OBJECTIVES: The main objective of this direct meta-analysis (DMA) was to quantitatively assess birth prior to, or soon after the start of the respiratory syncytial virus infection (RSV) season (age) and presence of school-age siblings (siblings) as risk factors for RSV infection requiring hospitalization in healthy preterm infants. A further objective was to assess the robustness of frequentist approaches, preferred for DMA, using Bayesian techniques.
METHODS: A RevMan frequentist DMA was performed and the results compared with those from a WinBugs Bayesian analysis. Both random and fixed effect models were developed using data identified from 1985 to 2014 by a systematic literature review (SLR).
RESULTS: 5 observational studies included data suitable for meta-analysis for either age or sibling risk factors. Odds ratio estimates (95% confidence limits) generated using a frequentist fixed effect model were 3.27 (2.67, 4.00) for age and 2.35 (1.92, 2.88) for siblings. Using a Bayesian fixed effect model, estimates were 3.27 (2.67, 4.01) and 2.35 (1.92, 2.88) respectively. Odds ratio estimates (95% confidence limits) generated using a frequentist random effect model were 3.30 (2.62, 4.17) for age and 2.36 (1.92, 2.91) for siblings. Using a Bayesian random effect model, estimates were 3.33 (1.84, 6.34) and 2.43 (1.66, 3.97) respectively.
CONCLUSIONS: DMA of the SLR evidence, using either a frequentist or Bayesian methods, supports the notion that age and siblings are important risk factors for RSV hospitalization. Estimates for frequentist and Bayesian fixed effects models were highly consistent for age and siblings indicating that the results for the frequentist fixed effect DMA for age and siblings were robust. However Bayesian parameter estimates for the random effect model, estimated a greater degree of uncertainty (wider confidence interval) within the data than the frequentist method. This inconsistency would suggest further investigation is required to identify which method is most accurate.
