neonatal sepsis?
William O. Tarnow-Mordi, BA (Cantab), MBChB DCH, MRCP(UK), FRCPCH
Objective: To identify strategies to enable randomized, con- substantially greater sample sizes are needed to limit the play of
trolled trials in neonatal sepsis to recognize therapies that in- chance. Prospective meta-analysis using individual patient data
crease disability-free survival. is a promising strategy. It requires researchers to obtain funding
Methods: Rapid literature review. from their national agencies to conduct similar trials according to
Results: Randomized, controlled trials are the gold standard an agreed protocol with prespecified hypotheses, interventions,
for testing a therapy because they minimize bias. However, ran- power calculations, data sets, and measures of outcome. Pro-
domized, controlled trials must also minimize random error if they spective meta-analysis combines the methodologic advantages of
are to detect moderate (and realistic) improvements in rates of a single “megatrial” with the practical advantages that financial
disability-free survival reliably. This requires surprisingly large burdens are spread internationally and different national funding
samples (i.e., thousands rather than hundreds). Against this per- cycles can be more flexibly accommodated.
spective, most neonatal trials have been too small to be conclu- Conclusions: Prospective meta-analysis using individual pa-
sive, so most neonatal therapies remain incompletely evaluated. tient data is a promising strategy for achieving large-scale, ran-
As in specialties like cardiology and obstetrics, achieving reliable domized evidence in neonatal sepsis. (Pediatr Crit Care Med 2005;
trials in neonatal sepsis will require international collaboration, 6[Suppl.]:S146 –S149)
simpler data sets, more cost-effective recruitment strategies, less KEY WORDS: randomized, controlled trials; neonatal sepsis; pro-
exclusive criteria for selecting collaborators and patients, and an spective meta-analysis
appreciation by clinicians and data-monitoring committees that
R andomized, controlled trials For example, statisticians at a recent con- ment effects. As a result, many modest
in pediatric and neonatal sep- sensus conference on clinical trials in but important treatment effects have
sis are too few and too small. adult sepsis pointed out that if investiga- probably been missed (6, 7), and most
The need for more random- tors wanted to detect a moderate risk neonatal therapies remain incompletely
ized, controlled trials and meta-analyses difference of 5% (e.g., from 40% to 35%) evaluated.
is well recognized in adult, pediatric, and reliably when control event rates are
perinatal medicine (1–5). Both ap- around 30%– 40%, adequately powered
proaches avoid bias, if appropriately ana- Large, Simple Trials Are
studies would need around 4,000 patients
lyzed. Yet, if we want reliable answers Necessary and Feasible
(1) (Table 1).
about whether therapies decrease mortal- Many authors “resolve” this problem Yusuf et al. (8) gave four reasons for
ity and disability, avoiding bias is not by designing smaller trials, with samples undertaking large, simple trials to answer
enough. The effects of most treatments of a few hundred patients or fewer. These certain important questions. First, treat-
are likely to be moderate. To avoid being trials have only enough power to detect ments with big effects on mortality are
seriously misled by the play of chance either unrealistically large differences rare, so looking for moderate effects on
when looking for moderate effects, trials (e.g., from 30% to 15%) in categorical mortality in common conditions is im-
must also minimize random error, and primary outcomes, like death or disabil- portant. This requires really large trials.
this requires surprisingly large numbers. ity, or to demonstrate moderate differ- Second, because death is simple to mea-
ences in a continuous primary outcome, sure, trials with mortality as the main
like blood pressure or plasma bilirubin outcome can be large, with simple fol-
From the Department of Neonatal Medicine, Uni- (6). In a recent survey of systematic re- low-up. Third, treatments used widely by
versity of Sydney, Westmead Hospital, and The Chil-
dren’s Hospital at Westmead, New South Wales, Aus-
views of neonatal trials in the Cochrane busy clinicians must be easy to adminis-
tralia. Library, there were only 90 reviews of ter, so really large trials can have very
This work was supported by the Mannion Family trials with a categorical primary outcome simple regimens. Lastly, quantitative dif-
Fund—Center for the Critically Ill Child, Division of (such as mortality or survival). Among ferences in the degree of effect between
Critical Care Medicine at Children’s Hospital Boston,
these, the median number of outcomes subgroups are common. However, unex-
the PALISI Network, and the ISF.
Copyright © 2005 by the Society of Critical Care per review was only 54, with a range from pected qualitative differences in the di-
Medicine and the World Federation of Pediatric Inten- 1 to 1,284 (7). Clearly, most neonatal rection of effect—so that treatment ben-
sive and Critical Care Societies trials are too small to provide reliable efits some subgroups but harms others—
DOI: 10.1097/01.PCC.0000161576.09029.56 evidence about any but the largest treat- are unlikely. Therefore really large trials