A descriptive analysis of risk of bias in neonatal trials and the association with effect estimates

Article type
Bialy L1, Lacaze-Masmonteil T2, Dryden D1, Ha C1, Armijo-Olivo S1, Vandermeer B1, Hartling L1
1Alberta Research Centre for Health Evidence, Canada
2Children’s Hospital of Eastern Ontario, Canada
Background: Research that quantifies the impact of different biases on effect sizes in neonatal randomized controlled trials (RCTs) has shown conflicting results. A meta-epidemiological study to quantify bias in neonatal RCTs will inform the design, conduct, and interpretation of research in this vulnerable population.

Objectives: To describe a sample of neonatal RCTs in terms of methodology, risk of bias, and associations with effect estimates.

Methods: We included all neonatal RCTs (n= 208) in the Cochrane Database of Systematic Reviews that examined the following key treatments: surfactant, corticosteroids, nitric oxide, indomethacin, ibuprofen, and head or total body cooling. Risk of bias was assessed on nine domains by two independent reviewers. Meta-epidemiological methods will be used to quantify the association between pre-specified methodological characteristics and effect estimates.

Results: The assessed RCTs were published between 1972 and 2009 with 52% conducted across multiple centres. A single or composite primary outcome was reported in 49%. Trials were supported by grants from academic institutions or governments (37%) or pharmaceutical industry (26%); 37% did not report a funding source. No RCTs had an overall 'low’ risk of bias assessment; 42% were assessed as high risk of bias and 58% were unclear. For allocation concealment and blinding of participants and investigators most RCTs received an unclear risk of bias (69%, 74%, and 64%, respectively). Nineteen percent were assigned a high risk for selectively reporting outcomes and 60% were unclear. Analyses of association between risk of bias and effect estimates are ongoing.

Conclusions: We have described the risk of bias for RCTs in 6 key areas of neonatal medicine. This body of literature shows methodological limitations in terms of selective reporting and documentation of blinding methods. It is critical to understand and quantify the potential impact of these biases on effect estimates in order to better ensure the delivery of optimal neonatal care.