Trends in gastric cancer mortality 1990–2019 in 36 countries worldwide, with predictions to 2025, and incidence, overall and by subtype

Abstract Background Gastric cancer (GC) incidence is declining heterogeneously worldwide. We aimed to calculate updated mortality trends for GC. Methods We investigated time trends for selected countries using the World Health Organization database. We computed age‐standardized mortality rates (ASMR) per 100,000 persons over the 1990–2019 period. We reported rates for the 2010–2014 and 2015–19 calendar periods, and the corresponding percent changes. We used joinpoint regression analysis to identify changes in the slope of mortality trends, and predict the number of deaths and rates for 2025. We also reported 2008–2012 incidence rates of cardia and noncardia GC. Results Mortality trends from GC have been favorable since 1990 for all countries analyzed and the European Union (EU 27), in both sexes and all ages. GC mortality is predicted to decline in all countries for both sexes, except for French and US women aged 35–64 years, and Canadian men aged 35–64. The highest proportions of cardia GC were observed in Northern and Central Europe while the lowest ones in Southern and Eastern Europe. Elsewhere, the highest proportions were registered in countries with low incidence and mortality rates, whereas high‐mortality countries showed lower proportions of cardia GC. Conclusion Observed and predicted GC mortality trends declined in most countries in both sexes, with few exceptions, likely due to the control of GC risk factors, in particular Hp infection.


| INTRODUCTION
Gastric cancer (GC) was the leading cancer worldwide until 1975, 1 it still accounts for over one million new cases (https://gco.iarc.fr), being the fourth cause of cancer mortality in 2020. 2 GC prognosis is poor, the disease is often diagnosed at an advanced stage where aggressive surgery is needed. 3 Consequently, prevention to control the burden of GC is crucial. Infection with Helicobacter pylori (Hp) is the major cause of GC and is widespread globally, with differences by country, socioeconomic status (SES), and birth cohort. 4 Infection prevalence has been declining progressively over the years, largely thanks to water sanitation. 5 Tobacco smoking, alcohol consumption, unbalanced diet, and other factors associated with low SES are major concomitant factors, while occupational and genetic factors play an additional role. 6 The burden of GC is larger among individuals with low SES and in low-and middle-income countries. 7 Asian countries experience a heavier burden of GC compared to other regions, with Mongolia, Japan and the Republic of Korea being the countries with the highest GC incidence in 2020. 8 Cardia and noncardia GC subtypes (this latter including fundus, corpus, large and lesser curvatures and antrum) representing 15% and 75% of cases, respectively, can be identified based on the proximal or distal origin of the neoplastic lesions. 9 Given that the anatomical locations are associated with different risk factors, with cardia GC sharing risk factors with esophageal adenocarcinoma such as obesity and gastroesophageal reflux while noncardia GC is strongly associated with Hp infection, GC shows different patterns across subgroups of the population and geographical areas. Conversely, the histological distinction of GC into intestinal and diffuse type does not translate into heterogeneity in GC epidemiology. 3 Upper-endoscopy is the gold standard for GC diagnosis, which is based on biopsy analysis. 10 We aimed at estimating and interpreting the mortality trends of GC worldwide between 1990 and 2019, with emphasis on the last 10 years, focusing on sex and age differences. We also showed the proportion of cardia to noncardia incidence subtypes. Moreover, we reported the projected mortality rates and deaths for the 2025 calendar year.

| Mortality
Using the WHO mortality database, 11 we analyzed official numbers of certified deaths from GC coded as C16 according to the 10th Revision of the International Classification of Disease, 12 in selected countries worldwide. The European Union as a whole was defined as the 27 member states, EU (27), as of January 2020. We restricted our analyses to countries with high-mortality coverage (i.e., over 90%), high data quality, and more than 5 million inhabitants identifying 36 countries worldwide plus the EU. 13 We extracted estimates of resident populations from the same WHO database 11 for European and Australasian countries and from the Pan American Health Organization (PAHO) database for American countries. 14 When population data were missing, we retrieved them from EUROSTAT 15 or the United Nations (UN) Population Division 16 databases. Thus, we derived figures for the calendar periods from 1990 up to the most recent available year.
For each country, calendar year, and sex, we calculated 5 years' age-specific rates and the age-standardized mortality rates (ASMRs) per 100,000 person-years, using the world standard population, 17 at all ages and for the 35-64 year age group. We then fitted joinpoint regression models 18 from 1990 until the most recent available year, for the most populous (≥10 million) countries and the EU (27), to identify significant changes in the linear slope (on a log scale) of ASMRs. For each identified linear segment, we estimated annual percent change (APC) and, for the entire study period, the average annual percent change (AAPC). 19,20 For countries with ≥20 million inhabitants (i.e., 15 countries plus the EU as a whole), we also predicted the number of deaths for the year 2025, by fitting a Poisson joinpoint regression model, with a maximum of 5 possible joinpoints 19 to the number of deaths in each 5-year age group and running a linear regression over the most recent trend segment identified, the joinpoint model was restricted to identifying models with the final segment having at least 5 data points. Using the matrices of predicted age-specific death counts and predicted populations, we computed predicted ASMRs and their 95% prediction intervals (PIs). The predicted populations were extracted using the PAHO database for North American and Latin American countries, the EUROSTAT database for European countries, and the UN database for Asian and Oceanian countries.

| Incidence
We obtained incidence data for GC for the 2008-2012 calendar period from the Cancer Incidence in Five Continents database, volume XI. 21 For countries with more than one cancer registry, we aggregated data to ensure the largest geographic coverage, and we restricted analyses to the longest common calendar period between registries. We presented, separately by sex, the number of cardia (C16.0) and noncardia (C16.1-6) incident cases of GC in various countries worldwide, in order to show the proportion of cardia cases.
We only considered publicly available data, so no clearance by Ethical Committee was necessary. For statistical analyses, we used R version 4.1.3 (R Development Core Team, 2017), SAS version 9.4 (SAS Institute Inc.), and Joinpoint Regression Program version 4.9.1. Table 1 shows the annual average number of deaths and ASMRs from GC per 100,000 person-years of all ages and both sexes in selected countries worldwide and the EU (27), for the periods 2010-14 and 2015-19, together with the corresponding percent changes. The ASMRs ordered from the highest to the lowest one during the most recent period displayed by bar plots for men and women are also shown in Figure 1. Except for Greek males, the change in rates was favorable in all countries and both sexes. During 2015-2019, the highest male rates were reported in the Russian Federation, Chile, and Belarus with rates higher than 15/100,000, followed by Japan, Ukraine, Romania, South Korea, and Portugal with rates higher than 10/100,000. The lowest rates were observed in the United States, Sweden, Australia, and Canada with rates around 2-3/100,000 males. Similarly for females, the Russian Federation, Chile, Belarus, Japan, Portugal, and Ukraine showed the highest ASMRs during the 2015-2019 calendar period (rates varied from 5.1 to 6.8/100,000). The lowest ASMRs were registered in the United States, Australia, Canada, Belgium, Norway, Sweden, and Finland; with rates around 1.5-1.7/100,000 females. Corresponding figures for truncated rates in the 35-64 years age group are reported in Table S1 and Figure S1. The change in rates from 2010-2014 to 2015-2019 was favorable in all countries considered and both sexes, except for Denmark (+3.3%), Greece (+5.9%), Norway (+10.0%), Switzerland (+1.4%), and Mexico (+0.4%) in males and the United States (+1.0%) in females, where the rates were higher in the most recent period. Table S2 and Figure S2 report ASMRs among order adults aged 65 or more. Among the elderly, the change in rates from 2010-2014 to 2015-2019 was favorable in all countries considered and both sexes. The ranking was similar to the all ages one with some differences, that is, Chile and Japan were the first-ranking countries for the elderly and the second and fourth one at all ages. This indicates steeper declines in young and middle age in those countries as compared to Russia and other eastern European countries. Figure 2 shows the trends in ASMRs from GC at all ages (full dots) and 35-64 years (empty dots) along with the corresponding joinpoint models (lines) for 23 selected countries worldwide and the EU-27, over the study period among males. The corresponding figures for females are shown in Figure 3. Results from the joinpoint analyses are reported in Tables S2 and S3, respectively, for males and females. For the largest countries and the EU (27), the predicted ASMRs at all ages and 35-64 years for the calendar year 2025 with the corresponding 95% PIs are also shown. Detailed results for the 2025 predicted deaths and ASMRs are reported in Table 2. Male ASMRs resulted in a decrease since 1990 in all countries considered and the EU (27) and for both all ages and truncated 35-64 years. The Republic of Korea showed the highest decrease over the considered period, with an AAPC of −5.2 for all ages. Czech Republic, Netherlands, Sweden, and the United Kingdom reported moderate decreases (AAPCs greater than −4% per year) too, while Greece, Argentina, Cuba, and Mexico reported less consistent but still significant decrements (<2% per year). The EU (27) reported a significant decrease over time, with an AAPC -3.3. Results for males aged 35-64 years were similar to those for all ages.

| RESULTS
The Republic of Korea reported the steepest fall in trend over time in females too (AAPC -5.2 for all ages). Belgium, the Czech Republic, and the United Kingdom showed favorable trends over time as well (AAPC around −4). Greece, Argentina, Brazil, Cuba, and Mexico had the smallest AAPCs-lower than 2-but still significant. The EU (27) showed a decrease over time with an AAPC of −3.3. Corresponding figures for females aged 35-64 were consistent with the decrease observed at all ages, albeit smaller.
Predicted rates for the year 2025 are downwards for all countries and the EU (27), for both sexes, and for all ages and adults aged 35-64, except for females aged 35-64 in France (% difference in rates versus 2015-19: +20.3) and in the United States (+7.3%), and for males aged 35-64 in Canada (+2.7%); however, these populations had comparatively low rates. Likewise, the predicted number of deaths in 2025 was favorable for all countries, sexes and ages except for Argentina, Brazil, Mexico, and the United States, for which the predicted deaths are higher than the ones observed in 2015-2019, due to population aging and growth. The most favorable predicted rates for 2025 are those in the Republic of Korea (% difference in rates versus 2015-19: −53.3 and −54.0, respectively, for men and women), followed by those calculated in Japan (−32.3% and −37.3%, respectively, for men and women). In the EU (27), the ASMR is predicted to decline in men and women, with, respectively, a −23.9% and −28.3% decrease from 2015-19 to 2025 for all ages and −26.6% and −22.7% for adults aged 35-64 years. Figure 4 gives the proportion of cardia and noncardia GC in men and women, respectively, for selected countries from Europe and other countries worldwide. For T A B L E 1 Annual average deaths and age-standardized (world population) mortality rates from gastric cancers per 100,000 person-years in 2010-14 and 2015-19 for both sexes, along with the corresponding change in rates.  23 Endoscopy and multiple biopsy sampling are the gold standard for GC diagnosis. Also, endoscopic ultrasound helps in identifying wall infiltrations, while endoscopic mucosal resection and endoscopic mucosal dissection provide reliable staging information, allowing in addition the treatment of superficial early-stage lesions. The endoscopy is subject to the compliance of the patients and the availability of both a trained specialist and an endoscope and endoscopy room. Indeed, it is an invasive and expensive test, and is limitedly available in less affluent countries. 10 Despite surgical resection may be curative, most patients relapse. The most recent ESMO Clinical Practice Guidelines indicate that stage IB-III disease requires radical gastrectomy. 24 Nodal dissection accompanying radical gastrectomy is indicated to cover at least 15 lymph nodes according to the AJCC/UICC TNM, which again implies the availability of large upper-GI surgical units and highly specialized peri-operative care centers, which are more common in Asia than in other countries. 24 Preoperative and postoperative chemotherapy is recommended in all ≥IB resectable GC, possibly including fluoropyrimidine, a platinum agent and docetaxel. 24 When focusing on locally advanced unresectable or metastatic disease, the prognosis is very poor, with less than 1 year survival. 24 However, combined chemotherapy has been demonstrated to be more effective than single-agent chemotherapy or supportive care only. Moreover, survival improvement has been reached by using chemotherapy combined with immunotherapy (nivolumab or trastuzumab). 24 Part of the mortality decline is due to screening programs and early diagnosis in high-incidence countries like Japan and Korea. Nearly three-quarters of new GC cases occur in Asia, but large differences can be found when observing GC trend in Asian countries. A recent study highlighted how the different socioeconomic development, the historical background of cancer control programs explain the different trends in incidence by birth cohorts, where Japan underwent a decline in younger cohorts while China did not. 22 These authors attributed this difference primarily to the high-quality cancer control policies undertaken in Japan 22 years earlier than in China. Our results corroborate previous analyses focused on GC trends during 1980-2011, and identified a decrease in GC mortality which was more pronounced for noncardia cancer in non-Asian countries. 22 The authors interpreted these results as the progress in Hp infection control matched with a birth-cohort effect characterized by lower prevalence of infection in subsequent generations, causing a lower rates of noncardia GC cases. Also, they described a gradually smaller decline in GC mortality trends in US and predicted no further reduction for France after 2015. 22 Our findings also agree with the projections to 2040 made in a recent study, which described the estimated incidence and mortality of GC worldwide to be particularly high in Asia and in men, with about two thirds of all cases occurring in the male sex. 23 When considering subtypes, noncardia is responsible for more cases and, assuming comparable case fatality between the two types, 25 it likely causes more deaths than cardia GC. Noncardia GC is more frequent than cardia in F I G U R E 2 Annual age-standardized (World standard population) mortality rates per 100,000 males from gastric cancer in major countries (>10 million inhabitants)worldwide and the EU (27), the resulting joinpoint regression models and predicted rates (only for countries with >20 million inhabitants and the EU)for the year 2025 with 95% prediction intervals. The Y axes in the figure (Deaths per 100,000) are not uniform.

Men
All ages 35-64 years most countries, with a ratio between 2:1 and 3:1. 26 Diet and obesity as well as gastroesophageal reflux disease are implicated in the increasing number of cardia cases in US and EU countries. 27,28 In addition, the ratio reflects past (given 20-30 years of latency for cancer development) prevalence of Hp. The high prevalence of infection is responsible for the high ratio in South America, East Asia, and Southern and Eastern Europe. When looking at the numbers of cardia GC, possible anatomical location misclassification should be taken into account. Indeed, the increased number of cancers of the esophageal junction and of the cardia in cancer registries may be partially due to differences in tumor classification because of discrepancies in the anatomic subsite classification. This discrepancy was observed in a study based on Swedish Cancer Registry, where the authors also noticed a bidirectional misclassification between cancer of the lower esophagus and cancer of the gastric cardia. 29 Another study also based on Swedish Cancer Registry estimated that true cardia GC incidence could be up to 45% higher or 15% lower than that reported, possibly explaining the increasing number of cardia GC cases in the country. 30 A subsequent analysis on Cancer Registry of the Swiss Canton of Vaud over the period from 1976 and 1997 aimed at interpreting the reported increasing incidence in cardia GC, did not find an absolute increase in cardia GC cases, but rather a decrease of noncardia GC cases, which resulted in an increase in the proportion of cardia GC. 31 However, as noted by several authors, 32,33 the classification of upper GI cancers remains problematic given the lack of a universally accepted and clearly reproducible anatomic landmark separating the gastric cardia from the lower esophagus, and even when the landmarks are defined, cancer often destroys the anatomy of the organ where it occurs, making landmarks become unrecognizable.
Based on our data, the reduction in GC mortality is similar in both sexes, with few exceptions (e.g., Denmark, Greece, and Norway, where declining rates were greater among women), although GC incidence and mortality remain higher in men. Previous studies explained this difference with the distribution of smoking habits and Hp infection, which are higher in men than in women: Countries where Hp infection shows larger gender gaps (e.g., USA and EU) have a stronger gender difference in mortality rates for GC as well. 34,35 It will be interesting to see if in countries where cardia GC and its risk factors are not so frequent (e.g., Asian countries) there will be a further decrease in GC overall, because of the low incidence of cardia coupled with a continuous decrease in noncardia GC. Whether the incidence of cardia GC will increase, following the increasing prevalence of its risk factors such as obesity 48 remains an open question.
For some countries, including the United Kingdom, Brazil, Mexico, Canada, and the United States, the overall reduction in GC mortality was less pronounced among younger subjects, furthermore rising incidence rates in young adults have been reported recently. 26,[34][35][36][37] Age-cohort effects would suggest a lower incidence of GC among the new generations, due to the birthcohort effect of Hp infection 38 -well known to be less prevalent in the young 39 -as well as smoking habits, declining progressively in subsequent birth cohorts. 40 These results are consistent with the findings of Wang and collaborators, who calculated GC incidence rates in the United States between 1999 and 2013: New GC cases have been reported to overall diminish, except for patients diagnosed under 50 years of age. 41 Moreover, the incidence of advanced noncardia cases has been observed among under-50 years old Hispanics. 42 These elements could justify the smaller decrease in GC mortality in the young population. The lower reduction in the young could also be referred as the overall small incidence rate in this population. Also, this can be due to more aggressive forms of cancer affecting this subgroup in these areas. 42 Besides this, we could not consider neither the stage and grade at diagnosis to explain this difference more in detail.
Mortality trends are intrinsically limited by death certification validity. For this reason, we only included countries with WHO death certification data with over 90% death certification coverage, with good data quality, as declared by the WHO, 13 and with a resident population of over 5 million, and over 20 for prediction analysis, in order to minimize issues of excessive variability. Predicted statistics for cancer mortality should be interpreted with caution, the selected prediction method is based on age-specific joinpoint models, as such it is apt for short to medium term projections based on recent trends. Very recent changes in trends, occurring in the final 2-3 years of available data, and major changes in cohort effect trends could be difficult to detect with this prediction model. However, our previous European cancer mortality predictions, that have been running for over 10 years, have proved to be reliable and valid. 38 F I G U R E 3 Annual age-standardized (World standard population) mortality rates per 100,000 females from gastric cancer in major countries (>10 million inhabitants) worldwide and the EU (27), the resulting joinpoint regression models and predicted rates (only for countries with >20 million inhabitants and the EU) for the year 2025 with 95% prediction intervals. The Y axes in the figure (Deaths per 100,000) are not uniform.
All ages 35-64 years T A B L E 2 Number of predicted deaths and mortality rates from gastric cancer, for both sexes, for the year 2025 and comparison figures for 2015-2019, for selected (>20 million inhabitants) countries worldwide and the EU as a whole, with 95% prediction intervals. Further, a previous age-period-cohort analysis displays monotonal descending cohort effects for the age groups and cohorts involved in these predictions, and comparing our previous prediction analysis on GC mortality published in 2014 with up to date registered data we found that the estimated projected ASMRs were accurate. 38,43 Broken line models, such as the joinpoint model used in this study, may also suffer from issues of overfitting. However, the NCI's joinpoint software is conservative in its model selection algorithms, and the frequentist permutation model selection method we use is the most conservative. 19 The previous age-periodcohort analysis found an asymptotic trend in GC mortality in subjects born since the 1940 s in some Nordic and Western countries, foretelling the lack of progress in French middle-aged women. 38 One issue is whether misclassification on death certificates has changed over time: If it has remained the same, it would not affect the results on trends. In any case, the surveillance of trends in younger adults is needed to monitor mortality changes by age and to plan potential preventive interventions towards this subgroup. 42 The analysis of 35-64 years truncated age group was performed for data quality and certification issues. That this truncated age group has higher mortality than the overall is expected, due to the weighting of the world standard population that gives younger age groups greater weights.

Men
A major strength of this analysis is the high number of countries included for which real data were available, making our results more reliable than those based on estimates. The data provided in this manuscript are based on large countries, those reducing statistical fluctuation. Moreover, we could present results for both GC anatomical sites, namely cardia and noncardia.

| CONCLUSION
Our findings confirm the declining mortality of GC in all countries with valid data, unfortunately this excludes China and other Asian, South American, and African countries of major interest for GC with lower data quality, limiting the representativeness of these results on a F I G U R E 4 Gastric cancer age-standardized (World standard population) incidence rate estimates for the year 2020 from Globocan in males and females, proportion of cardia (black) and noncardia (white) incident cancers from Cancer Incidence in 5 Continents Volume XI. global population scale. 22 The pattern of reduction is not homogeneous by geographical area and seems to follow the prevalence of Hp infection. The reduction in mortality rates is lower in US and EU countries, where the prevalence of infection has already reached a stable point, while high-risk areas such as Eastern Europe and Asia have undergone an important decline in GC mortality, and will likely see a further decrease in 2025. Indeed, aggressive programs of screening and eradication of Hp infection may lead to a dramatic decrease in GC mortality in a relatively short time, in high-risk populations. 44 The distribution by anatomical site showed that the proportion of cardia and noncardia are getting close in many areas. Again, this may reflect the distribution of GC risk factors, with prevention of Hp infection and its eradication leading to a promising downscaling of noncardia GC in the EU and US, where cardia remains linked to lifestyle factors and obesity. The development of predictive models, the identification of high-risk subjects, and the early "test-and-treat" of Hp at the population level may represent important preventive tools.