Principal findings
The present large-scale umbrella review explored the strength and
validity of evidence in 509 meta-analyses of environmental exposures and
risk of developing 10 different types of childhood cancer. None of the
509 meta-analyses were supported by strong evidence, whereas only 2.4%
of the associations were considered to have highly suggestive evidence
inferred by strongly statistically significant results. These
associations were mainly confined to increased risk of overall leukemia,
especially ALL, in relation to high birthweight, paternal smoking and
exposure to pesticides, particularly insecticides, as well as decreased
risk of ALL related to maternal vitamin supplementation and
breastfeeding (>6 months). There was also highly suggestive
evidence for increased risk of CNS tumors, and especially astrocytoma in
relation to high birthweight, as well as increased risk of testicular
cancer in relation to testicular microlithiasis and isolated
cryptorchidism.
Despite the abundance of published evidence on environmental exposures
and childhood cancer risk, 40% of the included meta-analyses reported a
nominally statistically significant summary effect estimate. Moreover,
though the reported associations seem biologically plausible, our study
showed that inherent biases might overestimate the suggested
associations, which is consistent with the results of previous umbrella
reviews on cancer epidemiology . When decreasing the threshold atp <10-6, only 4% of the summary
random effects estimates remained statistically significant.
Considerable or substantial heterogeneity (>50%) was
reported in 26% of meta-analyses, albeit the 95% prediction intervals
which further account for heterogeneity included the null value in most
associations (76%). There was indication for small study effects in
13% of the meta-analyses, while 7% of the associations had evidence of
excess significance bias. Lastly, most of the included meta-analyses had
a moderate- and high-quality rating based on the AMSTAR quality
assessment tool.