Childhood cancer incidence and survival in Japan and England: A population‐based study (1993‐2010)

The present study aimed to compare cancer incidence and trends in survival for children diagnosed in Japan and England, using population‐based cancer registry data. The analysis was based on 5192 children with cancer (age 0‐14 years) from 6 prefectural cancer registries in Japan and 21 295 children diagnosed in England during 1993‐2010. Differences in incidence rates between the 2 countries were measured with Poisson regression models. Overall survival was estimated using the Kaplan–Meier method. Incidence rates for Hodgkin lymphoma, renal tumors and Ewing sarcomas in England were more than twice as high as those in Japan. Incidence of germ cell tumors, hepatic tumors, neuroblastoma and acute myeloid leukemia (AML) was higher in Japan than in England. Incidence of all cancers combined decreased in Japan throughout the period 1993 to 2010, which was mainly explained by a decrease in registration of neuroblastoma in infants. For many cancers, 5‐year survival improved in both countries. The improvement in survival in chronic myeloid leukemia (CML) was particularly dramatic in both countries. However, 5‐year survival remained less than 80% in 2005‐2008 in both countries for AML, brain tumors, soft tissue sarcomas, malignant bone tumors and neuroblastoma (age 1‐14 years). There were significant differences in incidence of several cancers between countries, suggesting variation in genetic susceptibility and possibly environmental factors. The decrease in incidence for all cancers combined in Japan was related to the cessation of the national screening program for neuroblastoma. The large improvement in survival in CML coincided with the introduction of effective therapy (imatinib).

The incidence of childhood cancer overall and by diagnostic subgroup has been reported in the International Incidence of Childhood Cancer (IICC) for many countries, including Japan and England. 2,3 In Europe, survival analysis has been performed to evaluate the quality of care for children with cancer in each country or region in several studies, including the Automated Childhood Cancer Information System (ACCIS) 4 and EUROCARE-5. 5 In 2012, the global surveillance of cancer survival program (the CONCORD-2 study), 6,7 which includes childhood leukemia, was initiated using population-based cancer registry data from 67 countries. In Japan, population-based studies for childhood cancer comparisons to other countries are scarce, although some recent studies show childhood cancer incidence 8,9 or survival, 10 and several cancer registries contributed to the IICC and CONCORD-2 studies. 2,7 In England, population-based incidence and survival for childhood cancer have been reported since 1980s. [11][12][13] In the current study, we compared incidence and time trends in survival for childhood cancer between Japan and England during the period 1993-2010, to gain insight into the progress against childhood cancer in both countries.

| Data
This study was based on data from population-based cancer registries in Japan and England. It included all children (0-14 years) diagnosed with cancer between 1993 and 2010 residing in 6 Japanese prefectures (Miyagi, Yamagata, Niigata, Fukui, Osaka and Nagasaki) 10 or in England. Japanese data were obtained from the Monitoring of Cancer Incidence in Japan (MCIJ) project and the Japanese Cancer Survival Information for Society (J-CANSIS) project, 10 while data for England were obtained from the Office for National Statistics. A standard set of variables included basic demographic data (age, sex and country), information on the tumor (date of diagnosis, site and morphology) and on follow-up (date of last contact and vital status). Follow-up information was available at least 5 years after diagnosis in Japan although the patient follow-up system differs for each cancer registry. 10 Within the Japanese data, vital status information was available for patients diagnosed during 1993-2008. In the English data, the vital status was last updated on 31 December 2015.
We included only records of malignant cancers (behavior code/3) defined in the International Classification of Disease for Oncology, 3rd edition (ICD-O-3). 14 Non-malignant or borderline central nervous system tumors such as craniopharyngioma, meningioma, ganglioglioma, benign teratoma and pilocytic astrocytoma were all excluded. Skin carcinomas were also excluded. Cancers were grouped into 12 main diagnostic categories according to the International Classification of Childhood Cancer, 3rd edition (ICCC-3). 15 We modified some subgroups of ICCC-3, based on the topography and morphology codes from ICD-O-3 (Table S1).
These data partially overlapped with the data used in the IICC or CONCORD-2, although for participating registries or study periods, inclusion criteria were not completely matched. There were few discrepancies in the incidence of each cancer between both datasets, with the exception of central nervous system (CNS) tumors and all cancers combined.

| Statistical analysis
Incidence rates were calculated as the average annual number of children newly diagnosed with cancer per million children. Agestandardized incidence rates (ASR) were calculated by the direct method, using the weights of the world standard population for the age groups under 15 years (0, 1-4, 5-9 and 10-14 years). 16,17 Changes in incidence rates over time were calculated using a Poisson regression model, divided into 3 time periods (1993-1998, 1999-2004 and 2005-2010) and adjusted for age-group, and expressed as average annual percent change (AAPC). Differences in incidence rates between the 2 countries were measured with Poisson regression models and expressed as the incidence rate ratio (IRR), using English data as the reference. These ratios were adjusted for time period and age group. Observed population-based survival was estimated by cancer type in each time period (1993-1996, 1997-2000, 2001-2004 and 2005-2008)

| Data quality
Analyses were based on 5192 cases in Japan and 21 295 cases in England between 1993 and 2010. Table 1 shows the quality criteria for validity and completeness of the data over time in each country.    Table 2 and Figure S1A). However, incidence for all cancers except neuroblastoma was stable (AAPC = 0.2%, [95% CI À0.4-0.8],  Figure S1B). In England, the incidence of all childhood cancers increased from 1993-  Table 3 and Figure 1 show ASR in the total period of 1993-2010 and the incidence rate ratios (IRR) for each cancer type between Japan and England (England reference), adjusted for time period and age group. We analyzed age-specific incidence rates by sex for some solid tumors (Wilms tumor, hepatoblastoma and GCT of each site) in each country (Table S2 and Figure S2). The peak age for Wilms tumor in Japan was infants aged under 1 year, whereas in England it was children aged 1-4 years. Hepatoblastoma was the most common type of hepatic tumor in both countries (N = 100 [88%] in Japan vs N = 211 [82%] in England) and the age distribution was similar between countries. Age-specific incidence rates for intracranial GCT were higher in Japan than in England for all age groups. Incidence of gonadal GCT in male infants in Japan was much higher than in England. However, the numbers were too small to perform any relevant statistical comparison.

Japan and England
We analyzed trends in 1-year, 5-year and 10-year survival for each cancer type and each period in both countries ( We included for incidence analysis, but excluded for survival analysis. T A B L E 2 Trends in incidence of childhood cancer (age 0-14 y) by period of diagnosis and cancer type in Japan and England To calculate 10-year survival in recent periods, we used a different approach (period approach) from the cohort approach, so there was divergence between 5-year survival and 10-year survival (higher survival in 10-year survival than 5-year survival) in some cancers (lymphomas, NBL, renal tumors, and unspecified cancers in Japan, and AML, CML, NBL infants, and GCT in England). Figure 2

| DISCUSSION
In this study, we compared incidence and trends in survival for each childhood cancer type in Japan and England. Incidence of all childhood cancers combined decreased in Japan throughout 1993-2010 ( AAPC, average annual percentage of change; ASR, age-standardized incidence rate (person per million-years); CNS, central nervous system; NBL, neuroblastoma; n.a., AAPCs of "NBL infants (age < 1 y)" were not calculated because models were not fitted. a Age-specific incidence rate. b Using records, population and world standard population in age 1-4, 5-9, 10-14 y. differences in incidence for these cancers. [23][24][25][26][27][28][29] Etiological factors of HL have been suggested by the bimodal age distribution, by elevated risks in males, by the occurrence of Epstein-Barr virus in HL tumor cells, and by identifying inherited susceptibility genes; however, the mechanism by which racial differences in incidence for HL occur is still unclear. Regarding renal tumors, previous studies which reported on differences in age distribution between countries for Wilms tumor showed the peak age for occurrence in East Asia to be infants (age <1 year), but among Caucasians in the USA the peak occurrence was older. 24,25 Our study supports these findings (Table S2 and  Wilms tumor between Japanese and Caucasians. 26,27 For Ewing sarcomas, one report showed that Japanese Ewing sarcoma patients have a higher frequency of loss of chromosome 19 than European Caucasian patients. 28 However, these tumors are rare and their etiology has not been sufficiently investigated to explain these differences in incidence. 29 Regarding the higher incidence of AML in Japan, Bessho reported the mis-classification of ALL to ANLL (AML), T A B L E 3 Age-standardized incidence rate (ASR) and incidence rate ratio (IRR, England reference) of childhood cancer (age 0-14 y) in Japan and England, 1993-2010 which overestimated the proportion of ANLL in the 1970s. 30 However, nowadays, diagnosis of leukemia has become much more accurate and the proportion of unknown leukemia subtype was only approximately 5% in our data ( Table 3). The ALL:AML ratio in our data was 2.4:1, which is similar to that found in the report of the Japanese pediatric leukemia study group (JPLSG), containing information on molecular abnormalities collected by pediatric oncologists (ALL: AML = 2.8:1). 31 On the IICC-3 website, the ASR of AML was around 10 per million person-years in Japan and Korea, 32 whereas the figure was around 7 per million person-years in an Austria-based study. 19 The CONCORD-2 study on cancer survival reported higher proportions of AML in Asia than in Europe. 7 In the US data, there are no large racial differences in incidence for AML. 33 Further research will be needed to clarify whether the differences we have observed are due to underlying ethnic difference in the incidence of AML.
When comparing incidence for each subgroup, the proportion of "unspecified" histology within each cancer group should be taken into account (Table 3). The proportion of "unspecified" lymphomas (ICCC-3 II-e; 18%) or "unspecified" CNS tumors (ICCC-3 III-f; 16%) in Japan was over 10% within each cancer group in the total period (1993-2010), although it decreased to under 10% in the most recent periods (data not shown).
Five-year survival for most cancer types improved in both Japan and England. (Table 4)

| Cancer strategy for childhood cancer
Since 1974, the Japanese Government has subsidized medical expenses for children and adolescents under 18 years of age with cancer. 40 The   Another limitation was the divergence between 5-year survival and 10-year survival (higher survival in 10-year survival than 5-year survival) in several cancers because we used the period approach to predict 10-year survival in recent periods. To improve surveillance and comparability, we need to keep collecting data widely and precisely, and follow up patients' vital status in the long term.
In conclusion, the incidence rates of the majority of childhood cancers differed significantly between Japan and England. Some of

ACKNOWLEDG MENTS
We would like to thank the cancer registries in Japan (Miyagi, Yamagata, Niigata, Fukui, Osaka and Nagasaki) and in England for providing data. We would like to thank Dr Julia Mortimer for helping us with the English language.

CONFLI CT OF INTEREST
The authors have no conflicts of interest to declare.