Methodological considerations
We acknowledge that the present study integrated data from different systematic reviews and meta-analyses that used non-uniform literature search criteria and different analytical approaches; thus, the potential of missing some studies cannot be excluded. In addition, an inherent limitation of the present umbrella review, as of any review, is that it is liable to become out of date almost as soon as it is finished; of note is that eight systematic reviews and meta-analyses have been published after the end of the literature search up to the submission of the present study . To overcome this limitation, we are planning to develop our work to a living umbrella review. Moreover, summarizing the evidence from large pooled analyses and several large cohort studies on environmental risk factors and childhood cancer was beyond the scope of the present umbrella review, which integrated data only on systematic reviews and meta-analyses. The combination of systematic reviews, meta-analyses, pooled analyses and cohort studies under the scope of a future “umbrella review” is being planned. In addition, though we included all study-specific associations reported in each meta-analysis, we have missed some additional subgroup meta-analyses due to insufficient study-specific estimates. Of note is that age-specific or sex-specific subgroup meta-analyses (i.e. sub-analyses on infant leukemia) were not retrieved due to the lack of study-specific effect estimates. The only available age-specific meta-analyses examined the association of environmental risk factors with early-onset leukemia (<5 years) and CNS tumors (<10 years). Thus, we could not further address at this stage the potentially differential effect of these factors in males and females, or the differential effect of these factors on infant leukemia, which is a distinct disease entity. Moreover, the assessment of the quality of the primary studies included in each meta-analysis was beyond the scope of this study; however, the AMSTAR tool allowed us to assess some trajectories of the quality of primary studies. However, though the fully adjusted meta-analysis-specific effect estimates were used whenever available, some meta-analyses were based on unadjusted primary study-specific estimates; the definition of the fully adjusted model also varied greatly from study to study and from exposure to exposure. Thus, issues of confounding may have hampered the results presented herein. In addition, owing to the rarity of childhood cancer, primary cohort studies represented a minority of this literature with case-control studies accounting for the overwhelming majority of primary studies included in each meta-analysis. Case-control evidence is prone to selection bias and is thus less robust to provide support to the causality of associations. Indeed, sensitivity analyses including only cohort or nested case-control studies provided weak level of evidence about specific exposures and childhood cancer risk. Moreover, though we examined the presence and extent of biases, the statistical tests implemented were not definitive in determining the exact source of these biases . Lastly, some inherent limitations of these tests, i.e., the possibility of false-positive results from the Egger’s test, should be taken account, although the effect magnitude is usually not substantial and the primary studies are not randomized controlled trials of interventions.