von Kernkraftwerken) 2 in the regression analyses of the inverse distance to the nearest nuclear power plant (NPP). He misinterprets numerically similar results that were reported on different scales: the results given on page 31 of our web appendix 2 are incidence rate ratios (IRRs) and not comparable to the KiKK study, which reported the untransformed regression coefficients. The regression coefficient in our study was 0.55 (95% confidence interval (CI) 1.22 to 2.32) in the resident cohort and 0.29 (2.36 to 1.79) in the birth cohort; clearly different from the 1.75 (95% lower confidence bound 0.65) in the German study. Schadelin also sees a discrepancy between our results for the resident cohort (using address of residence at diagnosis and showing a slight decrease of risk closer to NPPs) and the birth cohort (using address at birth and showing a slight increase of risk), but ignores the wide confidence intervals (which both include the null). We agree with Schadelin that the 1/distance model makes strong assumptions about the relation between dis- tance and risk. Due to the sharp increase of the function as distances approach zero, the few cases in close proximity to the NPPs will strongly influence regression parameters. Whereas the assumption that potential effects of radioactive emissions are limited to their immediate proximity is plausible, the precise functional form between distance and cancer incidence is unknown. Korblein pooled our data with results from two studies from Germany and the United Kingdom to suggest that our study 'confirms the excess of leukae- mia' observed in the latter studies. Given the large number of previous studies, the three studies included by Korblein represent a highly selected subset. For example, it is unclear why the French study by Laurier 3 was not included. This study reported a standardized incidence ratio (SIR) for the 0- to 5-km zone around NPPs of 0.96 (95% CI 0.31-2.24). In the UK the most recent figures (20 observed and 16.35 expected cases) published by the Committee on Medical Aspects of Radiation in the Environment (COMARE) 4 differ from those used by Korblein. Of note, COMARE reported a combined SIR of 1.07 (95% CI 0.92-1.26) from a random-effects meta- analysis of 37 estimates from five countries. 4 Korblein rightly points out that the 95% CI of the rate ratio estimate from the main analysis of child- hood leukaemia in 0- to 4-year olds just includes the point estimate (2.19) of the German KiKK study. However, there is in fact considerable disagreement between the two studies. Assuming that the true IRR for leukaemia in 0- to 4-year olds comparing the 0- to 5-km zone with the 415-km zone around NPPs is indeed 2.2 (as estimated in the KIKK study), the probability of observing an IRR of 1.2 (as in our study) is 0.03 (Figure 1) and the power of rejecting the null hypothesis of no association at the 5% significance level is 76%. Chirga argues that estimates of mean annual radiation doses originating from nuclear power plants (NPPs) are uncertain and that actual doses might fluctuate over time. He suggests that short spikes in emissions could increase the incidence of childhood cancer. Whereas we cannot exclude this possibility, we reiterate that our study provides little evidence that the rate of childhood cancer is higher in the proximity of NPPs. The few cases of cancer occur- ring in excess of the expected number of cases among
Read full abstract