Time trends of esophageal cancer in Hong Kong: Age, period and birth cohort analyses
Article first published online: 27 NOV 2006
Copyright © 2006 Wiley-Liss, Inc.
International Journal of Cancer
Volume 120, Issue 4, pages 853–858, 15 February 2007
How to Cite
Tse, L.-a., Yu, I. T.-s. and Mang, O. W.-K. (2007), Time trends of esophageal cancer in Hong Kong: Age, period and birth cohort analyses. Int. J. Cancer, 120: 853–858. doi: 10.1002/ijc.22382
- Issue published online: 27 DEC 2006
- Article first published online: 27 NOV 2006
- Manuscript Accepted: 13 SEP 2006
- Manuscript Received: 15 JUL 2006
- time trends;
- esophageal cancer;
- age-period-cohort modeling;
This study was to examine the time trend of the incidence rates of esophageal cancer during the period 1979–2003 in Hong Kong and to identify the effects of year of diagnosis (period) and year of birth (cohort) on the observed time trends using regression models. Cancer incidence data were obtained from Hong Kong Cancer Registry and population data were from the Census and Statistics Department. Age-standardized incidence rates were computed by the direct method using the World population of 1966. Annual percentage change (APC) in incidence rate was calculated using nonlinear regression. Period and cohort effects were assessed using 2 separate Poisson regression models after adjusting for age. During the period 1979–2003, a steady decrease in the age-standardized incidence rate was observed for both males (APC = −3.38%, 95% confidence interval [CI]: −2.89%, −3.86%) and females (APC = −3.92%, 95% CI: −3.15%, −4.69%). The incidence rates were consistently higher among males than females. After the adjustment for age and with the period 1989–1993 or birth cohort of 1934–1938 as reference, the relative risk of more recent periods or birth cohorts significantly decreased. The age-cohort model provided a better description of the data than the age-period model. Given reasonable latency between exposures and esophageal cancer incidence, the declining birth cohort effects in the recent generations were in line with the increased intakes of fresh vegetables and decreased consumptions of alcohol drinking, tobacco smoking, and preserved foods observed in population, thus supported their importance in influencing the burden of esophageal cancer. © 2006 Wiley-Liss, Inc.
Esophageal cancer is relatively infrequent in most of the Western Europe and North American but overall is the 6th most common cancer in the world.1 Esophageal cancer incidence is also high in Hong Kong. According to the report from Hong Kong Cancer Registry, there were 445 new cases having occurred in 2003.2 Esophageal cancer is currently the 9th leading cause of cancer among Hong Kong males.2
Recent reports from endemic and some less endemic areas consistently showed downward trends on the overall incidence rates of esophageal cancer,3, 4, 5 while an increasing trend has still be seen in Africa Transkei where the socioeconomic class appears to be low.6 In many non-endemic areas, the time trends of overall esophageal cancer incidence rates appeared relatively stable during the last 2 decades with the exception of France where an annual 2% reduction in the incidence rate was observed.5, 7, 8 The overall stable or even a slightly increased incidence of esophageal cancer among these non-endemic areas was limited primarily to an increase in adenocarcinoma.5, 8
People inhabiting Linxian and Cixian of North China and Nanao Island of South China, previously identified to be at the highest risk for esophageal cancer in China, have shown a progressive decrease in the incidence since the mid-1970s and early 1980s, respectively.3, 4 The earliest decreases in incidence occurred in urban Shanghai in the early 1970s.3, 9 Such decline was paralleled by a 10-year lagged increased consumption of fresh fruits and vegetables among Shanghai population.10 A hospital-based case-control study showed that Singapore-born Chinese have only about 40% of the risk of esophageal cancer of that born in China; moreover, significantly elevated risk was associated with Teochew and Hokkien dialect groups.11 The striking geographic, racial and temporal variations in the occurrence of esophageal cancer could be attributable to environmental risk factors alone or their interactions with genetics.
The time trends of esophageal cancer incidence in Hong Kong have not been analyzed previously. It would be of interest and importance to examine how the time trends were influenced by the effects of birth cohort and calendar period of diagnosis. Understanding the underlying reasons behind the changing trends of the incidence rate will generate useful etiological hints for the prevention of esophageal cancer in Hong Kong and other places.
Material and methods
Source of data
Newly diagnosed esophageal cancer cases during the period 1979–2003 were obtained from the Hong Kong Cancer Registry. Registered esophageal cancer cases corresponded to the code of 150 from Ninth Revision of the International Classification of Disease (ICD-9) before 2000 and C15 from ICD-10 revision thereafter.
Population data during the corresponding period were obtained from Hong Kong Census and Statistics Department. Mid-year population data were employed in the calculation of the incidence rate.
Annual standardized incidence rate for esophageal cancer was calculated by the direct method using the World Health Organization (WHO) 1966 world standard population as the reference. Annual percentage change (APC) in incidence rate was calculated by using nonlinear regression under the assumption of same rate of change throughout the study period.
To study the effects of age, period and cohort effects on the time trends, esophageal cancer cases were regrouped into 14 five-year age groups, ranging from 20–24 to 85 or above, as very few cases of esophageal cancer were diagnosed at ages below 20. Multiple Poisson regressions were based on the period 1979–2003 and thus, covered 5 five-year periods. Based on the period and age groups, we calculated 18 partially overlapping 5-year birth cohorts from 1884–1888 to 1979–1983. Multiple Poisson regressions were fitted for each gender to examine the age, period and birth cohort effects on esophageal incidence using the SAS procedure GENMOD.12 The reference categories are the 50–54 years age group, the 1989–1993 period and 1934–1938 birth cohort. To correct for the possible underestimation of the true standard error, which may be caused by the phenomenon of over-dispersion that can exist in Poisson distribution data, we choose the ‘pscale’ option in the GENMON function.12 Three models, age alone, age-period and age-cohort, were generated. A model including all 3 time variables was not computed because a full age-period-cohort model may involve serious methodological difficulties.13 The goodness of fit of the models was assessed by the deviance. The closer the deviance to the degree of freedom, the better the model fit.13, 14
During the period 1979–2003, a total of 13,407 new cases of esophageal carcinomas were diagnosed (10,740 males and 2,667 females). Forty-eight cases (36 males and 12 females) with unknown age were excluded from further analyses.
Time trends of age-standardized and age-specific incidence rates
Age-standardized incidence rates of esophageal cancer among males and females are plotted in Figure 1. A steadily downward trend was seen among males from the rate of 19/100,000 in 1979 to 7.65/100,000 in 2003. Females also experienced a decline from 5.03/100,000 to 1.51/100,000 during the same period. The decline was reflected in all age groups (by 10-year interval) throughout the 25-year period, as shown in Figure 2.
The visual declines in the trends of the age-specific and age-standardized incidence rates were confirmed by the calculation of APC (Table I). The age-standardized incidence rates among males decreased annually by 3.38% (95% confidence interval [CI]: −2.89%, −3.86%) during the 25-year period. There was a bit faster declining for females with an APC of −3.92% (95 %CI: −3.15%, −4.69%). APC was significant in most age groups but shown somewhat larger variations among females. The most marked changes occurred in the age group 30–34 for both genders.
|APC||95% CI||APC||95% CI|
|20–24||0.14||13.00, −11.25||−1.38||8.34, −10.22|
|25–29||−6.90||3.92, −16.60||−3.56||3.82, −10.42|
|30–34||−7.41||−2.44, −12.13||−12.78||−1.73, −22.58|
|35–39||−6.63||−2.49, −10.60||−1.89||5.14, −8.45|
|40–44||−4.88||−2.73, −6.99||−3.32||2.42, −8.74|
|45–49||−4.94||−3.12, −6.73||−2.30||0.74, −5.26|
|50–54||−4.14||−3.01, −5.25||−5.12||−2.07, −8.08|
|55–59||−4.16||−3.44, −4.88||−4.69||−3.04, −6.31|
|60–64||−3.13||−2.31, −3.94||−6.01||−3.73, −8.23|
|65–69||−3.00||−2.22, −3.79||−2.96||−1.08, −4.80|
|70–74||−2.84||−1.71, −3.95||−4.03||−2.70, −5.35|
|75–79||−2.63||−1.47, −3.77||−4.04||−2.32, −5.72|
|80–84||−2.32||−0.70, −3.91||−0.77||0.69, −2.20|
|85 or above||−1.57||0.70, −3.80||−3.58||−1.78, −5.34|
|Age-standardized incidence rate||−3.38||−2.89, −3.86||−3.92||−3.15, −4.69|
Age, period and birth cohort modeling
Table II shows the results of the Poisson regression on the changes in deviance with the different models. The age-period and age-cohort models were highly statistically significant for both genders. The deviance of the age-cohort model was smaller than that of the age-period, suggesting a better model fit of the data in the age-cohort model. The pattern of deviance statistics among males was quite similar to that among females.
|Variables in the model||DF||Deviance||Deviance/DF||p-Value for χ2-square|
|Age + period||52||94.09||82.31||1.81||1.58||<0.0001||<0.0001|
|Age + cohort||39||55.41||45.14||1.42||1.16||<0.0001||<0.0001|
The relative risks (RR) of esophageal cancer incidence rates by time period and birth cohort were estimated based on 2 separate models with the adjustment of age and depicted in Figure 3. From the period 1979–1983 to 1999–2003, the RRs dropped dramatically by 51% for males and 57% for females. With the birth cohort 1934–1938 as reference, the RRs of consecutive birth cohorts reduced noticeably for both genders with a slightly greater cohort effect being observed among females. The rate of decrease appeared to be flatted for more recent birth cohorts and a slight increase in the latest generation; nevertheless, it should be noted that there were less data available for these groups.
Figure 4 displays graphically the age-specific incidence rate of esophageal cancer by successive birth cohorts of males and females respectively. The parallelism in the curves confirms the strong cohort effects on the overall incidence rates of esophageal cancer.
Differences by gender
The age-standardized incidence rates among males were 4.52–5.56 times high as that among females with an average rate ratio of 4.75 throughout the entire 25-year period. Male to female rate ratios were relatively stable during the period 1979–1997, but it appeared to rise since 1998 (Table III). On the other hand, the male to female rate ratio generally increased with age up to the peak age group of 55–59, and thereafter the differences by gender gradually diminished.
|85 or above||1.76||2.07||2.30||2.01||2.23||2.04|
|Age-standardized incidence rate||4.52||4.84||4.66||4.63||5.56||4.75|
A clear downward trend in age-standardized incidence rate of esophageal cancer was observed for both genders in Hong Kong during the period 1979–2003, as in many other prosperous countries or even less affluent areas.3, 4, 5, 9 The incidence was falling rapidly in all age groups and birth cohorts with more rapid decline in females.
Could the trends of the incidence rate of esophageal cancer reflect changes in diagnostic accuracy in Hong Kong? With the improved diagnostic ability, cancer previously attributed to the esophagus, especially in the gastroesophageal junction, may now be diagnosed as stomach cancer. A significantly decreased period effect seen in the overall incidence rates of esophageal cancer could have resulted from this. If there had been misclassifications in the diagnoses between cancers from esophagus and stomach in Hong Kong, one would expect that stomach cancer would have shown an increase or a lower rate of decline than that of esophageal cancer. Given similar time trends in the incidence rates and a relatively constant rate ratio from esophageal to gastric cancer over the same calendar periods,15 the decreasing trend seen in the observed esophageal cancer was probably genuine, as misclassification into gastric cancer in recent years could have led to a diminished rate ratio between esophageal and gastric cancer. On the other hand, the proportion of morphologically verified esophageal cancer cases among Hong Kong males increased from 73.4% (66% among females) in 1980 to 95.5% (95.5% among females) in 2003. Nevertheless, the proportion of squamous cell carcinoma (86–88% in males and 76–82% in females) and adenocarcinoma (6–7.5% in males and 9–17% in females) for both genders were relatively stable over the entire 25-year study period based on limited data available. Increasing morphologically verified cases and medical seeking behavior could account for the increasing trends of esophageal cancer but could not explain a decreased period effect.
The changes in the overall pattern of esophageal cancer incidence might be related to a shift on ICD coding system, as 2 revisions came into force in Hong Kong during the study period: the code 150 from ICD-9 before 2000 and code C15 from ICD-10 thereafter for the malignant tumors of esophagus. Nevertheless, misclassification of esophageal cancer due to such shifting was less likely because migration from ICD-9 to ICD-10 coding system totally matched in Hong Kong. Moreover, an observed steady decline in the time trends over study period and a consistency in the results from age-period model supports that the changes in the incidence of esophageal cancer were unlikely the results from such errors. Being a full member of the International Association of Cancer Registries under the WHO, there should be little doubt on the quality and comparability of cancer data between Hong Kong Cancer Registry and other member.16
Cohort effects reflect carcinogenic exposures in early life and would affect the cancer risk of the birth cohort throughout the whole lifetime. If changes in these exposures occur in the population over time, successive birth cohorts will go through changing degrees of exposures in early life, leading to a progressive decrease or increase in the risk of cancers throughout their life spans. The age-cohort model fitted well for esophageal cancer in both genders, suggesting changes in specific dietary and lifestyles as well as other environmental factors are likely to be involved. The significantly decreasing risk in the recent periods was more likely a reflection of the birth cohort effects than being genuine, as genuine period effects from better detection/diagnosis in recent periods would have resulted in significantly increasing trend rather than the decreasing trend seen in this study.
Heavy alcohol drinking and tobacco smoking have long been recognized as the 2 main risk factors for esophageal cancer for both squamous cell and adenocarcinoma subtypes.17, 18, 19, 20, 21, 22, 23 The risk of esophageal cancer due to alcohol drinking alone was much higher than smoking alone,18, 24 and their combined effect showed a very strong synergy in the development of esophageal cancer.25 A steadily decreasing prevalence of daily smoking among Hong Kong population was observed since the mid-1970, but an increasing prevalence among the younger age groups (15–29) should raise some concerns.26 The absolute proportions of alcohol-related drinks (including beer, Chinese wine and foreign-style alcoholic drinks) to the overall monthly expenditure by Hong Kong households decreased from 0.8% in 1963–1964 to 0.33% in 1994–1995 by 59% reduction.27 Given a reasonable latency period between exposures and esophageal cancer incidence, the progressive protective cohort effects in more recent generations could be associated with the population decreasing levels of alcohol drinking and tobacco smoking among Hong Kong males and females. However, if smoking was the major cause of oncogenesis of esophageal cancer, given certain years of lagged exposures, the incidence rate of esophageal cancer would be expected to continue to decrease in the coming decades, but a rebound in the younger generations would be possible.
Potential changes in dietary habits of population have been believed to play an important role in the etiology of esophageal carcinoma.4, 8, 18 With the economic advances in the recent few decades, the dietary habits of the Hong Kong population have changed substantially. Fresh meat and vegetables have become widely available and much more affordable, and pickled vegetables and salted fish favored by Cantonese in the past have no longer been a major component of the regular diet. High consumptions of fruits and vegetables have been associated with a reduction on the risk of esophageal cancer in many epidemiological studies.11, 18, 19 A nutritional intervention trial of multiple vitamin and minerals in Linxian during 1985–1991 revealed that a combination of beta-carotene, vitamin E and selenium supplements could effectively lowing the risk of esophageal and gastric cardia cancer.28, 29 It was estimated that the combined attributable risk due to dietary risk factors, smoking and alcohol drinking were 89% according to a case-control study in Hong Kong.18 During the period 1973–1995, the trends in yearly kilograms of intake per person consumption of fresh vegetables significantly increased by 95% in Hong Kong population.27 The progressive protective effects in more recent birth cohorts suggest that the decline in incidence rates over the past 25 years might be linked with changes in these dietary risk factors, alcohol drinking and smoking.
Genetic risk factors shifting across birth cohorts might influence the trends the incidence of esophageal cancer.20 People from either Fukkien (Fujian) province or Teochew (Chaozhou) in Southeast China have been recognized to be at high risk of esophageal cancer.11 However, there has been no substantial changes in the place of original (ethnic origin) among Hong Kong residents in recent decades,30, 31, 32 implying a stable genetic background in the Hong Kong population. The esophagus has been shown the only site where the China-born have a substantially higher risk according to a case-control study in Singapore.11 It implies that the temporal and geographical variations on the incidence of esophageal cancer are the results of environmental, genetic and the interactive effects between them.
Other risk factors, such as Barrett's esophagus and occupational exposure to silica dust,8, 33 are likely in the etiology of esophageal cancer. However, to date, evidence on these aspects is very sparse and in demand.
So far, little has been known about the role of sex chromosomes in the etiology of oncogenesis. Our study showed that male population on the whole increased 4.5–5.6 times the risk of esophageal cancer when compared with female population (Table III). Such male excess was observed in almost all age groups throughout the entire study period except for the youngest population. The highest differences occurred in the middle aged working population (30–59) but diminished with the increase of age. Given more males being exposed to tobacco smoking and alcohol drinking and/or occupationally exposed to silica dust than females in middle aged group, our study suggests that the differences in esophageal cancer incidence by gender could be better explained by the different pattern of occupational and/or environmental risk factors exposures rather than by sex chromosome itself.
In conclusion, given a reasonable latency between exposures and esophageal cancer incidence, the declining birth cohort effects in the recent generations were in line with the increased intakes of fresh vegetables and decreased consumptions of alcohol drinking, tobacco smoking and preserved foods observed in population, thus supported their importance in influencing the burden of esophageal carcinoma. The hypotheses generated from this ecological study need to be tested by subsequent analytic studies.
- 2Hong Kong Cancer Registry. Hong Kong Cancer Stat 2003. Hong Kong: Hong Kong Cancer Registry, Hospital Authority, 2006.
- 9Cancer incidence trends in urban Shanghai, 1972–1999. Tumor 2004; 1: 11–15 (in Chinese)., , , , , , , , .
- 10A study of association between cancer incidence and diet in Shanghai. Tumor 1987; 7: 68–70 (in Chinese)., .
- 12SAS Institute Inc. SAS/STAT user's guide, Release 6.03. Cary, NC: SAS Institute, 1988.
- 26Tobacco Control Office. Hong Kong smoking prevalence. http://www.tobaccocontrol.gov.hk/chi/loadframe.html?id=102. Accessed on 10 January 2006.
- 29Prevention of esophageal cancer: the nutrition intervention trials in Linxian, China. Linxian Nutrition Intervention Trials Study Group. Cancer Res 1994; 54 ( 7 Suppl): S2029–S31., , , , , , .
- 30Hong Kong Census and Statistics Department. Hong Kong statistics 1947–1967. Hong Kong: Government Printer, 1969. 22 p.
- 31Hong Kong Census and Statistics Department. Hong Kong 1981 census: Main report. Hong Kong: Government Printer, 1982.
- 32Hong Kong Census and Statistics Department. Hong Kong 1991 population census. Hong Kong: Government Printer, 1992. 49 p.