A Simulation Study of Lognormal Measurement Error Effect −Discrimination Problem of Radon and Thoron−

Several case-control studies have indicated an increased risk of lung cancer linked to indoor radon exposure. For a precise evaluation of radon-related lung cancer risk, however, contribution of thoron should be considered. There are a lot of studies in which passive type radon detectors are used without thoron discrimination techniques. Therefore, these passive type radon detectors may be strongly affected to the presence of thoron in case they are installed near a wall or floor as potential thoron sources. The thoron effect we consider here is an increase of radon signal in the radon detectors without the discrimination technique. This problem is classified as a part of measurement error problem in statistical literatures and the thoron is considered as a possible source of measurement error. In general, concentrations of radon and thoron follow a lognormal distribution. The effects of measurement error following normal distribution have been studied well, but there are few studies on measurement error following non-normal distribution. In order to evaluate the effect of measurement error due to thoron, we conducted a simulation study. We assumed a case-control study of lung cancer and indoor radon with hypothetical data where radon concentrations were measured with and without discrimination of thoron concentrations. We also assumed that logistic regression was used, and that the concentrations of radon and thoron were correlated, following lognormal distribution. The thoron disturbance in radon measurement resulted in an approximately 90% downward bias in the effect of radon, and this bias was almost constant when the parameters were varied. The downward bias was consistent with results from previous studies taking measurement error following normal distribution into account. It was confirmed that the effect of lognormal measurement error is concordant with the normal measurement error in this case.