UC Berkeley

This dissertation examined several methodological aspects of case control studies potentially influencing the association of environmental exposures with the risk of childhood leukemia. These aspects included the role of SES and racial distribution on subjects' participation, the role of selection bias on the association between EMF and childhood leukemia, as well as reproducibility of self-reports of household pesticide exposure. The research presented in this dissertation was made possible by the Northern California Childhood Leukemia Study, an ongoing population-based case-control study, which commenced in 1995. Results of meta-analyses suggest that in interview based case-control studies, cases were more likely to be non-White than White with an overall OR of 1.37 (95% CI 1.13, 1.67). In contrast, in the record-based studies cases, compared to controls, were less likely to be non-White than White, with an overall OR of 0.81 (95% CI 0.72,0.91). Results also indicated that SES was inversely associated with childhood leukemia while using interview based study design while positive association was observed while using record based study design.

Chapter Four of this dissertation examined the association between childhood leukemia and extremely low frequency of magnetic fields (measured by wire coding) with an emphasis on selection bias. Results indicated that the observed risk estimate depends on the selected control group. The odds ratios (OR) for developing childhood leukemia in the high-current configurations category were 1.43 (95% confidence interval (CI): 0.91, 2.26) compared to the first choice participant controls, while no associations were observed when compared to non-first choice participant controls (OR=1.06, 95% CI: 0.71-1.60) or first choice non-participant controls (OR=1.06, 95% CI: 0.71-1.57). Overall, no association was found between childhood leukemia and EMF as measured by wire configuration codes. Ideal (participating and non-participating) cases assigned to high-current configurations experience a non-significant increased risk of childhood leukemia, when compared to the ideal controls (OR=1.18, 95% CI: 0.85-1.64).

Chapter Five assessed the reproducibility of maternal-reported household use of pesticides and potential differential recall between cases and controls. Results indicated that the Kappa statistics ranged from 0.31 to 0.61 (fair to substantial agreement), with 9 out of the 12 tests indicating moderate agreement. The percent positive agreement ranged from 46-80% and the percent negative agreement from 54-95%. Results indicated that the reliability of self-reported exposures for all pesticide categories using the three reliability measures did not differ markedly for cases and controls as confirmed by bootstrap analysis.