In this study, indoor radon concentration was collected and analyzed for 3 months during Korean winter season. Between October 28, 2015 and March 30, 2016, 518 radon measurements at 263 residences were obtained. The subjects who voluntarily participated in 30 cities and counties were surveyed. Alpha-track detectors (Raduet Model RSV-8, Radosys Ltd., Budapest, Hungary) were used as a passive radon measuring device. The average concentration of radon in the indoor air was calculated from two points within the household. The measurement points were selected from the living room and a bedroom, spaces where residents of a household primarily spend most of their time. The measuring devices were positioned away from household electrical appliances, windows, and sealed drawers. The measurement period was 3 months.

As measurements started in different months for targeted households, we analyzed mean radon concentrations with standard deviations according to installation month: 119.5±21.0 Bq/m³ in October (n=2), 123.7±88.0 Bq/m³ in December (n=292), and 111.8±82.2 Bq/m³ in January (n=224). There were no statistical differences in the mean radon concentrations for each month of installation (p=0.296 by ANOVA, data not shown).

Questionnaires on the residential environment of 263 residences were administered. The questionnaires assessed residential location, type of house, building materials, ventilation, the foundation of the residence, groundwater usage, and information on the position of the radon measuring equipment. According to the Environmental Geographic Information Service (https://eais.me.go.kr) maintained by the Ministry of Environment in Korea, residential locations were converted to greenery ratios and impermeable layer ratios [10,11]. Greenery and impermeable layer ratios reflect the area of green and impermeable layer space in comparison to the administrative area by state/province. Green area corresponds to forest and grassland area; agricultural space, such as rice fields, is not included in green area. Impermeable layer refers to the area covered with pavement or a building, where rainwater does not infiltrate the ground. Based on natural radionuclides in groundwater by the Korea Institute of Geoscience and Mineral Resources and Soil and Groundwater Division, National Institute of Environmental Research, 222-radon groundwater concentrations were assigned to each residence [12-14]. The study protocol with residence questionnaires was reviewed and approved by the Korean Ministry of Environment (Protocol no. 2015001350004).

Indoor radon concentrations are presented as means with standard deviations and geometric means. To consider repeated measures within the same residence, a generalized estimating equation with a linear model was applied according to individual residential environments. To identify factors affecting higher indoor radon concentrations, we divided residences into five groups according to radon concentrations based on indoor radon reference levels as international standards (<74 Bq/m³, 74100 Bq/m³, 100-148 Bq/m³, 148-200 Bq/m³, and ≥200 Bq/m³) and applied a generalized estimating equation based on a multinomial probability distribution with a cumulative probit link function. Spearman’s rho was used to analyze correlations among greenery ratio, impermeable layer ratio, and indoor radon concentrations. Decision tree analysis was conducted to highlight factors and to suggest optimal cut point of each parameter affecting high indoor radon concentrations by exhaustive chisquared automatic interaction detection (CHAID), as a growing method. Also, random validation was performed with 80% training and 20% testing sets. To construct an affected model of indoor radon concentrations, structural equation modeling (SEM) as confirmatory multivariate analysis was applied with AMOS 21.0 (SPSS Inc., Chicago, IL, USA). Although the χ² to degrees of freedom ratio has no exact interpretation, it has been suggested that values of χ²/df<2 are indicative of an acceptable fit for a hypothetical model [15]. SPSS 23.0 (SPSS Inc., Chicago, IL, USA) was used for statistical analysis except SEM.

Table 1 presents radon concentrations according to residential structure, building materials, lifestyles of the resident, and the positioning of measuring equipment. European-style, detached houses showed the highest indoor radon concentrations among all housing types (p=0.026 vs. semi-detached houses [≤661.157 m²]). Residences at ground level showed significantly higher indoor radon concentrations than residences above ground level (p=0.006). Although indoor radon concentrations were relatively lower in residences that used concrete/cement and no gypsum board as building materials, there were no statistical differences therein, compared to homes that did not (p>0.05, Table 1). Daily ventilation reduced mean indoor radon concentrations to 39.7 Bq/m³, remarkably lower than concentrations in homes that were not ventilated once a day (p<0.001). There were no differences in concentrations according to groundwater use (p>0.05). Interestingly, indoor radon concentrations were higher when there was cracks around measuring equipment and when measuring devices were placed in bedrooms (p=0.013 and p= 0.010, respectively).

Table 2 presents odds ratios for factors likely to affect indoor radon concentrations. Higher greenery ratio and lower impermeable layer ratio were associated with a significantly greater risk of a residence having higher indoor radon concentrations (i.e., ≥200 Bq/m³) (p=0.017 and p=0.043, respectively). Also, a residence at ground level, ventilation less than once a day, long-term heating over half a year, a crack around measuring equipment, and bedrooms significantly increased the rates of houses likely to show high indoor radon concentrations, from <74 Bq/ m³ to ≥200 Bq/m³ (72.4% to 89.1% [p=0.003], 43.8% to 71.7% [p<0.001], 16.2% to 31.5% [p=0.031], 32.9% to 50.0% [p=0.022], and 48.6% to 55.4% [p=0.013], respectively), compared to their counterparts. A negative correlation was noted between greenery and impermeable ratio (r_spearman =-0.872, p<0.001, data not shown). Meanwhile, greenery and impermeable layer ratios for each residence were positively correlated with indoor radon concentrations (r_spearman=0.196, p<0.001 and r_spearman=-0.192, p<0.001). Use of groundwater indoors and radon levels in groundwater exhibited no statistical association with higher indoor radon concentrations group (p=0.097 and p=0.604). Among residences that use of groundwater indoors, those with high levels of radon in the groundwater tended to show higher indoor radon concentrations, although statistical significance was lacking (p=0.740).

To clarify factors affecting high indoor radon concentrations, decision tree analysis was performed (Figure 1). Therein, daily ventilation and greenery ratio were identified as significant nodes from which to identify residences with higher indoor radon concentrations by exhaustive CHAID (p<0.001 and p=0.012, respectively). Daily ventilation increased the probability of radon concentrations in homes being <74 Bq/ m³ by 15.5% than ventilation less than once a day. Meanwhile, lack of daily ventilation and having a greenery ratio (≥65%) increased the chances of homes showing radon concentration over 200 Bq/m³ (15.3% vs. daily ventilation and 8.4% vs. lack of daily ventilation and having a greenery ratio [<65%], Figure 1).

Our final multivariate model was constructed using SEM (Figure 2). To predict indoor radon concentrations, several affected factors were identified from the univariate analysis (p<0.05). In total, seven factors were selected and included in the affected model, including foundation type, a crack around measuring equipment, greenery ratio, radon levels in groundwater, long-term heating, number of ventilations in a week, and measurement in a bedroom. Greenery ratio was selected for inclusion in the affected model, because greenery ratio did not reflect agricultural area. Moreover, we found that impermeable layer ratio, greenery ratio, and indoor radon concentration were highly correlated with each other, which implied that radon amounts in the ground underneath a building could be an important factor affecting indoor radon concentrations. Accordingly, we assumed that radon levels in groundwater would be an indirect predictor of radon amounts in the ground beneath a building.

Figure 2 outlines the constructed model, in which rectangles and circles represent observed variables and error terms (e1 to e8), respectively. Single-headed arrows represent the impact of one variable on another, and double-headed arrows represent covariance between two variables. The final model obtained acceptable fit with χ²/df=1.192 (GFI=0.990 and RMSEA=0.019). A crack was positively correlated with residences at ground level and radon levels in groundwater (correlation: r=0.223 and r=0.138, respectively). A residence at ground level and crack were positively associated with greenery ratio (p<0.001 and p<0.001), while radon levels in groundwater were negatively associated with daily ventilation (p=0.001). We noted direct positive relationships for greenery ratio (p<0.001) and inverse relationships for daily ventilation (p<0.001) with indoor radon concentrations. Greenery ratio also showed indirect relationships with long-term heating and ventilation in relation to indoor radon concentrations. Bedrooms held no statistical significance in this model.