Sara van den Berg

66 Chapter 3 Figure 5. Effect of CMV serostatus on response to influenza vaccination. Results of the DerSimonian–Laird random effects model meta-analysis of five studies that included numbers of responders and non-responders to influenza vaccination. Odds ratios (diamonds) of the effect of CMV serostatus on responders to influenza and their 95% CI error bars (width of diamonds) are shown. Studies are split by age of the study population (<60 or ≥60) and definition of responder that was used in the study: either ≥four-fold increase or a four-fold increase in combination with a post-vaccination titer ≥40 hemagglutinating units (HAU). The influenza strain, number of study participants and overall study quality are noted for each study.  I 2 (the percentage of variation across studies that is due to heterogeneity rather than chance), Q  (the weighted sum of squared differences between individual study effects and the pooled effect across studies) and p  values (to determine whether significant heterogeneity exists) are calculated for every subgroup separately and for all studies together. Arrows indicate error bars on the odds ratio extending beyond the scale. Funnel-plot analysis suggests a publication bias in meta-analysis Because a potential positive effect of CMV-seropositivity on influenza vaccine responses was only recently considered, we assessed whether there was any evidence for a publication bias in the studies included in our meta-analysis by performing a funnel-plot analysis [37]. Such an analysis is based on two assumptions. It assumes (1) that the OR of studies with a large study population are close to the true average OR, since they have the highest precision, while (2) the OR of studies with low precision (smaller study populations) should, based on chance, be spread evenly on both sides of the average OR. If this is not the case, there is a sign of bias in studies reaching publication. Funnel-plot analysis of the studies in our meta- analysis revealed that the low precision studies reported significantly more often a negative