Published online November 1, 2007
PEDIATRICS Vol. 120 No. 5 November 2007, pp. 1224-1225 (doi:10.1542/peds.2007-2268)

This Article Extract Full Text (PDF) Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Services E-mail this article to a friend Similar articles in this journal Similar articles in Web of Science Similar articles in PubMed Alert me to new issues of the journal Add to My File Cabinet Download to citation manager Request Permissions Citing Articles Citing Articles via CrossRef Citing Articles via Google Scholar Google Scholar Articles by Senn, S. Articles by Rizzi, G. Search for Related Content PubMed PubMed Citation Articles by Senn, S. Articles by Rizzi, G. Related Collections Premature & Newborn Social Bookmarking                         What's this?

LETTER TO THE EDITOR

Meta-Analysis Combining 2 Previously Reported Trials on Respiratory Distress Syndrome in Neonates

Stephen Senn, PhD
Department of Statistics
University of Glasgow
Glasgow G12 8QQ, United Kingdom

Marco Costantini, MSc
Giada Rizzi, BSc
Biometrics Unit
Chiesi Farmaceutici SpA
43100 Parma, Italy

To the Editor.—

In the recent article "One-Year Follow-up of Very Preterm Infants Who Received Lucinactant for Prevention of Respiratory Distress Syndrome: Results From 2 Multicenter Randomized, Controlled Trials,"¹ Moya et al reexamined 2 different trials (Surfaxin Therapy Against Respiratory Distress Syndrome [STAR]² and Safety and Effectiveness of Lucinactant Versus Exosurf in a Clinical Trial [SELECT]³) on respiratory distress syndrome and performed a meta-analysis by pooling the 2 studies together.

Although meta-analysis can be a valuable way of summarizing evidence, it suffers from the disadvantage, compared with the prespecified analysis of individual randomized clinical trials, of being retrospective. This is well understood in a regulatory context. For example, the Committee for Proprietary Medicinal Products guidelines state:

To minimize the opportunity for a retrospectively specified meta-analysis protocol to be data dependent, the primary specifications and definitions set up in the individual studies should be followed. The credibility of the meta-analysis will depend on the degree of adherence to the specifications in the individual study protocols.⁴

Of course, standards for publication are not as rigorous as for regulatory submission. Nevertheless, when authors exchange a prespecified analysis with one determined after seeing the data, it behooves them to choose wisely and justify their choice scientifically.

We believe that this meta-analysis failed on these counts.

First, from a clinical point of view, it is irrelevant that a premature infant survives a few more weeks in a treatment group or in another, because the key point, rather, is whether the infant survives the critical period of vulnerability.

To the extent that some reasonable follow-up time can be agreed upon by which, if a child is alive, he or she is highly likely to survive to adulthood, a binary outcome (alive/dead) at this time point becomes an appropriate end point on which to focus. Therefore, an appropriate way of analyzing such data is to perform a logistic regression, not a survival analysis.

Typically, survival analysis is used when faced with data for which, by a reasonable time horizon, most subjects would die, and longer survival, therefore, is important. In such cases the focus is on time to event rather than the probability of the event itself occurring. Under such circumstances, the use of a survival analysis will permit recovery of relevant information that otherwise would be lost.

The same authors in the original publication (ie, STAR study²), followed a strategy consistent with this argument and used the binary outcome of alive/dead at 28 days as primary clinical outcome to assess treatment differences. Also, in the long-term, as reported in the revised article, differences that were not statistically significant were found, as the fixed time-point estimates of mortality at 1 year corrected age (imputing loss to follow-up as a death) were 19.4% for lucinactant and 24.2% for poractant (odds ratio: 0.64 [95% confidence interval: 0.32–1.27]). Of course, these confidence intervals are wide and do not exclude a benefit for lucinactant, but they also do not exclude a possible benefit for poractant. In other words, the results in terms of death at 1 year were inconclusive.

Furthermore, the choice of the Wilcoxon test is unusual, because the default is the log-rank test. This choice of test is particularly questionable if there is neither evidence that the use of the Wilcoxon test was specified beforehand (eg, at the time of the study plan) nor scientific justification for that choice.

As the authors themselves claimed in their discussion, the Wilcoxon test emphasizes earlier treatment differences, and it is more powerful when early differences are larger than late differences. As highlighted above, this approach does not address the main clinical message and, in any case, such gain in power is illegitimate if the choice of test is based on an inspection of the data.

In short, we believe that the conclusions drawn by the authors exceed what are justified by the STAR study² results and seem to be in contrast with the considerations that the same authors raised in their original article, in which they stated that the 2 surfactants were equally effective in preventing respiratory distress syndrome and had comparable incidences of common complications, which suggests similar efficacy and safety.

FOOTNOTES

Financial Disclosure: Professor Senn acts as a consultant for Chiesi Farmaceutici SpA, and Mr Constantini and Ms Rizzi are employees of Chiesi Farmaceutici SpA.

REFERENCES

1. Moya F, Sinha S, Gadzinowski J, et al. One-year follow-up of very preterm infants who received lucinactant for prevention of respiratory distress syndrome: results from 2 multicenter randomized, controlled trials. Pediatrics. 2007;119(6) . Available at: www.pediatrics.org/cgi/content/full/119/6/e1361
2. Sinha SK, Lacaze-Masmonteil T, Valls I, et al. A multicenter, randomized, controlled trial of lucinactant versus poractant alfa among very premature infants at high risk for respiratory distress syndrome. Pediatrics. 2005;115 :1030 –1038[Abstract/Free Full Text]
3. Moya FR, Gadzinowski J, Bancalari E, et al. A multicenter, randomized, masked, comparison trial of lucinactant, colfosceril palmitate, and beractant for the prevention of respiratory distress syndrome among very preterm infants. Pediatrics. 2005;115 :1018 –1029[Abstract/Free Full Text]
4. Committee for Proprietary Medicinal Products. Points to Consider on Application With 1. Meta-analyses; 2. One Pivotal Study. London, United Kingdom: European Evaluation Agency; 2001. Available at: www.emea.europa.eu/pdfs/human/ewp/233099fen.pdf. Accessed September 12, 2007

---

PEDIATRICS (ISSN 1098-4275). ©2007 by the American Academy of Pediatrics

CiteULike    Connotea    Del.icio.us    Digg    Facebook    Reddit    Technorati    Twitter    What's this?

This Article Extract Full Text (PDF) Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Services E-mail this article to a friend Similar articles in this journal Similar articles in Web of Science Similar articles in PubMed Alert me to new issues of the journal Add to My File Cabinet Download to citation manager Request Permissions Citing Articles Citing Articles via CrossRef Citing Articles via Google Scholar Google Scholar Articles by Senn, S. Articles by Rizzi, G. Search for Related Content PubMed PubMed Citation Articles by Senn, S. Articles by Rizzi, G. Related Collections Premature & Newborn Social Bookmarking                         What's this?