Bayesian Mapping of Cancer Mortality in Japan: A Small Area Analysis Megumi Hori1, Eiko Saito1, Tomoki Nakaya2, Kota Katanoda1 1 Division of Cancer Statistics Integration, Center for Cancer Control & Information Services, National Cancer Center, Japan 2 Department of Frontier Sciences for Advanced Environment, Graduate School of Environmental Studies, Tohoku University, Japan Thank you for your introduction. I’m very happy to be able to talk to you. Today, I’d like to talk about the small-area mapping in cancer mortality in Japan using Bayesian method. ★First of all, let me talk about our motivations to study a small-area mapping and apply the Bayesian approach. 1 1
Motivation - 1 Geographic disparities in incidence and mortality of cancer In Japan, Cancer incidence and mortality have large variation across prefectures. Incidence in 2015 Mortality in 2015 All cancer, Males All cancer, Males Please look at these maps. The map on the left shows standardized incidence rates(SIR) by prefectures for males all cancers, And the map on the right shows standardized mortality rate(SMR) by prefectures for males all cancers. A prefecture is similar to a state or a district. What can we know from these maps? As you can see, ★in Japan, cancer incidence and mortality have large variation across prefectures. SIR SMR ≧120 <80 80 - 100 100 - 120 ≧120 <80 80 - 100 100 - 120
Motivation - 1 Monitoring of disparities by small areas The pattern of exposure to cancer risk has differed across smaller units than prefectures. Education, occupation and general lifestyle behaviors(ie, dietary patterns, physical activity and smoking status) may vary by small regions Population-based screening programs have been conducted by municipal authorities(wards, cities, towns, and villages), with government support The number of prefectures and municipalities in Japan, 2015 Prefectures Municipalities No. 47 1902 Pop: Max 13,515 K 903 K Pop: Min 573 K 337 We have already known that there are prefectural disparities in cancer incidence and mortality in Japan. However, we have to solve many more issues. Because, ★the pattern of exposure to cancer risk has differed across smaller units than prefectures. Education, occupation and general lifestyle behaviors (ie, dietary patterns, physical activity and smoking status) may vary by small region. And, in Japan, municipal authorities (wards, cities, towns, villages) have conducted population-based cancer screening programs with government support. This table shows the number of prefectures and municipalities in Japan, in 2015 Japan has 47 prefectures, and All prefectures are further subdivided into municipalities There are about a thousand and nine hundred municipalities in Japan. Prefectures range in population is from 5 hundred thousand to 13milion, while Municipalities Range in population is from 3 hundred to 9 hundred thousand To monitor the small area cancer mortality is important issue and the next step for us.
Motivation - 2 Reliability in small population statistics General issues of small-area estimation extreme risks might be shown in low populated areas significant risks tend to be shown in highly populated areas areas next to each other might show completely opposite risks Bayesian approach has been widely used since the 2000’s and has proved to be robust borrowing information from neighboring areas with a spatial smoothing process enabling to generate more stable and accurate estimates of rates for geographic areas Next, I’d like to talk about the reason for using Bayesian method. When we do the small area estimation, we have to solve the small population problems. For example, extreme risks might be shown in low populated areas. Significant risks tend to be shown in highly populated areas. Areas next to each other might show completely opposite risks. To solve these problems, we applied the Bayesian approach in this study. Bayesian approach has been widely used since the 2000’s In the approach, we borrow information from neighboring areas with a spatial smoothing process. By this process, we can obtain the more stable and accurate estimates of rates for small area.
Aim To promote a better understanding the geographic disparities To estimate the small-area cancer mortality based on Bayesian approach To develop an interactive mapping tool for cancer mortality Considering these reasons, the objectives of our study were to estimate the small-area cancer mortality based on Bayesian approach and to develop an interactive mapping tool for cancer mortality.
Data Sources Using open data and GIS Mortality and population Vital statistics in Japan Between 2008-2012 By municipality By sex (Male, Female) By site (stomach, colon, liver and lung)* *only for mortality Available from: https://www.e-stat.go.jp/stat- search/files?page=1&layout=datalist&toukei=00450013&tstat=00000106368 0&cycle=0&cycle_facet=cycle&second=1&second2=1&tclass1val=0 GIS data and map in Japan ESRI shapefiles data For municipality boundary map Available from: https://www.esrij.com/products/japan-shp/ Here, I’ll show the using data for our study. We used Vital statistics in Japan for Mortality and population. From the data We obtained cancer mortality and population from 2008 to 2012 by municipality, by sex, and by site. For the GIS data and map in Japan, We used Esri shapefiles data. Esri is an international company of geographic information system (GIS) software. In the next few slide, I’ll introduce the method of the estimation and development of mapping tool.
Method-1 Empirical Bayes estimation of SMR Empirical Bayesian(EB) approach with nearest neighbor procedure 𝑠𝑚𝑟 𝑖 𝐸𝐵 = 𝑜 𝑖 + 𝛽 (𝑖) 𝑒 𝑖 + 𝛼 (𝑖) = 𝑤 (𝑖) 𝑠𝑚𝑟 𝑖 +(1− 𝑤 𝑖 ) 𝑠𝑚𝑟 (𝑖) Hyperparameters were estimated by the MLE method. 𝑖 : municipality (𝑖) : 10-nearest neighbors of municipality 𝑖 , 𝑠𝑚𝑟 𝑖 𝐸𝐵 : standardized mortality ratio based on EB approach for 𝑖, 𝑜 𝑖 : observed number of cancer death, 𝑒 𝑖 : expected number of cancer death 𝛼 (𝑖) , 𝛽 (𝑖) : hyperparameters (𝛼>0, 𝛽>0), 𝑤 (𝑖) : weights reflecting the population size of 𝑖 , where 𝑒 𝑖 / 𝑒 𝑖 + 𝛼 (𝑖) 𝑠𝑚𝑟 (𝑖) : standardized mortality ratio for 𝑖 , where 𝛼 (𝑖) / 𝛽 (𝑖) To show the small area cancer mortality and the disparities by small-area, we applied Empirical Bayes estimation of SMR with nearest neighbor procedure. In this method, we borrow the information of 10-nearest neighbors of municipality to get more reliable region-specific estimates. Here, i means municipality, parenthesis i means 10-nearest neighbors of municipality 𝑖 and w means the weights reflecting the population size of each municiparity. In this procedure, when the population size of target region is large, the weight of 10-nearest neighbors is small.
Method-2 Interactive mapping tool for cancer mortality Features of the tool Create and customize municipality level of cancer mortality number, rate, age-adjusted rate, standardized rate by year, site, gender, View graph and report data by area Download the data as a csv file, including GIS file (shapefile ) Show the location of hospitals Here, I’ll show the interactive mapping tool we developed. This slide shows the Features of the Interactive mapping tool for cancer mortality we developed. We developed the mapping tool so that you can Create and customize municipality level of cancer mortality So, we can see the number of death, rate, ASR, and standardized rate by year, site and gender at any area unit Our mapping tool can also View graph and report data by area, Download the data as a csv fail including GIS file and show the location of hospitals. In the next few slides, I’ll be showing you some results of small-area cancer mortality
Results - A1 Empirical Bayes Estimate of SMR: All cancer Cancer mortality varied at the municipality level even within same prefecture. Japan Prefectures Map Please look at these maps. The map to the left shows the standardized mortality ratio by municipality in Hokkaido/Tohoku region for male, all cancers, and the map to the right shows that for females, all cancers. Some prefectures in these areas were known as high cancer mortality area in Japan. However, we can see that the ★cancer mortality varied at the municipality level even within same prefecture from these maps. high low Male Female Hokkaido and Tohoku Region Hokkaido and Tohoku Region
Results - A2 Empirical Bayes Estimate of SMR : Stomach cancer Large spatial variation in stomach cancer mortality. Higher rates were seen in the Western part of Tohoku region. Japan Prefectures Map Next, the map to the left shows the SMR by municipality in Hokkaido/Tohoku region for male, stomach cancers, and the map to the right shows that for females, stomach cancers. As you can see, there are ★large spatial variations in stomach cancer mortality. Higher rates were seen in the Western part of Tohoku region. high low Male Female Hokkaido and Tohoku Region Hokkaido and Tohoku Region
Results - A3 Empirical Bayes Estimate of SMR : Lung cancer SMR of lung cancer varied at the smaller area unit. For female, higher SMRs were seen in urban areas. Japan Prefectures Map Next, the map to the left shows the SMR by municipality in Chugoku region for male, lung cancers, and the map to the right shows that for females, lung cancers. ★SMR of lung cancer varied at the smaller area unit. Especially, For female, higher SMRs were seen in urban areas. high low Male Female Chugoku Region Chugoku Region
Results - A4 Empirical Bayes Estimate of SMR : Liver cancer Higher rates were seen in the Western Japan. Especially, SMRs were high on the sea side. Japan Prefectures Map Next, the map to the left shows the SMR by municipality in Kinki region for male, Liver cancers, and the map to the right shows that for females, liver cancers. This region were known as high liver cancer mortality area in Japan. When we estimated the SMR by municipalities, ★Especially SMRs were high on the sea side even within the same prefecture. In the next slide, I’ll introduce Interactive mapping tool for cancer mortality high low Male Female Kinki Region Kinki Region
Results – B1 Interactive mapping tool for cancer mortality Prefecture-level map Report data by prefecture Bar graph by prefecture This is a Prefecture-level map of standardized cancer mortality. You can create a report that contains detailed demographic information about a pointed prefecture on a map. And We can see the bar graph by prefecture on prefecture level map.
Results – B2 Interactive mapping tool for cancer mortality Municipality-level map Location of designated cancer care hospitals Bar graph by municipality Next, This is a munisiparity-level map of standardized cancer mortality. Here, we can see the Location of designated cancer care hospitals as point. And We can see the bar graph by municipality on municiparity level map.
Results – B3 Interactive mapping tool for cancer mortality Municipality-level map Moreover, our mapping tool allows you to change the base map. Our tool includes several types of basemaps, including aerial imagery, terrain, streets, and topographic data. Changing the base map
Enter your footer text here Discussion We showed small-area variation in cancer mortality by mapping visualization based on Bayesian estimation. The Japanese government introduced a national cancer registry (NCR) in 2016 and started providing the data in 2018 NCR enables us to obtain highly reliable incidence data We will be able to develop an interactive mapping tool that show small-area cancer incidence The empirical Bayesian mapping will be useful for planning more focused cancer control policies. In this study, we showed small-area variation in cancer mortality by mapping visualization based on Bayesian estimation. The Japanese government introduced a national cancer registry in 2016 and started providing the data in 2018 So today, We can obtain highly reliable incidence data and will be able to develop an interactive mapping tool for small-area cancer incidence. Our mapping tool will be useful for planning more focused cancer control policies. Enter your footer text here
Thank you for your attention! E-mail: mhori@ncc.go.jp 【結果全般】 Cities that have large population would show little differences between observed mortality and Bayesian estimated mortality. 【罹患の地域差の理由】 The high incidence of hepatitis C of the western region could contribute to high mortality rate of liver cancer. The reason for high incidence of hepatitis C is The Tohoku district has the highest salt consumption, also sees the highest incidence of gastric cancer. 機能の拡張について I still cannot prepare that now. In the future, we want to try to add the functions as viewing infection rates, screening rate, and proportion of early detection and other useful information for cancer control to our mapping tool. Drug abuse Bay 17 17