Predicting indoor radon-222 concentration

Radon, a cause of lung cancer among miners, is being investigated as a source of lung cancer in the general population due to long-term low-level exposures in residences. Assessment of cumulative residential radon exposure entails measurements in past residences, some of which no longer exist or are not accessible. Estimates of radon concentrations in these missing homes are necessary for analysis of the radon - lung cancer association.;Various approaches have been used by researchers attempting to predict the distribution of radon measurements in homes from specified geological and building characteristics. This study has modelled the set of basement radon measurements of 3788 Connecticut homes with several of these approaches, in addition to a descriptive tree method not previously utilized, and compared their validity on a random subset of homes not used in model construction.;Each geographical and geological variable was more predictive of radon concentration than any of the housing characteristics. The single variable which explained the largest fraction of the variability in radon readings was the mean radon concentration for the zipcode area in which the house was located (R;Multiple regression models, both additive and multiplicative, were not able to explain more than 22% of the variation in radon readings. Variables found to be significant in these models were zipcode mean, residential radon mean of bedrock unit, building age, type of foundation walls, type of water supply, aeroradioactivity reading, and lithology of the bedrock.;A site potential index, based upon a classification of the bedrock underlying the house, was a better predictor of indoor radon level than other single geological variables, yet only explained 8% of the radon variability. Site-specific indices for predicting indoor radon concentration, which included housing characteristics, were able to explain only slightly more variation (12-15%).;In order to separate houses into subgroups having similar geological and housing characteristics, a descriptive tree design was employed. The mean indoor radon concentration for houses in each terminal branch was used to predict a radon reading for a house with similar characteristics. These models explained up to 22% of the radon variation.;The prediction for a single house in the validation set varied as much as a factor of 2.38 due to error. The utility of these models for supplying estimates of missing data in epidemiological studies is very limited. These limitations apply to a lesser extent to the prediction of radon potential in a region. The ability to discover areas with a higher proportion of houses having elevated radon levels is improving with the use of more comprehensive geographic information systems.