Introduction Tobacco smoking is a major risk factor for many diseases, including cardiovascular disease (CVD), respiratory disease, and cancers of the lung and multiple other sites [1],[2]. In the US and many other Western countries, the epidemic of tobacco smoking started in men in the early 1900s and reached its peak in the 1960s; a similar epidemic occurred among women ∼40 y later [3]–[5]. The main increase in tobacco-related deaths in these countries was not seen until the second half of the 20th century [3],[6]–[8]. By the 1990s, tobacco smoking accounted for an estimated one-third of all deaths and >50% of cancer deaths in adult men [3],[6]–[8]. With increasing awareness of smoking-associated risks and heightened anti-smoking campaigns, tobacco use has steadily declined in the US and many other developed countries over the past 20–30 y [3]–[5],[9],[10], resulting in a recent decrease in lung cancer and other smoking-related diseases in these countries [3],[11]. In Asia, where ∼60% of the world population lives, tobacco control programs are less well developed, particularly in low- and middle-income countries including China and India, the two most populous countries in the world. Inadequate public awareness of smoking risks, combined with aggressive marketing by tobacco companies, has resulted in a sharp increase in tobacco smoking among men in many Asian countries over the past few decades [3],[11],[12]. Smoking prevalence in women was traditionally very low but has increased in recent decades in some Asian countries [3],[11],[12]. More than 50% of men in many Asian countries are smokers [12],[13], approximately twice the level in many Western countries. Despite a recent decline in smoking prevalence in several high-income Asian countries [11],[13], tobacco use in most Asian countries remains very high. Indeed, Asia is now considered the largest tobacco producer and consumer in the world. More than half of the world's 1.1 billion smokers live in Asia [3],[13]. Because many Asian countries are in the early stages of the tobacco epidemic, it is likely that the burden of diseases caused by tobacco smoking will continue to rise over the next few decades, and much longer if the tobacco epidemic remains unchecked. The size of the effect of tobacco smoking on risk of death, typically measured using smoking-associated relative risks (RRs), varies across countries because of differences in characteristics of smokers, smoking behaviors, and tobacco products. Over the past 15 y, several studies have investigated associations between smoking and selected health outcomes in certain Asian populations and have estimated smoking-associated population attributable risk (PAR) [14]–[21]. Some studies estimated burden of disease due to smoking in a specific Asian country/region [14],[16],[17],[19],[20]. However, most of these estimates were derived from either a single cohort study or studies using a less-than-optimal research design. In this study, we first estimated RRs of overall and cause-specific mortality associated with tobacco smoking as well as smoking prevalence, using data from ∼1 million participants recruited in 21 prospective cohort studies in seven countries/regions that account for ∼71% of Asia's total population. We then used these estimates and mortality data from the World Health Organization [22] to quantify deaths attributable to tobacco smoking in these Asian populations. Methods This study was approved by the ethics committees for all the participating studies and of the Fred Hutchinson Cancer Research Center. This study utilized resources from a recent pooling project of prospective cohort studies conducted as part of the Asia Cohort Consortium that quantified the association between body mass index and risk of overall and cause-specific mortality in Asians [23]. Cohorts included in the current analysis were in Bangladesh, India, mainland China, Japan, Republic of Korea, Singapore, and Taiwan. A brief description of each of the participating cohort studies is provided in Text S1. All of the cohort studies collected baseline data on demographics, lifestyle factors, body mass index, and history of tobacco smoking, which included current smoking status, duration, and amount and types of tobacco products. Data on all-cause and cause-specific mortality were ascertained through linkage to death certificate data or active follow-up. Additional data were collected on other baseline variables, including education, marital status, alcohol consumption, physical activity, and previous diagnosis of selected diseases, including diabetes, hypertension, cancer, and CVDs. Individual-level data from all participating cohorts were collected and harmonized for statistical analysis. The association between tobacco smoking and risk of death was examined using Cox proportional hazards regression models, employing a categorical representation of tobacco smoking as the predictor variable. Lifetime nonsmokers were used as the reference for estimating hazard ratios (HRs)—as measures of RR of death for the exposed versus the non-exposed population—and 95% confidence intervals associated with ever, former, and current smoking, as well as pack-years smoked, after adjusting for potential confounders including baseline age, education, urban/rural residence, body mass index, and marital status. All analyses were conducted separately for men and women because of large differences in smoking prevalence. Analyses were country-specific unless otherwise noted. To improve the stability of point estimates in the analyses of pack-years of smoking and for risk of death due to site-specific cancer, as well as types of CVD and respiratory diseases, cohorts were combined into broad ethnic groupings: South Asians (Indians and Bangladeshis) and East Asians (Chinese [including cohorts from mainland China, Singapore, and Taiwan], Japanese, and Koreans), and categorized further among East Asians into Chinese/Koreans and Japanese. No smoking-associated HR was estimated for Bangladesh separately because of the small sample size. The number of Koreans in this study was small, and, thus, they were combined with Chinese individuals in some analyses. Bidi smoking is common in India and Bangladesh; thus, information regarding bidi smoking was incorporated to construct smoking variables, including pack-years smoked (4 bidis = 1 cigarette based on approximately 0.25 and 1.0 g of tobacco per bidi and cigarette, respectively). In the models, the effect of tobacco smoking on mortality was assumed to be cohort-specific. For each cohort, we assumed that the log-HR for tobacco smoking has a fixed-effect component that is common to all cohorts within each country and a random effect that is cohort-specific. Random effects for log-HRs were assumed to be normally distributed, with mean zero; that is, we assumed that , the estimated log-HR for the j-th smoking level in the i-th cohort, has distribution , where is the within-study variance of as estimated from the Cox regression model and is the between-cohort variance of [24],[25]. Parameter βj and 95% CIs were estimated in the meta-analysis. Age at study entry and exit was used to define the time-to-event variable in the Cox models. Age at study exit was defined as age at date of death or end of follow-up, whichever occurred first. Cox model estimation for each cohort was performed using the PHREG procedure in SAS version 9.2. Meta-analysis estimation was performed using the SAS MIXED procedure. To estimate PAR, we used the following formula: PAR = P(RR−1)/[P(RR−1)+1], where smoking prevalence and smoking-associated RR are denoted as P and RR (measured using HR in this analysis), respectively. PARs for overall mortality and major causes of death associated with tobacco smoking were estimated for each cohort and then combined using meta-analyses to derive summary PARs per country. To estimate PARs for East Asians (Chinese, Japanese, and Koreans), South Asians (Bangladeshis and Indians), or all seven countries/regions combined, we used the population size of each country/region as a weight to derive weighted HR and smoking prevalence values. To estimate the number of deaths attributable to tobacco smoking, we used World Health Organization age-specific death rates for 2004 for each country. Most of the cohort studies enrolled participants after the mid-1980s; therefore, smoking prevalence rates estimated in this study reflect smoking status in the 1990s (Table 1). Given the long latency of chronic diseases—typically 15 y and longer—it is reasonable to use smoking prevalence rates assessed in the 1990s to estimate number of deaths due to tobacco smoking in 2004. 10.1371/journal.pmed.1001631.t001 Table 1 Characteristics of participating cohorts in the Asia Cohort Consortium. Cohort Number of Participantsa Study Entry Mean Years of Follow-Up Women (Percent) Mean Age at Entry Ever-Smokers (Percent) Number of Deaths Cause of Death (Percent)b Men Women Cancer CVD Respiratory Diseases Other India Mumbai 120,055 1991–1997 5.3 36.4 53.4 31.8 0.5 10,839 8.5 45.0 14.4 32.2 Trivandrum 103,942 1995–2002 7.8 59.6 52.7 60.1 1.8 9,406 10.6 36.6 12.8 40.0 Bangladesh 4,572 2000–2002 6.7 41.0 46.8 83.0 15.5 206 13.7 51.2 10.2 24.9 Mainland China CHEFS 137,460 1990–1992 7.8 50.9 54.9 63.9 13.4 14,776 23.4 44.8 5.0 26.8 SCS 18,010 1986–1989 16.4 0.0 55.2 57.2 NA 4,902 39.6 33.9 10.7 15.9 SMHS 54,707 2001–2006 3.1 0.0 55.1 69.6 NA 596 53.1 25.7 5.4 15.7 SWHS 67,245 1996–2000 8.7 100.0 51.3 NA 2.7 1,921 48.2 23.5 2.6 25.7 Taiwan CBCSP 22,961 1991–1992 15.4 50.1 47.2 56.4 1.0 2,400 38.1 19.5 5.9 36.4 CVDFACTS 4,170 1990–1993 15.0 55.8 50.7 54.9 1.3 711 27.5 26.1 10.7 35.7 Singapore (SCHS) 57,714 1993–1999 11.7 56.1 56.1 57.1 8.4 8,234 36.7 33.1 14.8 15.4 Japan 3 Pref Aichi 29,316 1985 12.1 50.6 56.3 84.3 17.5 5,330 32.4 35.0 11.9 20.7 Ibaraki 91,847 1993–1994 11.6 66.3 58.5 77.8 5.6 9,545 NA NA NA NA JACC 74,465 1988–1990 12.9 56.4 57.0 79.1 6.6 10,099 38.6 29.1 11.4 20.9 JPHC1 40,574 1990–1992 14.7 52.2 49.6 75.7 7.3 3,007 45.0 24.6 6.0 24.3 JPHC2 52,838 1992–1995 11.7 52.9 54.1 75.7 7.6 4,708 44.6 24.1 8.7 22.6 3 Pref Miyagi 18,951 1984 12.0 53.4 56.2 77.1 12.0 3,307 31.0 38.5 11.0 19.5 Miyagi 38,560 1990 12.9 45.2 51.5 81.5 11.1 2,932 54.9 25.9 6.3 12.9 Ohsaki 37,884 1995 10.5 47.0 59.5 81.1 11.0 5,093 37.4 30.7 12.9 19.0 RERF 47,532 1963–1993 22.0 59.2 51.6 86.2 15.5 24,128 27.4 37.2 13.3 22.2 Republic of Korea KMCC 13,446 1993–2004 6.6 62.5 57.9 79.1 10.0 1,036 29.3 24.8 8.6 37.3 Seoul 13,680 1992–1993 14.7 0.0 49.2 77.3 NA 799 53.6 16.8 3.0 26.7 Total 1,049,929 1963–2006 10.2 51.4 54.3 65.1 7.1 123,975 29.8 35.0 10.8 24.3 a Including only participants eligible for the current analysis. b Deaths from unknown causes are not included. 3 Pref, Three Prefecture Cohort Study; CBCSP, Community-Based Cancer Screening Project; CHEFS, China National Hypertension Survey Epidemiology Follow-Up Study; CVDFACTS, Cardiovascular Disease Risk Factor Two-Township Study; JACC, Japan Collaborative Cohort Study, JPHC, Japan Public Health Center-Based Prospective Study; KMCC, Korea Multi-Center Cancer Cohort; NA, not available; RERF, Radiation Effects Research Foundation; SCHS, Singapore Chinese Health Study; SCS, Shanghai Cohort Study; SMHS, Shanghai Men's Health Study; SWHS, Shanghai Women's Health Study. The number of deaths from a particular disease attributable to tobacco smoking was calculated by multiplying the PAR for that disease by the total number of deaths in the population from that disease. Analyses also were performed to estimate the number of deaths from a particular disease due to smoking for age groups 45–59, 60–69, and ≥70 y using age-specific HRs and smoking prevalence and then summing these age-specific estimates to obtain the overall number of deaths due to smoking for that disease. This age-specific method yielded similar results to the one without age-specific estimates, and, thus, the latter method was used, as it provides a tighter 95% CI than the age-specific method. Results A total of 1,223,092 participants were included in the 21 participating cohorts for this study. Because most studies were conducted among adults aged ≥45 y, participants (n = 70,812) who did not contribute person-years in the age group ≥45 y were excluded from this analysis. Also excluded (not mutually exclusively) were participants with prior history of cancer or CVD at baseline (n = 47,585), with missing data on tobacco smoking (n = 38,898) or vital status (n = 451), or with less than 1 y of observation after baseline survey (n = 30,039). After these exclusions, 1,049,929 participants (510,261 men; 539,668 women) remained (Table 1). Overall, the mean prevalence of tobacco smoking was 65.1% for men and 7.1% for women. Over a mean follow-up of 10.2 y through roughly the mid-2000s for most cohorts, a total of 123,975 deaths were identified in these cohorts. Compared with never-smokers, a 1.44-fold higher risk (95% CI = 1.37–1.51) of deaths from all causes was observed among male ever-smokers in pooled analyses of all cohorts (Table 2). The estimated HRs related to smoking were slightly higher in Singapore, Republic of Korea, Japan, and Taiwan than in India and mainland China, although 95% CIs overlapped in some of these point estimates (heterogeneity test: p 2-fold elevated risk for all-cause mortality is typically reported for current smokers [1]–[3],[5],[9],[10]. Among specific causes of mortality evaluated in this study, lung cancer showed the strongest association with tobacco smoking, with estimated HRs of 3.0 to 4.0, approximately one-third of the risk observed in most studies conducted in Western countries [1]–[3],[5],[9],[10]. The smaller effect of smoking on mortality in Asia compared with Western countries could be partly explained by the fact that widespread tobacco smoking in most Asian countries began several decades later than in Europe and North America, and thus many Asian countries are still in the early stages of a tobacco epidemic; many smokers in the population started smoking tobacco at a late age and smoke a small number of cigarettes daily [3],[12]. In the British Doctors Study, a 1.6-fold elevated risk of all-cause mortality was observed among smokers in early years of follow-up (1951–1971) [26], close to the effect size estimated in this study. In later follow-up (1971–1991), the RR rose to 2.1. A recent Japanese study showed a clear birth-cohort effect: male smokers born before 1890 started smoking at a later age and smoked fewer cigarettes daily than those born in 1940–1945 [27]. As a result, the association of smoking with risk of all-cause mortality was weaker in the older cohort (RR = 1.24) than the younger cohort (RR = 1.92). Our study showed a clear dose–response relationship between pack-years of smoking and risk of all-cause and cause-specific mortality. It is likely that, with maturation of the tobacco epidemic in Asia and lack of effective tobacco control, more smokers will accumulate much higher pack-years of smoking, and, thus, smoking-associated RRs will rise, mirroring the trend in the US and Europe. Several previous studies have estimated the burden of disease due to tobacco smoking in a specific Asian country/region [14],[16],[17],[19],[20] (Table S2). However, most previous estimates for smoking-associated RRs and PARs were derived from either a single cohort study [14],[19] or a retrospective case–control study [16]–[18]. Not all previous studies had detailed demographic and risk-factor information to adequately adjust for potential confounders when estimating risks. Three previous studies conducted in mainland China and Taiwan provided somewhat lower estimates of total male deaths due to smoking than our estimates, perhaps because these studies were conducted during even earlier stages of the tobacco epidemic in these populations, resulting in smaller PARs [14],[16],[19]. For India, however, the estimate of male deaths attributable to tobacco smoking from a previous study (20% of total male deaths) [17] was substantially higher than the estimate from our study (11.5% of total male deaths). To our knowledge, no study has been previously conducted in Bangladesh, the Republic of Korea, or Singapore; thus, our study provides, for the first time, direct estimates of deaths due to tobacco smoking in these countries. Despite methodological differences between this and previous studies, all studies conducted to date have shown that an alarming proportion of deaths are caused by tobacco smoking. In this study, some estimates among women are unstable because of very low smoking prevalence. Although not all participating cohorts are representative of the general population, smoking-associated RRs estimated in this study, are, in general, comparable to those from previous studies. Furthermore, smoking-associated RRs estimated in multiple cohorts within the same country are, in general, comparable. It is difficult to find national survey data consistent with the definitions, time period, and age groups of our study for all seven countries/regions in our analysis. Many national surveys used a smaller sample than our study, providing unstable smoking prevalence estimates. Therefore, we chose to use smoking prevalence estimates from our own study to estimate PARs: smoking-associated RRs were estimated based on exposure history of the same group of individuals, which should provide better estimates of disease burden due to tobacco smoking in the study population than using data from external sources. Smoking prevalence has declined recently in several high-income Asian countries. However, given the long latency of chronic diseases, typically 15 y and longer, it is reasonable to use smoking prevalence rates assessed in the 1990s to estimate number of deaths due to tobacco smoking in 2004. As most of the cohort studies included in this study were conducted among adults aged ≥45 y, we were unable to estimate the impact of active tobacco smoking in people younger than 45 y old. Again, because of the long latency of chronic diseases, most of the smoking-related diseases tend to occur later in life. We estimated smoking-associated PARs and numbers of deaths due to tobacco smoking in 2004. As many Asian countries, such as China and India, are still in the early stage of tobacco epidemics, the number of deaths due to tobacco smoking in more recent years in these countries is likely to be larger than that estimated in this study. Data on secondhand tobacco-smoke exposure was not available in this study. Secondhand smoke has been linked to an elevated risk of multiple chronic diseases [2],[28],[29]. It has been estimated that approximately 603,000 deaths worldwide may be due to secondhand smoke [29]. We also were unable to evaluate smokeless tobacco, a risk factor for oral cancer and several other chronic diseases [30],[31]. Smokeless tobacco use is common in India and Bangladesh, especially among women in these countries. Some individuals who had secondhand tobacco-smoke exposure or used smokeless tobacco may be included in the reference group, which may result in an underestimate of the risk associated with active tobacco smoking. Furthermore, in our study, RRs associated with tobacco smoking were estimated primarily based on the time period from the early 1990s to the mid-2000s. Because smoking-associated risk of death is likely to increase with the maturation of the tobacco epidemic, the total number of deaths due to smoking in 2004 may be underestimated using this set of RRs. Therefore, the true impact of tobacco smoking on mortality in these Asian countries is likely to be even larger than estimated here. Despite some limitations mentioned above, our study provides perhaps the best estimates of tobacco-associated deaths to date in these Asian countries/regions. Over the past 50 years, the landscape of tobacco smoking has changed dramatically around the world. Smoking prevalence has declined sharply in many high-income countries, resulting in a recent decrease in smoking-related deaths, particularly among men [3],[8]. Conversely, prevalence of tobacco use remains high in China, India, and other low- and middle-income countries. As the tobacco epidemic grows in these countries, we anticipate that an increasing number of deaths will be attributable to tobacco smoking in Asia in the coming years. Even in more well-developed Asian countries such as Japan and Republic of Korea, where smoking rates have recently declined, the full impact of tobacco smoking on mortality is unlikely to be seen soon because, as noted above, smokers in recent birth cohorts tend to smoke more and start smoking earlier, elevating their risk of smoking-associated deaths, and because of the long latency of the diseases associated with smoking, these deaths will not accrue immediately. Our study shows that tobacco smoking is a major cause of death in Asia, accounting for ∼1.6 million deaths of adults ≥45 y in 2004 in the seven countries/regions in this analysis. If the remaining 29% of the Asian population is experiencing a tobacco epidemic similar to that of these seven countries/regions, we estimate that, in 2004, >2 million deaths in Asia were attributable to tobacco smoking. Thus, of the 5 million deaths currently attributable to active tobacco smoking worldwide [32], nearly 45% occur in Asia. Our study provides sobering evidence that stresses the urgency of implementing comprehensive tobacco control programs in Asia, as recommended by the WHO Framework Convention on Tobacco Control [33]. Tobacco control should be among the top priorities in Asia to reduce the burden of disease. Supporting Information Table S1 Association of tobacco smoking status (former or current) with risk of death from all causes in selected study populations in Asia. (DOC) Click here for additional data file. Table S2 Population-attributable risk and number of deaths due to smoking in major Asian populations estimated in previous studies. (DOC) Click here for additional data file. Text S1 Descriptions of participating cohorts. (DOC) Click here for additional data file.
