Document KJOXQEJzomD9oMYJYrq6e5Lar

Int. J. Cancer: 116, 963-971 (2005) 2005 Wiley-Liss, Inc. Role of smoking in global and regional cancer epidemiology: current patterns and data needs Majid Ezzati1*, S. Jane Henley2, Alan D. Lopez3 and Michael J. Thun2 1Department of Population and International Health, Harvard School of Public Health, Boston, MA, USA 2Epidemiology and Surveillance Research, American Cancer Society, Atlanta, GA, USA 3School of Population Health, University of Queensland, Brisbane, Australia Although smoking is widely recognized as a major cause of cancer, there is little information on how it contributes to the global and regional burden of cancers in combination with other risk factors that affect background cancer mortality patterns. We used data from the American Cancer Society's Cancer Prevention Study II (CPS-II) and the WHO and IARC cancer mortality databases to estimate deaths from 8 clusters of site-specific cancers caused by smoking, for 14 epidemiologic subregions of the world, by age and sex. We used lung cancer mortality as an indirect marker for accumulated smoking hazard. CPS-II hazards were adjusted for important covariates. In the year 2000, an estimated 1.42 (95% CI 1.27-1.57) million cancer deaths in the world, 21% of total global cancer deaths, were caused by smoking. Of these, 1.18 million deaths were among men and 0.24 million among women; 625,000 (95% CI 485,000-749,000) smoking-caused cancer deaths occurred in the developing world and 794,000 (95% CI 749,000 840,000) in industrialized regions. Lung cancer accounted for 60% of smoking-attributable cancer mortality, followed by can cers of the upper aerodigestive tract (20%). Based on available data, more than one in every 5 cancer deaths in the world in the year 2000 were caused by smoking, making it possibly the single largest preventable cause of cancer mortality. There was signifi cant variability across regions in the role of smoking as a cause of the different site-specific cancers. This variability illustrates the importance of coupling research and surveillance of smoking with that for other risk factors for more effective cancer prevention. 2005 Wiley-Liss, Inc. Key words: smoking; cancer mortality; cause of death; risk assess ment; lung cancer Global cancer mortality rose from 6 million in 19901 to 7 mil lion in 2000.2 Aging of the world's population is a major factor contributing to the increase in cancer mortality because the inci dence of most cancers increases with age. This demographic change is coupled with an epidemiologic shift, in which age-spe cific mortality has increased for some cancers (e.g., lung and breast) and decreased for others (e.g, stomach and cervix in many industrialized countries). This epidemiologic shift is in turn driven by changes in risk factors for cancers of different sites, including diet and the environment, infectious agents (e.g, for cervix uteri, liver and stomach cancers), behavioral risks such as alcohol and tobacco use and reproductive risks and behaviors. Tobacco smok ing is currently the most widespread source of exposure to known carcinogens in the world and is causally associated with at least 15 types of cancer, as summarized in a review by the IARC and the 2004 report of the U.S. surgeon general.3-6 Recent estimates of global cancer mortality attributable to smoking were based on cancers of multiple sites combined.7 These estimates are important for motivating global tobacco-con trol efforts such as the FCTC. Combined estimates do not, how ever, address regional differences in the background risks of spe cific cancer sites. This limitation arises because smoking affects cancer incidence and mortality in combination with other behavio ral and environmental factors, all with important population-spe cific dynamics and potentially different biologic mechanisms. For example, coal is used for household heating and/or cooking in many regions of China8,9 and has given rise to high background mortality from lung cancer, which is magnified by smoking.10 Fuel use patterns in China are, however, changing, with wealthier coastal communities shifting from coal to cleaner fuels (e.g., natu ral gas) and rural inland communities switching from wood to coal to avoid deforestation.9 Chewing betel-quid with tobacco, com mon in many parts of southern Asia, is a risk factor for oral cancer in combination with smoking and has been affected by changing social norms.11-14 Alcohol consumption, an important risk factor for oral, esophageal and liver cancers, is currently increasing in some developing countries.15 Cancers with infectious risk factors also exhibit population-specific dynamics, ranging from a consis tent decline in stomach cancer in many countries during much of the 20th century to an increase in cervix uteri cancer in some developing regions (e.g., sub-Saharan Africa). Addressing the large and increasing GBD from cancers there fore requires an understanding of the role of smoking in combina tion with other risk factors. Data for multirisk factor models are extremely rare, especially in developing countries. In this report, we use 2 unique data sources, the American Cancer Society CPSII as well as the WHO and the IARC cancer mortality databases, to estimate site-specific cancer mortality caused by smoking in the year 2000 for 8 clusters of site-specific cancers and 14 epidemio logic subregions. Estimates include cervix uteri, liver and stomach cancers, which have been identified as causally related to smoking in the recent IARC review and the 2004 report of the U.S. surgeon general.3-5 This report also provides estimates of site-specific can cer mortality for other cancers caused by smoking in different world regions, thus updating a similar attempt made in 1985 based on cancer incidence.16 New estimates are important because smoking increased in most developing countries over the last quarter of the 20th century, with an estimated 930 million of the world's 1.1 billion smokers currently living in the developing world.17,18 A second contribution of this report is the use of RRs that are systematically and consistently adjusted for potential confounding covariates such as alcohol, diet and occupational exposures. Pre vious estimates of smoking-caused cancer mortality, whether in aggregate or site-specific estimates from 1985,7,16,19 used hazard estimates with no or partial adjustments for major covariates. The Abbreviations: AFR, African region; AMR, region of the Americas; CI, confidence interval; Clc, age- and sex-specific lung cancer mortality rate in the study population; CPS-II, Cancer Prevention Study phase II; EMR, eastern Mediterranean region; EUR, European region; FCTC, Framework Convention on Tobacco Control; GBD, global burden of disease; HBV/ HCV, hepatitis B/C virus; HPV, human papillomavirus; Nlc, age- and sexspecific lung cancer mortality rate for never-smokers in the study popula tion; N*lc, age- and sex-specific lung cancer mortality rate for neversmokers in the reference population; P, prevalence; PAF, population-attrib utable fraction; RR, relative risk; SEAR, Southeast Asia region; SIR, smoking impact ratio; S*lc, age- and sex-specific lung cancer mortality rate for current smokers in the reference population; WPR, Western Pacific region. Grant sponsors: National Institute on Aging. Grant number: PO1AG17625; CPS-II is funded by the American Cancer Society. *Correspondence to: Harvard School of Public Health, 665 Huntington Avenue, Boston, MA 02115, USA. Fax: 1617-566-0365. E-mail: mezzati@hsph.harvard.edu Received 25 November 2004; Accepted after revision 26 January 2005 DOI 10.1002/ijc.21100 Published online 4 May 2005 in Wiley InterScience (www.interscience. wiley.com). Publication oftheInternational Union Against Cancer K<c&sJ cancer control 964 EZZATI ET AL. TABLE I - RR OF MORTALITY FROM DIFFERENT CANCERS FOR CPS-II CURRENT SMOKERS RELATIVE TO NEVER-SMOKERS1 Cancer site (ICD-9 code) Additional covariates adjusted for Male Female Trachea, bronchus and lung cancer (162) Upper aerodigestive tract cancer (140-150) Stomach cancer (151) Liver cancer (155) Pancreatic cancer (157) Cervix uteri cancer (180) Bladder cancer (188) Myeloid leukemia (205)2 Kidney and other urinary cancers (223)3 Colorectal cancer (153-154)4 Occupational exposure to asbestos Alcohol consumption Family history of stomach cancer, intake of high-fiber grain foods and aspirin use frequency and duration Alcohol consumption Alcohol consumption and aspirin use Alcohol consumption, aspirin use, age at menarche, number of live births, oral contraceptive use and menopausal status Alcohol consumption, aspirin use and history of kidney stones -- Alcohol consumption and aspirin use Family history of colorectal cancer, intake of high-fiber grain foods, alcohol consumption, aspirin use frequency and duration, body mass index and exercise 21.3 (17.7-25.6) 8.1 (5.7-11.7) 2.2 (1.8-2.7) 2.3 (1.5-3.8) 2.2 (1.7-2.8) N/A 3.0 (2.1-4.3) 1.89 (1.3-2.9) 2.5 (1.8-3.6) 1.3 (1.2-1.5) 12.5 (10.9-14.3) 6.0 (4.3-8.5) 1.5 (1.2-1.9) 1.5 (0.8-2.7) 2.2 (1.8-2.8) 1.50 (0.9-2.6) 2.4 (1.5--4.1) 1.30 (0.8-1.8) 1.5 (1.0-2.1) 1.4 (1.3-1.6) 'Data from Thun et al.,22-24 and reanalysis of CPS-II data. RRs were estimated from Cox proportional hazard models, with never-smokers as the reference group (RR = 1.0 for never-smokers). Numbers in parentheses are 95% CIs. All risks were adjusted for age, race, education, marital status, ``blue-collar'' employment in most recent or current job, weekly consumption of vegetables and citrus fruit and vitamin (A, C and E) use.22-24 Additional covariates adjusted for are given in column 2.-2The RR for myeloid leukemia was applied to all leukemias (ICD-9 204 208) because the WHO GBD database provides estimates of mortality for all leukemias. In regions where detailed cause of death data exist (established market economies), myeloid leukemia was 40-55% of all leukemias.-3In the WHO GBD database, kidney and other urinary cancers are a part of the residual category ``other cancers'', which also includes cancers of multiple other sites. For this reason, no estimates were made for these cancers using the WhO cancer mortality database; estimates were made using the IARC cancer mortality database.-4Evidence of a causal association between smoking and colorectal cancer was classified as suggestive in the 2004 report of the U.S. surgeon general.5 There fore, colorectal cancer was not included in the main estimates. use of unadjusted, multicancer hazards necessitated the use of arbitrary hazard correction factors to ensure that the risks of tobacco were not overestimated.7,19 This increased uncertainty in the estimates of cancer mortality caused by smoking. By using a comparable measure of exposure and systematically adjusted haz ards, the estimates presented here provide a consistent basis for assessing the consequences of smoking in different regions of the world for global cancer epidemiology, and for assessing how global and regional tobacco-control efforts, such as those initiated under the FCTC, may contribute to reducing smoking-caused can cer mortality. Material and methods RRs for site-specific cancer mortality We used data from the American Cancer Society CPS-II to esti mate the hazards (RRs) for mortality from multiple site-specific cancers caused by smoking, including adjustment for important covariates (Table I). CPS-II is a prospective study of mortality in 1.2 million Americans aged 30 and older who completed a ques tionnaire on tobacco and alcohol use, diet and multiple other risk factors in 1982, with the latest published follow-up in 1998. A complete description of the CPS-II study design is provided elsewhere.20-22 In 1992, when the first 6-year (1982-1988) results were obtained, mortality follow-up was virtually complete for the first 2 years and about 98-99% complete for the next 4. Analyses of smoking hazards in CPS-II were based on the first 6 years of follow-up (1982-1988) to maximize the number of deaths avail able for analysis, especially in never-smokers, while minimizing misclassification of exposure due to cessation of smoking during follow-up.21,22 Analyses for stomach and colorectal cancers were based on the 1996 follow-up, to increase the number of deaths.23,24 This may result in underestimation of the hazard if some baseline current smokers quit smoking during follow-up. RRs for mortality from different cancers among CPS-II current smokers, relative to never-smokers, are provided in Table I. Measuring exposure to accumulated smoking hazards The accumulated hazards of smoking for cancers--which have relatively long latency--depend on factors such as the age at which smoking began or stopped, duration of smoking, number of ciga rettes smoked per day, whether the smoked tobacco product was in the form of cigarettes or in other forms such as bidis or cigars, ciga rette characteristics and smoking behavior such as degree of inhalation.25 Many of these factors vary over time because of changes in the socioeconomic determinants of smoking, including income, tobacco-control efforts such as taxation and advertising laws and cul tural norms. Therefore, current smoking prevalence or tobacco con sumption alone would be an insufficient indicator of the accumulated hazards of smoking, even if detailed data were available in all coun tries. This is particularly important in low- and middle-income coun tries, where smoking has increased over the past few decades.17,18 To capture the accumulated hazard of smoking, we used the SIR method of Peto et al,19 adapted to the specific conditions of devel oping countries.26 The method uses lung cancer mortality data, which are available or can be estimated using various methods, as an indirect indicator of the accumulated hazards of tobacco smok ing. Background-adjusted SIR is defined as population lung cancer mortality in excess of never-smokers relative to excess lung cancer mortality for a known reference group of smokers, adjusted to account for differences in never-smoker lung cancer mortality rates across populations (equation 1).26 SIR is calculated by age and sex. Age groups were 30-44, 45-59, 60-69, 70-79 and 80+. No deaths before the age of 30 were attributed to smoking because there are few cancer deaths before 30 and RRs are unstable. SIR SLc Nlc x NLc NLc Nlc (1) Conceptually, by using excess lung cancer mortality as the indi cator of the accumulated hazards of smoking in both study and reference populations, SIR converts the current smokers in the SMOKING AND GLOBAL CANCER EPIDEMIOLOGY 965 TABLE II - GBD 2000 EPIDEMIOLOGIC SUBREGIONS1 WHO region Mortality stratum Countries Population (thousands) African Region D Algeria, Angola, Benin, Burkina Faso, Cameroon, (AFR) Cape Verde, Chad, Comoros, Equatorial Guinea, Gabon, Gambia, Ghana, Guinea, Guinea-Bissau, Liberia, Madagascar, Mali, Mauritania, Mauritius, Niger, Nigeria, Sao Tome and Principe, Senegal, Seychelles, Sierra Leone, Togo E Botswana, Burundi, Central African Republic, Congo, Cote d'lvoire, Democratic Republic of the Congo, Eritrea, Ethiopia, Kenya, Lesotho, Malawi, Mozambique, Namibia, Rwanda, South Africa, Swaziland, Uganda, United Republic of Tanzania, Zambia, Zimbabwe Region of the A Canada, Cuba, United States of America Americas (AMR) B Antigua and Barbuda, Argentina, Bahamas, Barbados, Belize, Brazil, Chile, Colombia, Costa Rica, Dominica, Dominican Republic, El Salvador, Grenada, Guyana, Honduras, Jamaica, Mexico, Panama, Paraguay, Saint Kitts and Nevis, Saint Lucia, Saint Vincent and the Grenadines, Suriname, Trinidad and Tobago, Uruguay, Venezuela D Bolivia, Ecuador, Guatemala, Haiti, Nicaragua, Peru Eastern Mediterranean B Bahrain, Cyprus, Iran (Islamic Republic of), Jordan, Kuwait, Lebanon, Region (EMR) Libyan Arab Jamahiriya, Oman, Qatar, Saudi Arabia, Syrian Arab Republic, Tunisia, United Arab Emirates D Afghanistan, Djibouti, Egypt, Iraq, Morocco, Pakistan, Somalia, Sudan, Yemen European Region A Andorra, Austria, Belgium, Croatia, Czech Republic, Denmark, Finland, France, (EUR) Germany, Greece, Iceland, Ireland, Israel, Italy, Luxembourg, Malta, Monaco, Netherlands, Norway, Portugal, San Marino, Slovenia, Spain, Sweden, Switzerland, United Kingdom B Albania, Armenia, Azerbaijan, Bosnia and Herzegovina, Bulgaria, Georgia, Kyrgyzstan, Poland, Romania, Slovakia, Tajikistan, the former Yugoslav Republic of Macedonia, Turkey, Turkmenistan, Uzbekistan, Yugoslavia C Belarus, Estonia, Hungary, Kazakhstan, Latvia, Lithuania, Republic of Moldova, Russian Federation, Ukraine Southeast Asia B Indonesia, Sri Lanka, Thailand Region (SEAR) D Bangladesh, Bhutan, Democratic People's Republic of Korea, India, Maldives, Myanmar, Nepal Western Pacific A Australia, Brunei Darussalam, Japan, New Zealand, Singapore Region (WPR) B Cambodia, China, Cook Islands, Fiji, Kiribati, Lao People's Democratic Republic, Malaysia, Marshall Islands, Micronesia (Federated States of), Mongolia, Nauru, Niue, Palau, Papua New Guinea, Philippines, Republic of Korea, Samoa, Solomon Islands, Tonga, Tuvalu, Vanuatu, Viet Nam 294,078 345,515 325,183 430,932 71,230 139,059 342,576 411,889 218,458 243,184 293,819 1,241,806 154,354 1,532,933 !Data from WHO.2-2A, very low child mortality and very low adult mortality; B, low child mortality and low adult mortality; C, low child mortality and high adult mortality; D, high child mortality and high adult mortality; E, high child mortality and very high adult mortality. study population, who may have different smoking histories, into equivalent current smokers in the reference population, where haz ards for other diseases (e.g., different cancers) have been meas ured.19 We used CPS-II as the reference population (S*LC and N*lc) because CPS-II data allowed separate hazard estimates for cancers of various sites and systematic adjustment for multiple covariates. SIR was calculated for each age group and sex and for individual countries, then averaged (population-weighted) in each of 14 epidemiologic subregions of the world (Table II). Most of the CPS-II current smokers were lifelong cigarette smokers, with a mean consumption of about 20 cigarettes/day.19 In applying the method to developing countries, we accounted for the role of coal, a common fuel in parts of the developing world, as an important determinant of lung cancer patterns among never-smokers. In populations that use coal in poorly vented stoves (e.g., China and parts of Southeast Asia), never-smoker lung cancer mortality rates (NLC in equation 1) were estimated based on data on lung cancer mortality among Chinese nonsmokers.10,26 For China, the observed rates in the retrospective mortality study of Liu et al.10 were used; for all other countries, a weighted average of CPS-II never-smoker and Chinese nonsmoker rates was used, with weights for Chinese rates equal to the preva lence of coal use in poorly vented stoves.8 The remaining risk fac tors for lung cancer mortality (e.g., ambient air pollution, occupa tional hazards, indoor air pollution from radon or biomass smoke) in various combinations affect a subset of the population of each country. It is unclear what the net impacts of these risks on the rel ative levels of lung cancer among never-smokers in different countries may be. Therefore, the impacts of other risk factors were considered as sources of uncertainty in never-smoker lung cancer mortality (Table III). Estimating tobacco-attributable mortality For each age, sex and cancer site, the fraction of deaths attribut able to smoking was estimated using the standard PAF relation ship (equation 2), with prevalence, P, set to the SIR for each age/ sex group and RR from Table I. For each cancer site, PAFs were multiplied by regional cancer mortality (from WHO or IARC database) to estimate site-specific cancer deaths caused by smok ing in 14 epidemiologic subregions of the world (Table II). P(RR - 1) P(RR - 1) + 1 (2) Analysis of uncertainty Despite the recent improvements in the data sources required for global estimates of smoking-caused mortality (e.g, mortality registration and adjusted hazards), there remains substantial uncer tainty about levels of smoking-attributable cancer mortality, espe cially in developing countries where data are more sparse. Some sources of uncertainty were addressed in a quantitative analysis: (i) uncertainty in the parameters used in calculating SIR, including 966 EZZATI ET AL. TABLE III - ESTIMATED CANCER MORTALITY AMONG PEOPLE 30 YEARS OF AGE AND OLDER ATTRIBUTABLE TO SMOKING IN DEVELOPING (301 POPULATION 1,874 MILLION) AND INDUSTRIALIZED (301 POPULATION 795 MILLION) REGIONS IN THE YEAR 20001 Male Female Total Developing2 Industrialized2 World WHO: 549,000 (435,000-637,000) IARC: 464,000 WHO: 630,000 (597,000-658,000) IARC: 664,000 WHO: 1,179,000 (1,058,000-1,278,000) IARC: 1,127,000 WHO: 75,000 (50,000-112,000) IARC: 64,000 WHO: 165,000 (151,000-182,000) IARC: 161,000 WHO: 240,000 (213,000-286,000) IARC: 225,000 WHO: 625,000 (485,000-749,000) IARC: 528,000 WHO: 794,000 (749,000-840,000) IARC: 824,000 WHO: 1,419,000 (1,272,000-1,565,000) IARC: 1,352,000 !In each cell, the first number shows estimates using the WHO cancer mortality database, num bers in parentheses show 95% CIs for these estimates and number after parentheses shows the esti mate using the IARC cancer mortality database. All estimates exclude colorectal cancer. -2Industrialized regions include AMR-A, EuR-A, EUR-B, EUR-C and WPR-A; developing regions include all other regions. This classification is based on WHO.2 See Table II for descriptions of regions. lung cancer mortality in each country based on its mortal ity reporting system (CLC in equation 1) levels of background lung cancer mortality among neversmokers as a result of exposure to other risk factors, such as occupational exposures and indoor air pollution from sources other than coal (NLC in equation 1) lung cancer mortality in CPS-II smokers and never-smokers (S*LC and N*LC in equation 1) (ii) statistical uncertainty (as a result of random error) in the estimated CPS-II RRs for site-specific cancer mortality among smokers (Table I) Uncertainty for each of the parameters was estimated sepa rately, as described elsewhere.7 Uncertainties for individual parameters were combined in a stratified sampling simulation27 to obtain the uncertainty distribution for the fraction of cause-spe cific mortality attributable to smoking. Uncertainty in lung cancer mortality (CLC) was assumed to be correlated across age/sex groups (i.e., underestimation of lung cancer in one age/sex group implies underestimation in other age/sex groups). Uncertainties of different parameters were assumed to be independent. The uncer tainty reported here is the 95% range of the combined distribution. The above quantitative analysis does not include a number of other sources of uncertainty. First, using lung cancer as the marker for accumulated smoking hazard for other cancers can introduce additional uncertainty because the role of factors such as lag in disease causation can vary across diseases and across cigarette types or other tobacco products (e.g., bidis or cigars). This source of uncertainty is likely to be smaller for cancers than for cardio vascular or respiratory diseases, where the mechanisms and time dimension of hazards may be fundamentally different. Second, there is uncertainty in WHO cancer mortality estimates to which the estimated PAFs are applied. This is discussed in detail elsewhere.28,29 Currently, 75 countries have reasonably complete vital registration and medical certification of deaths.28, Another 51 countries have incomplete vital statistics or use sample registration and/or surveillance systems. The remaining 65 coun tries, mostly in sub-Saharan Africa, have no reliable data on adult mortality. WHO, through its GBD project, uses standard demo graphic techniques30-32 to estimate all-cause death rates by age for these populations. Cause-of-death models are then used to esti mate the total number of cancer deaths for countries with poor data. The distribution of cancer deaths by site is based on regional incidence or mortality patterns from cancer registries reporting to the IARC33 or on cancer survival models when such data are unavailable. The IARC also provides estimates of cancer mortality by site for most WHO member states. The differences between the 2 reflects the direct use of reported cancer incidence or mortality from registries by the IARC and model-based corrections for underreporting by the WHO. The impact of these different approaches is discussed in detail elsewhere, including an assess ment of consistency and validity of WHO and IARC estimates.28,29 The net effect is that the WHO estimates of global cancer mortality are 11% higher than those of the IARC, with larger differences in the AFR, EMR and SEAR regions. Regional and cause-specific cancer mortality rates attributable to smoking in this report are pre sented using both databases (Tables III-V), both of which are subject to uncertainty. A third source of uncertainty involves the extrapolation of haz ards from CPS-II to other populations that have different exposure distributions for other cancer risk factors and background disease patterns. The nature of this source of uncertainty and its implica tions for research and surveillance are described in detail below (see Discussion). Results The estimated number of cancer deaths caused by smoking is summarized in Table III and presented by region and sex in Table IV. Deaths by region and cancer site are shown in Figure 1. Globally, an estimated 1.42 (95% CI 1.27-1.57) million cancer deaths were caused by smoking in the year 2000, 21% of total global adult (age 301) cancer mortality. The proportions of adult cancer mortality caused by smoking were 32% for males and 8% for females and 16% and 29% for developing and industrialized countries, respectively. The total number of smoking-caused can cer deaths using the IARC database was an estimated 1.35 million, or about 95% of the global total estimated using the WHO figures. Regional patterns The regions with the largest number of cancer deaths caused by smoking were the industrialized regions of western Europe (EURA; 272,000 deaths, 27% of all cancer deaths) and North America (AMR-A; 211,000 deaths, 34% of all cancer deaths) and the developing regions of the Western Pacific (WPR-B, dominated by China in terms of population; 244,000 deaths, 14% of all cancer deaths) and Southeast Asia (SEAR-D, dominated by India in terms of population; 190,000 deaths, 24% of all cancer deaths), followed by the remaining industrialized regions. The large number of can cer deaths attributable to smoking in WPR-B and SEAR-D is due to a combination of 2 factors: (i) the large number of total cancer deaths, due to both population size and other cancer risk factors, and (ii) accumulated hazards of smoking. For example, while the fraction of total cancer mortality was lower in WPR-B (14%) than SEAR-D (24%), the absolute number of smoking-caused cancer deaths was higher in the former (244,000 vs. 190,000), due to higher background cancer deaths. Higher background cancer deaths in turn were due to the larger population in WPR-B and other risk factors that affect background cancer levels (e.g., indoor coal smoke). SMOKING AND GLOBAL CANCER EPIDEMIOLOGY 967 TABLE IV - ESTIMATED CANCER MORTALITY ATTRIBUTABLE TO SMOKING BY REGION IN THE YEAR 20001 Region Adult (30+) population (millions) MF Smoking-attributable cancer mortality M F % Total adult (30+ ) cancer mortality M F M+F African region D 41 43 WHO: 10,000 (6,000-16,000) WHO: 1,000 (200-3,000) 91 IARC: 5,000 IARC: 400 African region E 47 50 WHO: 23,000 (17,000-31,000) WHO: 5,000 (3,000-8,000) 17 4 IARC: 12,000 IARC: 3,000 Region of the 91 98 WHO: 131,000(119,000-143,000) WHO: 80,000 (71,000-89,000) 42 26 Americas A IARC: 141,000 IARC: 80,000 Region of the 85 92 WHO: 48,000 (38,000-58,000) WHO: 12,000 (7,000-17,000) 27 6 Americas B IARC: 52,000 IARC: 12,000 Region of the 12 12 WHO: 2,000 (1,000-3,000) WHO: 300 (100-1,000) 61 Americas D 2,000 200 Eastern Mediterranean 25 22 WHO: 12,000 (8,000-16,000) WHO: 2,000 (400-4,000) 30 7 region B IARC: 12,000 IARC: 2,000 Eastern Mediterranean 52 52 WHO: 26,000 (16,000-35,000) WHO: 3,000 (1,000-6,000) 28 3 region D IARC: 24,000 IARC: 2,000 European region A 125 137 WHO: 225,000 (209,000-241,000) WHO: 47,000 (38,000-55,000) 40 10 IARC: 234,000 IARC: 43,000 European region B 50 54 WHO: 72,000 (62,000-81,000) WHO: 9,000 (7,000-12,000) 44 8 IARC: 70,000 IARC: 8,000 European region C 63 80 WHO: 133,000 (122,000-141,000) WHO: 11,000 (8,000-15,000) 49 5 IARC: 143,000 IARC: 12,000 Southeast Asia 61 62 WHO: 45,000 (22,000-63,000) WHO: 2,000 (200-6,000) 43 4 region B IARC: 32,000 IARC: 2,000 Southeast Asia 244 235 WHO: 174,000 (66,000-249,000) WHO: 16,000 (3,000-47,000) 43 4 region D IARC: 124,000 IARC: 11,000 Western Pacific 47 51 WHO: 69,000 (61,000-80,000) WHO: 18,000(14,000-22,000) 36 13 region A IARC: 76,000 IARC: 19,000 Western Pacific 374 367 WHO: 209,000 (156,000-262,000) WHO: 35,000 (11,000-61,000) 20 5 region B IARC: 201,000 IARC: 31,000 5 10 34 17 3 21 16 27 29 29 24 24 27 14 !In each cell, the first number shows estimates using the WHO cancer mortality database, numbers in parentheses show 95% CIs for these esti mates and the number after the parentheses shows estimates using the IARC cancer mortality database. All estimates exclude colorectal cancer. Figure 1 - Adult (age 30+) cancer mortality due to smoking in GBD epidemiologic subregions, by cancer site. Estimates are based on the WHO cancer mortality data base. See Table II for description of subregions. UADC, upper aerodigestive tract cancer. The proportions of total adult cancer deaths caused by smoking were generally highest in industrialized regions (27-34%), where smoking has a longer history (Table IV). At the same time, the rise in smoking over the last quarter of the 20th century in a number of developing regions, including parts of Latin America, the eastern Mediterranean and Southeast Asia, has resulted in 15% or more of all current adult cancer deaths being caused by smoking--higher among men (Table IV). Current cancer mortality due to smoking was larger among men than women (Tables III, IV), with 79% of smoking-attributable cancer deaths in industrialized regions and 88% in developing regions occurring among men. The exception to this pattern was North America (AMR-A), where women have smoked for several decades and female deaths accounted for 38% of total smokingattributable cancer mortality (note that women in AMR-A had a higher number of cardiovascular deaths attributable to smoking34). 968 EZZATI ET AL. TABLE V - ESTIMATED CANCER MORTALITY ATTRIBUTABLE TO SMOKING BY SITE IN THE YEAR 2000 AND TO OTHER MAJOR RISK FACTORS1 Cancer type Deaths caused by smoking % Caused by smoking Other major risk factors (PAF)2 Trachea, bronchus WHO: 848,000 (781,000-940,000) and lung cancer IARC: 800,000 Upper aerodigestive WHO: 287,000 (245,000-327,000) tract cancer IARC: 291,000 Stomach cancer Liver cancer Pancreatic cancer Cervix uteri cancer Bladder cancer Leukemia3 Kidney and other urinary cancers4 Colorectal cancer5 WHO: 94,000 (82,000-111,000) IARC: 85,000 WHO: 69,000 (51,000-91,000) IARC: 65,000 WHO: 47,000 (42,000-54,000) IARC: 44,000 WHO: 6,000 (3,000-13,000) IARC: 7,000 WHO: 47,000 (40,000-54,000) IARC: 39,000 WHO: 21,000 (14,000-29,000) IARC: 22,000 No estimates Other cancers6 0 71 39 11 12 21 3 27 12 No estimates 0 Low fruit and vegetable consumption (12%), urban air pollution (5%), indoor air pollution from household use of solid fuels (coal only) (1%), selected occupational carcinogens (8%) Low fruit and vegetable consumption (12% for esophageal cancer), alcohol use (16% for mouth and oropharygeal cancers, 27% for esophageal cancer) Low fruit and vegetable consumption (18%) Alcohol use (24%), contaminated injections in health-care settings (as an outcome of HBV and HCV infection) (27%) None of the other 25 analyzed risk factors Unsafe sex (defined as unprotected sex with an infected partner) (100% by definition) None of the other 25 analyzed risk factors Selected occupational carcinogens (3%) None of the other 25 analyzed risk factors Overweight and obesity (12%), low fruit and vegetable consumption (2%), physical inactivity (15%) 3In each cell, the first number shows estimates using the WHO cancer mortality database, numbers in parentheses show 95% CIs for these estimates and the number after the parentheses shows estimates using the IARC cancer mortality database.-2Source: Reanalysis of data on 26 major global risk factors.43 The figure shows the estimated cancer mortality for each risk factor considered individually. These risks act jointly with one another. Consequently, the burden due to groups of risk factors will usually be less than the sum of individual risks.38-3The RR for myeloid leukemia (ICD-9 205) was applied to all leukemias (ICD-9 204-208) because the WHO GBD database provides estimates of mortality for all leukemias. In regions where detailed cause of death data exist (established market economies), myeloid leukemia was 40-55% of all leukemias.-4In the WHO GBD database, kidney and other urinary cancers are a part of the residual category ``other cancers'', which also includes cancers of multiple other sites. For this reason, no estimates were made for these cancers using the WHO cancer mortality database; estimates using, the IARC cancer mortality database indicate approximately 20,000 kidney cancer deaths attributable to smoking.-5Evidence of a causal association between smoking and colorectal cancer was classified as suggestive in the 2004 report of the U.S. surgeon general.5 There fore, colorectal cancer was not included in the main estimates in Tables III and IV and Figures 1 and 2. If the evidence on causality is confirmed, the RR figures in Table I give an estimated 43,000 colorectal cancer deaths attributable to smoking in 2000.-6This group also includes cancers of the nasal cavity and paranasal sinuses (ICD-9 160) and larynx (ICD-9 161), for which there is sufficient evidence of causality.3,4 In developing regions, 67% of smoking-attributable cancer deaths were between the ages of 30 and 69 compared to 52% in industri alized regions. Cancer deaths at younger ages result in a larger loss of potential life years, as described in detail elsewhere.1 Site-specific patterns Globally, the leading cancers caused by smoking were lung can cer (848,000 deaths, 60% of all smoking-attributable cancer deaths) and upper aerodigestive tract cancer (287,000, 20% of all smoking-attributable cancer deaths) (Table V). The site distribu tions of smoking-attributable cancer deaths varied across develop ing and industrialized regions because of variations in both the site distributions of total cancer deaths and the fractions attributable to smoking (Figs. 1, 2). In developing regions, the 328,000 lung and 188.000 upper aerodigestive tract cancer deaths caused by smok ing comprised 53% and 30% of all smoking-attributable cancer deaths, respectively. In industrialized regions, 520,000 lung and 100.000 upper aerodigestive tract cancer deaths were estimated to have been due to smoking, comprising 65% and 12% of all smok ing-caused cancer deaths, respectively. The contribution of lung cancer to total smoking-attributable cancer deaths varied from a low of 30-40% in sub-Saharan Africa and parts of the eastern Mediterranean region to a high of 60-70% in North America and Europe (Fig. 1). Upper aerodigestive tract cancer had a considerably larger proportional contribution to total smoking-caused cancer deaths in some developing regions, espe cially in SEAR-D, a region dominated by India in terms of popula tion (43% of all smoking-caused cancer deaths). In this region, oral tobacco use and dietary factors result in high background levels and/or interact with smoking to result in increased risk for this group of cancers.11 The relatively large estimated PAF for upper aerodigestive tract cancer attributable to smoking in this work is consistent with recent direct estimates of Gajalakshmi et al.,35 who estimated that 52% of mouth, laryngeal and pharyn geal cancers in India and 36% of esophageal cancers for men between 25 and 69 years of age were caused by smoking. Because most smokers in SEAR-D are also oral tobacco users, many of these deaths are attributable to both risk factors. Oral tobacco use also results in upper aerodigestive tract cancer deaths above and beyond those which are affected jointly by smoking. Upper aerodigestive tract cancer also had a relatively large smoking-attribut able fraction in sub-Saharan Africa (32-49%) because of high background levels. Cancers of the stomach and liver, 2 sites identified in the recent IARC review and the 2004 report of the U.S. surgeon general as important new targets for studying the hazards of smoking,3-5 made relatively large contributions to smoking-attributable cancer deaths in a number of industrialized and developing regions: stom ach cancer represented 10% or more of all smoking-caused cancer deaths in parts of Latin America, the Western Pacific and eastern Europe and liver cancer, 9-13% in sub-Saharan Africa, the West ern Pacific and parts of Southeast Asia. Other sites with relatively large contributions to smoking-caused cancer mortality include the pancreas in all industrialized regions and parts of Latin America and the bladder in the eastern Mediterranean and parts of Europe. Cervix uteri cancer, also identified by the IARC and U.S. sur geon general reports as causally related to smoking, made a small contribution to total smoking-caused cancer deaths. This was 970 EZZATI ET AL. The estimated number of global cancer deaths attributable to smoking in this analysis (1.42 million, excluding colorectal cancer) was practically identical to the number obtained by applying a con stant correction factor of 30% to unadjusted hazards for multiple cancers combined (1.47 million).7 The differences in estimates from the 2 methods across regions and sexes ranged from a 15-20% underestimation among females in parts of sub-Saharan Africa, Latin America and the Western Pacific to about a 10-20% overesti mation among males in North America and other parts of the West ern Pacific, when covariate-adjusted site-specific hazards were not used. Using the IARC database of global cancer mortality instead of that of the WHO reduced the total number of global cancer deaths attributable to smoking by approximately 5%. This global decrease corresponds to a 16% decrease in developing countries and a 4% increase in industrialized countries because the difference between the WHO and IARC estimates of total cancer deaths is larger for the developing world. In the vast majority of subanalyses (e.g., by region or by cancer site), the IARC-based estimates were within the 95% CIs of the WHO-based estimates, suggesting that the implica tions of the differences in global and regional descriptive epidemiol ogy for assessing the contribution of smoking are minor. The validity of smoking-attributable mortality estimates using the indirect SIR method has been confirmed against alternative methods that account for accumulated hazards of smoking.36 Uncertainty nonetheless remains, especially in developing countries where com plete mortality records and detailed studies of cancer risk factors are less common. Some sources of uncertainty were described earlier. Two additional and important sources of uncertainty in the current estimates with implications for future research are (i) the correlation between smoking and other risks for any cancer within populations and (ii) biologic interactions between smoking and other risks, which would result in uncertainty when extrapolating CPS-II RRs to other populations. For many diseases, multiple risk factors magnify one another's total risks, as represented in standard proportional hazard models. If multiple risks for the same cancer are concentrated (posi tively correlated) among specific population subgroups, total (abso lute) hazards would be larger than those estimated here, even if RRs remain unchanged (and smaller with negative correlation).37,38 Higher consumption of alcohol among smokers is one example of such positive correlation among exposures; lower smoking rates among females in rural areas of developing countries, who also have more limited access to HPV screening and interventions and hence higher risk of cervical cancer, is an example of negative correlation. In addition to correlation of risk factor exposures, multiple risks for the same cancer may modify one another's RR. Examples include tobacco and alcohol as risk factors for oral cancer.39 Both issues may be particularly relevant for cancers whose risk factors include infec tious and noninfectious agents, particularly cervical (HPV), liver (alcohol, HBV and HCV infections, aflatoxin exposure) and stomach (alcohol, diet and Helicobacterpylori infection) cancers.40,41 The evi dence indicates that extrapolation of hazards from CPS-II to other populations is a valid option. For example, the Chinese retrospective study of smoking and mortality illustrated that background mortality rates, which varied by up to a factor of 10 for some cancers, did not change the RR of mortality as a result of smoking (Figs. 4 and 5 in Liu et al.10). Similarly, the differences in RRs for stomach cancer among smokers in various geographic regions of the world were not statistically significant.42 Despite these assurances, extrapo lation of CPS-II RRs is a source of additional uncertainty but neces sary in the absence of local epidemiologic studies in developing countries (at any point in time, only recent epidemiologic studies can be used for direct estimates because of changes in smoking patterns over time). The above data gaps and uncertainties also emphasize 2 impor tant research and surveillance needs in order to better understand how tobacco affects global and regional cancer epidemiology together with other risks. First, there is a need for epidemiologic and toxicologic research on how smoking interacts with other fac tors to affect hazards for any cancer. This includes identifying whether a proportional hazard model is appropriate, as observed for exposure to coal and tobacco smoke in China,10 or whether smoking and other infectious and toxic cofactors potentiate each others' hazards, hence resulting in modified population-specific RRs. For example infectious agents may result in subtypes of stomach cancer that are affected differently by smoking compared to stomach cancer in CPS-II subjects, which may be caused more by noninfectious risk factors (carcinoma of the corpus vs. carci noma of the cardia). Parallel to this individual-level analysis, there is a need in population-level risk factor surveillance for data on correlation of cancer risks in specific population subgroups based on factors such as income, education, race/ethnicity or place of residence. Data on multiple risk factor interaction and correlation would allow better estimation of total health effects of cancer risk factors. Equally, data on correlated multiple-risk exposure would be essential for targeting preventive interventions or treatment toward subgroups at highest risk. Currently, cancer mortality attributable to smoking is higher in industrialized regions than in the developing world (Table III). The data on smoking trends, however, suggest that smoking has increased in much of the developing world over the past few decades,17 with the likely implication that the attributable cancer burden has also. Accumulated hazards in developing countries coupled with the shifting demographic and epidemiologic patterns suggest that cancer mortality as a consequence of smoking will continue to rise in the near future in the developing world. Effective interventions and policies (e.g, through enforcing the FCTC) that reduce smoking among males and prevent increases among females in the develop ing world can curb and eventually reverse this increase, as has been seen in a number of industrialized nations. Our estimates of the cur rent level and patterns of smoking-attributable cancer mortality pro vide an important baseline for evaluating how FCTC and related national tobacco-control efforts may contribute to reducing the global and regional burden of cancers. 1. Murray CJL, Lopez AD. The global burden of disease. Cambridge, MA: Harvard School of Public Health, on behalf of the WHO and the World Bank, 1996. 2. WHO. World health report 2002: reducing risks, promoting healthy life. Geneva: WHO, 2002. 3. IARC. Monographs on the evaluation of carcinogenic risks to humans. Tobacco smoke and involuntary smoking. vol.83. Lyon: IARC, 2004. 4. Vineis P, Alavanja M, Buffler P, Fontham E, Franceschi S, Gao YT, Gupta PC, Hackshaw A, Matos E, Samet J, Sitas F, Smith J, et al. Tobacco and cancer: recent epidemiological evidence. J Natl Cancer Inst 2004;21:99-106. 5. US Department of Health and Human Services. The health con sequences of smoking: a report of the surgeon general. Washing ton DC: USDHHS, Public Health Service, Centers for Disease Control and Prevention, National Center for Chronic Disease Pre vention and Health Promotion, Office on Smoking and Health, 2004. 6. Hecht SS. Tobacco carcinogens, their biomarkers and tobacco- induced cancer. Nature Rev Cancer 2003;3:733-44. 7. Ezzati M, Lopez AD. Estimates of global mortality attributable to smoking in 2000. Lancet 2003;362:847-52. 8. Smith KR, Mehta S, Maeusezahl-Feuz M. Indoor air pollution from household solid fuel use. In: Ezzati M, Lopez AD, Rodgers A, Mur ray CJL, eds. Comparative quantification of health risks: global and regional burden of disease attributable to selected major risk factors. Geneva: WHO, 2004.1435-93. 9. Florig HK. China's air pollution risks. Environ Sci Technol 1997;31:274A-79A. 10. Liu BQ, Peto R, Chen ZM, Boreham J, Wu YP, Li JY, Campbell TC, Chen JS. Emerging tobacco hazards in China: 1. Retrospective propor tional mortality study of one million deaths. BMJ 1998;317:1411-22. 11. Dikshit RP, Kanhere S. Tobacco habits and risk of lung, oropharyn geal and oral cavity cancer: a population-based case-control study in Bhopal, India. Int J Epidemiol 2000;29:609-14.