of olanzapine and haloperidol in combination with benzotropine tofor treating schizophrenia. These findings, however, do not agreefit with the main results of a Cochrane Systematic Review, which currently included 20 randomized controlled trials (RCTs). In this Review, olanzapine has advantages whenas compared to First Generation Antipsychotics in terms of clinical improvement in negative symptoms.3
In Rosenheck's trial, offrom a total of 4386 subjects were screened, andand 2141 were eligible for inclusion, and only 309 were randomized. This restrictive inclusion enrollment process limits the generalizability of the studys findings and resulted in a sample of chronic patients with longer duration of diseases, aged 45 years in average (in olanzapine trials, the mean age of patients is around 35 years). In a more chronic population with schizophrenia it is expected that smaller differences between two treatments are to can be found.4 Therefore, lack of statistical power could be another explanation for their negative results.
However, the critical point in this paper is a missing and simple principle: just because of chanceit is expected that some trials will find no significant differences in one or more outcome measures only by chance. According to the Central Limit Theorem,5 it is expected that 5% of the total set of studies will find extreme results (more than two standard deviations from the mean), or 3216% will stand beyond about one standard deviation from the actual mean.
It is crucial that high impact journals like JAMA publish trials with negative results readers can have then a real sense about how different samples of patients (in RCTs) can produce different results. If a pharmaceutical company sponsors the trial, this is even crucial.
General rules of medical statisticals, such as estimations of samples, heterogeneity of populations, and the selection process, must always be considered. For the best care of individual patients, when assessing scientific information, negative results should be more than welcome by both publishers and readers, but theirits conclusions need to be considered in a more comprehensivewide view, in the context of other similar studies.
References
1. Sutton AJ, Duval SJ, Tweedie RL, Abrams KR, Jones DR. Empirical assessment of effect of publication bias on meta-analyses. British Medical Journal 2000;320(7249):1574-7.
2. Rosenheck R, Perlick D, Bingham S, Liu-Mares W, Collins J, Warren S, et al. Effectiveness and cost of Olanzapine and Haloperidol in the treatment of schizophrenia. A randomized controlled trial. JAMA 2003;290:2693-702.
3. Breier A, Schreiber JL, Dyer J, Pickar D. National Institute of Mental Health longitudinal study of chronic schizophrenia. Prognosis and predictors of outcome. Arch Gen Psychiatry 1991;48(3):239-46.
4. Kirkwood B. Confidence interval for a mean. In: Kirkwood B. Essential of Medical Statistics. London: Blackwell Scientific Publications; 1988.
Conflito de interesses: Maurício Silva de Lima é Gerente Médico de Neurociências do Laboratório Eli Lilly do Brasil.
Publication Dates
-
Publication in this collection
27 Oct 2004 -
Date of issue
June 2004
