Sara van den Berg

67 3 CMV on influenza vaccination: a systematic review than a positive effect of CMV (p=0.019, Figure 6A ), indicating a publication bias. Importantly, influenza vaccine responses were initially investigated based on the assumption that CMV enhances immunosenescence and a positive effect of CMV on influenza antibody responses was only recently considered [28]. As a result, positive associations between CMV and the influenza vaccine response may have remained unpublished. Based on the same two assumptions, the trim and fill method [38] was performed which removes the smaller studies causing the asymmetry, calculates the ‘true center’ of the studies and next replaces the omitted studies and missing ‘counterparts’ around the centre (filling). The trim and fill analysis estimated that five studies that revealed a positive effect of CMV infection on the influenza vaccination response have not been published in the literature ( Figure 6B ). Of note, in addition to the studies in our analysis, two more studies exist, of which the data was not published, but in which it was stated that no difference was observed [26, 47]. Interestingly, when the five hypothetical ‘missing studies’ were included in our analysis, the OR shifted up from 0.65 to 1.0, suggesting that in fact there is no effect of CMV-serostatus on the response to influenza vaccination. Together, these analyses suggest that publication bias underlies the trend to a negative effect of CMV-seropositivity on the response to influenza vaccination reported in the literature ( Figure 6 ). Correlation and regression models of CMV antibody levels and influenza antibody titers suggest a negative effect of CMV infection (outcome c) The potential influence of latent CMV infection on the antibody response to influenza vaccination was also investigated by extracting the association between the CMV antibody level and the influenza antibody titer. Five studies reported this outcome variable, either as a correlation between the CMV antibody level and the influenza antibody titer post-vaccination or by using a regression model, in which additional factors than only CMV antibody level, such as age, were taken into account [25, 42, 47, 55, 56]. The five studies together reported 10 individual records ( Supplementary Figure 4 ) and included younger and older adults (18-97 years) and different influenza strains ( Table 4 ). Unfortunately, different outcomes for associations were reported ( Table 3 ) and raw data could not be extracted from studies to standardize the outcomes to a comparable outcome. Instead, reported associations were tabulated ( Table 4 ). Of note, 6 out of 8 correlation outcomes came from a single study and were based on post-vaccination titers instead of the fold-increase, which can be influenced to a large extend by pre-vaccination titers. For one correlation outcome, influenza antibody data was divided for time elapsed since immunization, which is not a generally accepted method. Also, both CMV-seropositive and CMV-seronegative individuals were included in all associations. For relations between anti- CMV antibody levels and influenza antibody levels, only CMV-seropositive individuals should be included in our opinion, to focus on the height of CMV antibody level as a surrogate marker of reactivation.