Bulletin of the World Health Organization

Smoking-attributable mortality in Bangladesh: proportional mortality study

Dewan S Alam a, Prabhat Jha b, Chinthanie Ramasundarahettige b, Peter Kim Streatfield a, Louis W Niessen a, Muhammad Ashique H Chowdhury a, Ali T Siddiquee a, Shyfuddin Ahmed a & Timothy G Evans c

a. Centre for Control of Chronic Diseases, International Centre for Diarrhoeal Disease Research Bangladesh, 68 Shaheed Tajuddin Ahmed Sarani, Mohakhali, Dhaka 1212, Bangladesh.
b. Centre for Global Health Research, St Michael’s Hospital at University of Toronto, Toronto, Canada.
c. Human Development Network, The World Bank, Washington, United States of America.

Correspondence to Dewan S Alam (e-mail: dsalam@icddrb.org).

(Submitted: 26 February 2013 – Revised version received: 12 June 2013 – Accepted: 13 June 2013 – Published online: 12 July 2013.)

Bulletin of the World Health Organization 2013;91:757-764. doi: http://dx.doi.org/10.2471/BLT.13.120196


Deaths attributable to smoking are projected to increase substantially throughout the 21st century and much of the increase will occur in low- and middle-income countries1 such as Bangladesh, whose population of 150 million makes it the seventh most populous country in the world. More than half of Bangladeshi men over the age of 25 years smoke cigarettes or bidis, small handmade cigarettes containing about one fourth the amount of tobacco found in cigarettes.2 A nationally representative case–control study in neighbouring India showed that in 2010 smoking caused about 20% of all deaths among males aged 30 to 69 years.3 Smoking cessation rates are relatively low in Bangladesh and India,2 but Bangladeshi men are, on average, younger than Indian men when they start smoking and they smoke more cigarettes or bidis daily than Indian men.2,4 Bidis account for most of the tobacco smoked in India, but in Bangladesh cigarettes represent about half of all the tobacco smoked.

According to recent studies in high-income countries, men and women who start smoking as young adults and do not quit have a threefold risk of dying relative to those who have never smoked.5,6 Prospective studies are required to determine whether the same extreme risks hold true for Bangladesh and other low- and middle-income countries. Of the estimated 21.9 million smokers in Bangladesh, 21.2 million are males and only 0.7 million are females. Thus, only the effects of smoking among men can be reliably studied at present.

The objective of this study was to assess the effects of smoking on cause-specific mortality among Bangladeshi men between the ages of 25 and 69 years. We conducted a retrospective case–control study using data on cause of death and smoking status for all men in this age group who died in the subdistrict of Matlab in rural Bangladesh between 2003 and 2010. Although tobacco chewing is common in Bangladesh, its effects will be reported in a separate paper since chewing causes fewer diseases than smoking, most notably oral cancer.7


Study design

As part of the INDEPTH Network, the International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b), has maintained a comprehensive Health and Demographic Surveillance System (HDSS) in Matlab, in the district (zilla) of Chandpur, since 1966.8,9 Matlab is a rural area (upazila) located about 55 km south-east of Dhaka, the capital city. It covers 184 km2 and has 142 villages, each with about 1500 people. The total population of Matlab is about 225 000. To track fatal events as part of the HDSS, trained field research assistants visit households where a death has occurred from 6 to 12 weeks after the death and administer a structured verbal autopsy questionnaire to any relative who lived with the deceased. The purpose is to obtain from these respondents information on the symptoms, signs and medical details surrounding the death. Field staff are instructed not to try to arrive at a diagnosis of the cause of death. Instead, causes of death are assigned by two trained physicians at icddr,b in accordance with the International statistical classification of diseases and related health problems, tenth revision (ICD-10) and using standardized disease coding guidelines developed for other INDEPTH sites.8,10


Cases were men aged 25 to 69 years of age who had died from causes strongly associated with tobacco smoking:1,7,11 cancers of the lung, mouth and larynx (ICD-10 codes: C00–14, C32–34); cancers of the digestive organs (C16–26, C48, D01, D12–13); all other cancers (rest of C00-D48, excluding C60–63 and C69–72); chronic lower respiratory diseases (J40–47); pulmonary tuberculosis (A15–19); all other respiratory diseases (rest of J00–99); stroke (I60–64); ischaemic heart disease (I20–25); all other vascular diseases (rest of I00–99); cirrhosis (K70–77); and other medical causes (rest of A00-R99, excluding the diseases not associated with tobacco smoking, which were the causes of death among controls). Controls were men who had died of causes not associated with tobacco, such as intestinal infections (A00–09); other bacterial diseases (A30–99); viral infections (B01–77); cancers of the male genital organs (C60–63); cancers of the brain and eye (C69–72); endocrine, nutritional and metabolic diseases (E00–07, E25–35, E40–46, E50–64, E70–90); mental and behavioural disorders (F00–99); diseases of the nervous system (G00–99), diseases of the eye and adnexa (H00–95); diseases of the appendix and hernias (K35–40); diseases of the skin (L00–99); diseases of the musculoskeletal system (M00–99); diseases of the genitourinary system (N00–99), and injuries (T00-Y99).

We excluded women from the analyses because the smoking prevalence among Bangladeshi women is very low. According to the nationally representative Global Adult Tobacco Survey, or GATS, in 2009 smoking prevalence among living Bangladeshi women 15 years of age or older was only about 1.5%.2 It was somewhat higher, at 5.7%, among deceased women aged 20 years of age or older in our study.

The field questionnaire contained questions on the smoking and alcohol consumption history of the deceased individual. Smoking history included the type of tobacco smoked (e.g. cigarette or bidi tobacco [combined into a single question]; tobacco for pipe or hookah) or chewed (e.g. betel leaf with or without tobacco, tobacco powder, rolled tobacco); the amount smoked; the age (in years) at initiation and cessation of smoking; and the duration of tobacco smoking or chewing. The field questionnaire did not separate the number of cigarettes from the number of bidis smoked, so we treated them together in the analysis. We excluded 156 men from the analysis because 34 of them drank alcohol, 11 smoked only pipe and the remainder had missing information on smoking status, education or cause of death. We compared “ever-smokers” – i.e. people who smoked at the time of the survey (“current smokers”) or who had smoked in the past (“former smokers”) – with “never-smokers”. This was appropriate because most former smokers (214/324) had smoked for more than 10 years. Moreover, excess hazard depends strongly not only on recent smoking habits but also on smoking habits in early adult life.1,7,11

Statistical methods

We used proportional mortality to relate bidi or cigarette smoking to various specific and general medical causes of death.12,13 The assumption behind proportional mortality is that the smoking patterns among the deceased in the control group are similar to those observed in the general population. However, in reality smoking prevalence was lower among the general population of men surveyed in 2009,2 than among our deceased controls, perhaps because smoking caused some excess deaths among controls. Thus, the net effect would be an underestimation of the differences between ever-smokers and never-smokers. Other reporting biases should equally affect the cases and controls.12

We calculated cause-specific odds ratios (ORs) (as approximations of relative risks [RRs]) for ever-smokers versus never-smokers using logistic regression, with adjustment for educational status (illiterate, below secondary school, secondary school or above), history of tobacco chewing (yes, no) and age (in continuous years). We calculated the smoking-attributable deaths among cases by multiplying the number of ever-smokers among the cases by: (1)where RR represents the ratio of the odds of death in ever-smokers to the odds of death in never-smokers adjusted for age, education and chewing tobacco status. To calculate the national attributable fractions we used the following standard formula: (2)with the RRs derived from the Matlab results and smoking prevalence (p) from the GATS survey in 2009.2 About 80% of the Bangladeshi population lives in rural areas similar to Matlab, so the use of the RRs estimated in this study does not introduce a major bias in the calculation of overall national attributable fractions. In Bangladesh, national cause-of-death statistics are based on self-reporting, which can be unreliable.14 For this reason, to generate national cause-of-death estimates we applied, to the total deaths reported by the United Nations for Bangladesh, the proportions of the major causes of death found in Matlab.15 In Chandpur zilla, the rates of all-cause mortality among adult males are comparable to the rates observed in the whole of Bangladesh, but child mortality rates are lower.14 We calculated the cumulative probability of surviving from 25 until 69 years of age for ever-smokers and never-smokers, with adjustment for any differences in age, education and use of chewing tobacco. To do so, we combined the RRs estimated from Matlab with national smoking prevalence1 and age-specific Bangladeshi death rates as reported by the United Nations,15 following methods described in previous studies.3,6



The Matlab HDSS recorded a total of 9708 deaths among individuals aged 20 years or older (5296 males, 4412 females). Of these deaths, 2474 occurred in men aged 25 to 69 years. Table 1 presents data for the 2213 cases and 261 controls. Cases were older than controls on average, whereas both groups had similar educational levels and similar rates of use of chewing tobacco. Among the deceased, smoking prevalence was highest among those who had been illiterate or between the ages of 40 and 60 years when they died (data not shown).

About 84% (1855/2213) of the cases were ever-smokers (Table 1), as compared with 73% (191/261 data not shown) of the controls. The OR for all causes of death, representing the odds in ever-smokers versus the odds in never-smokers, was 1.7 (99% confidence interval, CI: 1.1–2.5) after adjusting for any differences in age, education and use of chewing tobacco (Table 2). This corresponds to 31% of all 2213 deaths among the cases, equivalent to an excess of 692 deaths among smokers between the ages of 25 and 69 years in Matlab. About 94% of the men who died from cancers of the lung, oral cavity or larynx were ever-smokers (162/172); 103 men died from lung cancer and of these men, 97 were ever-smokers. The ORs were notably lower for digestive organ cancers than for other cancers. The ORs and the proportions of deaths due to smoking were highest for cancers; they were lower or similar for respiratory diseases, vascular diseases and other medical causes of death.

In Bangladesh, men aged 25 to 69 years initiate smoking at an average age of 18.6 years (Appendix A, available at: www.cghr.org/tobacco) and nearly all smokers report having initiated smoking before the age of 25 years.2 As shown in Fig. 1, in our study, those who started smoking before this age had greater odds of dying than never-smokers (OR: 1.9) and than those who started smoking after the age of 25 years (OR: 1.4; χ2 test for trend 12.1; P < 0.0001). The mean daily number of cigarettes or bidis smoked by males who died was 11. A dose–response analysis (comparing 1757 ever-smokers with 438 never-smokers for whom information on smoking amount was available) showed that those smokers who consumed from 20 to 50 (mean: 23) cigarettes or bidis per day had a much higher odds of dying (OR: 2.7) when compared with never-smokers, than those who consumed 10 to 19 cigarettes or bidis (mean: 12) daily (OR: 1.6) or than those who smoked 1 to 9 (mean: 6) cigarettes or bidis a day (OR: 1.3; χ2 test for trend 16.3; P < 0.001).

Fig. 1. Odds ratios (ORs) and excess deaths due to smoking, by age of smoking initiation and daily number of cigarettes or bidis smoked by men aged 25 to 69 years, Bangladesh, 2003–2010
Fig. 1. Odds ratios (ORs) and excess deaths due to smoking, by age of smoking initiation and daily number of cigarettes or bidis smoked by men aged 25 to 69 years, Bangladesh, 2003–2010
a ORs represent the odds of dying among ever-smokers relative to the odds among never-smokers, adjusted for age, education and tobacco chewing status.
b The figure in parentheses is the median age of smoking initiation.
c The figure in parentheses is the mean number of cigarettes or bidis smoked.

Applying the Matlab relative risk to the national death totals suggests that about 42 800 Bangladeshi men aged 25 to 69 years died in 2010 from smoking-attributable diseases. This was equivalent to about 25% of the 172 200 deaths from all diseases in men in this age group in 2010 (Table 3). We did not attribute to smoking any deaths from cirrhosis, ill-defined conditions or injury. Of the smoking-attributable deaths, about 50% (21 700/42 800) were caused by vascular disease and about 26% by cancer (11 400).

The cumulative probability of dying between the ages of 25 and 69 years was much higher for ever-smokers than for never-smokers (Fig. 2). At this death rate, 32% of 25-year-old Bangladeshi male ever-smokers would die before the age of 70 years, versus only 19% of 25-year-old never-smokers. Marked differences in the risk of death between ever-smokers and never-smokers were seen among those as young as 60 years – 14% dead by that age among smokers; only 8% dead among never-smokers. These data suggest that the average 25-year-old Bangladeshi male smoker currently loses an average of seven years of life because of smoking.

Fig. 2. Cumulative probability of dying between the ages of 25 and 69 years among male ever-smokers and never-smokers in Bangladesh, at 2010 death rates
Fig. 2. Cumulative probability of dying between the ages of 25 and 69 years among male ever-smokers and never-smokers in Bangladesh, at 2010 death rates
Note: The horizontal arrow shows that the death rates experienced by never-smokers at the age of 70 years are experienced seven years earlier by ever-smokers. Thus, smokers lose seven years of life on average. This combines a loss of zero years among smokers not killed by smoking with a loss of far more than seven years among those killed by smoking.


Bangladeshi men between the ages of 25 and 69 years who have smoked at some point have a risk of dying of a tobacco-related disease that is 70% higher than the risk among never-smokers of similar age, educational level and tobacco chewing status. Our cases were men who had died from diseases that have been causally associated with smoking in other studies1,3,5-7,11-12 and we observed a well-defined dose–response relationship between age at smoking initiation and the number of cigarettes or bidis smoked and the risk of dying. Admittedly, however, not all smokers died of diseases associated with tobacco-attributable diseases and not all such diseases occurred among smokers. Thus, smoking is an important cause of most, but not all, of the excess deaths among smokers in Bangladesh. Cumulative survival analysis revealed that a typical Bangladeshi smoker currently loses about seven years of life because of smoking. This combines an average loss of zero years of life for some smokers not killed by smoking with a loss far in excess of seven years in some smokers who were killed by smoking.

The ORs for ever-smokers versus never-smokers are slightly higher among Bangladeshi men than among Indian men (OR: 1.6).3 The proportion of all deaths in men aged 25 to 69 years that is due to smoking is also higher in Bangladesh (25%) than in India (20%). Subtle but potentially important differences in smoking patterns between the two countries might have an influence on the risk of specific diseases. Bangladeshi bidis are wrapped in cigarette paper rather than the tendu leaf commonly used to wrap Indian bidis. Cigarette smoking is much more common among Bangladeshi males than Indian males.2,4 In our study, lung cancer caused more than 10% of the deaths among smokers in Bangladesh (5300/42 800) but a much smaller proportion of the deaths among smokers in India.3 By contrast, pulmonary tuberculosis accounted for only 4% (2000) of the deaths among smokers in our study but for a much higher proportion of the deaths in Indian smokers.3 This may be because prompt short-course tuberculosis treatment is available in Matlab.16

Bangladeshi men have higher smoking cessation rates than Indian men but initiate smoking at a younger age and smoke more sticks on a given day.2,4 Smoking may well have caused some of the deaths we defined as our controls, such as deaths from diabetes. Thus, we may be underestimating the true risk of death from smoking in Bangladeshi men. Moreover, there was a higher proportion of ever-smokers among our controls (73%) than among the males of the same age in the GATS (66%).2

One of the strengths of the proportional mortality method is that most biases, such as recall bias, would apply to cases and controls equally.12 Indeed, we observed few differences between cases and controls in terms of education and tobacco chewing. Few Bangladeshis report drinking alcohol17 and we excluded any self-reported drinkers. However, the 121 deaths from cirrhosis suggest that in Bangladesh drinking is underreported for cultural reasons. The elevated risk of dying from cirrhosis observed in ever-smokers relative to never-smokers suggests that most of the heavy alcohol drinkers also smoked and vice versa. The combination of drinking and smoking should also exist among controls, many of whom died of injuries. Thus, we might be slightly underestimating the risks of smoking. However, the exclusion of men who drank alcohol does not explain the marked differences in mortality observed between ever-smokers and never-smokers. Deaths in the control group (n = 261) were fewer than ideal. Consequently, the ORs associated with specific diseases had wide CIs. However, the cause-specific ORs in our study are consistent with those seen in India3,18,19 and other Asian countries.12 Similarly, some deaths may have been assigned to the wrong cause on verbal autopsy.20,21 This would tend to raise the ratio of the risk in smokers to the risk in non-smokers in the case of some diseases and to lower this ratio in the case of others. The proportion of vascular deaths among all deaths in men in Matlab (about 41%) is higher than the proportion seen in crude national cause-of-death patterns in Bangladesh or the proportion seen in India in men of comparable age.3,14 Thus, we might be overestimating the absolute number of smoking-attributable deaths from vascular disease (but not the odds). However, the computed ORs and the absolute number of smoking-attributable deaths from any cause are affected less, if at all, by the misclassification of specific diseases. Most of the misclassification on verbal autopsy is confined to people who die in old age.20 Our study focused on men aged 25 to 69 years and fewer than 4% of the deaths among these men were assigned to ill-defined causes.

Bangladesh already has about 20 million male smokers and it is likely that they will experience a loss in lifespan of more than the seven years that we estimated in this study. Bangladeshi men still smoke fewer cigarettes per day and initiate smoking at a later age than men in most high-income countries. Urban male smokers in Bangladesh consume more cigarettes than bidis and smoke more per day than rural smokers.2 Recent increases in income in Bangladesh have also made smoking relatively more affordable.22 For cultural reasons, few women smoke in Bangladesh, but there is no guarantee that this will hold in the future. Recent prospective studies in high-income countries have shown that male or female lifelong smokers who start smoking as young adults and do not quit have an excess risk of dying 200% as high – corresponding to an RR three times as high – as ever-smokers and live, on average, one decade less than never-smokers.5,6 Indeed, Indian men who smoke cigarettes in these large quantities already appear to be losing a decade of life.3 Of Bangladeshi men aged 45 to 64 years, only 15% are former smokers and 63% are current smokers.2 By contrast, in the United States of America, where smoking cessation has become common, former smokers are about three times more numerous than current smokers among men this same age.6

About one quarter of the deaths in Bangladeshi men between the ages of 25 and 69 years are due to smoking. Tobacco control in Bangladesh is still at an early stage; the Smoking and Tobacco Products Usage (Control) Act, which was passed in 2005, has been implemented only partially.22 Efforts to accelerate tobacco control in Bangladesh are warranted. Cigarette prices in the country are among the lowest in the world; bidis are even cheaper than cigarettes.22 Hence, the most effective way to encourage smoking cessation in Bangladesh would probably be to impose a substantial excise tax on tobacco products. Prominent warning labels using graphic images, bans on smoking in public places, strict restrictions on tobacco advertising and promotion, and expanded access to smoking cessation services are other interventions that can be implemented in Bangladesh to encourage men to stop smoking and deter women and youth from taking up smoking.1,23


This paper is dedicated to the late Sir Richard Doll (1912–2005), who would have turned 100 on 28 October 2012. The opinions expressed here are those of the authors and do not necessarily represent those of the institutions where the authors are employed.


Funding for this study was provided by the Fogarty International Centre of the US National Institutes of Health (grant TW007939-01), the Canadian Institute of Health Research (IEG-53506), the Bill & Melinda Gates Foundation (Grant 51447) and the Oxford Health Alliance Vision 2020 (Grant 5444447). PJ is supported by a University of Toronto Endowed Chair. DA is supported by icddr,b.

Competing interests:

None declared.