Time trends incidence of both major histologic types of esophageal carcinomas in selected countries, 1973–1995

Authors


Abstract

The purpose of our study was to examine the incidence patterns of 2 major histologic types of esophageal cancer, in selected countries world-wide and to identify components of birth cohort, period and age as determinants of observed time trends using regression modeling. The roles of temporal changes in specification of histology of tumors and of classification of cancers at the gastroesophageal junction as esophageal or gastric in origin were taken into consideration. In all, 56,426 esophageal cancer cases were included. The results indicate that the incidence rate of squamous cell carcinoma of the esophagus has been relatively stable in most of the countries analyzed, although increasing trends were observed in Denmark and the Netherlands (Eindhoven) among men and in Canada, Scotland and Switzerland among women. There was a significant increase in the incidence of esophageal adenocarcinomas in both sexes in the United States (among whites and blacks), Canada and South Australia and in 6 European countries (Scotland, Denmark, Iceland, Finland, Sweden and Norway). In France the increase was limited to men and in Switzerland the increase was observed only in women. Modeling was unable to distinguish which trends were the results of changes in risk between generations (as cohort effects), or changes in all age groups simultaneously (as a period effect). © 2002 Wiley-Liss, Inc.

Esophageal cancer is the 8th most common cancer in the world1 and one of the most lethal. Squamous cell carcinoma (SCC) is the predominant histologic type of esophageal cancer worldwide. Several studies using population-based cancer registry data have suggested that incidence rates of adenocarcinomas of the esophagus have been increasing since 1970 and that this has been accompanied by an increase in incidence of cancer at the gastric cardia. For example, significant increases at both sites have been reported from registries in Australia,2 the United States,3–9 New Zealand,10 the United Kingdom,11, 12 Norway,13 Denmark,14 Sweden15 and Finland.16 This increase of incidence rates for cancers at the esophagogastric junction was also demonstrated in a recent study of 9 European countries.17

The purpose of the present study was to examine, on a worldwide basis, time trends in the incidence of esophageal cancers by histologic type and to identify the relative importance of the components of birth cohort, period and age as determinants of observed time trends, using regression modeling. The contribution of changes in the proportion of cases with specified histology (and/or correctly allocated to the different subsites) was also assessed. The present analyses make use of incidence data provided by 34 cancer registries in 12 countries over a median 15-year period.

MATERIAL AND METHODS

Selected cancer registries

Population-based cancer registries fulfilling the following criteria provided the original data:

  • 1The registry had to be included in the (peer-reviewed) series of publications known as Cancer Incidence in Five Continents in at least 3 consecutive volumes: vol. IV (1973–1977), vol. V (1978–1982), vol. VI (1983–1987) and vol. VII (1988–1992).18–21 Quality indices for these data are given in Table I.
  • 2Esophageal and stomach cancer cases with histologic diagnoses confirmed by histology had to be 80% or more of the total registrations.
  • 3The percentage of cancers of the esophagus registered through death certificate only (DCO) had to be 5% or less.
  • 4All datasets had to include histologic diagnosis compatible with the coding of the International Classification of Disease for Oncology (ICD-O).22
Table I. Data Quality Indices For Esophageal Cancer In The Selected Countries1
CountryVol. IV (1973–77)Vol. V (1978–82)Vol. VI (1983–87)Vol. VII (1988–92)
MaleFemaleMaleFemaleMalesFemalesMalesFemales
HV2DCOM/IHVDCOM/IHVDCOM/IHVDCOM/IHVDCOM/IHVDCOM/IHVDCOM/IHVDCOM/I
  • 1

    From Cancer Incidence in Five Continents, IARC.18–21

  • 2

    DCO, death certificate only; HV, histological verification of diagnosis; M/I, mortality/incidence ratio.

North America
 USA, SEER (9 registries)
  White95195942879619495288
  Black96179980829808396180
  San Francisco
   White91210094181941106952879529590789
   Black944102935929818498170
  Connecticut
   White902939118292292931849528494180961107962100
   Black930739031009437296074
   Metropolitan Detroit
    White91295861105911958528795195952809718896274
    Black9587979096078920779707798173
  Hawaii: White990113900909706789067
  New Mexico: White97376596093494295962989428096267
  Seattle: White92295861105931999018597090961919708895085
  Utah: White8941139398794298950629309496074971102910109
  Metropolitan Atlanta
   White91089874838818998076991969801139717990080
   Black9638792810889287932709308095093
 Canada (7 registries)
  Newfoundland90898955120874131891110694011287013097010378996
  Nova Scotia8501158801389601139309391610889593
  Prince Edward Island94039900770108920108955100500150
  Saskatchewan82989920129882102990929809894485
  Manitoba6793777908608767010875176831113840113
  British Columbia81279275277973097760828871008899493610194490
  New Brunswick891194881211294011287013095212986797
Europe
 Scotland (5 registries)82699789948341008317386210581496
  East88989210996710397119477610083173
  North7810583979239891010791110381389
  north-east8386819188410878410587010678472
  South East82110819187410882810288213487299
  West74891679947679373108881590775838438978589
 Denmark829107791197921922942116903105951108942104
 Iceland910930970114830839906799052
 Finland83293803938729183189918690919318792292
 Sweden93102889797949186909199969894
 Norway89195851959118519690931849729593395
 Netherlands; Eindhoven969990899112190115971359076
 Switzerland (2 registries)
  Geneva93197756949719293010798187930969628996489
  Basel96096990150990929901179908199082
 France (4 registries)
  Bas-Rhin9158949698909411298889988
  Doubs849992868786868696849685
  Calvados887993839511596849685
  Isere8678921039713398589359
Australia
 South Australia941099477980799507580978285

According to these criteria, 67 populations were eligible.

Using the European Cancer Incidence and Mortality (EUROCIM) database, all cases of esophageal and stomach cancer recorded in the registries in 9 European countries since 1973 were extracted. EUROCIM is a computerized cancer incidence and mortality database developed by the European Network of Cancer Registries (ENCR; http://www-dep.iarc.fr/encr.htm). For the United States the data were extracted from the public use Surveillance, Epidemiology and End Results (SEER) data by the CD-ROM edited by the National Cancer Institute (NCI).23 Statistics Canada provided the Canadian dataset. The first step in the verification process was the establishment of a frequency distribution for each variable (age, sex, site, data of birth, year of diagnosis, histology). Registries were contacted if mistakes were found and asked for their comments on possible misunderstanding in the fields. Thirty-two registries were excluded because the proportion of esophageal cancers for which the coded histology was “cancer unspecified” (ICD-O 8000–8004) or “carcinoma unspecified” (ICD-O 8010–8034) was more than 25% of cases registered. Finally, 34 cancer registries for which incidence data of comparable quality were available for a minimum of 15 years (within the period 1973–1995) were included.

Statistical analysis

Each participating registry coded esophageal cancers by histologic type according to the 4 digits of the ICD-O.23 Cancer cases were grouped according to Parkin et al.24 as follows:

  • 1Carcinomas: squamous cell carcinoma (ICD-O codes 8050–8076), adenocarcinoma (ICD-O codes 8140–8141, 8190–8231, 8260–8263, 8310, 8430, 8480–8490, 8560, 8570–8572), other specified carcinomas (ICD-O 8041–8042, 8082, 8120–8123, 8144–8145, 8320, 8550), unspecified carcinomas (ICD-O 8010–8011).
  • 2Sarcomas (ICD-O codes 8800–8811, 8830, 8840–8920, 8990–8991, 9040–9044, 9120–9133, 9150, 9540–9581).
  • 3Other specified cancer (ICD-O 9672, 9683, 9684)
  • 4Unspecified cancer (ICD-O codes 8000–8004).
  • 5For simplicity in summarizing the results, we grouped other specified carcinomas, unspecified carcinomas, sarcomas, other specified cancer and unspecified esophageal cancer cases into a group that we termed other and not specified (Other/Nos).

Cases were grouped into 5-year intervals of diagnosis and 5-year age groups, between the ages of 25 and 85 years. Age-specific rates were calculated using midperiod population denominators for each age group and summary age-adjusted incidence rates were calculated by direct standardization using the world standard population.

Subsite of stomach cancers was not specified for all time periods in 4 datasets (Australia, Finland, Norway and Sweden). In the remainder, stomach cancers were grouped into the following 3 anatomic subsites: cardia (ICD-O, C16.0), other sites (ICD-O, C16.1–C16.8) and unknown or unspecified subsite (ICD-O, C16.9).

The unknown/unspecified (C16.9) subgroup comprised the majority of registered cases in most registries; however, the percentage of cases recorded in this way generally showed a decrease over time. For this reason, in order to evaluate trends in the specified subsites unaffected by improvements in ascertainment, recording or coding, the unknown/unspecified cases were proportionately redistributed among the other 2 categories, within each sex and age group (<40, 40–49, 50–59, 60–69, ≥70 years). Age-specific and age-standardized rates were calculated, as for esophageal carcinomas.

On the assumption that the number of new cases in each age group and period would be distributed as a Poisson variable, a log-linear model was used to assess the effect of age, period and cohort as independent predictor variables. Bearing in mind the problem of identifiability of the parameters, we used cohort-effect curvature analysis as proposed by Holford25 and a restriction of the possible range of cohort-effect slopes. Models were elaborated by smoothing the different effects in a stepwise procedure. First, age effects were fitted to explain the variability in incidence due to age. Second, drift effects adjusted for age were explored (when age-period or age-cohort effects were significant) and period or cohort effects were fitted for age to determine the curvature or deviation from linearity of each effect.26 Finally, period or cohort effects were fitted in those registries in which age-cohort or age-period was not enough to explain most of the variability in incidence. The fit of the models was evaluated in terms of deviance and the log-likelihood ratio test was used to evaluate the statistical significance of the terms. In the testing of various nested models against each other, the difference in deviance was assumed to have a Chi-square distribution. p-values of <0.01 difference between the models were considered statistically significant. When overdispersion was detected, a negative binomial distribution was used.27

The statistical models used offer considerable advantages over simple descriptive methods because the data can be described by a smooth function and yet address the dispersion of cases observed in some registries. However, a difficulty arises when adjustment is attempted, since the variables used (age, year of diagnosis and year of birth) are mathematically interrelated. This means that an infinite number of estimates can be computed for the same model and it is difficult to differentiate cohort from period effects.

For simplicity, we calculated the annual percentage change over the interval from the coefficient of the linear component for the cohort and period effect (net drift), adjusted for age.

Separate models were fitted for each cancer registry. However, for some countries the data from several cancer registries were pooled, provided they presented similar patterns in the data from individual registries, in order to provide larger numbers for analysis. Thus, the cancer cases were combined for the registries of the SEER program in the United States (9 registries: Connecticut, Iowa, New Mexico, Utah, Hawaii, San Francisco-Oakland, Metropolitan Detroit, Metropolitan Atlanta and Seattle), Canada (7 registries: Newfoundland, Nova Scotia, Prince Edward Island, Saskatchewan, Manitoba, British Columbia and New Brunswick), Scotland (5 registries: East, North, North-East, South East and West), Switzerland (2 registries: Geneva and Basel) and France (4 registries: Bas-Rhin, Doubs, Calvados and Isère).

RESULTS

In all, 56,426 cases of esophageal cancer, from registries in 12 countries, were included in the analysis (Table II). The percentage of cases with histology “other and unspecified” in both sexes ranged from 5.1% in the Swiss registries to 23.4% in the Scottish registries. SCCs were the predominant histologic type of esophageal cancer in most populations, with the exception of males in the registries of Scotland. However, the ratio between the 2 histologies varied from 1:1 (equality) in South Australian men to 34:1 in U.S. blacks.

Table II. Observation Period, Number Of Cases And Age-Standardized Incidence Rates (Asrs) For Esophageal Carcinomas By Histology In Selected Countries
CountryPeriodMenWomen
SCCAdenocarcinomasOther/NOSTotal1SCCAdenocarcinomasOther/NOSTotal1
No.%ASR2No.%ASR2No.%No. (100%)No.%ASR2No.%ASR2No.%ASR2
  • 1

    Number of cases of esophageal cancers (all histologies) registered in the interval.

  • 2

    Age standardized rates for the entire time period. SCC, squamous cell carcinoma.

North America
 USA, SEER (9 registries)
  White1973–955,467(52.5)2.23,720(35.7)1.51,237(11.9)10,4243,068(70.5)0.9638(14.6)0.3651(14.9)4,357
  Black1973–952,658(89.0)12.982(2.7)0.4247(8.2)2,987933(88.0)3.623(2.2)0.1105(9.9)1,061
 Canada (7 registries)1981–931,590(54.0)2.0968(32.9)1.3385(13.1)2,943864(68.7)0.8191(15.2)0.2203(16.1)1,258
Europe
 Scotland (5 registries)1981–951,899(36.7)3.42,147(41.5)3.91,121(21.7)5,1672,141(50.2)2.61,033(24.2)1.11,087(25.5)4,261
 Denmark1978–951,420(48.7)2.11,103(37.9)1.5391(13.4)2,914786(59.2)0.9322(24.2)0.3220(16.6)1,328
 Iceland1978–96104(65.4)3.339(24.5)1.816(10.1)15966(79.5)1.611(13.3)0.36(7.2)83
 Finland1982–971,027(62.9)1.8323(19.8)0.8284(17.4)1,6341,135(74.8)2.3100(6.6)0.5282(18.6)1,517
 Sweden1977–963,173(67.8)2.1925(19.8)0.6585(12.5)4,6831,447(72.7)0.9240(12.1)0.1304(15.3)1,991
 Norway1978–961,190(65.0)2.1381(20.8)0.6261(14.2)1,832477(68.2)0.997(13.9)0.1125(17.9)699
 Netherlands Eindhoven1978–92167(70.8)2.347(19.9)0.622(9.3)23656(60.2)0.621(22.6)0.216(17.2)93
 Switzerland (2 registries)1978–96451(77.2)4.8107(18.3)1.126(4.5)584154(76.2)2.233(16.3)0.415(7.4)202
 France (4 registries)1978–953,919(83.4)13.6316(6.7)1.0465(9.9)4,700353(75.6)0.960(12.9)0.152(11.2)465
Australia
 South Australia1979–93212(39.4)1.4210(39.0)1.4116(21.6)538198(63.9)1.042(13.5)0.270(22.6)310
Total23,277(60.0)10,368(26.7)5,156(13.3)38,80111,678(66.2)2,811(15.9)3,136(17.8)17,625

Table II shows, for each population, the age-standardized incidence rates (ASRs) for both major histologic types of esophageal cancer for the entire period in both sexes. Incidence rates of esophageal cancer are higher in males than females irrespective of histologic type. Males have an approximately 3- to 4-fold greater risk than females for SCC and a 7- to 10-fold higher risk for adenocarcinomas.

There is a 7-fold difference in incidence of SCCs of the esophagus among men in European populations, from a high in France to a low in Finland. Within the United States, the rate in black men is almost 6 times that for white men.

The ASR for adenocarcarcinomas of the esophagus is below 5/100,000 in most registries. In the European populations, the incidence rates are high in Scotland compared with other countries analyzed. The lowest incidence rate for men is that in the black population of the United States and for women in France. In the United States, the incidence rate in white men is almost 4 times that for black men.

Figure 1 shows the trends in ASRs (all ages; world standard population) for the different histologic types (and the Other/Nos group) of esophageal cancers. In addition, we present the observed and adjusted ASRs for carcinomas of gastric cardia in selected countries. ASRs for Other/Nos esophageal carcinomas are also presented. Tables III and IV show the trends in esophageal cancer (by histologic type) and cancer of the gastric cardia (and adjusted data) expressed as the annual percent change over the interval, with a 95% confidence interval. A negative sign indicates a decreasing trend. The results of fitting Poisson models to data for esophageal cancer by histologic type are also presented.

Figure 1.

Trends in age-standardized incidence rates (world population) of both major histologic types of esophageal carcinomas and cancer of gastric cardia (observed and adjusted data) by sex in selected countries, 1973–1995.

Table III. Men: Percent Change Per Year Of Incidence Rate And Results Of Fitting Models To The Asrs Of Esophageal Carcinomas (By Histology) And Cancer Of Gastric Cardia (Observed And Adjusted' Rates)
CountryEsophageal carcinomasCancer of gastric cardia
SquamousAdenocarcinomaObservedAdjusted1
Δ % (95% CI)2Best model3Δ % (95% CI)2Best model3Δ % (95 % CI)2Δ % (95 % CI)2
  • 1

    Adjusted for unspecified gastric cancers (see text).

  • 2

    Percent change and 95% confidence intervals (CI) per year derived from the linear term of the Poisson model. A negative sign denotes a decreasing trend.

  • 3

    The best fitting model for each dataset derived to compare age (A), age-drift (A+D), age-period, age-cohort and age-cohort-period (A+C+P). p-value < 0.01.

  • 4

    Data not available by subsite of stomach in EUROCIM dataset.

North America
  USA, SEER (9 registries)
   White−1.5 (−2.1; −1.0)A+D8.6 (8.0; 9.1)A+D2.8 (2.3; 3.3)0.3 (−0.3; 1.0)
   Black−0.6 (−1.1; −0.0)A+C+P4.1 (0.7; 7.7)A+D3.0 (1.3; 4.8)1.1 (−0.4; 2.6)
  Canada (7 registries)−0.1 (−1.3; 1.2)A4.6 (2.9; 6.4)A+D1.5 (0.2; 2.8)−0.8 (−1.5; −0.1)
Europe
 Scotland (5 registries)0.7 (−0.4; 1.8)A3.1 (1.9; 4.4)A+D2.4 (0.4; 4.5)−0.5 (−1.5; 0.5)
 Denmark2.6 (1.6; 3.6)A+D7.9 (6.6; 9.2)A+D−0.8 (−1.6; −0.1)−2.8 (−3.4; −2.3)
 Iceland0.7 (−2.7; 4.3)A7.7 (1.3; 14.6)A+D−4.3 (−7.6; −0.8)−2.0 (−3.7; −0.1)
 Finland−1.1 (−1.7; −0.5)A7.2 (5.6; 8.8)A+D44
 Sweden−0.5 (−1.2; 0.1)A+C+P2.3 (1.1; 3.5)A+D44
 Norway−0.2 (−0.9; 0.5)A8.3 (6.7; 10)A+D44
 Netherlands, Eindhoven4.3 (0.4; 8.3)A+D−1.6 (−8.2; 5.5)A1.7 (−1.1; 4.7)−1.7 (−4.2; 0.8)
 Switzerland (2 registries)−1.5 (−3.3; 0.2)A4.2 (−0.1; 8.5)A0.6 (−1.7; 2.9)−0.5 (−2.3; 1.4)
 France (4 registries)−1.5 (−2.4; −0.6)A2.8 (0.6; 5.1)A+D0.9 (−0.5; 2.4)−0.7 (−1.7; 0.4)
Australia
 South Australia1.6 (−1.6; 5.0)A6.4 (2.9; 10.0)A+D44
Table IV. Women: Percent Change Per Year Of Incidence Rate And Results Of Fitting Models To The Asrs Of Esophageal Carcinomas (By Histology) And Cancer Of Gastric Cardia (Observed And Adjusted Rates)
CountryEsophageal carcinomas
SquamousAdenocarcinomaObservedAdjusted1
Δ % (95% CI)2Best model3Δ % (95% CI)2Best model3Δ % (95 % CI)2Δ % (95 % CI)2
  • 1

    Adjusted for unspecified gastric cancers (see text).

  • 2

    Percent change and 95% confidence intervals (CI) per year derived from the linear term of the Poisson model. A negative sign denotes a decreasing trend.

  • 3

    The best fitting model for each data set derived to compare: age (A), age-drift (A+D), age-period, age-cohort, and age-cohort-period (A+C+P). p-value < 0.01.

  • 4

    Data not available by subsite of stomach in EUROCIM dataset.

North America
  USA, SEER (9 registries)
   White0.1 (−0.5; 0.7)A+C+P6.8 (5.5; 8.1)A+D2.6 (1.8; 3.3)−0.3 (−1.4; 0.8)
   Black−0.3 (−1.3; 0.6)A+C+P13.8 (5.1; 23.3)A3.8 (1.0; 6.6)2.2 (0.0; 4.5)
  Canada (7 registries)2.8 (1.1; 4.6)A+D6.0 (2.2; 9.9)A+D1.1 (−1.2; 3.4)−1.5 (−2.6; −0.4)
Europe
 Scotland (5 registries)2.8 (1.8; 3.9)A+D4.8 (3.2; 6.4)A+D0.8 (−0.9; 2.5)−2.0 (−3.6; 0.5)
 Denmark1.3 (−0.1; 2.6)A4.6 (2.4; 6.8)A+D−1.3 (−2.7; 0.1)−3.7 (−4.5; −2.8)
 Iceland−2.2 (−6.3; 2.1)A18.6 (2.5; 37.1)A+D−5.8 (−12.3; 1.2)−2.5 (−5.7; 0.7)
 Finland−0.4 (−1.0; 0.2)A+C+P6.1 (3.6; 8.5)A+D44
 Sweden0.4 (−0.5; 1.3)A3.1 (0.7; 5.5)A+D44
 Norway0.8 (−0.4; 2.0)A5.5 (2.5; 8.6)A44
 Netherlands, Eindhoven2.6 (−3.9; 9.6)A0.8 (−9.3; 12.0)A+D4.6 (−0.3; 9.8)1.3 (−3.0; 5.8)
 Switzerland (2 registries)8.5 (4.5; 12.2)A+D12.1 (2.8; 22.2)A+D1.8 (−1.9; 5.7)−3.0 (−5.7; −0.2)
 France (4 registries)1.0 (−1.1; 3.1)A5.5 (−0.4; 11.7)A+D−0.6 (−3.3; 2.2)−2.7 (−4.6; −0.8)
Australia
 South Australia5.9 (2.3; 9.6)A9.2 (1.1; 17.8)A44

In men squamous cell carcinoma rates declined during the period in most of the countries analyzed. However, an upward trend was found in Denmark and in the Netherlands (Eindhoven). A significant increase was also noted among women in Canada, Scotland, Switzerland and South Australia.

For esophageal adenocarcinomas, significant increases in the ASRs have been observed in both sexes in the United States (among both black and white populations), in Canada and in South Australia. There was also a significant increase in both sexes in Scotland, Denmark, Iceland, Finland, Sweden and Norway. In France the increase was limited to men and in Switzerland the increase was observed only in women. No significant trends weree observed in the registry of Eindhoven (the Netherlands).

For the most part, the temporal trends for cancers of the gastric cardia were nonsignificant. However, a significant increase for cancer of gastric cardia was observed in both sexes in the United States (both black and white populations) and among men in Canada and in Scotland. Adjusting these trends to take into account the increasing proportion of gastric cancer cases for which subsite was specified rendered these changes nonsignificant, whereas there were declines in incidence of gastric cardia cancers in several of the registries showing increased rates of esophageal adenocarcinomas (Tables III, IV).

The statistical model that best fits esophageal carcinomas (by histologic type) for each defined population is also indicated in Tables III and IV. Most datasets were well fitted by simple age-drift models; only for a few registries was a saturated age-period cohort model necessary to explain the variability of the data.

Figure 2a and b shows the concurrent annual percent change in the ASRs of esphageal adenocarcinomas, as well as esophageal cancer Other/Nos. The increased incidence of adenocarcinoma was generally greater than any decline in the Other/Nos category. Figure 2c and d shows concomitant change in rates of esophageal adenocarcinomas and cancers of the gastric cardia; for the most part (particularly in men) both rates were increasing (right upper quadrant of the figure). However, when change in esophageal adenocarcinoma is compared with change in adjusted ASRs of gastric cardia cancer (Fig. 2e,f), it can be seen that there is some correlation between the 2, an increase in adenocarcinoma of the esophagus being accompanied by a decline in gastric cardia cancers.

Figure 2.

Bivariate distribution of the annual percent change in the age-standardized incidence rates (ASRs) of esophageal adenocarcinomas and Other/Nos esophageal carcinomas (a,b), as well as cancers of the gastric cardia (observed [O, c,d] and adjusted [A, e,f] data) and esophageal adenocarcinoma in selected countries. Solid circles, A+, O+: significant change in both esophageal adenocarcinonas and Other/Nos esophageal cancers (a,b); esophageal adenocarcinoma and cancer of gastric cardia (observed data in c,d or adjusted data in e,f). ×,A+, O: significant change only in esophageal adenocarcinomas. Open circles, A, O+: significant change only in cancer of gastric cardia (observed or adjusted data). Open squares, A, O: nonsignificant change in either carcinoma type (esophageal adenocarcinoma and Other/Nos (a,b); esophageal adenocarcinomas and gastric cardia (observed data in c,d or adjusted data in e,f).;3>

DISCUSSION

The results of our descriptive study confirm the existence of substantial variations in the incidence rates for both major histologic types of esophageal carcinomas. The incidence rate of adenocarcinoma of esophagus is increasing in both sexes in the United States, Canada, South Australia, Scotland, Denmark, Iceland, Finland, Norway and Sweden, whereas the rates of SCC are relatively stable or decreasing. However, there were some exceptions: in Denmark and the Netherlands (Eindhoven), the incidence rates for SCCs are increasing in both sexes. In Canada, Scotland, Switzerland and South Australia, the increase for SCCs was limited to women.

The linear drift of the regression models implies that the observed trends are uniform and that the relative curvature of birth cohort effects and the period-specific risk cannot be distinguished. The drift also implies that the existing trends will continue for some time into the future.

Several studies conducted in the United States show that there has been a steady increase in the incidence of adenocarcinomas of the esophagus since the early 1970s, but not of SCCs of the esophagus. Between 1976 and 1987, the incidence of adenocarcinoma of the esophagus in white males in the United States doubled, an increase that was higher than for most other cancers.4, 7, 28 Similar trends were reported in Australia1, 29 and New Zealand.10

European investigators have reported similar changes, but increases have been more modest than in the United States. In Scotland, the incidence of adenocarcinoma increased from 0.8/100,000 in 1976 to 1.1/100,000 in 1989.11 Increasing trends for adenocarcinomas of esophagus and cancers of the gastric cardia have been observed in some registries in England.12 In the Nordic countries, increasing trends were observed in Norway, with average annual increases in adenocarcinoma incidence of 16.6% in males and 14.4% in females during 1983–1992.13 In Denmark, an increase in incidence was noted for both major histologic types of cancer in men, but the increase was most pronounced for adenocarcinomas, which increased from 0.78 to 1.29/100,000 during 1978–1987.14 In Finland during 1978–1995, the incidence of esophageal adenocarcinoma increased in both sexes.16 An increasing incidence of esophageal adenocarcinomas was noted in France, but it remains more moderate than elsewhere.30 In a population-based study from the Swiss canton of Vaud, an increase in incidence for esophageal adenocarcinoma occurred, whereas the incidence of carcinomas of the gastric cardia was relatively constant.31

In a recent study using EUROCIM data from 10 European countries, increases in incidence of adenocarcinomas of the esophagogastric junction were found in some European countries (Denmark, Italy, England and Scotland), but not in others (France, Iceland, Southern Ireland, Switzerland and the Netherlands).17

Quality of data

All the registries used in our study had appeared in the series Cancer Incidence in Five Continents.18–21 This constitutes the most reliable source of international cancer incidence statistics available and peer-reviewed procedures are applied to decide on the inclusion or otherwise of the data. Nevertheless, there are several potential sources of bias in studies of trends of esophageal cancer by histologic subtype.

First, a proportion of esophageal cancers are diagnosed (and registered) without a histologic verification of diagnosis. A decrease in this proportion over time (due, for example, to improved diagnostic techniques, especially the introduction of flexible esophagoscopy) will lead to an apparent increase in the incidence of histologically specified subtypes. However, the datasets used for the present study had been selected so that histologic verification of diagnosis was high (>85%) and the fraction of cases registered on the basis of information contained on death certificates alone was low (<5%). Moreover, there were no significant changes in the proportion of esophageal cancers with histologic verification of diagnosis. The high mortality/incidence ratios (>100) observed for some populations reflect inaccuracy in certification of cause of death.20

A second possible explanation for increasing trends of esophageal adenocarcinoma lies in improvement in the proportion of cases specified as adenocarcinoma or SCC, rather than simply “carcinoma” (the Other/Nos category). The latter generally comprises less than 20% of esophageal cancers and the changes in incidence were nonsignificant and too small to account for observed trends in adenocarcinomas.

Finally, it is possible that an increasing tendency to classify cancers located at the gastroesophageal junction as adenocarcinoma of the esophagus (rather than gastric cardia) accounts for some or all of the observed increases. Our analysis suggests that a substantial part of the apparent increase in incidence of cancer at the gastric cardia in several countries is due to improvement in the specification of subsite of stomach cancers over time and that, when this is taken into account, incidence of cardia cancers was quite probably decreasing in most centers (Tables III and IV). Ekstrom et al.,32 in a careful analysis of data from Sweden, have also shown that there may be considerable misclassification between cardia and distal gastric cancers, even when subsite is specificed; this would also tend to give the impression of rising incidence over time. The allocation of adenocarcinomas at the gastroesophageal junction to cardia or distal esophagus may also be changing over time, perhaps influenced by increasing awareness of the “epidemic” of Barrett's esophagus and esophageal adenocarcinoma. It does seem from our results (Fig. 2c–f) that increases in incidence of adenocarcinoma of esophagus have been accompanied, in some cancer registries, by decreases in incidence of cancer at the gastric cardia, particularly if (Fig. 2e,f) changes in the proportion of gastric cancers with subsite specified are taken into account.

Etiologic clues from trends analyses

As well as artifactual changes, due to improving histologic verification of tumors of the esophagus and stomach, improvement of precision of histologic diagnosis and changes in misclassification between gastric cardia and esophagus over time, some of the increase in incidence of adenocarcinomas of the esophagus might be real and might be explained by the introduction of a “new” environmental risk factor.

Case-control studies of patients with esophageal adenocarcinoma have suggested cigarette smoking and possibly, heavy alcohol consumption as risk factors, although not to the extent observed for SCC of the esophagus.33–35 There is however, some biologic plausibility for an effect of smoking and alcohol on adenocarcinoma of esophagus, because both are known to decrease lower esophageal sphincter pressure and predispose to gastresophageal reflux.35

The marked increase in the incidence of esophageal adenocarcinoma has been linked to factors favoring gastresophageal reflux such as the increased prevalence of obesity.33, 36, 37 Several epidemiologic studies have shown a 3- to 6-fold excess risk among overweight individuals.37, 38 A recent population-based case-control study of British women concluded that a high body mass index in early adulthood and low consumption of fruit are important risk factors for adenocarcinoma of the esophagus.37 Obese individuals have a predisposition to gastresophageal reflux disease and to an increased prevalence of hiatal hernia.39

In this context, distal esophageal adenocarcinomas have been associated with the presence of Barrett's esophagus.40 Although Barrett's esophagus is the consequence of severe gastresophageal reflux, it is difficult to know which constituents of the reflux are important in causing intestinal metaplasia and cancer. Some investigators believe that duodenal contents (bile salts) are as important as acid and pepsin41 and suggest that the increasing prevalence of acid-suppressive medical therapy may be contributing to the epidemic of Barrett's esophagus and adenocarcinoma of the esophagus, by increasing the pH. The increase in pH may be inimical to Helicobacter pylori infection and there is some evidence that infection with the cagA+ strains of H. pylori is inversely associated with cancers of the gastric cardia and esophagus.42

The use of a variety of other medications, such as anticholinergics, asthma drugs, tricyclic antidepressants, narcotics and antihistamines, all of which can relax the lower esophageal sphincter, has greatly increased in this period and may also have had an impact on Barrett's esophagus and adenocarcinomas.39

The similarities in the temporal trends (and in some of the etiologic factors) for cancers of the lower esophagus and gastric cardia have led several authors to conclude that they represent a single entity.43 A separate chapter for carcinoma of the esophagogastric junction is included in the new WHO Classification of Tumours of the Digestive System.44 In fact, it is very difficult to determine the site of origin of tumors arising at this level.45

On the other hand, it appears that there are some differences in the origin and characteristics of intestinal metaplasia preceding or accompanying the carcinomas of the gastroesophageal junction. Intestinal metaplasia in the lower esophagus is considered to be a sequel of chronic inflammation caused by gastroesophageal reflux, whereas intestinal metaplasia in the gastric cardia might be the result of carditis caused by H. pylori infection, gastroesophageal reflux or local trauma.46 Furthermore, some differences in pattern of cytokeratins have been reported: the cytokeratin 7 and 20 pattern is essentially restricted to intestinal metaplasia of Barrett's esophagus but is absent or rare in intestinal metaplasia of the gastric cardia.47

Analytical epidemiologic studies are needed to identify risk factors responsible for the observed increase in esophageal carcinomas. The common epidemiologic features of esophageal adenocarcinoma and carcinomas of the gastric cardia suggest a common etiology. However, the different characteristics of intestinal metaplasia may imply a different carcinogenic process at the 2 sites, which would have implications for primary prevention.

Acknowledgements

We express our gratitude to all collaborating registries from the European Network of Cancer Registries that contributed data to the EUROCIM database. Dr. N. Muñoz and Dr. S. Franceschi (IARC) provided us with advice and valued comments.

Ancillary