You are here

Article Text

Article menu

This article has a correction. Please see:

Oesophagus
Original article
A global assessment of the oesophageal adenocarcinoma epidemic
  1. Gustaf Edgren1,2,
  2. Hans-Olov Adami1,2,
  3. Elisabete Weiderpass2,3,4,5,
  4. Olof Nyrén2,6
  1. 1Department of Epidemiology, Harvard School of Public Health, Harvard University, Boston, Massachusetts, USA
  2. 2Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden
  3. 3Cancer Registry of Norway, Oslo, Norway
  4. 4Department of Community Medicine, University of Tromsø, Tromsø, Norway
  5. 5Samfundet Folkhälsan, Helsinki, Finland
  6. 6Department of Medicine, Vanderbilt University Medical Center, Nashville, Tennessee, USA
  1. Correspondence to Dr Gustaf Edgren, Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm SE-171 77, Sweden; gustaf.edgren{at}ki.se

Abstract

Objective About 20 years ago, the scientific community was first alerted to an enigmatic increase of oesophageal adenocarcinomas in the UK and USA. Subsequently, a virtual epidemic—still unexplained—was confirmed in several western countries. Detailed descriptive data might provide clues to its causes.

Design We collected data on incident cases of oesophageal adenocarcinoma from population-based cancer registries in Australia, Europe, North America and Asia. We calculated age-standardised incidence rates and fitted log-linear Poisson models to assess annual rate of increase and to disentangle age-period-cohort effects, linear spine models to estimate rate of increase since 1985, and Joinpoint models to identify possible inflection points.

Results With considerable between-registry variation in magnitude and timing, we found a consistent dramatic increase in incidence with an observed or estimated start between 1960 and 1990. The average annual increase ranged from 3.5% in Scotland to 8.1% in Hawaii with similar proportional increase among men and women in most registries and a maintained three to sixfold higher incidence among men. Generally, calendar period was a more important determinant of incidence trends than birth cohort. Where possible to conduct, Joinpoint analyses indicated that the onset of the epidemic varied considerably even between neighbouring countries.

Conclusions Given the preponderant period effect and the abrupt onset observed or inferred in most populations, the epidemic appears to be caused by some exposure that was first introduced around 1950. At least 30 years' variation in estimated time of onset opens prospects for hypothesis-generating ecological analyses.

  • Oesophageal cancer
  • adenocarcinoma
  • cancer epidemiology
  • epidemiology
  • cancer
  • screening
  • biostatistics
  • cancer prevention
  • cancer vaccines
  • gastric adenocarcinoma
  • gastric cancer
  • Helicobacter pylori

https://doi.org/10.1136/gutjnl-2012-302412

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    Significance of this study

    What is already known about this subject?

    • ▸ The first observations that the incidence of oesophageal adenocarcinoma was increasing rapidly were made in the latter half of the 1980s.

    • ▸ The incidence is still increasing in much of the developed world, but data from the developing world are scarce.

    • ▸ The rapid increase cannot be fully explained.

    What are the new findings?

    • ▸ We found a consistent dramatic increase in incidence in all the considered countries, which persists even as late as 2009.

    • ▸ Eyeball inspection and joinpoint analyses place the start of the epidemic at some point between 1960 and 1990, and indicate that it varies considerably even between neighbouring countries and states.

    • ▸ Calendar period seems to be a more important determinant of incidence trends than birth cohort.

    • ▸ On account of similar rates of increase in men and women, on a relative scale, sex ratios remained virtually constant throughout follow-up with men having three to sixfold higher incidence than women.

    How might it impact on clinical practice in the foreseeable future?

    • ▸ The obvious variation in time of onset of the epidemic and the uneven distribution among sexes may provide important clues as to the aetiology of this disease which, in turn, may enable preventive efforts.

    Introduction

    During the latter half of the 1980s it was realised that oesophageal adenocarcinoma—until then considered a rare histological type in the squamous mucosa-lined oesophagus—was on the rise in the USA and the UK.1–6 Since then, numerous studies have shown that the incidence of oesophageal adenocarcinoma has been increasing rapidly in many western populations.7–12 In several cancer registers the incidence of this histological type has overtaken that of the previously more common squamous cell carcinoma.8–10 ,13

    Although many have speculated about the causes behind this ominous development, none of the explanations offered fits perfectly with the descriptive epidemiology. Hence, the epidemic remains essentially unexplained. To understand its causes, information about the time of onset is key, but the scarcity of data and complex patterns of occurrence have impeded the interpretation. In an international comparison of incidence data from 1980 to 1997,11 it was evident (although not commented by the authors) that adenocarcinoma incidence curves for men from some countries made an abrupt turn from a low and presumably stable background rate to a monotonic and unambiguous upward trend. Apparently, the graphs had captured the start of the epidemic in these countries, while the curves from other countries were either way above the presumed background rates and rising, or remained unaltered on a low level throughout the relatively short observation period. We therefore hypothesised that the start was distinct enough to be pinpointed with reasonable precision, and that there might be informative between-population variation in time of onset.

    Our point of departure was that further analysis of this epidemic, in greater detail and over a longer time span, might provide useful insights for further aetiological research. To this end, we assembled all potentially informative and accessible incidence data from around the world in order to locate possible inflection points representing onset in different populations, and to disentangle calendar period effects from age- and birth-cohort effects.

    Methods

    Data sources

    We collected all available data on incident cases of oesophageal adenocarcinoma from population-based cancer registers in Australia, Denmark, England, Finland, Norway, Scotland, Sweden and the United States SEER (Surveillance, Epidemiology and End Results) registers with the longest registration (ie, Atlanta, Connecticut, Detroit, Hawaii, Iowa, New Mexico, San Francisco-Oakland, Seattle-Puget Sound and Utah).14 Initially we intended to limit the analyses to registers with data going back to the early 1970's, or further. However, to increase the geographic range of the data we decided to include Australia also despite its more limited temporal range. Additionally, we obtained supplementary data from the International agency for Research on Cancer (IARC) for a few smaller registers, some representing non-Western populations—Canada (British Columbia, Alberta and Ontario); France (four registers combined: Bas-Rhin, Calvados, Isere and Somme); Israel; Singapore; Japan (Miyagi and Osaka prefectures); Philippines (Manila); and India (Chennai and Mumbai). Data from the respective registers were obtained by directly contacting the registers, with contact persons listed in the acknowledgements section.

    The classification of oesophageal adenocarcinoma varied somewhat between the different registers (table 1). For Australia, Finland, Norway, Scotland and USA—where tumour site was recorded using international classification of disease, revision 10 (ICD-10), and histopathological type using ICD-O-3—we used codes C15 and 8140–8573 for site and histology, respectively. For Denmark, a local variant of ICD-7, code 150.1 was used between 1943 and 1977 and thereafter the combination of ICD-10 (C15) and different variants of ICD-O, code 8140 for histology. In Sweden, the ICD-7-code 150 in combination with WHO/24.1-histology code 096 was used for the entire study period. In addition, we also extracted all data, ranging from 1958 to 2002, on this histological sub-type from Cancer Incidence in Five Continents.21

    View this table:
    Table 1

    Details of cancer registers included in analyses*

    In addition to data on the number of newly diagnosed cancers, calendar year (1-year increments), age group (0 to 80 in 5-year increments, and 85+) and sex, we also extracted the corresponding size of the respective background population at risk. Although data from some of the Nordic countries dates back as far as 1943 (Denmark) and 1951 (Norway), the first few years of data was excluded to minimise the effect of coding irregularities and allow comparison between countries. Thus, in the final analyses, we used data dating from 1955 (Denmark, Finland and Norway) and 1960 (Sweden). As none of the data sources available explicitly listed year of birth, this was imputed by subtracting the age midpoint (ie, 62.5 for the age interval 60–64) from the midpoint of the respective calendar years. Thus, persons aged 60–64 in 1990 were assumed to have been born on 1 January 1928.

    Statistical analyses

    We first calculated age-standardised incidence rates using the direct method with the European standard population as background. The rate of increase of the incidence was estimated both on an additive and a relative (log-linear) scale. The average increase on an additive scale was assessed using Poisson regression, by modelling the age standardised incidence as a function of calendar period (expressed as decade) with an identity link function under the assumption of homoscedasticity. For the log-linear increase, the Poisson models were fitted assuming multiplicative effects on the logarithm of the incidence (ie, with a logarithmic link function), and with the logarithm of the size of the population at risk as offset. For both models we used a linear term for calendar year of observation.

    Recognising the varying length of registration for the different registers, we also fitted linear spline models testing for the log-linear rate of increase since 1985, at which point we had data from all registers. These models allowed us to assume different rates of increase before and after 1985 with loss of only one degree of freedom. We only included persons age 40 years and above in the models. Except for in the drift model, where the gradual change over time was fitted as a simple linear term, age, calendar period of observation and birth year were all treated as restricted cubic splines with five knots, constrained to be linear beyond the end knots, as these allow a smooth, flexible characterisation of these three time axes.22 We assessed different rates of increase in men and women by including an interaction term between sex and drift. Wherever analyses were not conducted separately for men and women, sex was included in the model as a categorical term. CIs for parameter estimates were computed using likelihood ratio tests.

    We fitted four different sub-models of the final age-period-cohort model, including (a) age only, (b) age-drift, (c) age-period and (d) age-cohort. Drift refers to the gradual, linear change of incidence over calendar time and is therefore equally well characterised by linear terms (on a log-scale) for period or birth year. In this instance, the age-drift model differs from the age-period and age-cohort models by fitting this gradual change over time as a simple linear term, as opposed the restricted cubic splines in the latter. On account of the linear interdependence of age, calendar period and birth year, it is not possible to reliably estimate the effects of all three in the same model, without imposing some constraints.23 Given the appearance of a stronger effect of period than of birth cohort, from visual appearance, the full age-period-cohort model was approached by first fitting a model with age and period and then, a model with cohort on the residuals of the first model, with the logarithm of the linear predictions extracted from the first model. This previously used approach,24 ,25 results in estimates of the log-incidence for the respective age groups, at an arbitrary calendar year reference point and the average departure before or after this reference point. In the full age-period-cohort model, the estimates for the cohort parameters can then be interpreted as the effects of cohort, conditional on the observed (and already estimated) age and period effects. p Values and statistical tests are two sided. Model fit was compared using deviance statistics and χ2-tests.

    Finally, to attempt to identify the point at which the incidence began increasing, we used the Joinpoint Regression program supplied by the National Cancer Institute (V.3.5.2, October 2011). As it seemed that the incidence of oesophageal adenocarcinoma began increasing already before cancer registration in some registers (figure 1), we restricted these analyses to Denmark, Finland, Norway and Sweden as well as the US registers in New Mexico, Utah and San Francisco. The age standardised incidence rates were fitted on an additive, that is non-logarithmic, scale. We assumed that there was a maximum of one joinpoint. p Values and CIs were computed by comparing the model with the joinpoint to the model without using permutation tests.26 For the registers where the increase seems to have begun before start of registration, approximate inflection points were estimated by eyeball extrapolation. Except for the joinpoint analyses, all statistical analyses were conducted using SAS (SAS Institute), V.9.2.

    Figure 1
    Figure 1
    Figure 1

    Age standardised incidence by calendar period for men (A) and women (B). Access the article online to view this figure in colour.

    Results

    Altogether we included data on 117 946 incident cases of oesophageal adenocarcinoma, 26 252 (22.3%) of which occurred in women. The vast majority, 116 742 (99% of all cases), was diagnosed at age 40 years or older. In men (figure 1A), the incidence roughly followed one of two calendar period patterns: (1) In New Mexico, San Francisco, Hawaii, Denmark, Norway, Sweden, Finland and possibly Detroit, there was a first phase with a reasonably steady incidence of around or less than 1 case per 100 000 person-years and a second phase of monotonic increase (absolute increment 1–2 cases per 100 000 person-years per decade) following a suggested inflection point perceptible at eyeball inspection; (2) in England, Scotland, Iowa, Connecticut, Seattle, Atlanta and Australia, there was a continuous increase (approximately 2–3 cases per 100 000 person-years per decade) throughout the period covered by the registration. It is noticeable that, with the exception of Atlanta, the incidence in group (2) was already at or above the first phase level described under (1) when the data series started. Due to smaller numbers with entailing chance variation, the pattern in Utah was difficult to assess. There was considerable variation in magnitude of the age-standardised incidence reached during the observation period. In year 2008 (the last year for which we had complete data from all registers), the incidence ranged from 1.9 (95% CI 1.5 to 2.4) cases per 100 000 person-years in Finland to 10.9 (95% CI 9.7 to 12.0) in Scotland.

    In women, the incidence rates were considerably lower and random fluctuations were greater, but the patterns described among men were largely similar among women. Age-standardised incidence in 2008 ranged from 0.1 cases per 100 000 person-years (95% CI 0 to 0.4) in Hawaii to 2.4 (95% CI 1.9 to 2.8) in Scotland. The sex ratio remained relatively stable with the incidence in men generally being three to sixfold higher than in women. However, intriguingly, the sex ratio was considerably higher in the US data, with ratios ranging from sevenfold to ninefold higher in men (online supplementary figure 1).

    Analyses of log-linear incidence increase revealed a universally rapid increase of the age-adjusted incidence (table 2). The estimates for average annual proportional increase ranged from 3.5% (95% CI 3.3 to 3.7) per year in Scotland to 8.1% (95% CI 6.4 to 10) per year in Hawaii. For most registers, the rate of increase did not differ between men and women. Tests for interaction between sex and calendar year, on a log-linear scale, revealed slightly higher rates of increase for men than for women in Denmark, Atlanta (USA), England and Scotland (Pinteraction=0.02, 0.04, 0.0002 and <0.0001, respectively), and a slightly higher rate of increase for women than for men in Sweden (Pinteraction=0.005). The rate of increase since 1985 also varied considerably, with the lowest rate of increase, on a relative scale, seen in Scottish women (2.3%; 95% CI 1.8 to 2.8) and the highest rates seen in Norwegian men and women (8.3%; 95% CI 7.5 to 9.0) (data not shown). When we instead considered the rate of increase on an additive scale, the pattern was largely the opposite, with the most pronounced increase seen in countries with a high incidence already at baseline, for example, Scotland where incidence increased by as much as 1.57 (95% CI 1.42 to 1.73) cases per 100 000 person-years per decade (table 2).

    View this table:
    Table 2

    Annual incidence increase, presented both on relative and additive scales, for men and women combined and stratified by sex, where relevant*

    Age-period-cohort models

    Table 3 presents model fit statistics for the age, age-drift, age-period, age-cohort and age-period-cohort models. Addition of a term for drift improved model fit significantly (p<0.0001) in all registers. When calendar period was fitted instead as a restricted cubic spline we also saw improved model fit, compared with the age-drift model, in all registers except for Hawaii (p=0.20). This was not the case for the age-cohort model, where we did not see a significant improvement compared with the age-drift model for a majority of the US registers (Hawaii, Iowa, New Mexico, San Francisco and Utah), and Norway (table 3). Finally, compared with the generally better fitting age-period model, the age-period-cohort model only conferred a statistically significantly improved fit for Australia, England and Finland.

    View this table:
    Table 3

    Goodness of fit of models including parameters for age, age-drift, age-period, age-cohort and age-period-cohort*

    Joinpoint analyses

    Finally, in table 4, we present results from Joinpoint analyses conducted in Denmark, Finland, Norway, Sweden and three selected US SEER registers. Best estimates for the point at which the incidence began increasing ranged from 1976 in Denmark to 1991 in Sweden. We could not find a statistically significant Joinpoint in New Mexico and San Francisco. On an absolute scale, we saw dramatic changes in the annual incidence increase before and after the Joinpoint. For example, whereas the incidence on average increased by only 0.05 (95% CI 0.01 to 0.10) cases per 100 000 person-years per decade before 1987 in Norway, it increased by 0.79 (95% CI 0.71 to 0.88) cases per 100 000 person-years per decade after 1987.

    View this table:
    Table 4

    Results from joinpoint analyses for selected registers*

    Supplementary data

    Upon eyeball inspection of the age-standardised incidence curves for men in additional registers, from which we deemed available data to be too limited to allow meaningful detailed analysis and/or which only had follow-up until 2002, we could discern patterns suggestive of pattern (1) mentioned above in Canada/Alberta, Israel, and Japan (online supplementary figure 2A). Pattern (2) was suggested in British Columbia, Ontario and France. Data was too scarce to allow conclusions about trends or patterns in men in India (Chennai and Mumbai) and Philippines/Manila, whereas an upward trend was suggested in data from Singapore. Due to lower incidence rates and greater chance fluctuations, patterns and trends were even less interpretable among women (online supplementary figure 2B). However, when the rate of increase was addressed using Poisson regression analyses, there was evidence of increasing incidence in all the considered registers, except for Chennai and Manila (online supplementary table 1).

    Discussion

    Taking advantage of 29–54 years of cancer registration in 16 registers from eight countries, we observed that (1) the incidence of oesophageal adenocarcinoma is increasing rapidly in all registers; (2) although the incidence among men has remained 3 to 9 times higher than among women, the proportional increase has been roughly the same in both sexes; (3) overall, age-standardised incidence curves provide unconvincing evidence of an abating trend; (4) the increase seems to be best explained by a calendar period effect; (5) a rather abrupt change from a low and seemingly stable baseline rate to a monotonic upward trend was captured in approximately half of the registers, while the incidence in most of the other registers was already above the presumed baseline rate when the time series started. Notwithstanding the inherent uncertainty about patterns when rates are low and chance variation high, it appears that the register-specific patterns observed in men were generally corresponded by similar patterns in women.

    Although the baseline rates may have been slowly increasing already during the apparent baseline phase,27 the inflection points were generally readily discerned, could be verified by Joinpoint analysis, and represented impressive alterations in incidence rate. Supported also by the preponderance of calendar period effects in age-period-cohort analyses, it is reasonable to assume that the inflection points signify the region-specific points in time when the increase in adenocarcinoma incidence started taking epidemic proportions. Interestingly, our best estimates of this point in time varied between registers by up to 15 years, with sizeable differences even between neighbouring countries like Sweden and Finland or between Sweden and Norway.

    It seems reasonable to assume that similar abrupt changes in incidence had occurred also in the populations covered by the other registers, whose rates exceeded the presumed ‘natural’ baseline rate of oesophageal adenocarcinoma already in the first available data point. Extrapolation of these incidence curves to the points where they would intersect the presumed baseline rate suggested that the earliest inflection points had occurred around 1960 or shortly thereafter (Scotland and England), while the earliest such point in the US seems to date back to the early or mid-1970s. The latest inflection point confirmed in our Joinpoint analyses was 1991 (Sweden). Hence, the starting points likely vary by approximately 30 years in the studied registers.

    As the main analyses were restricted to countries and regions with reliable and, for the most part, long-standing cancer registration, it seems unlikely that more than a minor portion of the dramatic increase has resulted from improvements in case ascertainment. While overdiagnosis and misclassification were indeed valid concerns initially, Pohl and Welch28 have presented persuasive arguments, based on data from the SEER database, against such explanations. Reclassification of squamous cell carcinoma is unlikely because its histological appearance is quite different from that of adenocarcinoma, and because the only sub site within the oesophagus with increasing cancer incidence is the lower third, the predilection site for the adenocarcinomas. Reclassification of adjacent gastric cancer is likewise unlikely because its incidence has increased, too.9 ,29 In fact, there is evidence that misclassification of proximal gastric cancer has rather led to underestimation of the magnitude of the incidence increase.30 In addition, given evidence that oesophageal and proximal gastric adenocarcinomas are aetiologically closely related, the increase of both might have been caused by the same factors.31 Since there has been little change in the proportion of patients with in situ or localised disease, and because oesophageal adenocarcinoma mortality has increased at similar rates as the incidence, overdiagnosis is a very unlikely explanation.28 Besides, uninterrupted linear increases of the observed magnitude over decades are essentially inconsistent with changes in diagnostic practices or classification. That said, however, it cannot be confidently ruled out that misclassification and reclassification of tumours that were previously reported as unspecified, has impacted especially the early incidence data. Thus, changes in classification of tumours might have led to misestimation both of the timing of the start of the dramatic increase and to overestimations of the scope of the increase but are unlikely to explain the observed trends.

    While most of the data here presented were analysed using robust methods, the disentangling of age-period-cohort effects requires comparatively advanced statistical analyses and, even then, such analyses must be interpreted with caution. It is indeed not possible to uniquely identify the effects of age, period and cohort in the same model. Therefore, we resorted to only assessing the additional contribution of each additional parameter when adding drift, calendar period and birth cohort to the simple age-adjusted model. For the most part, we saw little or no additional improvement of model fit when adding terms for birth cohort to the model with age and period, suggesting that the increasing incidence is chiefly a period effect. Although birth cohort could only be imputed from age and calendar year of observation, with entailing imprecision, the transition point phenomenon did not emerge as clearly for cohort as for calendar period.

    We emphasise that our results do not contradict those of previous reports.1–3 6–9 ,11 ,12 ,28 ,29 ,32–34 In fact, the raw data largely overlapped, but the longer and/or wider register coverage in this study facilitated pattern recognition. In particular, the rather abrupt change in incidence in registers exhibiting a pre-epidemic baseline phase was not fully recognised previously, nor was it realised that the data do not contradict a similarly abrupt start in any of the other registers. Moreover, our age-period-cohort analyses are the first to identify calendar period as more important than birth cohort in explaining the variability of the incidence data, somewhat departing from similar analyses with data up to 1995.32

    The implication is potentially alarming. A decisive exposure may have been widely introduced within a short time span in the respective populations; however, at points in time that differ by up to 30 years between populations. Although the induction and latency times can only be conjectured, we speculate that the exposure might have been first introduced in the early 1950s (in the UK), and about 10 years later it was spread to certain—but not all—regions of the North American continent. Soon thereafter it reached Australia and ultimately the remainder of the North American states (including Hawaii, which was one of the last). In Europe, Denmark, Finland and possibly France were tentatively affected around 1970, while the inflection points in Sweden and Norway suggest a late arrival of the critical exposure, possibly around 1980. With other assumptions regarding the induction time these arrival times could change, but the sequence of arrival and the absolute differences in timing between populations are likely to remain valid.

    Of all well-established risk factors for oesophageal adenocarcinoma, gastro-oesophageal reflux disease and its more severe manifestation, Barrett's oesophagus, are the strongest known to date.34 ,35 While increases in the prevalence of the former of these conditions have been repeatedly cited, these increases mostly seem to have been modest and, in light of a lacking uniform definition of reflux, varying diagnostic interest and no longstanding prospective investigations, the extent of the increase has defied precise quantification.36 At the same time, although similar methodological issues—particularly problems concerning varying indications for endoscopic workup and referral patterns—have hampered also investigations of changes in the prevalence of Barrett's oesophagus, there is clear evidence that it has been increasing at least since the early 1990s.37 ,38 It is thus possible that both the increasing occurrence of reflux disease and of Barrett's oesophagus have contributed to the rising incidence of oesophageal adenocarcinoma.

    Rising prevalence proportions of obesity—linked to an increased tendency for gastro-oesophageal reflux but also in itself an independent risk factor for oesophageal adenocarcinoma39 ,40—have been reported from most Western populations.41 With due caution against variations in sampling and measurements, it appears that the international pattern of the obesity epidemic does not match that of the oesophageal adenocarcinoma epidemic, though. First, available data suggest no more than a tripling of the prevalence of obesity (body mass index ≥30) in the worst affected countries, with no evident inflection points. Second, it seems as if the UK, the presumed starting point of the oesophageal adenocarcinoma epidemic, lags approximately 10 years behind the USA in regard to obesity prevalence and reached the US 1960's levels first around 1990. Third, in countries like Australia and Denmark, the increases in obesity prevalence have been unimpressive, despite sharply rising adenocarcinoma rates.41 Fourth, although obesity is one of the strongest established risk factors for oesophageal adenocarcinoma, it only conveys relatively modest risk increases (two to fourfold).40 Lastly, the obesity epidemic has affected men and women similarly, while the male predominance among oesophageal adenocarcinoma patients is stronger than for any other cancer in organs present in both sexes.42 This enigmatic sex difference8 ,43 remains largely unexplained. In most studies, the strength of the association between obesity and oesophageal adenocarcinoma has been similar between sexes.40 Thus, it seems that the obesity epidemic may have contributed to the increase of oesophageal adenocarcinoma incidence, but seems unlikely to explain more than a fraction. It would perhaps be relevant to assemble data on known oesophageal adenocarcinoma risk factors, and simulate their combined effect in order to get a better understanding of the extent to which these can explain the observed changes.

    Among individuals with Barrett's oesophagus, presumably near the end of a path of successive reflux-related precursor states, the risk for malignant transformation into adenocarcinoma is 2–3 times higher among men than among women.34 ,44 ,45 Similarly, the male predominance among oesophageal adenocarcinoma patients is similar in smokers and non-smokers.46 Except for the finding of a reduced risk linked to breast-feeding,47 and a suggested role of oestrogen in protecting against gastro-oesophageal reflux,48 the search for reproductive and sex hormonal factors that would explain the protection afforded by female gender has so far largely been unrewarding.

    An inverse association between infection with CagA-positive Helicobacter pylori and risk of oesophageal adenocarcinoma has been noted in several studies.49 The well-established decline in H pylori prevalence in essentially all Western populations could thus have contributed to the increase in oesophageal adenocarcinoma incidence.50 ,51 However, as judged from the H pylori seroprevalence in successive birth cohorts, reflecting the all-dominant paediatric incidence of this life-long infection, the decline has been slow and monotonic during the entire 20th century.50 Moreover, the sex differences in H pylori seroprevalence are no more than marginal.52

    While changes in the exposure prevalence to known major risk factors and mis- or reclassification of other malignancies in the oesophagus and proximal part of the stomach have likely contributed to an increasing incidence of oesophageal adenocarcinoma, they are unlikely to explain neither the abrupt change nor the astounding rate of increase. Therefore, it seems reasonable to hypothesise that effects of a strong, highly prevalent and yet unidentified causal factor – first introduced in the UK in the middle of the 20th century—are superimposed on the effects of known risk factors. The obvious variation in time of onset of the epidemic and the uneven distribution among sexes may provide important clues as to the nature of this exposure. Ecological comparisons, fuelled by hints from epidemiologists, toxicologists and basic scientists, may provide hypotheses that can subsequently be tested in case-control or cohort studies.

    Acknowledgments

    The authors are greatly thankful for assistance and contribution of data from the following persons: Drs Mikael Hartman and Min-Han Tan (Singapore), Drs Risto Sankila and Timo Hakulinen (Helsinki, Finland), Mr. Jan Ivar Martinsen (Norway), Ms Paula McClements (Scotland), Ms Natalie Jakomis (England), Dr Mark Short and Ms Wendy Ho (Australia), as well as Dr David Forman and Mr Jacques Ferlay (IARC). Dr Forman and Mr Ferlay also graciously provided comments on an earlier draft of the manuscript.

    References

    1. ↵
      1. Wang HH,
      2. Antonioli DA,
      3. Goldman H
      . Comparative features of esophageal and gastric adenocarcinomas: recent changes in type and frequency. Hum Pathol 1986;17:482–7.
      OpenUrlPubMedWeb of Science
    2. ↵
      1. Yang PC,
      2. Davis S
      . Epidemiological characteristics of adenocarcinoma of the gastric cardia and distal stomach in the United States, 1973-1982. Int J Epidemiol 1988;17:293–7.
      OpenUrlAbstract/FREE Full Text
    3. ↵
      1. Yang PC,
      2. Davis S
      . Incidence of cancer of the esophagus in the US by histologic type. Cancer 1988;61:612–17.
      OpenUrlCrossRefPubMedWeb of Science
    4. ↵
      1. Hesketh PJ,
      2. Clapp RW,
      3. Doos WG,
      4. et al
      . The increasing frequency of adenocarcinoma of the esophagus. Cancer 1989;64:526–30.
      OpenUrlCrossRefPubMedWeb of Science
    5. ↵
      1. Powell J,
      2. McConkey CC
      . Increasing incidence of adenocarcinoma of the gastric cardia and adjacent sites. Br J Cancer 1990;62:440–3.
      OpenUrlPubMedWeb of Science
    6. ↵
      1. Blot WJ,
      2. Devesa SS,
      3. Kneller RW,
      4. et al
      . Rising incidence of adenocarcinoma of the esophagus and gastric cardia. JAMA 1991;265:1287–9.
      OpenUrlCrossRefPubMedWeb of Science
    7. ↵
      1. Trivers KF,
      2. Sabatino SA,
      3. Stewart SL
      . Trends in esophageal cancer incidence by histology, United States, 1998-2003. Int J Cancer 2008;123:1422–8.
      OpenUrlCrossRefPubMedWeb of Science
    8. ↵
      1. Cook MB,
      2. Chow WH,
      3. Devesa SS
      . Oesophageal cancer incidence in the United States by race, sex, and histologic type, 1977-2005. Br J Cancer 2009;101:855–9.
      OpenUrlCrossRefPubMedWeb of Science
    9. ↵
      1. Lagergren J,
      2. Mattsson F
      . No further increase in the incidence of esophageal adenocarcinoma in Sweden. Int J Cancer 2011;129:513–16.
      OpenUrlCrossRefPubMedWeb of Science
    10. ↵
      1. Schmassmann A,
      2. Oldendorf MG,
      3. Gebbers JO
      . Changing incidence of gastric and oesophageal cancer subtypes in central Switzerland between 1982 and 2007. Eur J Epidemiol 2009;24:603–9.
      OpenUrlCrossRefPubMedWeb of Science
    11. ↵
      1. Bosetti C,
      2. Levi F,
      3. Ferlay J,
      4. et al
      . Trends in oesophageal cancer incidence and mortality in Europe. Int J Cancer 2008;122:1118–29.
      OpenUrlCrossRefPubMedWeb of Science
    12. ↵
      1. Steevens J,
      2. Botterweck AA,
      3. Dirx MJ,
      4. et al
      . Trends in incidence of oesophageal and stomach cancer subtypes in Europe. Eur J Gastroenterol Hepatol 2010;22:669–78.
      OpenUrlPubMedWeb of Science
    13. ↵
      1. Brown LM,
      2. Devesa SS,
      3. Chow WH
      . Incidence of adenocarcinoma of the esophagus among white Americans by sex, stage, and age. J Natl Cancer Inst 2008;100:1184–7.
      OpenUrlAbstract/FREE Full Text
    14. ↵
      National Cancer Institute. Joinpoint Regression Program, Version 3.5 - April 2011. Statistical Methodology and Applications Branch and Data Modeling Branch, Surveillance Research Program National Cancer Institute, 2011.
      1. Storm HH,
      2. Michelsen EV,
      3. Clemmensen IH,
      4. et al
      . The Danish Cancer Registry—history, content, quality and use. Dan Med Bull 1997;44:535–9.
      OpenUrlPubMedWeb of Science
      1. Swerdlow AJ
      . Cancer registration in England and Wales: some aspects relevant to interpretation of the data. J R Stat Soc 1986;149:146–60.
      OpenUrl
      1. Teppo L,
      2. Pukkala E,
      3. Lehtonen M
      . Data quality and quality control of a population-based cancer registry. Experience in Finland. Acta Oncol 1994;33:365–9.
      OpenUrlCrossRefPubMedWeb of Science
      1. Larsen IK,
      2. Småstuen M,
      3. Johannesen TB,
      4. et al
      . Data quality at the cancer registry of Norway: an overview of comparability, completeness, validity and timeliness. Eur J Cancer 2009;45:1218–31.
      OpenUrlCrossRefPubMedWeb of Science
      1. Brewster DH,
      2. Crichton J,
      3. Harvey JC,
      4. et al
      . Completeness of case ascertainment in a Scottish regional cancer registry for the year 1992. Public Health 1997;111:339–43.
      OpenUrlCrossRefPubMedWeb of Science
      1. Barlow L,
      2. Westergren K,
      3. Holmberg L,
      4. et al
      . The completeness of the Swedish cancer register: a sample survey for year 1998. Acta Oncol 2009;48:27–33.
      OpenUrlCrossRefPubMedWeb of Science
    21. ↵
      1. Curado MP,
      2. Edwards B,
      3. Shin HR,
      4. et al.
      , eds. Cancer Incidence in Five Continents, Vol. IX. Lyon: IARC Scientific Publications, 2007.
    22. ↵
      1. Durrleman S,
      2. Simon R
      . Flexible regression models with cubic splines. Stat Med 1989;8:551–61.
      OpenUrlCrossRefPubMedWeb of Science
    23. ↵
      1. Clayton D,
      2. Schifflers E
      . Models for temporal variation in cancer rates. II: age-period-cohort models. Stat Med 1987;6:469–81.
      OpenUrlCrossRefPubMedWeb of Science
    24. ↵
      1. Carstensen B
      . Age-period-cohort models for the Lexis diagram. Stat Med 2007;26:3018–45.
      OpenUrlCrossRefPubMedWeb of Science
    25. ↵
      1. Bray F,
      2. Carstensen B,
      3. Møller H,
      4. et al
      . Incidence trends of adenocarcinoma of the cervix in 13 European countries. Cancer Epidemiol Biomarkers Prev 2005;14:2191–9.
      OpenUrlAbstract/FREE Full Text
    26. ↵
      1. Kim HJ,
      2. Fay MP,
      3. Feuer EJ,
      4. et al
      . Permutation tests for joinpoint regression with applications to cancer rates. Stat Med 2000;19:335–51.
      OpenUrlCrossRefPubMedWeb of Science
    27. ↵
      1. Hansson LE,
      2. Sparén P,
      3. Nyrén O
      . Increasing incidence of both major histological types of esophageal carcinomas among men in Sweden. Int J Cancer 1993;54:402–7.
      OpenUrlPubMedWeb of Science
    28. ↵
      1. Pohl H,
      2. Welch HG
      . The role of overdiagnosis and reclassification in the marked increase of esophageal adenocarcinoma incidence. J Natl Cancer Inst 2005;97:142–6.
      OpenUrlAbstract/FREE Full Text
    29. ↵
      1. Pohl H,
      2. Sirovich B,
      3. Welch HG
      . Esophageal adenocarcinoma incidence: are we reaching the peak? Cancer Epidemiol Biomarkers Prev 2010;19: 1468–70.
      OpenUrlAbstract/FREE Full Text
    30. ↵
      1. Lindblad M,
      2. Ye W,
      3. Lindgren A,
      4. et al
      . Disparities in the classification of esophageal and cardia adenocarcinomas and their influence on reported incidence rates. Ann Surg 2006;243:479–85.
      OpenUrlCrossRefPubMedWeb of Science
    31. ↵
      1. McColl KE,
      2. Going JJ
      . Aetiology and classification of adenocarcinoma of the gastro-oesophageal junction/cardia. Gut 2010;59:282–4.
      OpenUrlFREE Full Text
    32. ↵
      1. Vizcaino AP,
      2. Moreno V,
      3. Lambert R,
      4. et al
      . Time trends incidence of both major histologic types of esophageal carcinomas in selected countries, 1973-1995. Int J Cancer 2002;99:860–8.
      OpenUrlCrossRefPubMedWeb of Science
    33. ↵
      1. Nyrén O,
      2. Adami HO
      . Esophageal cancer. In: Adami HO, Hunter D, Trichopoulos D, eds. Textbook of Cancer Epidemiology. New York: Oxford University Press, 2008:196–238.
    34. ↵
      1. Hvid-Jensen F,
      2. Pedersen L,
      3. Drewes AM,
      4. et al
      . Incidence of adenocarcinoma among patients with Barrett's esophagus. N Engl J Med 2011;365:1375–83.
      OpenUrlCrossRefPubMedWeb of Science
    35. ↵
      1. Lagergren J,
      2. Bergström R,
      3. Lindgren A,
      4. et al
      . Symptomatic gastroesophageal reflux as a risk factor for esophageal adenocarcinoma. N Engl J Med 1999;340: 825–31.
      OpenUrlCrossRefPubMedWeb of Science
    36. ↵
      1. El-Serag HB
      . Time trends of gastroesophageal reflux disease: a systematic review. Clin Gastroenterol Hepatol 2007;5:17–26.
      OpenUrlCrossRefPubMedWeb of Science
    37. ↵
      1. van Soest EM,
      2. Dieleman JP,
      3. Siersema PD,
      4. et al
      . Increasing incidence of Barrett's oesophagus in the general population. Gut 2005;54:1062–6.
      OpenUrlAbstract/FREE Full Text
    38. ↵
      1. Coleman HG,
      2. Bhat S,
      3. Murray LJ,
      4. et al
      . Increasing incidence of Barrett's oesophagus: a population-based study. Eur J Epidemiol 2011;26:739–45.
      OpenUrlCrossRefPubMed
    39. ↵
      1. O'Doherty MG,
      2. Freedman ND,
      3. Hollenbeck AR,
      4. et al
      . A prospective cohort study of obesity and risk of oesophageal and gastric adenocarcinoma in the NIH-AARP Diet and Health Study. Gut. Published Online First: 15 December 2011. doi:10.1136/gutjnl-2011-300551
    40. ↵
      1. Lagergren J
      . Influence of obesity on the risk of esophageal disorders. Nat Rev Gastroenterol Hepatol 2011;8:340–7.
      OpenUrlCrossRefPubMed
    41. ↵
      WHO Global Database on Body Mass Index. http://apps.who.int/bmi/index.jsp
    42. ↵
      1. Edgren G,
      2. Liang L,
      3. Adami HO,
      4. et al
      . Enigmatic sex disparities in cancer incidence. Eur J Epidemiol 2012;27:187–96.
      OpenUrlCrossRefPubMed
    43. ↵
      1. Lepage C,
      2. Rachet B,
      3. Jooste V,
      4. et al
      . Continuing rapid increase in esophageal adenocarcinoma in England and Wales. Am J Gastroenterol 2008;103: 2694–9.
      OpenUrlCrossRefPubMedWeb of Science
    44. ↵
      1. Bhat S,
      2. Coleman HG,
      3. Yousef F,
      4. et al
      . Risk of malignant progression in Barrett's esophagus patients: results from a large population-based study. J Natl Cancer Inst 2011;103:1049–57.
      OpenUrlAbstract/FREE Full Text
    45. ↵
      1. de Jonge PJ,
      2. van Blankenstein M,
      3. Looman CW,
      4. et al
      . Risk of malignant progression in patients with Barrett's oesophagus: a Dutch nationwide cohort study. Gut 2010;59:1030–6.
      OpenUrlAbstract/FREE Full Text
    46. ↵
      1. Freedman ND,
      2. Derakhshan MH,
      3. Abnet CC,
      4. et al
      . Male predominance of upper gastrointestinal adenocarcinoma cannot be explained by differences in tobacco smoking in men versus women. Eur J Cancer 2010;46:2473–8.
      OpenUrlCrossRefPubMed
    47. ↵
      1. Cronin-Fenton DP,
      2. Murray LJ,
      3. Whiteman DC,
      4. et al
      . Reproductive and sex hormonal factors and oesophageal and gastric junction adenocarcinoma: a pooled analysis. Eur J Cancer 2010;46:2067–76.
      OpenUrlCrossRefPubMed
    48. ↵
      1. Masaka T,
      2. Iijima K,
      3. Endo H,
      4. et al
      . Gender differences in oesophageal mucosal injury in a reflux oesophagitis model of rats. Gut. Published Online First: 27 January 2012. doi:10.1136/gutjnl-2011-301389
    49. ↵
      1. Islami F,
      2. Kamangar F
      . Helicobacter pylori and esophageal cancer risk: a meta-analysis. Cancer Prev Res (Phila) 2008;1:329–38.
      OpenUrlAbstract/FREE Full Text
    50. ↵
      1. Banatvala N,
      2. Mayo K,
      3. Megraud F,
      4. et al
      . The cohort effect and Helicobacter pylori. J Infect Dis 1993;168:219–21.
      OpenUrlAbstract/FREE Full Text
    51. ↵
      1. Pounder RE,
      2. Ng D
      . The prevalence of Helicobacter pylori infection in different countries. Aliment Pharmacol Ther 1995;9(Suppl 2):33–9.
      OpenUrlPubMedWeb of Science
    52. ↵
      1. de Martel C,
      2. Parsonnet J
      . Helicobacter pylori infection and gender: a meta-analysis of population-based prevalence surveys. Dig Dis Sci 2006;51:2292–301.
      OpenUrlCrossRefPubMedWeb of Science

    Supplementary materials

    • Supplementary Data

      This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

      Files in this Data Supplement:

    • Press release

      Press release

      Files in this Data Supplement:

    Footnotes

    • Some of the data in this article are from the Cancer Registry of Norway. The Cancer Registry of Norway is not responsible for the analysis or interpretation of the data presented.

    • Funding This work was supported by Karolinska Institutet, Distinguished Professor Award to H-OA (grant number: 2368/10-221). GE is supported by a postdoctoral grant from Svenska Sällskapet för Medicinsk Forskning (SSMF).

    • Competing interests None.

    • Ethics approval The analyses were solely based on publicly available data of population sizes and aggregate number of cancer cases and as such, Institutional Review Board approval was not deemed to be necessary.

    • Provenance and peer review Not commissioned; externally peer reviewed.

    • Data sharing statement Almost all of the data is publicly available from IARC.

    Linked Articles

    • Digest
      Highlights from this issue
      Emad El-Omar William Grady Alexander Gerbes Thomas Rösch
      Gut 2013; 62 i-ii Published Online First: 14 Sep 2013. doi: 10.1136/gutjnl-2013-305968
    • Correction
      Correction
      BMJ Publishing Group Ltd and British Society of Gastroenterology
      Gut 2013; 62 1820-1820 Published Online First: 07 Nov 2013. doi: 10.1136/gutjnl-2012-302412corr1