Statistical Analysis of Geographical Features of Lung Cancer Mortality in Japan

Transcription

1 Jpn J Clin Oncol 000;30(1) Statistical Analysis of Geographical Features of Lung Cancer Mortality in Japan Hiromi Kawasaki 1, Kenichi Satoh, Teruyuki Nakayama 3, Naohito Yamaguchi 4 and Megu Ohtaki 1 Department of Nursing, Hiroshima Prefectural College of Health Sciences, Mihara, Hiroshima, Department of Environmetrics and Biometrics, Research Institute for Radiation Biology and Medicine, 3 Department of Health Physics, Japan Atomic Energy Research Institute, Tokai, Ibaraki and 4 Cancer Information and Epidemiology Division, National Cancer Center Research Institute, Tokyo, Japan Received April 0, 000; accepted October 1, 000 Background: There is concern about the increase in death rates from lung cancer and that it may become the leading cause of death in the near future. Considerable geographical variations exist in lung cancer mortality in Japan. It seems that the difference cannot be explained by smoking rates, although no data are available to prove this. The evidence implies some unknown risk factors in the development of lung cancer. As an alternative approach, we used geographical or demographic information according to municipality. To explore other factors in lung cancer development, the geographical features of trends in lung cancer mortality in Japan were examined using long-term data. Methods: We summarized the 0-year municipality-specific standardized mortality ratios (SMRs) by fitting a straight line to the scatter plot of the SMR versus calendar year for each municipality, resulting in an average and a rate of change of SMR. Using the average or rate of change in the SMR as the response variable, we carried out a multiple linear regression analysis for all of Japan and a non-parametric regression analysis for the distance from the municipal office to the nearest coastline, using the ACE algorithm. Results: The average SMR decreased with distance from the coastline and increased with increasing population size of the municipality. The average SMR was high at or near the coastline irrespective of the direction of the nearest coastline. Conclusions: It was considered that something related to coastline other than population size may be associated with the development of lung cancer. Key words: lung cancer mean risk level multiple regression analysis municipality-specific data rate of change standardized mortality ratio INTRODUCTION Lung cancer is the leading cause of death among males in Japan (1) and it may become the leading cause for females in the near future (). It is said that smoking is the major risk factor of lung cancer, therefore prevention is sought by promoting non-smoking (3). To decrease the lung cancer mortality, periodic check-ups and improvement of individual lifestyle including prohibition of smoking are recommended (3). It is also reported that considerable geographical variations exist in lung cancer mortality in Japan: for instance, it is high in Osaka and Okinawa prefectures and low in Nagano prefecture (4,5). It seems that the difference cannot be explained by smoking rates, For reprints and all corespondence: Hiromi Kawasaki, Hiroshima Prefectural College of Health Sciences, 1 1, Gakuen-machi, Mihara , Japan although no data are available to prove this. The evidence implies that some unknown factors associated with geographical features may be related to the risk of developing lung cancer. To make reference information useful for preventing lung cancer deaths, we examined the regression analysis of its geographical differences by using the municipality-specific standardized mortality ratio (SMR) as the response variable and demographic/geographical data as the explanatory variables. MATERIALS Our study was based on two sets of data for 3331 municipalities, which were provided by the Japanese government. One was the number of deaths due to lung cancer obtained from the population-based registry in Japan between 1975 and The other was the population by age and gender, which was obtained from the five national censuses carried out in 1975, 000 Foundation for Promotion of Cancer Research

2 558 Geographical features of lung cancer mortality 1980, 1985, 1990 and To reduce the possibility of misdiagnosed lung cancer, individuals under 40 or over 75 years of age were eliminated from this study. METHODS AND RESULTS SUMMARIZATION OF SMRS The analysis consisted of two steps. First, we calculated the SMR by gender separately by municipality for each year between 1975 and Second, for each municipality, a straight line was fitted to the scatter plot of SMR versus calendar year by applying the following simple regression model. Let y 1i,,y 0i be the logarithmic values of the year-specific SMRs of lung cancer for the ith municipality. We assumed here that y ji = > 0i + > 1i t j + A ji, j =1,,0, i = 1,, 3331 (1) where t j is centered time (i.e. t j = j 10.5). Then the leastsquares estimates of > 0i and > 1i were calculated. Thus, we summarized the information for 0 SMRs into the two indices >ˆ 0i and >ˆ 1i, the intercept and the slope of the fitted straight line. These are the mean risk level and mean rate of change of logarithmic SMRs during the 0-year period. That is, the mean of SMRs for each municipality, SMR i, is given by taking the exponential value of >ˆ 0i. We also visualized the geographical features by mapping the estimated mean of SMRs and slopes. Based on the distributions of the two indices on the map over the period, we selected four environmental factors which may influence lung cancer mortality: the distance from the coastline, the population in logarithmic form represented by the size of the municipality, the longitude and the latitude. The relation between the each index, >ˆ 0, >ˆ 1 and the selected factors was examined by multiple regression and nonparametric regression. GEOGRAPHICAL DISTRIBUTION OF AVERAGE SMRS We carried out multiple linear regression analysis to clarify the relationship between mean risk level and two environmental factors: distance from coastline and population size. The adopted regression model for mean risk level for each municipality is given by E[log( )] = = 0 + = ( j) 1 p ji + = ( j) q ji + = ( SMR j) i 3 I ji () where E denotes the operation of taking the statistical expectation, p i denotes the distance from the municipal office to the nearest coastline, q i denotes the logarithm of the population size in 1980 and I ji is an indicator variable defined by 1, ifmunicipality i \belongs \to r j I ji = (3) 0, else where r j s are geographical regions (see Table 1). The regression coefficients = (1) 1, = () 1, = (1), = (), = (1) 3,, = (8) 3 were estimated by the weighted least-squares method 8 using the population size as weight. Table 1 shows the estimated coefficients and their 95% coefficient bounds. When municipalities <5 km from the nearest coastline were compared with the others, =ˆ 1 (1) was positive (p < 0.01). The municipalities with population > had positive =ˆ () (p <0.01). The geographical regions with highest risk of local blocks were Hokkaido and Kinki. It was noted that the mean risk level decreased with the distance from the coastline and increased with increasing population. The regression coefficients differed by geographical region. The coefficient of determination, R, was for males and for females. In order to consider in more detail the relationship between the mean risk level and the explanatory variables, we carried out a non-parametric regression analysis by direction of the nearest coastline. The explanatory variables were the distance from the municipal office to the nearest coastline (x 1 ), the natural logarithm of the population size (x ), the longitude (x 3 )and thelatitude(x 4 ). The nonparametric regression model can be expressed as E[log( SMR i )] = B 1 (x 1i )+B (x i )+B 3 (x 3i )+B 4 (x 4i ) (4) The component functions B j wereestimatedbytheacealgorithm (6). Fig. 1 shows the estimated components Bˆ 1 by four directions of the nearest coastline, north, east, south and west, representing the effects of the distance from the municipal office to the coastline, which were adjusted for the effect of x, x 3 and x 4, respectively. The function Bˆ 1 (x 1 ) decreased with x 1 monotonically irrespective of the direction. GEOGRAPHICAL DISTRIBUTION OF RATE OF CHANGE IN SMRS The relationship between the rate of change in SMR and the factors was examined as in the previous section. The model, in which the rate of change in SMR is denoted RATE, is expressed as E(RATE) =C 0 + C 1 ( j) p ji + C ( j) q ji + C 3 ( j) I ji (5) Again, by the weighted least-squares method with using the population size as weight, the coefficients C 1 (1), C 1 (), C (1), C (), C 3 (1),,C 3 (8) were estimated. The result was different between males and females. For males, the regression coefficient was positive for municipalities >50 km from the coastline (p < 0.05). The municipalities with population <0 000 have a positive regression coefficient (p < 0.01). The coefficient of determination (R ) was For females, the regression coefficient was negative for municipalities >50 km from the coastline (p < 0.01). The municipalities with populations <0 000 and > had a negative regression coefficient (p < 0.01). Hokkaido and Tokai had significant positive regression coefficients (p < 0.05). The municipalities which showed the highest rates of change were those with small population size for males and medium population size for females. In a way similar to the nonparametric regression analysis for the mean risk level, the rate of change was also analyzed with 8

3 Jpn J Clin Oncol 000;30(1) 559 Table 1. Estimated regression coefficients for averaged SMR Gender Variable name Estimate a 95% C.I. Males Constant =ˆ ** (0.6615, ) Distance from coastline: <5 km =ˆ 1 (1) ** (0.057, ) 50 km =ˆ 1 () ** ( , ) Population size (1000): <0 =ˆ (1) 0.09 ** ( , ) 300 =ˆ () ** (0.0118, 0.056) Geographical region: Hokkaido =ˆ 3 (1) * (0.0033, ) West-Tohoku =ˆ 3 () ( 0.083, 0.008) Kanto =ˆ 3 (3) ** ( , 0.009) Tokai =ˆ 3 (4) ** ( , 0.090) Kinki =ˆ 3 (5) ** (0.048, 0.07) Chugoku =ˆ 3 (6) * ( , ) Shikoku =ˆ 3 (7) ** ( , 0.008) Kyushu =ˆ 3 (8) * (0.0006, ) Î =0.081 R = Females Constant =ˆ ** (0.67, 0.667) Distance from coastline: <5 km =ˆ 1 (1) ** (0.05, ) 50 km =ˆ 1 () ** ( , ) Population size (1000): <0 =ˆ (1) ** ( 0.405, 0.0) 300 =ˆ () ** (0.0478, ) Geographical region: Hokkaido =ˆ 3 (1) ** (0.058, ) West-Tohoku =ˆ 3 () ** ( , ) Kanto =ˆ 3 (3) ( , ) Tokai =ˆ 3 (4) ** ( 0.059, 0.014) Kinki =ˆ 3 (5) ** (0.0455, 0.084) Chugoku =ˆ 3 (6) * ( , ) Shikoku =ˆ 3 (7) ** ( 0.073, 0.015) Kyushu =ˆ 3 (8) ( , ) Î =0.107 R = a **p < 0.01; *p <0.05. regard to distance from the nearest coastline. From this analysis no noteworthy feature was found. DISCUSSION We found from this study that significant relationships existed among the two indices (average and rate of SMR) and the two environmental factors: distance from the coastline and population size. The relationship between the mean risk level and the geographical factors requires clarification. Some studies found that regional or indoor air pollution was one of the most important risk factors for the development of lung cancer (7,8). Our results indicate that the areas with the highest mean risk level were located in large cities and also that the coastal areas had

4 560 Geographical features of lung cancer mortality Figure 1. Estimated profile of SMR as a smooth function of distance from the nearest coastline. higher mean risk levels of lung cancer. There were other highrisk areas on the coasts of the Sea of Japan in our special maps, although there are no large cities or heavy industrial areas on the coast of the Sea of Japan. The risks for the municipalities at or near the coastline were high irrespective of the direction to the sea. It was considered that something related to coastline other than population size may be associated with one of the causes of lung cancer. In fact, it is well known that Okinawa prefecture, consisting of many small islands, has a high risk of lung cancer mortality, whereas the mortality from cancer of other sites is very low there. The differences in the coefficient of determination between males and females and the differences in the risks by area may be explained in part by differences in rates of smoking, the major known cause of lung cancer (9,10). However, no published data on rates of smoking in each municipality are available yet. Further studies on the relationship between municipality-specific life-style, including rate of smoking, and the development of lung cancer are needed. The map of the rate of change in SMRs indicates that the municipalities with increasing lung cancer mortality were located mainly in the suburbs, particularly the inland areas. That was shown most clearly for males. Patients must be exposed to the risk factors for a long time prior to developing clinically detectable chronic diseases such as lung cancer (3). Watanabe pointed out that at least 0 years was necessary for dioxin, one carcinogen, to enhance the risk of lung cancer (11). The duration of environmental exposure may be important for lung cancer induction. The relationship between the increasing rate of lung cancer mortality and the factors of distance from the nearest coastline and population size were not clear, because of the very small coefficient of determination for the rate of change. The graphical display of spatial trends and time trends in the distribution of disease mortality/incidence allows us to examine its geographical features, such as clusters of high- or lowrisk areas. The long-term rate of change of disease rate should be examined further, because the characteristic factors within each town, the conditions related to its location, may be important in the development of lung cancer. CONCLUSION It is suggested from our analysis that distance from the nearest coastline is one of the risk factors for lung cancer development. Our findings emphasize that high risks of lung cancer are not observed only in large cities but also in rural areas along the coastline. Acknowledgments The authors thank Dr J. B. Cologne for valuable comments. This study was supported in part by a Grant-in-Aid from the Ministry of Health and Welfare of Japan. References 1. Health and Welfare Statistics Association. J Health Welfare Stat 1998;45:5 3 (in Japanese).. Tsukuma H, Kitagawa T, Hanai A, Fujimoto I, Kuroishi T, Tominaga S. Incidence of cancer predictions in Japan up to the year 015. Gan No Rinsho 199;38:1 10 (in Japanese). 3. Sobue T. Epidemiology of lung cancer and prevention strategy in Japan. Nippon Eiseigaku Zasshi 1996;51:641 7 (in Japanese). 4. Fujimoto I, Hanai A, Oshima A, Hiyama T, Tsukuma H, Murakami R, et al. Cancer incidence and mortality in Osaka Tokyo: Shinohara 1993; Sobue T, Tsukuma H, Oshima A, Genka K, Tamori H, Nishizawa N, et al. Lung cancer incidence rates by histologic type in high- and low-risk areas; a population-based study in Osaka, Okinawa and Saku Nagano, Japan. J Epidemiol 1999;9: Breiman L, Friedman J. Estimating optimal transformations for multiple regression and correlation. J Am Stat Assoc 1985;80: Yokoyama E. Assessment of air pollution health effects on respiratory organs. Nippon Eiseigaku Zasshi 199;47: (in Japanese). 8. Morita M. Exposure level to dioxins and furans. Gan No Rinsho 1998;44: (in Japanese). 9. Hirohata T. Cancer and lifestyle: new findings and future prospects. Gan No Rinsho 1998;44:5 9 (in Japanese).

5 Jpn J Clin Oncol 000;30(1) Sobue T, Suzuki T, Fujimoto I, Matsuda M, Doi O, Mori T, et al. Case-control study for lung cancer and cigarette smoking in Osaka, Japan: comparison with the results from Western Europe. Jpn J Cancer Res 1994;85: Watanabe S. Carcinogenicity of dioxin. Soshiki Baiyo Kogaku 1998;4: 5 6 (in Japanese).