Absence of evidence for threshold departures from linear-quadratic curvature in the Japanese A-bomb cancer incidence and mortality data
The recently released data on cancer incidence and mortality in the Japanese A-bomb survivors are analysed using a variety of relative risk models which take account of errors in estimates of dose to assess the dose-response at low doses. If a relative risk model with a threshold (the dose-response being assumed linear above the threshold) is fitted to the solid cancer incidence or mortality data, a threshold of more than about 0.25 Sv is inconsistent with both the incidence and mortality data, whereas this data is consistent with there being no threshold. If a relative risk model with a threshold (the dose-response being assumed linear above the threshold) is fitted to the leukaemia incidence or mortality data, a threshold of more than about 0.4 Sv is statistically inconsistent with the data. In contrast to the solid cancer data, the best estimate for the threshold level in the leukaemia incidence data is statistically significantly different from zero, even when allowance is made for a possible quadratic term in the dose-response, albeit at borderline levels of statistical significance (p = 0.04). However, the best estimate of the threshold is not statistically significantly different from zero in the mortality data (p = 0.16). The statistical significance of the threshold effect for leukaemia incidence is lost if the assumed geometric standard deviation of the random error in the dose estimates is increased from (the central estimate of) 35% to 45%. Given the similarity of the leukaemia mortality and incidence datasets, the reasons for this difference are not entirely clear, but the most likely explanation is the finer disaggregation of dose groups in the publicly available version of the leukaemia mortality data compared with the incidence data.
Keywords: A-bomb survivors, threshold, leukaemia, dosimetric error, low radiation