Estimates of Cancer Incidence in China for 2000 and Projections for 2005

Routine mortality surveillance is ongoing in China, conducted by the Center for Health Information and Statistics, Ministry of Health of China. These data, together with the results from the national mortality survey of 1990 to 1992, have been used to study cancer mortality trends (1) and to make estimates and projections at the national level (2) Because of population growth, aging of the population, and generally increasing incidence rates, cancer is becoming a huge health burden in China; if current cancer mortality trends are continued, the number of deaths from cancer will increase from 1.5 million in 2000 to 1.8 million in 2005. (2). However, although mortality does reflect one aspect of cancer burden (its most serious consequence), information on cancer incidence and survival is also essential for cancer control purposes in providing insight into the burden on society and the need for programs for prevention, early diagnosis, treatment, and aftercare (3).

Information on cancer incidence in China is rather sparse. Population-based cancer registries produce statistics on the incidence (and outcome) of cancer in a defined population and provide a framework for assessing and controlling the impact of cancer on the community (4). In a survey of Chinese cancer registration practices in 2002 (5), 48 population-based cancer registries were identified. They covered only 5.7% of the national population. The registries were mostly located in big cities, relatively developed areas, and in some high-risk areas for certain cancers. Many of the registries were established rather recently. Substantial variations in data sources, collection methods, data management, and quality control between the registries were found. The information from existing registries cannot, therefore, be simply pooled in order to estimate national patterns and trends of cancer incidence. However, it is possible to prepare an estimate by using data on incidence and mortality from the registries of mainland China meeting the criteria of quality for inclusion in the Cancer Incidence in Five Continents (6), together with the national mortality data described above.

Two sets of data were used in this study:

1. Estimated and projected national cancer mortality rates for the years 2000 and 2005 (2). These were based on a projection of cancer mortality rates from the second National Mortality Survey of 1990 to 1992 (7), using annual rates of change for 10 cancers and all other cancers combined, by age, sex, and urban/rural residence, derived from log-linear regression modeling of data for 1991 to 1999 from the surveillance system conducted by the Center for Health Information and Statistics, covering about 10% of the national population (8).

2. Cancer incidence and mortality information for the period 1993 to 1997 from seven population-based cancer registries in China as published in Cancer Incidence in Five Continents (ref. 6; Table 1): Beijing, Shanghai, Wuhan, Qidong, Jiashan, Cixian, and Changle (Tianjin Registry was excluded in the analysis since the registry carries out some correction of the mortality data it receives from the death registration system).

The data from all seven cancer registries were used in estimating the incidence for cancers of the esophagus and stomach; for other cancer sites, only data from five registries (without Cixian and Changle) were used. The number of cases and deaths, by sex and age group (0-44, 45-54, 55-64, 65-74, and 75+), were aggregated for each type of cancer. The number of cases in each registry i were weighed by the reciprocal of the square root of the average population at risk

As a first step, incidence was estimated for the 10 tcommon cancer sites included in the national mortality database (nasopharynx, esophagus, stomach, colon-rectum, liver, lung, female breast, cervix, leukemia, and bladder), as well as for “all other sites combined”.

The number of cases (I_Nij) for these cancers were calculated as the product of the estimated (or projected) national mortality (the age- and sex-specific number of deaths; M_Nij) for years 2000 and 2005 (2) and the corresponding incidence/mortality ratio (I_Rij/M_Rij) based on the aggregated cases and deaths (weighed as described) from the above cancer registries for the years 1993 to 1997.

where i refers to the age group and j to the sex.

A generalized log-linear model with Poisson error was fitted for the I_Rij/M_Rij (the incidence data offset by corresponding mortality) of the aggregated data for the selected cancer (10). The models were fitted by GLIM (11) and included explanatory terms for sex and age. For the category “all cancer sites”, estimates were obtained by summing the number of cases at the 10 major cancer sites, and the “all other cancers” category.

The validity of choosing a set of I_Rij/M_Rij based on pooled data from all cancer registries to estimate incidence from national mortality was examined in a sensitivity analysis, comparing the results obtained for esophageal and stomach cancer, with those from two variant methods:

1. (a) A stratified analysis, using two sets of I_Rij/M_Rij, one based on the registries in the large cities of Beijing, Shanghai, and Wuhan (“urban”), and the other for the registries serving the rural counties of Cixian, Changle, Jiashan, and Qidong (“rural 4”), together with the mortality from urban and rural areas of China (2).

2. (b) As for (a), but with the rural I_Rij/M_Rij based on the registries of Jiashan and Qidong alone (“rural 2”)—as would be available for all cancer sites other than esophagus and stomach.

To estimate the incidence of cancers other than the 10 sites for which national mortality estimates were available, the “all other cancers” category was partitioned according to the proportions observed in the aggregated cancer registry data set (weighed as described). Proportions, by age and sex, were obtained for 19 cancers [oral cavity, “other pharynx”, gall bladder and extrahepatic bile duct, pancreas, larynx, bone, melanoma of skin, other (nonmelanoma) skin cancer, corpus uteri, uterus NOS, ovary and uterine adnexa, prostate, testis, kidney or other unspecified urinary organs, brain and nervous system, thyroid, Hodgkin disease, non-Hodgkin lymphoma, multiple myeloma] plus “other” [shown in Table 2, classified according to the 10th revision of the International Classification of Disease (ICD-10); ref. 12].

The predicted age-specific number of new cases was applied to the 2000 census population (13) and the 2005 population projected by the UN (14) to obtain the estimated or projected age-specific incidence rates for the corresponding year.

Cumulative risks (up to 74 years of age) were calculated to give the net risk (as a percentage) which an individual would have of developing the cancer in question before the age of 75 in the absence of other causes of death (4).

where r_i = age group (i) specific incidence rate per 100,000.

The difference in the number of cases between 2000 and 2005 could be divided into a component due to changing cancer risk and a component due to changes in the population size and age structure. The component due to change in risk was calculated as the difference between the predicted cases in 2005 and the number of cases that would have been expected if the incidence rates in 2000 were applied to the population in 2005 (15).

For each cancer site in Table 2, the estimated age-specific and age-standardized incidence rates and the cumulative risks (up to age 74) in 2000 and 2005, by age and sex, are shown in Table 3. Overall cancer incidence (all sites combined) is predicted to increase slightly from 2000 to 2005 in both sexes, from 209.2 to 210.8 per 100,000 in men, and from 133.6 to 140.6 per 100,000 in women. The increases are more marked in the older age groups (over age 65). The cumulative risks were about 21% for men and 14% for women during the 5-year period.

The five leading cancers in terms of incidence in the year 2000 were cancers of the lung, stomach, liver, esophagus, and colon-rectum for men, and cancers of the breast, stomach, lung, liver, and esophagus for women. In the year 2005, the second and third ranks are reversed in both sexes, the others are the same as in 2000. During these 5 years, increasing incidence rates are anticipated for cancers of lung, liver, colon-rectum, prostate and leukemia in men, and for breast, lung, liver, colon-rectum, and cervix in women. Decreases in incidence are predicted for stomach, esophagus and nasopharynx in men, and for esophagus, stomach, nasopharynx, and leukemia in women (Table 3).

The estimated and projected number of new cases for the more common cancers in the year 2000 and 2005 are shown in Table 4. The two components (changes in cancer risk and in population size and age structure) that contribute to the difference in the number of cases between the 2 years are also shown in Table 4. The total estimated number of new cancer cases increased by 11.7% in men (from 1.3 to 1.4 million) and 19.3% in women (from 0.8 to 1.0 million) between 2000 and 2005. Only cancers of the esophagus (for both sexes) and stomach (in men) showed a decline in the number of cases during these 5 years.

Very large increases in the number of cases of lung cancers (26.9% increase in men and 38.4% in women) and in female breast cancers (+38.5%; Table 4) are predicted. The age-specific incidence and mortality rates for males and females estimated for these two cancers in the years 2000 and 2005 are shown in Figs. 1, 2,and 3. The rates increase in all age groups for both incidence and mortality for female breast cancer, but the increase is especially marked at ages 45 to 64. This large increase in cancer risk is responsible for a 27.5% increase in the number of cases, whereas population growth and aging contribute a further 11% increase. The result is that breast cancer shows the biggest proportional increase in the number of new cancer cases in women during the 5-year period, and remains the leading cancer of women. For lung cancer, increases were found in both incidence and mortality rates among all age groups (except for age group 55-64) in both sexes, the oldest age group (over age75) had the highest rates for both incidence and mortality. According to our estimates, an additional 120,000 new lung cases will occur between 2000 and 2005 (from 0.38 million in 2000 to 0.50 million cases in 2005), and the total number of lung cancer cases will increase 26.9% in men and 28.4% in women (Table 4).

Table 5 shows the results of the sensitivity analysis. The incidence of esophageal and stomach cancer, estimated using national mortality rates, and a model of incidence/mortality based on all registries [pooled (7)] is only slightly greater than the estimate based on a stratified analysis, with four rural registries [rural (4)] contributing to the rural model. The incidence estimates based on a stratified analysis with only Jiashan and Qidong contributing to the rural model [rural (2)] is clearly very different, with incidence estimates considerably below the other two models, for most sex-site combinations. Even the result using only two rural registries in a combined model [pooled (5)] provides results closer to the full stratified model [urban + rural (4)] for two of the four sex-site combinations than does the urban + rural (2) model; this is the choice available for all sites other than esophagus and stomach.

With one-fifth of the world population, the cancer incidence and mortality profile in China plays a very important role in determining the cancer burden at a global level. The cancer incidence estimates for the year 2000 in China published in Globocan 2000 (16) were directly derived from three cancer registries (Shanghai, Tianjin, and Qidong) from 1988 to 1992 (17) and the Center for Health Information and Statistics mortality data for the year 1998. However, the areas contributing mortality data to Center for Health Information and Statistics are by no means a representative sample of the national population (18), and use of data from a single year adds to the uncertainty of the estimate. Furthermore, in the last decade, large changes in cancer risks have occurred in China (1). In this paper, we make use of past trends in mortality, and a larger number of cancer registries, to prepare a more accurate evaluation of cancer incidence in China.

From the theoretical point of view, incidence and prevalence can be estimated from mortality and survival data (19). However, this is virtually impossible for China, where population-based survival statistics are available for very few populations, and the data quality uncertain. The methodology used in this paper is consistent with previous studies estimating cancer burden in the European Community, and has been shown to estimate the incidence accurately (10, 20, 21). It is also regularly used for global cancer burden evaluation (22). The Poisson regression model for estimating the number of cases has been shown to be superior to methods that assume a normal distribution for the rates (23). However, as discussed by Jensen et al. (4), these evaluations are conservative according to the considerable extra-Poisson variation. Nevertheless, the overdispersion may have no practical influence on the validity of the average relation obtained and the extra-Poisson variation does not bias the regression coefficients as shown in previous studies in Europe (10, 21, 24).

National mortality estimates combined with the aggregates of cancer registries' incidence and mortality data were used to achieve stable incidence estimates at the national level. Ideally, if reporting the cause of death were completely accurate, the ratio of mortality to incidence would equal (1-survival probability) in a steady state, with constant incidence and survival. Incidence rates and survival are, however, changing in China, furthermore, no information on the validity of the cause of death is available for the routine mortality data used in our study. If there were systematic differences in classifying the causes of death between the areas covered by the cancer registries included in the model (“equation area”), and the national-level (“prediction area”), the predictions would be distorted; if the incidence/mortality ratio is higher in the prediction area than in the equation area, the incident cases would be underestimated. On the other hand, an overestimation would result if the ratio was higher in the equation areas (10).

The cancer mortality information we used here was derived from both mortality data used by cancer registries and estimates of national mortality (2) based on the national mortality survey of 1990 to 1992 and mortality trends from the routine mortality reporting system (Center for Health Information and Statistics data, 1991-1999). The mortality sources were the same in both cancer registry and routine death reporting systems.

The registry data used in this study were from the seven populations for which results were published in Cancer Incidence in Five Continents (6). The data pertain to the period 1993 to 1997; therefore, if there had been marked changes in survival in recent years, the incidence/mortality ratios may not have been appropriate for estimating incidence for mortality data from 2000 or 2005. There is very little information on trends in cancer survival in China, but it seems unlikely that this would be a major source of error in our estimates.

On the other hand, three of these registries were in large, modern conurbations (Beijing, Shanghai, and Wuhan), and although their results were weighed to reduce their influence on the pooled incidence/mortality ratios, it seems likely that the health care system and health facilities in these areas (equation areas) are probably superior to the national average level (prediction area); our estimation may therefore overestimate the national cancer incidence profile. The distortion of the registries might be reduced by stratifying the analysis, and preparing separate incidence estimates for urban and rural populations because corresponding urban/rural mortality estimates were available (2). This was done for two sites—esophagus and stomach in a sensitivity analysis, and the results suggest that the pooled model does indeed slightly (0.5-3%) overestimate incidence. However, for most sites, there were only two cancer registries covering counties classified as “rural” (Jiashan and Qidong). That such a small sample is no more likely to provide a representation of rural populations than the pooled (crude) analysis is shown by the sensitivity analysis, a stratified model based on data from these two registries gave results for esophagus and stomach quite different from those using all four rural registries, and that were no better than those from the model we used for all other cancer sites, in which the data from five registries were simply pooled (after weighing for population size).

Finally, we do not know whether the age- and sex-specific proportions of different cancers among the “other sites” in the cancer registries are reasonably representative of the situation in China as a whole. The two rural registries make a relatively small contribution to the proportions, even after weighing, although this may be no disadvantage—even though these registries cover all cancer sites, the motivation behind their establishment was to observe populations with known high incidence rates of specific cancers—liver cancer in Qidong and colon cancer in Jiashan.

The results suggest that there will be an overall increase in the number of cancer cases of 11.7% in men and 19.3% in women between 2000 and 2005. Even without any change in cancer risk, there would still have been a substantial increase of 10.7% in men and 13.2% in women due to the rapid population growth and aging in China. The number of cases of lung cancer (in both sexes) and of female breast cancer are expected to increase dramatically during the 5-year period, with the biggest component the consequence of the sharply increased risks. These two diseases will be a major focus of cancer control programs in China in the next decade.

The epidemic of tobacco smoking in China is certainly a major factor underlying past and future increasing trends of lung cancer incidence and mortality. It will also contribute to changes for the other smoking-related cancers, such as stomach, esophagus, liver, bladder, and leukemia (25-27). It has been estimated that if current smoking patterns persist in China, about if 100 million Chinese men now aged 0 to 29 would be killed by smoking, and this figure would reach 3 million a year by the year 2050 (25, 26). Controlling the tobacco epidemic might well be the greatest long-term challenge for public health in China at the beginning of the 21st century.

Liver cancer is predicted to be the second most common cancer in China in 2005, 59.6% of cases are caused by chronic infection with hepatitis B (28). Vaccination against hepatitis B virus was introduced nationwide as a routine infant immunization project from 2002 (29). The protective efficacy against the development of disease or the carrier state is often 95% to 99% in cohorts of immunized infants (30). A direct reduction in liver cancer in cohorts of immunized children has already been shown in Taiwan (31).

In China, especially in urban areas, lifestyles are becoming more westernized, with respect to diet (and alcohol consumption), reduction in physical activity, obesity (especially among the young), late age at childbearing, and low fertility. One or more of these factors are probably responsible for the increasing incidence of cancers of the large bowel, breast and prostate. For breast cancer, early detection and effective treatment seem to offer the most realistic approach to reducing mortality. Screening programs have been introduced in a few urban areas, mostly confined to special occupational groups, due to high costs involved. A randomized trial of screening based upon breast self-examination in textile mills in Shanghai suggested that the technique was ineffective in reducing mortality (32). The efficacy and cost-effectiveness of different approaches to early diagnosis of breast cancer need to be investigated further in China. Other screening programs for colorectal cancer, prostate cancer, lung cancer, and stomach cancer are even less feasible at present because of doubtful cost-effectiveness and lack of appropriate resources.

Even those cancers for which incidence are declining—stomach and esophagus—will remain a significant burden in the year 2005; stomach cancer will be the third most common malignancy and esophageal cancer the fourth. Studies in China, as elsewhere, have suggested an important role for diet (33-35) and infection with Helicobacter pylori (36-38) in the etiology of stomach cancer, and provide a lead to potential protection programs. On the other hand, some locally popular customs, such as drinking green tea (39, 40) or consuming tofu, and ginger have been suggested to have a possible protective effect on stomach cancer (41, 42). Esophageal cancer has a very uneven geographic distribution in China, with foci of high incidence in central and western provinces, especially around the Taihang Mountains (43-45). A large amount of research has suggested that micronutrient deficiencies in these areas may be responsible (46, 47). Screening methods based on balloon cytology (48) and endoscopic examination (49) remain an essentially experimental approach. A strategy for esophageal cancer control in this high-risk area, called “Taihang Anti-Cancer Campaign”, focusing on secondary prevention, has recently been proposed as a means of decreasing mortality from esophageal cancer in the future (44).

Estimation and projection of the cancer burden is clearly an essential step in planning an allocation of resources, but the methods used in our study must be considered only a surrogate for measuring incidence by means of the systematic registration of all cancer cases arising in the population. Expansion of the existing registry network, so that it covers a more representative sample of the national population, would increase the validity of the estimates. A 10% sample is a reasonable target (5). For the moment, this remains a long-term aspiration—in the meantime, we believe that the estimates based on the results of the few existing registries with high-quality data provide a fair overview of the problem, and indicate the priorities for cancer control at the national level.

We thank Profs. Risto Sankila and Anssi Auvinen, Tampere School of Public Health, Finland, for their helpful suggestions, and Freddie Bray, IARC/WHO, Lyon for advice on the methodology, and Chantal Déchaux, IARC/WHO, for assisting in the preparation of the manuscript.