Andersen Claus E, Raaschou-Nielsen Ole, Andersen Helle Primdal, Lind Morten, Gravesen Peter, Thomsen Birthe L, Ulbak Kaare
Risø National Laboratory, DK-4000 Roskilde, Denmark.
Radiat Prot Dosimetry. 2007;123(1):83-94. doi: 10.1093/rpd/ncl082. Epub 2006 Jul 25.
A linear regression model has been developed for the prediction of indoor (222)Rn in Danish houses. The model provides proxy radon concentrations for about 21,000 houses in a Danish case-control study on the possible association between residential radon and childhood cancer (primarily leukaemia). The model was calibrated against radon measurements in 3116 houses. An independent dataset with 788 house measurements was used for model performance assessment. The model includes nine explanatory variables, of which the most important ones are house type and geology. All explanatory variables are available from central databases. The model was fitted to log-transformed radon concentrations and it has an R(2) of 40%. The uncertainty associated with individual predictions of (untransformed) radon concentrations is about a factor of 2.0 (one standard deviation). The comparison with the independent test data shows that the model makes sound predictions and that errors of radon predictions are only weakly correlated with the estimates themselves (R(2) = 10%).
已开发出一种线性回归模型，用于预测丹麦房屋内的室内氡气（²²²Rn）。在一项关于住宅氡气与儿童癌症（主要是白血病）之间可能关联的丹麦病例对照研究中，该模型为约21,000所房屋提供了氡气浓度的代理值。该模型是根据3116所房屋的氡气测量数据进行校准的。一个包含788所房屋测量数据的独立数据集用于模型性能评估。该模型包括九个解释变量，其中最重要的是房屋类型和地质情况。所有解释变量均可从中央数据库获取。该模型拟合于对数转换后的氡气浓度，其决定系数R²为40%。与（未转换的）氡气浓度的个体预测相关的不确定性约为2.0倍（一个标准差）。与独立测试数据的比较表明，该模型做出了合理的预测，并且氡气预测误差与估计值本身的相关性较弱（R² = 10%）。
