Sara van den Berg

56 Chapter 3 No restrictions were placed on study design or publication date. Only human studies with in vivo influenza vaccination were included. No restriction was placed on the age of the study population. Studies on primary CMV infection, CMV disease or immunocompromised participants were not included, because the immune system is expected to operate differently in those cases. Data extraction and risk of bias assessment Data collected from the studies included study design, study population, the type of influenza vaccine and the reported outcomes. These data were extracted via a data-extraction form, which was developed by KW based on the Cochrane Data Extraction and Assessment Template [33]. The form was pre-tested on several articles by KW and MK and refined accordingly. The final form can be found in Supplementary Table 2 . If influenza vaccine response outcomes of several studies were reported in one article, the studies were assessed as separate studies. The quality of each individual study was investigated by assessment of the risk of bias based on the Newcastle Ottawa scale for cohort studies [34]. According to these guidelines, studies were awarded with “stars” for high quality choices in three categories: “selection of cohorts” (max 4*), “comparability of cohorts” (max 2*) and “assessment of outcome” (max 3*). Based on all the acquired information, a study could acquire a maximum of 9 stars and the overall quality of the study was rated as high (+) (≥8 stars), intermediate (+/-) (7 stars) or low (-) (≤ 6 stars). Data analysis: statistical and narrative synthesis As the outcome variables were heterogeneous, a combination of narrative and statistical approaches to data synthesis was applied. Three influenza antibody outcomes were systematically extracted from the studies ( Figure 1 ), in line with European Medicine Agency (EMA) guidelines for handling of influenza antibody data. Following the EMA guidelines, whenever possible, we separately extracted the outcome variables per influenza strain. Since age is the most important confounder, we also extracted the three outcome variables separately for young and old individuals. Findings were reported per outcome variable. The principal outcome variable that we studied was the influenza-specific geometric mean titer ratio (GMR) pre-/post-vaccination (outcome a) with corresponding 95% confidence interval (CI) in CMV-seropositive and CMV-seronegative participants. Studies that reported this outcome or reported the data required to calculate the outcome were included in a figure per age subgroup. Per study the definition of young and old individuals differed, leading to two age groups in this analysis: young (<65 years of age) and old individuals (>60 years of age). Studies of which corresponding 95% CI could not be extracted or calculated reliably, were summarized in a separate figure, also per age subgroup (young and old adults).