Selection bias in case-control studies on household exposure to pesticides and childhood acute leukemia.

The goal of this study was to investigate the potential for selection bias in published case-control studies on household exposure to pesticides and childhood acute leukemia; most studies have reported positive findings. Items to evaluate the potential for selection bias were first developed. They focused on the source populations that gave rise to cases and controls, the probabilistic selection of subjects from the source, and the losses of the subjects actually selected. A quantitative assessment of bias was also carried out. Potential sources of selection bias were found in all the studies, but none of them were observed across all the studies. Main sources of potential bias were a non-concurrent selection of controls with respect to cases, the use of control diagnoses possibly caused by pesticide exposure in hospital-based studies, and non-participation of selected eligible subjects. A quantitative assessment of bias concluded that non-participation alone could not explain the reported positive associations. We conclude that overall, selection bias, as a likely source of bias in these studies, does not seem to explain their positive findings. Our analysis provides arguments strengthening the conclusions on associations reported in earlier studies.