Association of Mesothelioma Deaths With Cumulated Neighborhood Exposures Due to a Large-Scale Asbestos Cement Plant in Amagasaki City, Japan: A Nested Case-control Study

Although a causal relationship between mesothelioma and asbestos exposure is well known, few studies have shown a relationship to non-occupational exposure, including neighborhood exposure, most likely because of the large effect size of occupational exposure. The aim of this study was to quantify the risk of malignant mesothelioma death associated with neighborhood asbestos exposure due to a large-scale asbestos-cement (AC) plant in Amagasaki, Japan, by properly adjusting for occupational exposure. This was a nested case-control study in which a xed population of 143,929 residents who had been living in Amagasaki City between 1975 and 2002 were followed from 2002 to 2015. All 133 cases and 403 matched controls were interviewed about their occupational, domestic, household, and neighborhood asbestos exposures. Odds ratios (ORs) for mesothelioma death associated with neighborhood exposure were estimated by a conditional logistic-regression model that adjusted for other asbestos exposures. We adopted cumulative indices that considered residence-specic asbestos (crocidolite) concentrations and durations during the potential exposure period of 1957-1975 to evaluate individual neighborhood exposures. neighborhood


Background
Although the causal relationship between mesothelioma and asbestos exposure is well known, most of the convincing evidence has been obtained in occupational contexts [1,2], and few studies have shown whether the environmental exposures, including neighborhood, domestic, and household exposure, would quantitatively elevate the risk of mesothelioma death [3][4][5]. In Amagasaki City on 27 June 2005, a newspaper report documented ve cases of mesothelioma among residents around a former large-scale asbestos-cement (AC) plant (Kubota Kanzaki Factory), which was considered the largest asbestos consumer in Japan [6]. From 1957 to 1975, a large amount of asbestos (annual average per year: 4,670 tons of crocidolite and 4,600 tons of chrysotile), corresponding to approximately 5 to 10% of all asbestos imported to Japan, was used in the AC plant, and no other plants that used more than 10 tons of crocidolite per year were identi ed as asbestos-related companies in Amagasaki City [7,8]. By using spatial epidemiological mapping at the municipality level across Japan, researchers revealed distinctive geographical concentrations of elevated risk of mesothelioma death in several cities including Amagasaki City, especially in women during the latest observation period between 2010 and 2014 [9].
In our previous study, we showed that standardized mortality ratios (SMRs) for malignant mesothelioma compared to nation-wide mesothelioma mortalities were 6.75 for males and 14.99 for females in Japan between 2002 and 2015 [10]. Previously, remarkable increases in SMRs had been demonstrated for malignant mesothelioma in a clearly dose-dependent manner that were associated with simulated relative concentrations of airborne asbestos due to an AC plant [11]. However, SMRs in that study, especially in male subjects, were considered to be underestimated because they excluded all asbestosrelated workers from the population eligible for analyses. In general, an effect size of occupational exposure is quite large compared to non-occupational exposures, including neighborhood exposure, so studies should carefully consider ways to control for the effect of occupational exposure as a confounder, and one way is to use a case-control study design for proper adjustment. The aim of this study was to elucidate the quantitative association between neighborhood exposure and mesothelioma deaths due to a large-scale AC plant (Kubota Kanzaki Factory) in Amagasaki City by more e ciently controlling for occupational exposure.

Methods
We conducted a population-based nested case-control study in Amagasaki City. The study population, who had continuously been living in the city before 1975 and at least until the beginning of 2002, were xed as the Amagasaki City long-term resident cohort, and were followed up from 1 January 2002 to 31 December 2015 for death and outmigration status using the Basic Resident Records. The cohort consisted of 143,929 residents aged 40 years or more at baseline, as described in our previous study [10].
A ow diagram of the study is depicted in Fig. 1.
Since 2010, the Amagasaki City Government has been conducting interviews with bereaved families of those who died of mesothelioma, based on the list of mesothelioma deaths provided by vital statistics. In municipal administrative reports, a total of 379 mesothelioma deaths were recorded during the period between 2002 and 2015, however, 110 of these diseases were not identi ed through the Basic Resident Records due to outmigration or absence of bereaved family. Furthermore, of the 269 identi ed mesothelioma deaths, 102 bereaved families refused to be interviewed or could not be contacted. Of the 167 mesothelioma deaths (62.1%) for which bereaved families accepted to be interviewed, 133 mesothelioma deaths met inclusion criteria of the Amagasaki City long-term resident cohort and were thus de ned as the cases in our study (Fig. 1).
Controls were selected from the Amagasaki City long-term resident cohort, were matched individually to each case by sex and age (birth date ± 12 months), and were alive on the exact date of death of their corresponding cases. In order to obtain three controls per case to have a 1:40 case-control ratio, we randomly selected 5,320 residents for controls. Of these, addresses of 4,191 residents and their closest family members were con rmed through the Basic Resident Records and were eligible for assessment ( Fig. 1). Letters for contact availability were sent to these 4,191 residents through the local government of Amagasaki City, and 2,720 residents (64.9%) responded. Of these 2,720 residents, 1,021 were eligible to receive postal questionnaires regarding approval for participation in the study. From these 1,021 eligible residents, consent was obtained for 403 residents (39.5%), who were consequently interviewed for the study. These 403 controls were matched to 133 cases. All interviews for the controls were completed between 2015 and 2017, using the same questionnaire that was applied for cases. Finally, the number of sets and ratios for matching were as follows: 9 sets for 1:1, 22 sets for 1:2, 74 sets for 1:3, 19 sets for 1:4, 5 sets for 1:5, 1 set for 1:6, and 3 sets for 1:7.
Well-trained nurses with expertise in asbestos-related health hazards conducted interviews of all individuals for both cases and controls between 2002 and 2017, using a standardized questionnaire. The questionnaire, which was conducted by either a structured visit or a telephone interview, consisted of the following sections on the relationship of respondent with the index subject (cases and controls themselves): demographic characteristics including smoking history; status of occupational compensation certi cation; lifetime occupational history; residential history during the considerable asbestos exposure period of 1957-1975, when crocidolite was de nitely used in the plant; and other possible environmental asbestos exposures such as domestic or household exposure. Environmental (non-occupational) exposure is generally divided into three sources according to the exposure pathway: neighborhood, domestic, or household. In our study, neighborhood exposure that resulted from living near an emission point of airborne asbestos bers was estimated from individual residential histories. Domestic exposure was estimated in relation to cohabitants occupationally exposed to asbestos in the same house, and household exposure was estimated as it related to any exposure to asbestos-containing materials in the home, such as re-proof sheets and asbestos-sprayed walls and ceilings. Because individuals were considered to have multiple exposure sources, we intended to evaluate four types of exposure separately, according to relevant rating procedures: occupational, domestic, household, and neighborhood.
We collected occupational histories, which included "company name", "address", "employment period", "job description", and "experience of handling directly any asbestos products (None/Yes)". For the assessment of occupational exposure, a subject who had any self-reported experience of direct exposure was classi ed as "De nite" exposure (Table 1). For further assessment, the remaining subjects classi ed as "Non-de nite" exposure were divided into three tiers: "Convincing", "Possible", and "None", which were evaluated for the probability of occupational exposure based on an individual's occupational history obtained from the interview. Three raters, who had many years of experience with judgement of asbestosrelated occupational compensations, assigned probabilities to all the obtained data related to occupational exposure, referring to publicly available materials and relevant documents provided by the Ministry of Health, Labor and Welfare. To gain expert consensus, we took the following steps: 1) each expert assigned the probability of occupational exposure in two rounds, 2) a summary including demographics was provided to experts with reasons of judgements, 3) experts were encouraged to revise their earlier assignments in light of the other members' comments, and 4) the nal assignment was determined by a reliable panel of experts. To con rm the "De nite" group for nal analyses, we excluded 32 cases and 61 controls because they could have had a large effect on the relationship between neighborhood exposure and mesothelioma risk, which was the main interest of this study.
We de ned domestic exposure as the exposure to asbestos bers brought home by workers on their clothing or in their hairs or through living in the same house with occupationally-exposed individuals, and coded it as a dichotomous category (None/Yes) according to the corresponding questionnaires. We de ned household exposure as having exposure to asbestos-containing materials used in home structures (e.g., roofs, insulation) or home improvement products. If there was at least one item in the list of considered asbestos-related products, we coded it as a dichotomous category (None/Yes) according to the corresponding questionnaires.
Focusing on airborne asbestos bers derived from the AC plant, we estimated the relative asbestos concentration (unit, 1/m 3 ) in each 100 m x 100 m grid, using diffusion equations and meteorological conditions at that time. The method assumed that an emission point of airborne asbestos was at the center of the premises of the plant and followed the diffusion equations as previously described in detail [11,12]. Figure 2 depicts the distribution of geographically simulated relative asbestos concentrations across Amagasaki City, which were downscaled into a 10 m x 10 m grid by using spline interpolation. The contour lines show isolines of relative asbestos concentration, ranging from 1 to 10 7 (unit, 1/m 3 ), and the color-coded concentric circles indicate crow-y distances from the center of the premises of the plant. Residential histories in Amagasaki City during the exposure period of 1957-1975 were con rmed for each individual address and duration as precisely as possible based on the interview data and the Basic Resident Records. Geocoding with the Mapple Address Matching Tool (Shobunsya Inc.) was performed to convert the address of each residence into a geographic coordinate, a longitude and latitude pair, which represented a point on the city block called "Gaiku" (61.3% for cases and 79.9% for controls) or in the neighborhood called "Choh-aza" (38.7% for cases and 20.1% for controls). When the geocoding of old address names failed, we referred to detailed city maps to manually convert the old address into the current one. By using the geographic coordinate, each residence was linked to the simulated relative asbestos concentration, as mentioned above. In order to evaluate the cumulated asbestos exposure, taking into account individual residential histories, we calculated the cumulative dose of neighborhood exposures by summing the residence-speci c exposures (unit, year/m 3 ) obtained from the relative asbestos concentration (unit, 1/m 3 ) multiplied by the duration (unit, year) ( Table 3).
Odds ratios (ORs) in relationship to neighborhood exposure (quintiles of cumulative dose of asbestos exposure in the controls) were estimated separately for men and women by conditional logistic regression analysis, using the STATA software (version 14.2/MP; Stata Corp., College Station, TX, USA).
Covariates for adjustment were assessed as three tiers of occupational exposure and dichotomous variables for both domestic and household exposures. We de ned tests with P < 0.05 as statistically signi cant.

Results
Demographics of the 133 cases and 403 matched controls were generally well-balanced, however, a few differences were observed (Table 1). For the matched controls, we interviewed both bereaved families and subjected persons, but the highest proportion of interviews was with subjected persons both for males (69.8%) and females (51.8%). In the male cases, the highest proportion of interviews was with the spouse (65.5%), while in the female cases, the highest proportion of interviews was with the child (73.5%).
Regarding asbestos exposure, the number of subjects with de nite occupational asbestos exposure who had directly experienced handling asbestos products was 26 (31.0%) for cases and 59 (22.5%) for controls in males, and 6 (12.2%) for cases and 2 (1.4%) for controls in females, with a slightly higher proportion among females (P = 0.001, Fisher's exact test) ( Table 1).
After the exclusion of 32 cases and 61 controls with a history of directly handling asbestos products, Table 2 presents the demographic distribution of 101 cases and 342 controls with occupational, domestic, and household asbestos exposure. The proportion of cases that had convincing occupational exposure was 31.0% for males and 4.7% for females, which was much higher than that seen in controls (9.4% for males and 1.4% for females). The proportion of subjects with no occupational exposure was 82.7% for controls and 53.5% for cases in females, and 66.0% for controls and 37.9% for cases in males, which demonstrated that more than half of the cases were occupationally exposed to asbestos regardless of the intensity, especially in males. Overall for domestic and household asbestos exposures, we observed relatively higher proportions in the cases than in the controls.
No substantial differences in the number of residences per person were observed between cases and controls ( Table 2). The characteristics of all residential records in the 101 cases and 342 controls during the potential exposure period of 1957-1975 are presented in Table 3. Although no substantial differences were observed in the duration lived at each single residence, statistically signi cant differences (P < 0.001; Fisher's exact test) between cases and controls were observed for the simulated relative asbestos concentrations. In addition, medians for both the simulated relative asbestos concentration and the residence-speci c exposure (calculated by concentration multiplied by duration at each residence) showed statistically signi cant differences between cases and controls (P < 0.001; Wilcoxson's rank sum test). Moreover, the distribution of residence-speci c exposure was signi cantly different between cases and controls, with the highest proportion observed in the lowest category in controls, and a broad distribution observed in cases. The median distance from the AC plant in the cases was lower than that in the controls, with a statistically signi cant difference (P < 0.001; Wilcoxson's rank sum test).

Discussion
This was a population-based nested case-control study to evaluate the risk of mesothelioma death associated with neighborhood asbestos exposure due to a large-scale AC plant in Amagasaki, Japan. The results demonstrated that ORs in the top quintile were 21.4 for all, 23.7 for males, and 26.0 for females compared to the bottom quintile, suggesting that there were no substantial gender differences in relationship to the magnitude of risk.
In our previous studies, SMRs associated with neighborhood exposure showed a substantial sex difference because the effect of occupational exposure was higher in males than females and not controlled appropriately. We reported SMRs of 6.75 for males and 14.99 for females [10] because the effect of occupational exposure was included in the national mortality rates used as reference rates, which resulted in lower SMR for males than females. In the study by Kurumatani and Kumagai (2008), SMRs were estimated further by excluding mesothelioma cases that had possible occupational exposure from the numerator only and not excluding occupational exposure from the denominator. As a result, SMR was 2.6 for males and 9.9 for females, showing a much greater difference. Such sex differences were variable, depending on the attributable proportion of occupational exposure. Thus, we adopted a nested case-control design, by adequately taking into account the large effect of occupational asbestos exposure on the evaluation for neighborhood asbestos exposure, collecting enough data related to an individual's occupational and non-occupational exposure within the whole cohort population.
A recent pooled analysis of 21 AC worker cohorts in Italy [13] showed a similar tendency in relation to gender difference, with signi cant increases in mortality from all causes (SMR: men 1. . It also showed that the rate ratio (RR) by Poisson regression analyses for pleural and peritoneal malignant neoplasms increased with cumulative exposure ( ber-type weighted index), while with the time since rst exposure it showed an increase in the rst four decades, followed by a plateau in both genders for pleural malignant neoplasm. A population-based case-control study in France [14] demonstrated that ORs for the highest occupational exposure were 13.2 for males and 18.2 for females compared to those never exposed, also indicating no substantial gender differences. Regarding the association with non-occupational exposure, ORs for the highest probability of exposure were 4.6 for males and 7.5 for females compared to those never exposed. Furthermore, the population attributable risk (ARp) for those non-occupational subjects was 20.0% in males and 38.7% in females, suggesting that the overall population-attributable risk of asbestos exposure in females was largely driven by non-occupational exposure, considering the di culty in assessing domestic or environmental exposure.
In order to adjust properly for occupational exposure, we adopted two independent steps, an exclusion of direct occupational exposure and a rating for the reliability of exposure, and consequently 32 cases and 61 controls were excluded from our analyses. There was no doubt that the experience of directly handling asbestos products could have a signi cant impact on the evaluation for non-occupational exposure, however, it could not be used for proper assessment of occupational exposure due to lack of information on exposure intensity. The remaining cases and controls after exclusion were classi ed into three mutually exclusive tiers that were based on the reliability of exposure and contributed to adequate adjustments by using a conditional logistic regression model. We observed that the risk magnitude of occupational asbestos exposure was comparably higher than that of non-occupational exposure, demonstrating that ORs of "Convincing" compared to "None" were 25.4 for all, 40.0 for males, and 6.9 for females to suggest that the sex differences seemed to be fair.
To assess neighborhood exposure, we used cumulative indices of individuals' residence-speci c asbestos exposures, calculated by the relative concentration of airborne asbestos bers multiplied by the duration at each residence during the potential exposure period (1957)(1958)(1959)(1960)(1961)(1962)(1963)(1964)(1965)(1966)(1967)(1968)(1969)(1970)(1971)(1972)(1973)(1974)(1975). Many studies have evaluated neighborhood asbestos exposure by distance or by the simulated relative concentration of airborne asbestos bers based on the nearest residence from the emission point. In order to clarify the effect of distance on the contribution to our cumulative indices, we chose only one residence, which addressed the maximum (highest) value of residence-speci c exposure and compared characteristics of the cases and controls. Table 5 shows the distribution by strati ed levels of the cumulative dose of neighborhood exposure (Levels 1 to 5) and indicates that approximately 90% of both cases and controls in Level 5 were classi ed to the same Level 5 of residence-speci c exposure. Similarly, in Level 4 of the cumulative dose of residence-speci c exposures, over 80% of both cases and controls were classi ed to the same Level 4 of residence-speci c exposure. These ndings suggested that just a single residence could contribute to the evaluation of cumulative neighborhood exposure in most subjects (approximately 80-90%). For the relative concentration and duration, there were no substantial differences in distribution between cases and controls in the same exposure levels. However, in relationship to the nearest residence in which a subject lived for at least one year, we did not observe an obvious association with cumulative exposure levels. Moreover, we observed that 26.5% of cases and 22.1% of controls had lived within 500 m of the AC plant for over a year, and 4.1% of cases and 2.9% of controls had lived more than 2 km from the AC plant for over a year in Level 5, which is supposedly the highest risk group.
Because it is di cult to evaluate various asbestos exposures independently, most studies report risks associated with several combined non-occupational exposures such as neighborhood, domestic, and household. Although few studies of neighborhood exposure have been reported, and there is large heterogeneity among results from those studies, a recent meta-analysis reported an increased risk of mesothelioma death by exposure types, with a summary relative risk estimate (SRRE) of 5.33 (95% CI: 2.53-11.23) from neighborhood exposure, 4.31 (95% CI: 2.58-7.20) from domestic exposure, and 2.41 (95% CI: 1.30-4.48) from household exposure [5]. Other researchers reported a clear increasing trend in relationship to the cumulative exposure index, demonstrating that the highest OR was 23.3 (95% CI: 2.9-186.9), but they did not assess the exclusive estimates that could be associated with neighborhood, domestic, and household exposures [15]. Magnani and colleagues conducted a case-control study in Casale, Italy, and showed a signi cant increased risk of mesothelioma in a population that had only lived in the city, with an OR of 20.6 (95% CI: 6.2-68.6) [16]. In the same city, another study estimated OR in relationship to distance from an AC plant, however, the results were thought to be due to combined exposure of the neighborhood and household [17]. In another study that looked at distance from an AC plant in Bari, Italy, the highest OR ( [19]. However, regardless of the different indicators and the range of categories in those studies, the magnitude of association indicated by OR seems to be fairly consistent with our results. There were some limitations to our study. First, we observed differences in the types of respondents between cases and controls ( Table 1). The highest proportion of respondents among controls was the exposed subject, while the highest proportion of respondents among cases was a spouse or child of the exposed subject. Case interviews were thought to contain critically important information for the Ministry of Environment to make decisions about compensation certi cation, hence, the bereaved families had detailed information about the cases' occupational histories and any other possibilities related to asbestos exposure at that time. In contrast, most families of the controls, even though they were spouses and children, were not well informed about details of the indexed subjects' occupational and residential histories throughout their lifetime. We therefore needed to interview control subjects as well as their families to collect data with the same accuracy as that for the cases.
A second limitation was that the consent rate for interviews of the bereaved families of cases seemed to be low (62.1%). However, the psychological effects on bereaved families due to the miserable epidemic of mesothelioma deaths within a small area in Amagasaki City would also have to be considered. In the municipal administrative reports, there were various reasons for non-consent: 1) unwilling to look back on the past, 2) not well informed about history of the indexed subject, 3) recognized as an industrial compensation, 4) no bereaved families in the city, and 5) other reasons. Although the consent rate was 64.9% among controls when Amagasaki City Government rst asked about contact availability, after we sent advance questionnaires to the controls, the consent rate dropped to 39.5%, which is more indicative of a consent rate likely to be observed in a survey of the general population.
Third, we obtained the list of mesothelioma deaths from Japanese vital statistics, which began to record the primary cause of death according to ICD-10 in 1995, and the cause of death was posted from a death certi cate. Although the mesothelioma diagnosis was not histologically con rmed, the frequency of misclassi cations on the cause of mesothelioma deaths quali ed by a specialist was thought to be negligible. Fourth, the simulated asbestos concentrations were determined from several assumptions, and the relative numbers ranged from 1 to 10 7 (unit, 1/m 3 ), based on a 10 m x 10 m grid resolution.
The nal limitation was that we used cumulative indices of residence-speci c exposures to evaluate neighborhood exposure, however, It might be necessary to consider more speci c working area of daily living, such as walking to school, gardening, and agricultural working.
A recent world-wide trend assessment for excess risk of mesothelioma deaths using a period analysis from 1996 to 2005 demonstrated a signi cant increasing rate in Japan of 3.9% annual percent change, with overall quite large disparities between countries [20]. Similarly, our previous studies in Amagasaki City demonstrated that mesothelioma deaths increased signi cantly in long-term resident cohorts both in males and females from 2002 to 2015, with consistent trends in the three periods from 2002-2006, 2007-2011, and 2012-2015. A recent paper suggested that non-occupational exposure mainly contributed to the overall population attributable risk, especially for females [14]. Although it was only a municipal administrative report in Amagasaki City, a gradual increase in mesothelioma deaths from nonoccupational exposures was observed for males (19% during 2002-2007, 35% during 2008-2015) and females (39% during 2002-2007, 73% during 2008-2015). One possible consideration might be an agecohort effect, in which the risk of mesothelioma remains and never diminishes for the people who were exposed in early life, even after removal of the AC plant. Therefore, we should continue to pay close attention to trends in the risk of mesothelioma death associated with low-dose asbestos exposures including neighborhood exposure.

Conclusions
We demonstrated a dose-dependent increase of mesothelioma death associated with the neighborhood asbestos exposure due to a large-scale AC plant in Japan by using cumulated neighborhood asbestos exposure and adjusting properly for occupational and other asbestos exposures. Our ndings suggest that the risk of mesothelioma death with neighborhood exposure persists and will not be diminished for many years, even though it has been decades since the AC plant closed. Flow diagram of study population enrollment in the Amagasaki City long term resident cohort; cases and matched controls.