In-service training for health professionals to improve care of seriously ill newborns and children in low-income countries

Opiyo, Newton; English, Mike

Included studies assessed different interventions and outcomes. Meta-analysis was therefore inappropriate. We undertook a structured synthesis of results. In Senarath 2007, a unit of analysis error occurred; hospitals were randomly assigned and performance at deliveries was analysed, without adjustment for clustering. In addition, outcomes in intervention and control groups were not directly compared (comparisons were made within comparison groups before and after the intervention). Re-analysis was possible for only one outcomepreparedness for resuscitation -for which baseline levels of resuscitation practices were comparable between study groups. In the re-analysis, we assessed training effect by computing mean differences in outcomes, using reported standard deviations to estimate standard errors. To account for clustering, we assumed an intracluster correlation coefficient (ICC) of 0.015 (with a design effect of 1.129) that was based on published data (Rowe 2002). Review authors (NO and ME) independently assessed the certainty of evidence for each outcome using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) system (Guyatt 2008) . This approach classifies the certainty of evidence (defined as 'the extent to which one can be confident that an estimate of effect or association is correct') into one of four categories ('high', 'moderate', 'low' or 'very low'). We resolved disagreements on certainty ratings by discussion. We did not exclude studies on the basis of their GRADE certainty ratings; we took into account the certainty of evidence when synthesising overall findings. We report the results of certainty assessments in the 'Summary of findings tables' section.