Incidence and mortality rates for gastric cancer have been decreasing in nearly all developed countries for several decades.1 In spite of this trend, wide geographic variability still persists. Variability in survival from gastric cancer is also high within European countries2 and globally.3
Table I shows, for European cancer registry (CR) areas involved in the EUROCARE-2 study,2 age-adjusted incidence rates for cancers of the stomach for all subsites, and for cancers arising in the gastric cardia together with the percent of gastric cancers with a cardia subsite4 and 5-year age-adjusted relative survival rates ranked by decreasing gastric cancer incidence. In spite of the considerable variability in stomach cancer incidence among European CR areas (with Romagna, central Italy, having a 4-fold increased rate compared with Denmark), the incidence of cardia cancer was relatively stable in all these European countries at a level of 2–4 per 100,000. Relative survival rates appear to be ranked by gastric cancer incidence, thus indicating an association between incidence and survival (Fig. 1) for those CR areas with age-adjusted 5-year relative survival data available in Table I.
Table I. Incidence and Survival of Gastric Cancer Cases in Europe, Males
The aim of the study was to identify reasons for this association between incidence and survival and obtain survival estimates and differences corrected for this variation, thus facilitating further interpretation by clinical factors such as stage and treatment.5, 6
MATERIAL AND METHODS
Gastric cancer survival data from the EUROCARE database (1978–1992 for 47 cancer registries in 17 European countries) were analyzed. The EUROCARE-2 protocol and data checks provided the basis for data definition with appropriate quality and degree of completeness indicators. The intention was to use sex, age, period, subsite and histologic type as the main variables for the analysis. Stage information was available for a small subset of CRs and was used here in a restricted analysis. Table II shows the distribution of gastric cancer cases by subsite and histologic subtype. Subsite was classified into 5 categories based on the fifth digit of ICD-9 code 151, i.e., 0, cardia; 1 and 2, antrum and pylorus; 3, fundus; 4 and 6, body + lesser and greater curve; 8 and 9, other + unspecified. For some CRs, gastric cancer subsite information was almost totally unavailable and we restricted our analysis to those CRs where the proportion of unclassified cases by subsite was less than 70%. Thirty CRs out of 47 contributing data (from 10 out of 17 countries; Table II) could be included. Histologic type was categorized as ADK type 1 (differentiated): ICDO-2 8050, 8144, 8200, 8210, 8211, 8260–8481, 8560, 8570, 8211; ADK type 2 (nondifferentiated): ICDO-2 8020, 8021, 8231, 8032, 8141, 8142, 8145, 8490, 8550, 8230, 8143, 8245, 8510; ADK not specified: ICDO-2 8140; non-ADK: ICDO 8041, 8070, 8071, 8240, 8241, 8246, 8800, 8801, 8890, 8891, 9540, 8810, 8896, 9560, 8072, 8076, 8720, 8830 8980, 9130, 9140, 9150; not specified: ICDO-8000, 8010, 8012, 8001, 9990.
Table II. Distribution (%) of Gastric Cancer Cases by Subsitea Based on Fourth Digit of ICD-9 Code and Histology Typeb
ADK type 1
ADK type 2
Sub-site was coded as based on ICD-9 4th digit: 0: cardia; 1,2 = Pylorus; 3 = Fundus of stomach; 4,6 = Body of stomach + Lesser and greater curve; 8,9 = other + unspecified.
Histotype was defined according to ICD-O codes as follow: adk Type 1 (differentiated): 8050, 8144, 8200, 8210, 8211, 8260–8481, 8560, 8570, 8211; adk Type 2 (non differentiated): 8020, 8021, 8231, 8032, 8141, 8142, 8145, 8490, 8550, 8230, 8143, 8245, 8510; adk unspec.: 8140; non adk: 8041, 8070, 8071, 8240, 8241, 8246, 8800, 8801, 8890, 8891, 9540, 8810, 8896, 9560, 8072, 8076, 8720, 8830, 8980, 9130, 9140, 9150; non specified: 8000, 8010, 8012, 8001, 9990.
Only 6 CRs in 5 countries had stage information available for more than 60% of their cases and these data were used in a restricted analysis. These CRs were Varese (Italy), Thames (U.K.), Côte d'Or (France), Slovakia, Basel and Geneva (Switzerland). Stage was classified according to the EUROCARE protocol in 4 categories: localized, regional, metastatic and unstaged. Tumors confined to the site of origin (equivalent to T0-3 N0M0) were recoded as localized. Tumors spread to immediately adjacent tissues and/or regional lymph nodes (equivalent to T4, any T N1-3M0) were recoded as regional. Tumors spread to distant organs (equivalent to any T, any N M1) were recoded as distant. Finally, the aggregation of tumors not confined to the site of origin but not specified as regional (equivalent to TxMxNx) or distant or no distant metastasis but not specified if localized or regional was recoded as unstaged or as having no information on stage.
A proportional hazard regression method7 was applied to the relative survival rates in order to obtain estimates of the geographic differences adjusted for all the considered factors and differences in life expectancy among European populations. The PHREG and GENMOD procedures from the SAS package8 were used to calculate risk ratios (RRs) for the proportional hazard regression analysis on observed and relative survival rates.
Data were categorized as follows: sex, age (15–44, 45–54, 55–64, 65–74 and > 75 years), period of diagnosis (1978–1980, 1981–1983, 1984–1986, 1987–1989 and 1990–1992), subsite [cardia, fundus, middle third (including body, greater curve, lesser curve), lower third (pylorus and antrum) and not specified], histologic subtype (ADK type 1, ADK type 2, ADK not specified, non-ADK, not specified) and stage (localized, regional, metastatic, unstaged).
For regression analyses, the reference category was taken as Italy, male sex, 55–64 years, 1978–1980, cardia subsite, ADK type 1 (differentiated) histologic subtype. Localized cancers were the reference category for stage in the restricted analysis. Follow-up was divided into 10 6-month intervals. Only the first 7 follow-up intervals, within which most of the variation in survival took place, were considered in the analysis.
In order to evaluate variability by subsite and histologic type, an explanatory survival analysis for gastric cancer patients from the Varese CR was conducted. The results are presented in Table III by sex, age, period of diagnosis, histologic type and subsite. Cardia and fundus subsites showed higher relative risks of death compared with more distal localization of cancer. Adenocarcinomas nondifferentiated (type 2) showed higher RR than ADK-differentiated. Five-year survival probability varied by subsite from less than 10% for cardia to nearly 25% for the greater and lesser curves.9 The RRs also increased with age and decreased with time period of diagnosis. All these factors could be expected, therefore, to play some role in explaining differences in gastric cancer relative survival among European countries.
Table III. Cox Survival Analysis of Gastric Cancer in Varese, 1978–1992, by Subsite, Sex, Age Class, Histologic Type and Period
Number of cases
5-year survival %
< 65 years
> 75 years
ADK type 1 (differentiated)
ADK type 2 (not differentiated)
A regression analysis of gastric cancer relative survival was conducted for the 30 CRs with relatively complete data. Table IV presents the model selection procedure. Age, subsite and histology type were important factors in explaining gastric cancer survival variability. Neither sex nor period of diagnosis played any major role.
Table IV. Regression Analysis of Gastric Cancer Relative Survival: Model Selection Procedure
With model 1
With previous model
Reference was Italy, men, 55–64, 1978–1980, cardias, ADK differentiated.
Country + sex
Country + sex + age
Country + sex + age + period
Country + sex + age + period + subsite
Country + sex + age + period + subsite + histology
Table V compares the relative excess death ratio (RER) by country between model 1 (crude) and model 6 (adjusted). The effect of adjusting for age, period of diagnosis, subsite and histology type was a substantive reduction of variability in gastric cancer survival. England, Scotland and Denmark, with the highest crude RERs, all showed reductions in the RRs after adjustment, while France, Spain and Switzerland, with low crude RERs, all showed increases after adjustment. The reference country (Italy) showed the lowest relative death rate after adjustment. Although much reduced, variability still persisted after adjusting for the cancer and demographic-related characteristics.
Table V. Relative Excess Death Ratio and Corresponding Confidence Intervals for Model 1 (Crude Differences) and Model 6 (Adjusted Differences)
An analysis including stage and restricted to the 6 CRs with relatively complete and comparable stage information was performed to explain the residual variability. In Table VI, the model selection procedure is presented. Stage was the most important factor explaining variability of gastric cancer differences in addition to age, subsite and histologic subtype. Table VII reports the results of the restricted analysis. Crude relative death rates in the selected CR areas changed a little for countries with no national CRs with respect to the previous extended analysis (model 7). Varese, the reference CR, showed the lowest death rate. The variability in relative death rates was reduced when adjusted for age, period, histology type and subsite (model 12) and reduced further when additionally adjusted for stage (model 13). The relative survival rates for Basel and Geneva became very close to those for Varese, while a residual difference, although reduced, remained for Thames, Côte d'Or and Slovakia.
Table VI. Regression Analysis of Gastric Cancer Relative Survival Restricted to CR Areas with Stage Information: Model Selection Procedure
With model 1
With previous model
Reference was Italy, men, 55–64, 1978–1980, cardia, ADK differentiated, localized.
Country + sex
Country + sex + age
Country + sex + age + period
Country + sex + age + period + subsite
Country + sex + age + period + subsite + histology
Country + sex + age + period + subsite + histology + stage
Table VII. Relative Excess Death Ratio and Corresponding Confidence Intervals for Model 7 (Crude Differences), Model 12 (Adjusted Differences) and Model 13 (Including Stage) in a Restricted Analysis to Registries with Information on Stage
In this study, we investigated the association between incidence rates and survival from gastric cancer and found this to be due largely to differential survival rates for different cancer subsite locations within the stomach. The histologic type of gastric cancer also played an additional role in this association.
The major limitation for this study was the lack of adequate subsite and stage information for many CR areas. This meant that we could not conduct a complete analysis of the EUROCARE database. Within the EUROCARE protocol and associated data quality checks, the completeness and quality of detailed information represents a critical issue for future improvement. In Table II, the proportion of patients with subsite unspecified varied from 18% for Calvados (a specialized digestive diseases CR) to 69% for Scotland. For Finland and Sweden, data were converted from ICD-7 to ICD-9; thus, the cardia was the only specified subsite. There are other major issues associated with the routine recording of cancers in the gastric cardia that further complicate crossregistry comparisons, e.g., variability in the assignation of such cancers to the lower esophagus.10
Although data on histology were generally more complete than for subsite, they were missing for Saarland (Germany). Data from Tyrol (Austria), Finland, Saarland (Germany), Florence (Italy), Iceland, Cracow and Warsaw (Poland), Slovenia, Sweden and Wessex (U.K.) were not reported in Table II and not included in the regression analysis due to the lack of sufficient subsite or histologic subtype information. Comparable data on stage were available only for a small group of CRs.
The crude relative risk of death for gastric cancer patients compared to Varese (Italy) ranged from 0.98 to 1.74 (76% variability). When adjusted by age, sex, period of diagnosis, subsite and histology, the range was reduced from 1.0 to 1.5 (50% variability). About 35% of the crude variability of gastric cancer survival could be attributed, therefore, to these case mix factors. Age, subsite and histologic subtype were all important explanatory factors, while sex and period of diagnosis, although significant, had a relatively minor role in explaining differences between countries. The relative ranking of countries by relative death rate for gastric cancer patients did not change considerably, with England, Scotland, Denmark and Poland still in a relatively lower position compared to Italy. Factors related to diagnosis and treatment may explain the residual variability. The restricted analysis including stage showed that variability of gastric cancer survival was further reduced by 26%, with major differences still observed for Thames, England (36% higher RR than for Varese, Italy), and only minor differences between the other countries. As this analysis was based on only a few CRs, we cannot generalize from these results and they should only be considered as exploratory. An ongoing analysis of the EUROCARE high-resolution study on gastric cancer survival will allow a fuller evaluation of stage and treatment to explain further survival differences.
We can draw at least 3 important conclusions from this study that have implications for public health policy. First, there is a need to improve the standardization of coding systems and registration practices to give greater comparability to cancer statistics in Europe. For registries still using the ICD-7 topographic coding system, we were not able to provide detailed information on stomach cancer subsite. Where subsite information is available, there will be some concerns about its consistency and comparability between CRs. Also, stage information is largely missing and its use in comparative analyses is highly problematic. The systematic use of TNM classification could greatly help in standardizing information on stage. Promoting the general use of updated coding systems in cancer registration in Europe (ICD-10, ICDO-3, TNM-3, etc.) is an important issue to pursue.
Second, there is a need to improve the completeness of relevant detailed information on cases recorded by CRs. Completeness of information on subsite was very variable for gastric cancer among European CRs, with the proportion of unclassified cases varying from 15% to 86% (excluding areas lacking such information completely) and higher than 50% in 19 out 38 CRs that used the ICD-9 classification. An important finding of this study is the evidence that subsite has an impact on both incidence and survival and is an important data item to collect for considering comparative gastric cancer. Often the reason for incompleteness of subsite information is lack of attention during data generation and recording processes involving gastroenterologists, surgeons, pathologists, medical records staff and cancer registries. Stage information is missing for the majority of patients and CR data within the EUROCARE database. It is well recognized that stage is the most important variable related to treatment and posttreatment survival for stomach cancer and a substantial proportion of patients (sometimes near to 50%) usually undergo surgical resection. It is therefore disappointing that so few European registries furnish good staging data.
Third, near 60% of the variability in gastric cancer survival in Europe was attributable to different case mix, while stage is the most important factor in further explaining the residual variability. Treatment is expected to play some role in the remaining differences, such as that observed for England with respect to Italy and France. Since the appropriate treatment is mainly surgical, the proportion of patients undergoing either curative or palliative surgical intervention is the factor most likely to help further interpretation of differences. Information on the type of surgery performed would be an important component for future comparative survival studies.
Improving control of gastric disease and facilitating early diagnosis of gastric cancer is expected to be the way to reduce variability in survival in Europe. Also, improving adherence to guidelines for treatment in Europe11 could improve the criteria for curative and palliative surgery intervention in gastric cancer cases.
Members of the EUROCARE Working Group for this study are as follows: Austria, W. Oberaigner (Cancer Registry of Tirol); Denmark, H. Storm (Danish Cancer Society); Estonia, T. Aareleid (Estonian Cancer Registry); Finland, T. Hakulinen (Finnish Cancer Registry); France, D. Potier (Calvados Digestive Cancer Registry), J Mace-Lesec'h (Calvados General Cancer Registry), J. Faivre (Côte d'Or Digestive Cancer Registry), P. Arveux (Doubs Cancer Registry), J. Estève (International Agency for Research on Cancer), C. Exbrayat (Isère Cancer Registry), N. Raverdy (Somme Cancer Registry); Germany, H. Ziegler (Saarland Cancer Registry); Iceland, L. Tryggvadottir, H. Tulinius (Icelandic Cancer Registry); Italy, F. Berrino (project leader), P. Crosignani, G. Gatta, A. Micheli, M. Sant (Lombardy Cancer Registry), E. Conti (Latina Cancer Registry), M. Vercelli (Liguria Cancer Registry, NCI, Genova), M. Federico, L. Mangone (Modena Cancer Registry), V. De Lisi (Parma Cancer Registry), R. Zanetti (Piedmont Cancer Registry), L. Gafà, R. Tumino (Ragusa Cancer Registry), F. Falcini (Romagna Cancer Registry), A. Barchielli (Tuscan Cancer Registry), R. Capocaccia, G. De Angelis, F. Valente, A. Verdecchia (National Institute of Health, Rome); Poland, J. Pawlega, J. Rachtan (Cracow Cancer Registry), M. Bielska-Lasota, Z. Wronkowski (Warsaw Cancer Registry); Slovakia, A. Obsitníkovà, I. Ples̀ko (National Cancer Registry of Slovakia); Slovenia, V. Pompe-Kirn (Cancer Registry of Slovenia); Spain, I. Izarzugaza (Basque Country Cancer Registry), C. Martinez-Garcia (Granada Cancer Registry), I. Garau (Mallorca Cancer Registry), E. Ardanaz, C. Moreno (Navarra Cancer Registry), J. Galceran (Tarragona Cancer Registry); Sweden, T. Möller (Southern Swedish Regional Tumor Registry); Switzerland, J. Torhorst (Basel Cancer Registry), C. Bouchardy, L. Raymond (Geneva Cancer Registry); The Netherlands, J.W.W. Coebergh (Eindhoven Cancer Registry), R.A.M. Damhuis (Rotterdam Cancer Registry); Scotland, A. Gould, R.J. Black (Scottish Cancer Registry); England, T.W. Davies, D. Stockton (East Anglian Cancer Registry), M.P. Coleman (London School of Hygiene and Tropical Medicine), E.M.I. Williams, J. Littler (Merseyside and Cheshire Cancer Registry), D. Forman (Northern and Yorkshire Cancer Registry and Information Service and Centre for Cancer Research), M.J. Quinn (Office for National Statistics), M. Roche (Oxford Cancer Intelligence Unit), J. Smith (South and West Cancer Intelligence Unit), H. Moller and J. Bell (Thames Cancer Registry), G. Lawrence (West Midlands Cancer Intelligence Unit).