- Email: [email protected]
Single-center trials in neonatology: Issues to consider
Seminars in Fetal & Neonatal Medicine xxx (2015) 1e5
Contents lists available at ScienceDirect
Seminars in Fetal & Neonatal Medicine journal homepage: www.elsevier.com/locate/siny
Review
Single-center trials in neonatology: Issues to consider Ian P. Sinha a, *, Sunil K. Sinha b a b
Respiratory Unit, Alder Hey Children's Hospital, Liverpool, UK Department of Neonatology, University of Durham and James Cook University Hospital, Middlesbrough, UK
s u m m a r y Keywords: Evidence-based medicine Neonatology Systematic review Trials
Single-center randomized controlled trials confer certain advantages over multi-center trials, in that they are cheaper and easier to design and conduct. However, recent research suggests that single-center trials are likely to overestimate treatment effects. There are notable examples in neonatology where results from multi-center trials have contradicted results of single-center studies. In this paper we discuss issues around external generalizability of single-center studies, and methodological issues that may cause bias. © 2015 Published by Elsevier Ltd.
1. Introduction In neonatology many drugs, interventions and practices have developed as a result of randomized controlled trials (RCTs) and meta-analyses. More trials need to be conducted, as there are uncertainties around treatments, and several drugs used in newborn infants are unlicensed [1]. These will need to be appraised and decisions will be made about whether the results should change practice. Some of these trials will be large multi-center studies, and others will be conducted in a single center. In this paper we consider issues around single-center RCTs. 2. The attractions of single-center studies The obvious advantage of single-center studies is that they are logistically easier, and usually cheaper, than multi-center studies. With increasing numbers of centers participating in a trial, it becomes more complex and expensive to conduct and monitor sitespecific training, administrative duties, and governance. It can also become more difficult for collaborators to reach agreement about the implications of the results and write the final report. Even though trial networks can foster collaboration and make this process easier, the cost of a multi-center study may be prohibitive in some situations. In low-income countries where there may not be a robust infrastructure for collaborative research, it may only be possible to conduct studies in one center.
* Corresponding author. Address: Respiratory Unit, Alder Hey Children's Hospital, Alder Road, Liverpool L12 2AP, UK. Tel.: þ44 151 2284811. E-mail address: [email protected] (I.P. Sinha).
Single-center trials are particularly appropriate for studies that may not require a large number of patients, such as early-phase, pilot or feasibility studies. Such trials can be invaluable for informing subsequent studies, such as testing validity and importance of outcome measures, evaluating study protocols, assessing participant recruitment rates, and generating data to inform future sample size calculations. Single-center studies can provide useful estimates of benefits and harms, especially when aggregated in meta-analyses to help guide therapies. For example, a Cochrane review [2] comparing volume-targeted ventilation (VTV) with pressure-targeted ventilation included 12 trials (eight single-center, four conducted in two centers). VTV was found to reduce the risk of the composite outcome of death or bronchopulmonary dysplasia (BPD) [risk reduction (RD): 0.12 (95% CI: 0.21 to 0.03); number needed to treat (NNT): 8 (95% CI: 5 to 33)], and other clinically important outcomes. 3. Single-center studies show larger treatment effects Estimates of treatment effect appear to be larger in single-center RCTs. Meta-epidemiological analyses using the summary statistic ratio of odds ratio (ROR) have systematically examined trials in several specialties; ROR <1 indicates a larger estimate of the intervention effect in single-center trials than multi-center trials. In two analyses [3,4], single-center trials were more likely to show greater benefit with regards to dichotomous outcomes {combined ROR: 0.73 (95% CI: 0.64 to 0.83) [3] and 0.64 (95% CI: 0.47 to 0.87) [4]}, and in one there was a trend towards this finding which was not statistically significant {ROR: 0.91 (CI: 0.79 to 1.04) [5]}. Another analysis reported that single-center trials showed larger
http://dx.doi.org/10.1016/j.siny.2015.08.003 1744-165X/© 2015 Published by Elsevier Ltd.
Please cite this article in press as: Sinha IP, Sinha SK, Single-center trials in neonatology: Issues to consider, Seminars in Fetal & Neonatal Medicine (2015), http://dx.doi.org/10.1016/j.siny.2015.08.003
2
I.P. Sinha, S.K. Sinha / Seminars in Fetal & Neonatal Medicine xxx (2015) 1e5
intervention effects for continuous outcomes {combined difference in standardized mean differences 0.09 (95% CI: 0.17 to 0.01, P ¼ 0.04) [6]}. A combination of various factors probably causes this phenomenon. Single-center studies tend to recruit fewer patients, and it is recognized that smaller studies show greater effects [7]. In some studies the sample size may be so small that it is purely due to chance that a treatment has shown a benefit. However, in one analysis, the overestimation of treatment effect in single-center studies appears to remain consistent even when results are adjusted for sample size [3]. Publication bias, or selective reporting of trials based on the results, is well recognized [8]. One strategy to improve publication rates of trials is to prospectively list them in a registry, but whether single-center studies are less likely to be registered has not been tested. One meta-epidemiological study [9] combined results from three reports that examined full publication of abstracts of RCTs that were initially presented at conferences and found that single-center studies were no less likely to be published [risk ratio (RR): 1.27 (95% CI: 0.95 to 1.70)]. Another analysis found that single-center studies were more likely never to have started after they had been planned, or to have been abandoned after they started [123/337 (36%) vs 28/163 (17%); P < 0.0001], but single-center status did not predict whether a study was published or not [10]. In this paper we consider two other reasons why single-center studies may show greater treatment effects. The first is the likelihood that there are particular features of the center or investigator that may affect the magnitude of treatment effect, and hence the external generalizability of the results. The second is that some aspect of the methodology in single-center studies makes it more likely that treatments are shown to be beneficial (in other words, single-center studies may be prone to bias). 4. Are the results of single-center studies externally generalizable? Single-center studies are conducted in a more homogeneous population than multi-center trials. They are better suited therefore to efficacy (explanatory) trials, which test whether an intervention works in optimum conditions in contrast to effectiveness (pragmatic) trials which measure the effect in ‘real world’ settings. When interpreting the conclusion of any RCT, it is important to determine why the study was conducted. A center may want to conduct a trial if a particular problem occurs frequently in its own institution. This has implications in areas of neonatology in which there are substantial differences in outcomes between institutions [11]. One notable example is the use of prophylactic fluconazole in preterm infants. A single-center study of 100 infants, randomized to intravenous fluconazole or placebo, showed that antifungal prophylaxis reduced fungal colonization [11/50 (22%) vs 30/50 (60%); RD: 0.38 (95% CI: 0.18 to 0.56); P ¼ 0.002] and invasive fungal infection [0/50 vs 10/50 (20%); RD: 0.20 (95% CI: 0.04 to 0.36); P ¼ 0.008] [12]. A Cochrane review of systemic prophylactic antifungal therapy for very-low-birthweight infants incorporated seven studies, of which five were in single centers, one in eight centres, and one in two [13]. Although the review found a significant reduction in invasive fungal rates among very-lowbirthweight infants [RD: 0.09 (95% CI: 0.14 to 0.05), NNT: 11], the authors highlight that this may reflect the high incidence of systemic candidiasis in the control groups (mean 16%). In the UK, where the typical rates of invasive fungal sepsis are only 1% [14], a large number of infants would be exposed to potential risks of fluconazole but only very few would be expected to benefit.
Another reason why a center may wish to conduct an RCT is that it has particular experience or expertise with an intervention or strategy. For example, single-center trials have shown clear benefits of VTV, but it is uncertain whether other institutions can replicate the success of these studies, particularly as they may lack the clinical and nursing expertise available at the study centers. 5. Theoretical reasons why single-center studies are more prone to bias Bias refers to aspects of the trial (other than the effects of the interventions) that give one treatment arm an unfair advantage over another. In the Cochrane Risk of Bias tool, aspects of methodology and reporting are categorized as low, high, or unclear risk of bias [15]. There are some specific important considerations around these domains in context of single-center studies. Evidence is lacking, however, as to whether single-center studies are more likely to be classed as having high risk of bias, so these are currently theoretical. 5.1. Selection bias In an RCT, the groups of participants should be identical, except for the intervention they receive. If there are methodological reasons why this may not happen, the study is prone to selection bias, and is more likely to show greater treatment effects [16]. Avoidance of selection bias relies on two things e first that the participants are allocated to groups at random, and second that those enrolling participants into the trial are unaware of the treatment group to which they will be allocated. There is no specific reason why the process of randomization should be flawed in single-center studies. For this aspect of the trial to be classed as low risk of bias, a method should be used that allocates patients completely at random (for example by using random number tables or computerized random number generators) rather than by using a systematic approach to randomization (such as allocating participants to groups on the basis of date of birth, day of admission, or patient identification number). However, it is possible that in some single-center studies the process of allocation concealment is less optimal. This is important as studies with inadequate allocation concealment can overestimate treatment effects by 40% [16]. To prevent investigators from knowing the allocated groups of subsequent participants in advance (and hence affecting whether they are enrolled in the study or not), a trial can use allocation procedures at a separate central location (which could be telephone, web-based, or pharmacy-controlled), or measures to hide the allocation until the patient is enrolled (such as sequentially numbered drug containers of identical appearance, or opaque, sealed envelopes). Although there is no specific reason why single-center studies should not employ such methods (other than cost) it is more likely that the randomization schedule (i.e. the groups to which patients will be allocated) will be held locally rather than centrally, so unless appropriate concealment measures are taken, the study is at high risk of bias. In a commentary about single-center studies in adult critical care, the author suggests that it may be more difficult to conceal the likelihood of randomization to a particular group if a trial is conducted in one center [17]. This relates to the use of block randomization (i.e. randomization within blocks of smaller numbers of participants, to try to ensure fairly equal numbers in each group). If an investigator knows the block size, in an unblinded single-center study, they may be able to foresee the next treatment allocation.
Please cite this article in press as: Sinha IP, Sinha SK, Single-center trials in neonatology: Issues to consider, Seminars in Fetal & Neonatal Medicine (2015), http://dx.doi.org/10.1016/j.siny.2015.08.003
I.P. Sinha, S.K. Sinha / Seminars in Fetal & Neonatal Medicine xxx (2015) 1e5
5.2. Performance and detection bias Once a participant is randomized, it is ideal if personnel involved in their care, and trial personnel, do not know the treatment group to which they are allocated. If caregivers know a participant's allocated treatment, the study is at risk of performance bias, and if it is known to the person assessing the outcomes (particularly if these are subjective) the study is at risk of detection bias. To prevent this, ideally, the intervention is ‘masked’, but in certain situations in neonatal intensive care this is neither feasible nor ethical. In itself, there is no reason why masking of interventions should be less optimal in single-center studies. In a Cochrane review of cooling for hypoxiceischemic encephalopathy, none of the eleven included trials (three single-center, eight multicenter) were blinded, because of the nature of the intervention [18]. In another review including five studies comparing air with oxygen for neonatal resuscitation [19] (three single-center, two multicenter) only two studies were blinded, both of which were conducted in a single center. However, as well as the impact on allocation concealment, there are other implications of unblinded single-center trials that are important to consider. One is that if there is particular local experience with a given intervention or strategy, then clinicians are well placed to detect a problem, or be able to deviate from the protocol if they consider this in the best interests of the patient. This has implications for the external generalizability of the study, as clinicians in other centers may not have this knowledge and expertise. Another concern relates to the fact that investigators in singlecenter studies, who also directly provide care for the patients, may not be in equipoise about the intervention at the start of the trial, particularly if they hope to show that their approach is best. The doctors and nurses on the unit may be consciously or subconsciously keen to show that an intervention is beneficial and safe, either to justify its use or to please the investigator, and this ‘Hawthorne effect’ can significantly change behavior or outcome [17,20]. 5.3. Missing data All patients who are randomized in the study should be included in the final analysis, as those who drop out may do so for a reason related to the efficacy or safety of the interventions under investigation. The intention-to-treat principle describes the analysis that occurs when all patients at the end of the trial are analysed according to the treatment group to which they were allocated (“analyse as randomized”). Although there is no specific reason why single-center studies might not use an intention-to treat analysis, issues may arise if there is either particular expertise or a lack of equipoise with regard to an intervention or strategy. Take, for example, a theoretical single-center trial of two ventilation strategies A and B, where the center has particular expertise in A, and hopes to show that it is better. If the study is unblinded, some infants randomized to strategy B may be switched to A if the clinicians feel that this would be better for them. Additionally, infants predicted to have a good outcome may also be switched to A, or infants with a bad projected outcome may be switched to B. It is crucial that all the infants are accounted for in the final analysis according to the group to which they were randomized. In these circumstances the trial report should describe details of the switching such as how frequently it occurred, whether it happened before or after the time point at which the primary outcome was measured, and whether there was an a-priori definition of “treatment failure” that would trigger this switch.
3
There are currently no guidelines on how trials in which one arm acts as a “rescue” therapy for the other should be reported. 5.4. Outcome reporting bias The results of all outcomes that are measured and analysed should be presented in the final report. Frequently outcomes are not reported, or the primary outcome in a trial may be changed as the study progresses [21]. If either of these things happens on the basis that the results are not statistically significant then the final report is biased. In unblinded single-center trials, the investigator has, in real time, some idea of the results of the study. This has been described as an “ongoing interim analysis” [17]. This means that in singlecenter studies, an investigator may change the primary outcome to one with more favorable results. It is possible that single-center studies are less likely to utilize strategies that may protect against selective reporting, such as data safety monitoring committees, or making the trial protocol available, but this has not been systematically assessed. It is also possible that single-center studies are less likely to utilize core outcome sets, which are a minimum set of outcomes that should be measured and reported in all clinical trials in a given condition [22], but this has also not been evaluated thoroughly. 6. Examples of multi-center studies contradicting singlecenter studies in neonatology 6.1. Intravenous immunoglobulin (IVIG) for the treatment of infection Intravenous immunoglobulin (IVIG) was routinely used for the treatment of infection in preterm infants, and the rationale for its use was very plausible. Extremely preterm infants are at risk of severe infection because they neither make their own immunoglobulins, nor receive them transplacentally from their mother. IVIG provides immunoglobulin G that can improve immune function in several ways, and prophylactic administration showed statistically significant reductions in rates of sepsis [23]. A Cochrane review in 2010 [24] combined the results of seven small trials (six single-center and one conducted in three centers), in which IVIG was compared with placebo as an adjunctive for the treatment of suspected sepsis in infants. The six single-center studies had all individually showed a non-statistically significant reduction in the risk ratio for mortality during the initial hospital stay, and the small multi-center study showed no improvement. Meta-analysis of the results of these studies showed a statistically significant improvement in all-cause mortality [RD: 0.10 (95% CI 0.03 to 0.18); NNT: 10 (95% CI: 6 to 33)]. The INIS trial, published a year later [25], was conducted at 113 hospitals in nine countries, and enrolled 3493 infants who were receiving antibiotics for suspected or proven infection. IVIG did not improve the composite primary outcome, of death or severe neurodisability at two years of age [686/1759 (39.0%) in IVIG group vs 677/1734 (39.0%) in placebo group; relative risk: 1.00 (95% CI: 0.92 to 1.08)]. When combined with results from the previous Cochrane review, mortality during hospital stay in infants with clinically suspected infection was not significantly different after IVIG treatment (RD: 0.01; 95% CI: 0.04 to 0.02) [26]. 6.2. Non-invasive positive pressure ventilation for respiratory distress syndrome There is a move to using respiratory support for preterm infants that is as minimally invasive as possible, given the well-described
Please cite this article in press as: Sinha IP, Sinha SK, Single-center trials in neonatology: Issues to consider, Seminars in Fetal & Neonatal Medicine (2015), http://dx.doi.org/10.1016/j.siny.2015.08.003
4
I.P. Sinha, S.K. Sinha / Seminars in Fetal & Neonatal Medicine xxx (2015) 1e5
risks of ventilator-induced lung injury and increased risks of BPD with mechanical ventilation. Trials showed that early nasal continuous positive airways pressure (nCPAP) reduces the risk of BPD, when compared with intubation and ventilation, but may fail in extremely-low-birth-weight infants; thus, non-invasive positive pressure ventilation (NIPPV) has been suggested as an alternative modality with physiological benefits [27]. Meta-analysis of three single-center trials comparing NIPPV with nCPAP [28] suggested that use of NIPPV reduced the need for intubation and mechanical ventilation within the first 72 h of life [RR: 0.60 (95% CI: 0.43 to 0.83)] but did not reduce rates of BPD [RR: 0.56 (95% CI: 0.09 to 3.49)]. On the basis of the results of these small trials, some centers adopted the use of NIPPV [29], and a survey of tertiary UK neonatal units in 2008 found that around 50% of institutions were routinely using this modality [30]. In a subsequent trial [31], conducted at 34 centers, 1009 infants with birthweight <1000 g and gestational age <30 weeks were randomized to NIPPV or nCPAP. There was no difference between groups with regard to the primary outcome of the proportion of infants who died or who developed BPD [38.4% vs 36.7%; adjusted OR: 1.09 (95% CI: 0.83 to 1.43); P ¼ 0.56], or in either of the composite constituents of death or BPD.
Practice points Single-center studies generate important evidence in neonatology, and can inform subsequent studies. When reading reports of single-center studies, however, it is important to be aware that they are more likely to report larger treatment effects. It is important to consider whether the results of singlecenter trials are applicable in other situations e particularly if the center conducting the trial has a high incidence of a disease, or has particular experience with an intervention or strategy. Single-center studies may be at higher risk of bias, but this has not been investigated in detail. The effects of this bias may be compounded if there is a lack of equipoise at the center conducting the trial. This is particularly important in unblinded studies. It is important to consider the totality of evidence available, in context of the quality of the evidence, and so Cochrane reviews remain important in neonatal medicine.
7. Discussion In clinical trials in specialties such as neonatology, in which participants are particularly vulnerable to risk of both disease and interventions, the balance between doing more good than harm can itself be delicate. For this reason alone, single-center trials clearly have a place in neonatal medicine. They have started the process of research into many important questions. When conducted as pilot, feasibility, or early-phase studies they confer logistical and financial benefits over multi-center trials. The results of single-center studies should not simply be discounted. If they are well designed, recruit sufficient participants, and are well reported they should be considered in the process of making decisions about treatments. No studies to date have rigorously examined whether these markers of quality are inferior in single-center RCTs in neonatology. The examples of IVIG, NIPPV, and oxygen saturations, however, remind us that where possible multi-center trials should be conducted. That results of single-center studies may not be replicated by larger multi-center ones may reflect issues around external generalizability of the results, intrinsic bias that may lead to singlecenter studies finding artificially exaggerated treatment effects, or may simply be an example of how explanatory trials in a homogeneous population may not reflect results in a pragmatic trial across different populations. There is no clear pattern in neonatology as to how RCTs inform practice. On the one hand, when there was clear evidence that room air was no less effective for delivery room resuscitation than 100% oxygen [32], many centers were reticent to change practice [33,34]. On the other hand, when there was minimal evidence from a few small trials that NIPPV may be beneficial when compared to nCPAP, many centers adopted this technology. Once this decision has been made, given the money already invested in change and training of staff, it can be difficult to reject a treatment that is already in use. The best approach is to consider the totality of evidence, where possible using systematic reviews such as those available in the Cochrane Library to inform decisions about treatments. Whether from single- or multi-center trials, clinicians and policy-makers should interpret evidence with due consideration as to whether it is of sufficiently high quality to inform practice, and whether the results are applicable in their own institutions.
Research directions Methodological research is needed to rigorously compare the quality of single-center neonatal studies with that of multi-center trials. Further work is needed to identify aspects of single-center studies that could be improved to reduce the risks of bias. Trialists should make every attempt to minimize bias in their studies e especially if it is not possible to mask the identity of the interventions.
Conflict of interest statement None declared. Funding sources None. References [1] Conroy S, McIntyre J, Choonara I. Unlicensed and off label drug use in neonates. Archs Dis Childh Fetal Neonat Ed 1999;80:F142e5. [2] Wheeler K, Klingenberg C, McCallion N, Morley CJ, Davis PG. Volume-targeted versus pressure-limited ventilation in the neonate. Cochrane Database Syst Rev 2010;(11):CD003666. [3] Dechartres A, Boutron I, Trinquart L, Charles P, Ravaud P. Single-center trials show larger treatment effects than multicenter trials: evidence from a metaepidemiologic study. Ann Intern Med 2011;155:39e51. [4] Unverzagt S, Prondzinsky R, Peinemann F. Single-center trials tend to provide larger treatment effects than multicenter trials: a systematic review. J Clin Epidemiol 2013;66:1271e80. } ni P, Pildal J, et al. Influence of [5] Savovi c J, Jones HE, Altman DG, Harris RJ, Ju reported study design characteristics on intervention effect estimates from randomized, controlled trials. Ann Intern Med 2012;157:429e38. [6] Bafeta A, Dechartres A, Trinquart L, Yavchitz A, Boutron I, Ravaud P. Impact of single centre status on estimates of intervention effects in trials with continuous outcomes: meta-epidemiological study. BMJ 2012;344:e813. [7] Nüesch E, Trelle S, Reichenbach S, Rutjes AWS, Tschannen B, Altman DG, et al. Small study effects in meta-analyses of osteoarthritis trials: metaepidemiological study. BMJ 2010;341:c3515.
Please cite this article in press as: Sinha IP, Sinha SK, Single-center trials in neonatology: Issues to consider, Seminars in Fetal & Neonatal Medicine (2015), http://dx.doi.org/10.1016/j.siny.2015.08.003
I.P. Sinha, S.K. Sinha / Seminars in Fetal & Neonatal Medicine xxx (2015) 1e5 [8] Hopewell S, Loudon K, Clarke MJ, Oxman AD, Dickersin K. Publication bias in clinical trials due to statistical significance or direction of trial results. Cochrane Database Syst Rev 2009;(1):MR000006. [9] Scherer RW, Langenberg P, Elm E. Full publication of results initially presented in abstracts. Cochrane Database Syst Rev 2007;(2):MR000005. [10] Stern JM, Simes RJ. Publication bias: evidence of delayed publication in a cohort study of clinical research projects. BMJ 1997;315(7109):640e5. [11] Moran C, Smith PB, Cohen-Wolkowiez M, Benjamin DK. Clinical trial design in neonatal pharmacology: effect of center differences, with lessons from the pediatric oncology cooperative research experience. Clin Pharmacol Therapeut 2009;86:589e91. [12] Kaufman D, Boyle R, Hazen KC, Patrie JT, Robinson M, Donowitz LG. Fluconazole prophylaxis against fungal colonization and infection in preterm infants. N Engl J Med 2001;345:1660e6. [13] Austin N, McGuire W. Prophylactic systemic antifungal agents to prevent mortality and morbidity in very low birth weight infants. Cochrane Database Syst Rev 2013;(4):CD003850. [14] Clerihew L, Lamagni TL, Brocklehurst P, McGuire W. Invasive fungal infection in very low birthweight infants: national prospective surveillance study. Archs Dis Childh Fetal Neonatal Ed 2006;91:F188e92. [15] Higgins JPT, Altman DG, Gøtzsche PC, Jüni P, Moher D, Oxman AD, et al. The Cochrane Collaboration's tool for assessing risk of bias in randomised trials. BMJ 2011;343:d5928. [16] Schulz KF, Chalmers I, Hayes RJ, Altman DG. Empirical evidence of bias: Dimensions of methodological quality associated with estimates of treatment effects in controlled trials. JAMA 1995;273:408e12. [17] Bellomo R, Warrillow SJ, Reade MC. Why we should be wary of single-center trials. Crit Care Med 2009;37:3114e9. [18] Jacobs SE, Berg M, Hunt R, Tarnow-Mordi WO, Inder TE, Davis PG. Cooling for newborns with hypoxic ischaemic encephalopathy. Cochrane Database Syst Rev 2013;(1):CD003311. [19] Tan A, Schulze A, O'Donnell CP, Davis PG. Air versus oxygen for resuscitation of infants at birth. Cochrane Database Syst Rev 2005;(2):CD002273. [20] Eckmanns T, Bessert J, Behnke M, Gastmeier P, Ruden H. Compliance with antiseptic hand rub use in intensive care units: the Hawthorne Effect. Infect Control Hosp Epidemiol 2006;27:931e4. [21] Dwan K, Gamble C, Williamson PR, Kirkham JJ. Systematic review of the empirical evidence of study publication bias and outcome reporting bias e an updated review. PLoS One 2013;8:e66844.
5
[22] Sinha I, Jones L, Smyth RL, Williamson PR. A systematic review of studies that aim to determine which outcomes to measure in clinical trials in children. PLoS Med 2008;5:e96. [23] Ohlsson A, Lacy JB. Intravenous immunoglobulin for preventing infection in preterm and/or low birth weight infants. Cochrane Database Syst Rev 2013;(7):CD000361. [24] Ohlsson A, Lacy J, Ohlsson A. Intravenous immunoglobulin for suspected or subsequently proven infection in neonates. Cochrane Database Syst Rev 2010;(3):CD001239. [25] INIS Collaborative Group. Treatment of neonatal sepsis with intravenous immune globulin. N Engl J Med 2011;365:1201e11. [26] Ohlsson A, Lacy JB. Intravenous immunoglobulin for suspected or proven infection in neonates. Cochrane Database Syst Rev 2015;(3):CD001239. [27] Sinha IP, Sinha SK. Alternative therapies for respiratory distress syndrome in preterm infants. Res Rep Neonatol 2011;1:67e75. [28] Meneses J, Bhandari V, Alves JG. Nasal intermittent positive-pressure ventilation vs nasal continuous positive airway pressure for preterm infants with respiratory distress syndrome: a systematic review and meta-analysis. Archs Pediatr Adolesc Med 2012;166:372e6. [29] Jackson JK, Vellucci J, Johnson P, Kilbride HW. Evidence-based approach to change in clinical practice: introduction of expanded nasal continuous positive airway pressure use in an intensive care nursery. Pediatrics 2003;111(Suppl. E1):e542e7. [30] Owen LS, Morley CJ, Davis PG. Neonatal nasal intermittent positive pressure ventilation: a survey of practice in England. Archs Dis Childh Fetal Neonatal Ed 2008;93:F148e50. [31] Kirpalani H, Millar D, Lemyre B, Yoder BA, Chiu A, Roberts RS. A trial comparing noninvasive ventilation strategies in preterm infants. N Engl J Med 2013;369:611e20. [32] Saugstad OD, Ramji S, Soll RF, Vento M. Resuscitation of newborn infants with 21% or 100% oxygen: an updated systematic review and meta-analysis. Neonatology 2008;94:176e82. [33] Whitby TM, Whitby V, Sinha I. Delivery room resuscitation in the UK: postsurvey follow-up. Archs Dis Childh Fetal Neonatal Ed 2013;98:F182e3. M, Buro n E, Salguero E, Aguayo J, Vento M. A survey of [34] Iriondo M, Thio neonatal resuscitation in Spain: gaps between guidelines and practice. Acta Paediatr 2009;98:786e91.
Please cite this article in press as: Sinha IP, Sinha SK, Single-center trials in neonatology: Issues to consider, Seminars in Fetal & Neonatal Medicine (2015), http://dx.doi.org/10.1016/j.siny.2015.08.003
Contents lists available at ScienceDirect
Seminars in Fetal & Neonatal Medicine journal homepage: www.elsevier.com/locate/siny
Review
Single-center trials in neonatology: Issues to consider Ian P. Sinha a, *, Sunil K. Sinha b a b
Respiratory Unit, Alder Hey Children's Hospital, Liverpool, UK Department of Neonatology, University of Durham and James Cook University Hospital, Middlesbrough, UK
s u m m a r y Keywords: Evidence-based medicine Neonatology Systematic review Trials
Single-center randomized controlled trials confer certain advantages over multi-center trials, in that they are cheaper and easier to design and conduct. However, recent research suggests that single-center trials are likely to overestimate treatment effects. There are notable examples in neonatology where results from multi-center trials have contradicted results of single-center studies. In this paper we discuss issues around external generalizability of single-center studies, and methodological issues that may cause bias. © 2015 Published by Elsevier Ltd.
1. Introduction In neonatology many drugs, interventions and practices have developed as a result of randomized controlled trials (RCTs) and meta-analyses. More trials need to be conducted, as there are uncertainties around treatments, and several drugs used in newborn infants are unlicensed [1]. These will need to be appraised and decisions will be made about whether the results should change practice. Some of these trials will be large multi-center studies, and others will be conducted in a single center. In this paper we consider issues around single-center RCTs. 2. The attractions of single-center studies The obvious advantage of single-center studies is that they are logistically easier, and usually cheaper, than multi-center studies. With increasing numbers of centers participating in a trial, it becomes more complex and expensive to conduct and monitor sitespecific training, administrative duties, and governance. It can also become more difficult for collaborators to reach agreement about the implications of the results and write the final report. Even though trial networks can foster collaboration and make this process easier, the cost of a multi-center study may be prohibitive in some situations. In low-income countries where there may not be a robust infrastructure for collaborative research, it may only be possible to conduct studies in one center.
* Corresponding author. Address: Respiratory Unit, Alder Hey Children's Hospital, Alder Road, Liverpool L12 2AP, UK. Tel.: þ44 151 2284811. E-mail address: [email protected] (I.P. Sinha).
Single-center trials are particularly appropriate for studies that may not require a large number of patients, such as early-phase, pilot or feasibility studies. Such trials can be invaluable for informing subsequent studies, such as testing validity and importance of outcome measures, evaluating study protocols, assessing participant recruitment rates, and generating data to inform future sample size calculations. Single-center studies can provide useful estimates of benefits and harms, especially when aggregated in meta-analyses to help guide therapies. For example, a Cochrane review [2] comparing volume-targeted ventilation (VTV) with pressure-targeted ventilation included 12 trials (eight single-center, four conducted in two centers). VTV was found to reduce the risk of the composite outcome of death or bronchopulmonary dysplasia (BPD) [risk reduction (RD): 0.12 (95% CI: 0.21 to 0.03); number needed to treat (NNT): 8 (95% CI: 5 to 33)], and other clinically important outcomes. 3. Single-center studies show larger treatment effects Estimates of treatment effect appear to be larger in single-center RCTs. Meta-epidemiological analyses using the summary statistic ratio of odds ratio (ROR) have systematically examined trials in several specialties; ROR <1 indicates a larger estimate of the intervention effect in single-center trials than multi-center trials. In two analyses [3,4], single-center trials were more likely to show greater benefit with regards to dichotomous outcomes {combined ROR: 0.73 (95% CI: 0.64 to 0.83) [3] and 0.64 (95% CI: 0.47 to 0.87) [4]}, and in one there was a trend towards this finding which was not statistically significant {ROR: 0.91 (CI: 0.79 to 1.04) [5]}. Another analysis reported that single-center trials showed larger
http://dx.doi.org/10.1016/j.siny.2015.08.003 1744-165X/© 2015 Published by Elsevier Ltd.
Please cite this article in press as: Sinha IP, Sinha SK, Single-center trials in neonatology: Issues to consider, Seminars in Fetal & Neonatal Medicine (2015), http://dx.doi.org/10.1016/j.siny.2015.08.003
2
I.P. Sinha, S.K. Sinha / Seminars in Fetal & Neonatal Medicine xxx (2015) 1e5
intervention effects for continuous outcomes {combined difference in standardized mean differences 0.09 (95% CI: 0.17 to 0.01, P ¼ 0.04) [6]}. A combination of various factors probably causes this phenomenon. Single-center studies tend to recruit fewer patients, and it is recognized that smaller studies show greater effects [7]. In some studies the sample size may be so small that it is purely due to chance that a treatment has shown a benefit. However, in one analysis, the overestimation of treatment effect in single-center studies appears to remain consistent even when results are adjusted for sample size [3]. Publication bias, or selective reporting of trials based on the results, is well recognized [8]. One strategy to improve publication rates of trials is to prospectively list them in a registry, but whether single-center studies are less likely to be registered has not been tested. One meta-epidemiological study [9] combined results from three reports that examined full publication of abstracts of RCTs that were initially presented at conferences and found that single-center studies were no less likely to be published [risk ratio (RR): 1.27 (95% CI: 0.95 to 1.70)]. Another analysis found that single-center studies were more likely never to have started after they had been planned, or to have been abandoned after they started [123/337 (36%) vs 28/163 (17%); P < 0.0001], but single-center status did not predict whether a study was published or not [10]. In this paper we consider two other reasons why single-center studies may show greater treatment effects. The first is the likelihood that there are particular features of the center or investigator that may affect the magnitude of treatment effect, and hence the external generalizability of the results. The second is that some aspect of the methodology in single-center studies makes it more likely that treatments are shown to be beneficial (in other words, single-center studies may be prone to bias). 4. Are the results of single-center studies externally generalizable? Single-center studies are conducted in a more homogeneous population than multi-center trials. They are better suited therefore to efficacy (explanatory) trials, which test whether an intervention works in optimum conditions in contrast to effectiveness (pragmatic) trials which measure the effect in ‘real world’ settings. When interpreting the conclusion of any RCT, it is important to determine why the study was conducted. A center may want to conduct a trial if a particular problem occurs frequently in its own institution. This has implications in areas of neonatology in which there are substantial differences in outcomes between institutions [11]. One notable example is the use of prophylactic fluconazole in preterm infants. A single-center study of 100 infants, randomized to intravenous fluconazole or placebo, showed that antifungal prophylaxis reduced fungal colonization [11/50 (22%) vs 30/50 (60%); RD: 0.38 (95% CI: 0.18 to 0.56); P ¼ 0.002] and invasive fungal infection [0/50 vs 10/50 (20%); RD: 0.20 (95% CI: 0.04 to 0.36); P ¼ 0.008] [12]. A Cochrane review of systemic prophylactic antifungal therapy for very-low-birthweight infants incorporated seven studies, of which five were in single centers, one in eight centres, and one in two [13]. Although the review found a significant reduction in invasive fungal rates among very-lowbirthweight infants [RD: 0.09 (95% CI: 0.14 to 0.05), NNT: 11], the authors highlight that this may reflect the high incidence of systemic candidiasis in the control groups (mean 16%). In the UK, where the typical rates of invasive fungal sepsis are only 1% [14], a large number of infants would be exposed to potential risks of fluconazole but only very few would be expected to benefit.
Another reason why a center may wish to conduct an RCT is that it has particular experience or expertise with an intervention or strategy. For example, single-center trials have shown clear benefits of VTV, but it is uncertain whether other institutions can replicate the success of these studies, particularly as they may lack the clinical and nursing expertise available at the study centers. 5. Theoretical reasons why single-center studies are more prone to bias Bias refers to aspects of the trial (other than the effects of the interventions) that give one treatment arm an unfair advantage over another. In the Cochrane Risk of Bias tool, aspects of methodology and reporting are categorized as low, high, or unclear risk of bias [15]. There are some specific important considerations around these domains in context of single-center studies. Evidence is lacking, however, as to whether single-center studies are more likely to be classed as having high risk of bias, so these are currently theoretical. 5.1. Selection bias In an RCT, the groups of participants should be identical, except for the intervention they receive. If there are methodological reasons why this may not happen, the study is prone to selection bias, and is more likely to show greater treatment effects [16]. Avoidance of selection bias relies on two things e first that the participants are allocated to groups at random, and second that those enrolling participants into the trial are unaware of the treatment group to which they will be allocated. There is no specific reason why the process of randomization should be flawed in single-center studies. For this aspect of the trial to be classed as low risk of bias, a method should be used that allocates patients completely at random (for example by using random number tables or computerized random number generators) rather than by using a systematic approach to randomization (such as allocating participants to groups on the basis of date of birth, day of admission, or patient identification number). However, it is possible that in some single-center studies the process of allocation concealment is less optimal. This is important as studies with inadequate allocation concealment can overestimate treatment effects by 40% [16]. To prevent investigators from knowing the allocated groups of subsequent participants in advance (and hence affecting whether they are enrolled in the study or not), a trial can use allocation procedures at a separate central location (which could be telephone, web-based, or pharmacy-controlled), or measures to hide the allocation until the patient is enrolled (such as sequentially numbered drug containers of identical appearance, or opaque, sealed envelopes). Although there is no specific reason why single-center studies should not employ such methods (other than cost) it is more likely that the randomization schedule (i.e. the groups to which patients will be allocated) will be held locally rather than centrally, so unless appropriate concealment measures are taken, the study is at high risk of bias. In a commentary about single-center studies in adult critical care, the author suggests that it may be more difficult to conceal the likelihood of randomization to a particular group if a trial is conducted in one center [17]. This relates to the use of block randomization (i.e. randomization within blocks of smaller numbers of participants, to try to ensure fairly equal numbers in each group). If an investigator knows the block size, in an unblinded single-center study, they may be able to foresee the next treatment allocation.
Please cite this article in press as: Sinha IP, Sinha SK, Single-center trials in neonatology: Issues to consider, Seminars in Fetal & Neonatal Medicine (2015), http://dx.doi.org/10.1016/j.siny.2015.08.003
I.P. Sinha, S.K. Sinha / Seminars in Fetal & Neonatal Medicine xxx (2015) 1e5
5.2. Performance and detection bias Once a participant is randomized, it is ideal if personnel involved in their care, and trial personnel, do not know the treatment group to which they are allocated. If caregivers know a participant's allocated treatment, the study is at risk of performance bias, and if it is known to the person assessing the outcomes (particularly if these are subjective) the study is at risk of detection bias. To prevent this, ideally, the intervention is ‘masked’, but in certain situations in neonatal intensive care this is neither feasible nor ethical. In itself, there is no reason why masking of interventions should be less optimal in single-center studies. In a Cochrane review of cooling for hypoxiceischemic encephalopathy, none of the eleven included trials (three single-center, eight multicenter) were blinded, because of the nature of the intervention [18]. In another review including five studies comparing air with oxygen for neonatal resuscitation [19] (three single-center, two multicenter) only two studies were blinded, both of which were conducted in a single center. However, as well as the impact on allocation concealment, there are other implications of unblinded single-center trials that are important to consider. One is that if there is particular local experience with a given intervention or strategy, then clinicians are well placed to detect a problem, or be able to deviate from the protocol if they consider this in the best interests of the patient. This has implications for the external generalizability of the study, as clinicians in other centers may not have this knowledge and expertise. Another concern relates to the fact that investigators in singlecenter studies, who also directly provide care for the patients, may not be in equipoise about the intervention at the start of the trial, particularly if they hope to show that their approach is best. The doctors and nurses on the unit may be consciously or subconsciously keen to show that an intervention is beneficial and safe, either to justify its use or to please the investigator, and this ‘Hawthorne effect’ can significantly change behavior or outcome [17,20]. 5.3. Missing data All patients who are randomized in the study should be included in the final analysis, as those who drop out may do so for a reason related to the efficacy or safety of the interventions under investigation. The intention-to-treat principle describes the analysis that occurs when all patients at the end of the trial are analysed according to the treatment group to which they were allocated (“analyse as randomized”). Although there is no specific reason why single-center studies might not use an intention-to treat analysis, issues may arise if there is either particular expertise or a lack of equipoise with regard to an intervention or strategy. Take, for example, a theoretical single-center trial of two ventilation strategies A and B, where the center has particular expertise in A, and hopes to show that it is better. If the study is unblinded, some infants randomized to strategy B may be switched to A if the clinicians feel that this would be better for them. Additionally, infants predicted to have a good outcome may also be switched to A, or infants with a bad projected outcome may be switched to B. It is crucial that all the infants are accounted for in the final analysis according to the group to which they were randomized. In these circumstances the trial report should describe details of the switching such as how frequently it occurred, whether it happened before or after the time point at which the primary outcome was measured, and whether there was an a-priori definition of “treatment failure” that would trigger this switch.
3
There are currently no guidelines on how trials in which one arm acts as a “rescue” therapy for the other should be reported. 5.4. Outcome reporting bias The results of all outcomes that are measured and analysed should be presented in the final report. Frequently outcomes are not reported, or the primary outcome in a trial may be changed as the study progresses [21]. If either of these things happens on the basis that the results are not statistically significant then the final report is biased. In unblinded single-center trials, the investigator has, in real time, some idea of the results of the study. This has been described as an “ongoing interim analysis” [17]. This means that in singlecenter studies, an investigator may change the primary outcome to one with more favorable results. It is possible that single-center studies are less likely to utilize strategies that may protect against selective reporting, such as data safety monitoring committees, or making the trial protocol available, but this has not been systematically assessed. It is also possible that single-center studies are less likely to utilize core outcome sets, which are a minimum set of outcomes that should be measured and reported in all clinical trials in a given condition [22], but this has also not been evaluated thoroughly. 6. Examples of multi-center studies contradicting singlecenter studies in neonatology 6.1. Intravenous immunoglobulin (IVIG) for the treatment of infection Intravenous immunoglobulin (IVIG) was routinely used for the treatment of infection in preterm infants, and the rationale for its use was very plausible. Extremely preterm infants are at risk of severe infection because they neither make their own immunoglobulins, nor receive them transplacentally from their mother. IVIG provides immunoglobulin G that can improve immune function in several ways, and prophylactic administration showed statistically significant reductions in rates of sepsis [23]. A Cochrane review in 2010 [24] combined the results of seven small trials (six single-center and one conducted in three centers), in which IVIG was compared with placebo as an adjunctive for the treatment of suspected sepsis in infants. The six single-center studies had all individually showed a non-statistically significant reduction in the risk ratio for mortality during the initial hospital stay, and the small multi-center study showed no improvement. Meta-analysis of the results of these studies showed a statistically significant improvement in all-cause mortality [RD: 0.10 (95% CI 0.03 to 0.18); NNT: 10 (95% CI: 6 to 33)]. The INIS trial, published a year later [25], was conducted at 113 hospitals in nine countries, and enrolled 3493 infants who were receiving antibiotics for suspected or proven infection. IVIG did not improve the composite primary outcome, of death or severe neurodisability at two years of age [686/1759 (39.0%) in IVIG group vs 677/1734 (39.0%) in placebo group; relative risk: 1.00 (95% CI: 0.92 to 1.08)]. When combined with results from the previous Cochrane review, mortality during hospital stay in infants with clinically suspected infection was not significantly different after IVIG treatment (RD: 0.01; 95% CI: 0.04 to 0.02) [26]. 6.2. Non-invasive positive pressure ventilation for respiratory distress syndrome There is a move to using respiratory support for preterm infants that is as minimally invasive as possible, given the well-described
Please cite this article in press as: Sinha IP, Sinha SK, Single-center trials in neonatology: Issues to consider, Seminars in Fetal & Neonatal Medicine (2015), http://dx.doi.org/10.1016/j.siny.2015.08.003
4
I.P. Sinha, S.K. Sinha / Seminars in Fetal & Neonatal Medicine xxx (2015) 1e5
risks of ventilator-induced lung injury and increased risks of BPD with mechanical ventilation. Trials showed that early nasal continuous positive airways pressure (nCPAP) reduces the risk of BPD, when compared with intubation and ventilation, but may fail in extremely-low-birth-weight infants; thus, non-invasive positive pressure ventilation (NIPPV) has been suggested as an alternative modality with physiological benefits [27]. Meta-analysis of three single-center trials comparing NIPPV with nCPAP [28] suggested that use of NIPPV reduced the need for intubation and mechanical ventilation within the first 72 h of life [RR: 0.60 (95% CI: 0.43 to 0.83)] but did not reduce rates of BPD [RR: 0.56 (95% CI: 0.09 to 3.49)]. On the basis of the results of these small trials, some centers adopted the use of NIPPV [29], and a survey of tertiary UK neonatal units in 2008 found that around 50% of institutions were routinely using this modality [30]. In a subsequent trial [31], conducted at 34 centers, 1009 infants with birthweight <1000 g and gestational age <30 weeks were randomized to NIPPV or nCPAP. There was no difference between groups with regard to the primary outcome of the proportion of infants who died or who developed BPD [38.4% vs 36.7%; adjusted OR: 1.09 (95% CI: 0.83 to 1.43); P ¼ 0.56], or in either of the composite constituents of death or BPD.
Practice points Single-center studies generate important evidence in neonatology, and can inform subsequent studies. When reading reports of single-center studies, however, it is important to be aware that they are more likely to report larger treatment effects. It is important to consider whether the results of singlecenter trials are applicable in other situations e particularly if the center conducting the trial has a high incidence of a disease, or has particular experience with an intervention or strategy. Single-center studies may be at higher risk of bias, but this has not been investigated in detail. The effects of this bias may be compounded if there is a lack of equipoise at the center conducting the trial. This is particularly important in unblinded studies. It is important to consider the totality of evidence available, in context of the quality of the evidence, and so Cochrane reviews remain important in neonatal medicine.
7. Discussion In clinical trials in specialties such as neonatology, in which participants are particularly vulnerable to risk of both disease and interventions, the balance between doing more good than harm can itself be delicate. For this reason alone, single-center trials clearly have a place in neonatal medicine. They have started the process of research into many important questions. When conducted as pilot, feasibility, or early-phase studies they confer logistical and financial benefits over multi-center trials. The results of single-center studies should not simply be discounted. If they are well designed, recruit sufficient participants, and are well reported they should be considered in the process of making decisions about treatments. No studies to date have rigorously examined whether these markers of quality are inferior in single-center RCTs in neonatology. The examples of IVIG, NIPPV, and oxygen saturations, however, remind us that where possible multi-center trials should be conducted. That results of single-center studies may not be replicated by larger multi-center ones may reflect issues around external generalizability of the results, intrinsic bias that may lead to singlecenter studies finding artificially exaggerated treatment effects, or may simply be an example of how explanatory trials in a homogeneous population may not reflect results in a pragmatic trial across different populations. There is no clear pattern in neonatology as to how RCTs inform practice. On the one hand, when there was clear evidence that room air was no less effective for delivery room resuscitation than 100% oxygen [32], many centers were reticent to change practice [33,34]. On the other hand, when there was minimal evidence from a few small trials that NIPPV may be beneficial when compared to nCPAP, many centers adopted this technology. Once this decision has been made, given the money already invested in change and training of staff, it can be difficult to reject a treatment that is already in use. The best approach is to consider the totality of evidence, where possible using systematic reviews such as those available in the Cochrane Library to inform decisions about treatments. Whether from single- or multi-center trials, clinicians and policy-makers should interpret evidence with due consideration as to whether it is of sufficiently high quality to inform practice, and whether the results are applicable in their own institutions.
Research directions Methodological research is needed to rigorously compare the quality of single-center neonatal studies with that of multi-center trials. Further work is needed to identify aspects of single-center studies that could be improved to reduce the risks of bias. Trialists should make every attempt to minimize bias in their studies e especially if it is not possible to mask the identity of the interventions.
Conflict of interest statement None declared. Funding sources None. References [1] Conroy S, McIntyre J, Choonara I. Unlicensed and off label drug use in neonates. Archs Dis Childh Fetal Neonat Ed 1999;80:F142e5. [2] Wheeler K, Klingenberg C, McCallion N, Morley CJ, Davis PG. Volume-targeted versus pressure-limited ventilation in the neonate. Cochrane Database Syst Rev 2010;(11):CD003666. [3] Dechartres A, Boutron I, Trinquart L, Charles P, Ravaud P. Single-center trials show larger treatment effects than multicenter trials: evidence from a metaepidemiologic study. Ann Intern Med 2011;155:39e51. [4] Unverzagt S, Prondzinsky R, Peinemann F. Single-center trials tend to provide larger treatment effects than multicenter trials: a systematic review. J Clin Epidemiol 2013;66:1271e80. } ni P, Pildal J, et al. Influence of [5] Savovi c J, Jones HE, Altman DG, Harris RJ, Ju reported study design characteristics on intervention effect estimates from randomized, controlled trials. Ann Intern Med 2012;157:429e38. [6] Bafeta A, Dechartres A, Trinquart L, Yavchitz A, Boutron I, Ravaud P. Impact of single centre status on estimates of intervention effects in trials with continuous outcomes: meta-epidemiological study. BMJ 2012;344:e813. [7] Nüesch E, Trelle S, Reichenbach S, Rutjes AWS, Tschannen B, Altman DG, et al. Small study effects in meta-analyses of osteoarthritis trials: metaepidemiological study. BMJ 2010;341:c3515.
Please cite this article in press as: Sinha IP, Sinha SK, Single-center trials in neonatology: Issues to consider, Seminars in Fetal & Neonatal Medicine (2015), http://dx.doi.org/10.1016/j.siny.2015.08.003
I.P. Sinha, S.K. Sinha / Seminars in Fetal & Neonatal Medicine xxx (2015) 1e5 [8] Hopewell S, Loudon K, Clarke MJ, Oxman AD, Dickersin K. Publication bias in clinical trials due to statistical significance or direction of trial results. Cochrane Database Syst Rev 2009;(1):MR000006. [9] Scherer RW, Langenberg P, Elm E. Full publication of results initially presented in abstracts. Cochrane Database Syst Rev 2007;(2):MR000005. [10] Stern JM, Simes RJ. Publication bias: evidence of delayed publication in a cohort study of clinical research projects. BMJ 1997;315(7109):640e5. [11] Moran C, Smith PB, Cohen-Wolkowiez M, Benjamin DK. Clinical trial design in neonatal pharmacology: effect of center differences, with lessons from the pediatric oncology cooperative research experience. Clin Pharmacol Therapeut 2009;86:589e91. [12] Kaufman D, Boyle R, Hazen KC, Patrie JT, Robinson M, Donowitz LG. Fluconazole prophylaxis against fungal colonization and infection in preterm infants. N Engl J Med 2001;345:1660e6. [13] Austin N, McGuire W. Prophylactic systemic antifungal agents to prevent mortality and morbidity in very low birth weight infants. Cochrane Database Syst Rev 2013;(4):CD003850. [14] Clerihew L, Lamagni TL, Brocklehurst P, McGuire W. Invasive fungal infection in very low birthweight infants: national prospective surveillance study. Archs Dis Childh Fetal Neonatal Ed 2006;91:F188e92. [15] Higgins JPT, Altman DG, Gøtzsche PC, Jüni P, Moher D, Oxman AD, et al. The Cochrane Collaboration's tool for assessing risk of bias in randomised trials. BMJ 2011;343:d5928. [16] Schulz KF, Chalmers I, Hayes RJ, Altman DG. Empirical evidence of bias: Dimensions of methodological quality associated with estimates of treatment effects in controlled trials. JAMA 1995;273:408e12. [17] Bellomo R, Warrillow SJ, Reade MC. Why we should be wary of single-center trials. Crit Care Med 2009;37:3114e9. [18] Jacobs SE, Berg M, Hunt R, Tarnow-Mordi WO, Inder TE, Davis PG. Cooling for newborns with hypoxic ischaemic encephalopathy. Cochrane Database Syst Rev 2013;(1):CD003311. [19] Tan A, Schulze A, O'Donnell CP, Davis PG. Air versus oxygen for resuscitation of infants at birth. Cochrane Database Syst Rev 2005;(2):CD002273. [20] Eckmanns T, Bessert J, Behnke M, Gastmeier P, Ruden H. Compliance with antiseptic hand rub use in intensive care units: the Hawthorne Effect. Infect Control Hosp Epidemiol 2006;27:931e4. [21] Dwan K, Gamble C, Williamson PR, Kirkham JJ. Systematic review of the empirical evidence of study publication bias and outcome reporting bias e an updated review. PLoS One 2013;8:e66844.
5
[22] Sinha I, Jones L, Smyth RL, Williamson PR. A systematic review of studies that aim to determine which outcomes to measure in clinical trials in children. PLoS Med 2008;5:e96. [23] Ohlsson A, Lacy JB. Intravenous immunoglobulin for preventing infection in preterm and/or low birth weight infants. Cochrane Database Syst Rev 2013;(7):CD000361. [24] Ohlsson A, Lacy J, Ohlsson A. Intravenous immunoglobulin for suspected or subsequently proven infection in neonates. Cochrane Database Syst Rev 2010;(3):CD001239. [25] INIS Collaborative Group. Treatment of neonatal sepsis with intravenous immune globulin. N Engl J Med 2011;365:1201e11. [26] Ohlsson A, Lacy JB. Intravenous immunoglobulin for suspected or proven infection in neonates. Cochrane Database Syst Rev 2015;(3):CD001239. [27] Sinha IP, Sinha SK. Alternative therapies for respiratory distress syndrome in preterm infants. Res Rep Neonatol 2011;1:67e75. [28] Meneses J, Bhandari V, Alves JG. Nasal intermittent positive-pressure ventilation vs nasal continuous positive airway pressure for preterm infants with respiratory distress syndrome: a systematic review and meta-analysis. Archs Pediatr Adolesc Med 2012;166:372e6. [29] Jackson JK, Vellucci J, Johnson P, Kilbride HW. Evidence-based approach to change in clinical practice: introduction of expanded nasal continuous positive airway pressure use in an intensive care nursery. Pediatrics 2003;111(Suppl. E1):e542e7. [30] Owen LS, Morley CJ, Davis PG. Neonatal nasal intermittent positive pressure ventilation: a survey of practice in England. Archs Dis Childh Fetal Neonatal Ed 2008;93:F148e50. [31] Kirpalani H, Millar D, Lemyre B, Yoder BA, Chiu A, Roberts RS. A trial comparing noninvasive ventilation strategies in preterm infants. N Engl J Med 2013;369:611e20. [32] Saugstad OD, Ramji S, Soll RF, Vento M. Resuscitation of newborn infants with 21% or 100% oxygen: an updated systematic review and meta-analysis. Neonatology 2008;94:176e82. [33] Whitby TM, Whitby V, Sinha I. Delivery room resuscitation in the UK: postsurvey follow-up. Archs Dis Childh Fetal Neonatal Ed 2013;98:F182e3. M, Buro n E, Salguero E, Aguayo J, Vento M. A survey of [34] Iriondo M, Thio neonatal resuscitation in Spain: gaps between guidelines and practice. Acta Paediatr 2009;98:786e91.
Please cite this article in press as: Sinha IP, Sinha SK, Single-center trials in neonatology: Issues to consider, Seminars in Fetal & Neonatal Medicine (2015), http://dx.doi.org/10.1016/j.siny.2015.08.003