The Progression of the Tobacco Epidemic in India on the National and Regional Levels, 1998-2016

Background: According to the smoking epidemic framework, societies undergo successive stages in which the prevalence of tobacco use rst increases, and then declines. This pattern was has been extensively documented for developed countries, but evidence that it has occurred in low- and middle-income countries remains fragmented. We assessed the progression of the tobacco epidemic in India on the national and regional levels by gender, including trends in the consumption of smokeless tobacco, to provide a comprehensive overview of tobacco use in India and its states. Methods: We use information on current tobacco consumption among Indians aged 15-49 from three rounds of the National Family Health Survey (NFHS) (1998-99, 2005-06, 2015-16) to estimate the age-standardized smoking and smokeless tobacco prevalence across India and its states. Results: Age-standardized tobacco consumption prevalence in India increased between 1998-99 and 2005-06, and declined from 2005-06 to 2015-2016, simultaneously for men and women. There are substantial spatial differences in the progression of the tobacco epidemic in India. In the north-eastern states in particular, tobacco consumption prevalence remains higher than the national average, and is still increasing. Conclusions: Our results suggest that India and the majority of its states experienced a ‘compressed tobacco epidemic’ in which the prevalence of tobacco consumption increased and decreased simultaneously for women and men over a comparatively short period of time. Despite the overall progress in reducing tobacco use India has made, further lowering tobacco consumption remains a public health priority, as the prevalence of smoking and/or smokeless tobacco use remains high in a number of states. We therefore conclude that tobacco regulations should be expanded with the aim of reducing the overall health burden associated with tobacco consumption across India. a comprehensive overview of tobacco use in India and its states, we include for the rst time trends in the consumption of smokeless tobacco. By incorporating smokeless tobacco into the classical smoking epidemic model, we not only account for different forms of tobacco consumption and how they have changed over time, but also for country-specic peculiarities in tobacco consumption patterns. In addition, our study is the rst to use a high-quality dataset to apply the tobacco epidemic model to India and its states in order to discern national and subnational patterns in the progression of tobacco prevalence. Various public health initiatives aimed at reducing tobacco use have been implemented in the country since 1975. We discuss whether these public health initiatives have affected female and male tobacco consumption over time, and conclude that tobacco regulations should be expanded with the aim of reducing the overall health burden of tobacco consumption across India.

substantial spatial differences in the progression of the tobacco epidemic in India. In the north-eastern states in particular, tobacco consumption prevalence remains higher than the national average, and is still increasing.
Conclusions: Our results suggest that India and the majority of its states experienced a 'compressed tobacco epidemic' in which the prevalence of tobacco consumption increased and decreased simultaneously for women and men over a comparatively short period of time. Despite the overall progress in reducing tobacco use India has made, further lowering tobacco consumption remains a public health priority, as the prevalence of smoking and/or smokeless tobacco use remains high in a number of states. We therefore conclude that tobacco regulations should be expanded with the aim of reducing the overall health burden associated with tobacco consumption across India.

Background
Tobacco consumption is the single most preventable cause of death and disability across the globe 12 . In 2010, WHO estimated that 4.9 million premature deaths per year are attributable to tobacco use, mostly in the form of smoking 2 . This number rose to 7.1 million in 2016, and is expected to reach eight million by 2030 if the current rate of tobacco consumption is unchanged 3 . The health burden of tobacco consumption is disproportionately high in developing countries. It has been estimated that in these countries, tobacco consumption will kill around 40 million people in total between 2005 and 2030 4 . Since smoking is the dominant form of tobacco consumption in developed countries, the progression of smoking prevalence and the expected health damage due to tobacco use have been framed within the smoking epidemic model 5,6 . According to this framework, societies undergo successive stages in which the prevalence of smoking rst increases, and then declines. It is, however, striking that evidence on the progression of the smoking epidemic is extensive for developed countries, but remains fragmented for low-and middle-income countries like India 5 . The movement of the smoking epidemic from high-income to low-and middle-income countries may lead to an unprecedented level of premature mortality, posing one of the biggest preventable public health threats to current and future world health 6 .
As the country with the second-highest level of tobacco consumption worldwide 7-9 , India faces a particularly large health challenge. In 2017, approximately 266.8 million adults in India used tobacco in one form or another 7 ; a gure that is more than twice as high as in the European Union 10 . Because of the health risks and health care costs, tobacco use has been framed as an epidemic in itself. Cigarette smoking is less common in India than it is in western countries and most of the tobacco consumed is in the form of bidi smoking 8 or smokeless tobacco 7 , making cigarette smoking only one component of overall tobacco consumption in the country 11 . According to a 2008 study by Jha et al. 12 , deaths of more than one million adults per year in India can be attributed to different forms of tobacco smoking. The use of smokeless tobacco is also considered a major health risk in India, and has been shown to be associated with an increased risk of death 13,1411,[15][16][17][18][19] .
Most existing studies on tobacco use in India have focused on tobacco prevalence at a single point in time, either for the country as a whole or for speci c localized settings. In addition, most of these studies were based on non-representative sample surveys, and did not consider different types of tobacco products consumed 15,[19][20][21][22][23][24][25] . Moreover, there are only two previous studies on trends in tobacco use over time and across states in India, and these studies focused on tobacco smoking 26 Our study contributes to the scienti c discussion by estimating the progression of the tobacco epidemic in India by gender and across all Indian states. To provide a comprehensive overview of tobacco use in India and its states, we include for the rst time trends in the consumption of smokeless tobacco. By incorporating smokeless tobacco into the classical smoking epidemic model, we not only account for different forms of tobacco consumption and how they have changed over time, but also for countryspeci c peculiarities in tobacco consumption patterns. In addition, our study is the rst to use a highquality dataset to apply the tobacco epidemic model to India and its states in order to discern national and subnational patterns in the progression of tobacco prevalence. Various public health initiatives aimed at reducing tobacco use have been implemented in the country since 1975. We discuss whether these public health initiatives have affected female and male tobacco consumption over time, and conclude that tobacco regulations should be expanded with the aim of reducing the overall health burden of tobacco consumption across India.

Design/Data
We use data from second, third, and fourth rounds (1998-2016) of the National Family Health Survey (NFHS) to estimate national-and state-level age-standardized tobacco prevalence for men and women aged 15-49. The NFHS is a nationally representative, cross-sectional household and individual sample survey that represents 99% of the Indian population living in 27 states and two union territories [28][29][30] . It is the only population survey in the country that provides reliable estimates for various socio-demographic, lifestyle and morbidity indicators, for females aged 15-49 and males aged 15-54. In the NFHS-2, conducted in 1998-99 with a response rate of 97%, 91,196 households were interviewed. In the NFHS-3, undertaken in 2005-06 with a response rate of 98%, 109,041 households were questioned. In the NFHS-4, conducted in 2015-16 with a response rate of 98%, 601,509 household were interviewed.

Measures
Information on current tobacco use of household members was collected through self-administered questionnaires that included the following four questions: 1) "Do you chew pan masala or tobacco?" "Do you currently smoke cigarettes or bidis?" "Do you currently smoke or use tobacco in any other form?" "In what other forms do you currently smoke or use tobacco?" These questions were answered by the head of the household in the 1998-99 round, and by each individual respondent in the other two rounds. We categorized individuals as "tobacco users" if the respondent answered "yes" to either one of the rst three questions, and thus combined tobacco smoking and smokeless tobacco use. In addition, we distinguished between tobacco smoking (smoking of cigarettes, bidis, pipes/hookah, and other items) and smokeless tobacco use (consumption of products like ghutka, pan masala, snuff, and khaini).

Analyses
We calculated the prevalence of age-speci c tobacco use, tobacco smoking, and smokeless tobacco use for men and women aged 15-49 years for the years 1998-99, 2005-06, and 2015-16 for India and its 27 states and two union territories. To account for differences in the population structure across states and time, we age-standardized sex-speci c tobacco prevalence using direct standardization, using the Indian Census population of 2011 as the standard population. We mapped differences in age-standardized tobacco prevalence across states and by sex. We classi ed our data according to ve equally large intervals. We tested whether the differences in the age-and sex-standardized prevalence between the states and in the country as a whole are signi cant 31

Results
The age-standardized tobacco use prevalence among men in India increased by 32% between 1998-99 and 2005-06 (from 27.90-36.81%), and declined by 26% between 2005-06 and 2015-2016 (from 36.81-27.31%), whereas the age-standardized prevalence of tobacco use among women continued to decline from 1998-99 to 2015-16 (Fig. 1a). Between 1998-99 and 2005-06, the age-standardized smoking prevalence increased by 25% for men (from 14.07-17.54%) and by 32% for women (from 0.93-1.24%), whereas the age-standardized prevalence of smokeless tobacco use also increased for men (by 39% from 13.83-19.28%), but declined slightly for women (by 9% from 5.06-4.62%). Between 2005-06 and 2015-16, the age-standardized smoking prevalence declined to 12.20% for men and to 0.40% for women, and the age-standardized prevalence of smokeless use tobacco declined to 15.12% for men and 2.87% for women (Fig. 1b). Despite these declines, the prevalence of tobacco use remains high; at a level comparable to that observed in 1998-99 for both men and women. Tobacco use is much less prevalent among women than men. Among both men and women, but particularly among women, the use of smokeless tobacco is more extensive than tobacco smoking.
At the state level, we also observe an overall pattern of an increase in the age-standardized prevalence of tobacco use (1998-99 until 2005-06) followed by a decline (2005-06 until 2015-16), with similar timing for men and women (Fig. 2). However, exceptions to these overall declining trends after 2005 can be observed for men in the north-eastern regions (states of Manipur, Mizoram, and Nagaland) and women in the north-eastern and the south-western regions (states of Gujarat and Manipur), where the prevalence of tobacco use still increases (Fig. 2). Among women in Gujarat, this increasing trend has been discernible even though the level of tobacco use in the state has been very low.
Similar trends were detected when we dichotomize tobacco use into tobacco smoking and smokeless tobacco ( Figure S1). Tobacco smoking is increasing among men in the northern and north-eastern regions (states of Himachal Pradesh, Arunachal Pradesh, Manipur, and Meghalaya) and among women in the southern regions (states of Karnataka and Tamil Nadu), whereas smokeless tobacco use is increasing among men in the north-eastern regions (states of Manipur, Mizoram, and Nagaland) and among women in the northern, north-eastern, and western regions (states of Rajasthan, Manipur, and Gujarat).
While in most of the Indian states, levels of tobacco consumption have recently been declining, we can observe clear regional differences in the current patterns of smoking and smokeless tobacco use (Fig. 3).
Among men and women in north-eastern states, both tobacco smoking and smokeless tobacco use are more prevalent than the national average. We also nd that in the southern Indian states, prevalence of tobacco consumption is signi cantly lower than the national average. In 2015-16, male smoking prevalence ranged from 5.71% in Maharashtra in the west to 37.70% in Mizoram in the north-east, while female smoking prevalence ranged from 0.01% in Kerala in the south to 8.45% in Mizoram in the northeast. Similarly, the prevalence of smokeless tobacco use among men ranged from 2.64% in Kerala in the south to 32.01% in Nagaland in the north-east, while the prevalence of smokeless tobacco use among women ranged from 0.02% in Himachal Pradesh in the north to 28.18% in Mizoram in the north-east.

Summary of results
Our analysis revealed that in India, the age-standardized prevalence of tobacco use at national level has been higher among men than women: i.e., it increased between 1998-99 and 2005-06 and declined from 2005-06 to 2015-2016 among men, whereas it declined continuously among women, resulting in lower levels in 2015-16 than in 1998-99. For most of the states we observed similar trends with an exception that in the north-eastern states, prevalence of tobacco use has been higher than the national average, and is still increasing.

Discussion of results
Our ndings that the prevalence of tobacco use rst increased and then declined over time, and that the prevalence of smoking was higher among men than women, are in line with the classical smoking epidemic theory 5 . According to the classical smoking epidemic model, the increase and later decline in smoking prevalence took more than 5-6 decades. occurring rst among men and only 2-3 decades later among women, and with a maximum smoking prevalence of approximately 55% among men and approximately 32% among women. Such an extended smoking epidemic pattern with clear differences in timing between men and women has been observed for the majority of European and North American/Australasian countries 32 , and in developing countries like China 6 .
India, on the other hand, has a pattern that is quite distinct from the classical smoking epidemic model. The maximum levels of smoking prevalence for both men and women in India (17.5% for men; 1.2% for women) are considerably lower than those anticipated in the classical smoking epidemic model. In addition, the approximately similar timing of the increase and decrease in tobacco smoking among men and women differs from the patterns found in western countries. Moreover, the rather low prevalence levels in 1998/99 seem to suggest that India underwent an increase followed by a decline in the prevalence of tobacco use over a comparatively short period of time. In light of these apparent deviations from the classical smoking epidemic model, we suggest that India has experienced a 'compressed tobacco epidemic'.
A potential explanation for the compression of the tobacco epidemic in India is that larger numbers of Indians did not take up smoking until the late 20th century, at a time when the hazards of tobacco consumption were well understood. By contrast, men in Europe and North America/Australasia rst took up smoking in the rst half of the 20th century, when the negative health consequences were largely unknown 5,32 . Thus, tobacco use was increasing among Indians at a time when the negative effects of tobacco consumption on health were already very apparent 7 , and prevention strategies had been established [33][34][35] . The lack of social acceptance of tobacco use among women and increases in the price of smokeless tobacco 36 are among other factors that may have contributed to the prevalence of both, tobacco smoking and smokeless tobacco use being substantially lower among women in India than their counterparts in other countries 37,38 . In addition, India stands out as one of the countries where prevalence of tobacco consumption has been substantially reduced since 2005 39 through the introduction of a large number of tobacco control policies ( Figure S2). For example, the country implemented the National Tobacco Control Programme (NTCP) in 2008 with the aim of reducing the burden of tobacco dependence 40 . More recently, India initiated an mCessation service, a mobile service designed to motivate people to quit smoking 41 . The implementation of these policies on a nationwide scale has helped to reduce tobacco consumption, which has, in turn, contributed to the compression of the tobacco epidemic.
We also expanded the classical smoking epidemic model by including the use of smokeless tobacco as an important, country-speci c form of tobacco consumption. On a national level in India, smokeless tobacco consumption follows the trends for tobacco smoking, and therefore ts into the classical smoking epidemic framework. Although the maximum levels of smokeless tobacco use prevalence (19.28% for men and 5.06% for women) have been substantially lower in India than in other countries 42 , they were higher than the prevalence of tobacco smoking for all of the time points, particularly for women. Thus, the use of smokeless tobacco cannot be ignored when studying the tobacco epidemic in India.
Although the patterns of tobacco use for men and women in most Indian states followed the classical smoking epidemic model of an increase followed by a decline, some important exceptions were detected that indicate that there were regional differences in these patterns. The most important exception we observed is that particularly the north-eastern states of the country the prevalence of tobacco smoking and smokeless tobacco use was increasing continuously among men and was declining slowly among women, resulting in tobacco use prevalence that was much higher than the national average. Among the potential explanations for this nding are that in these states, cultural and societal acceptance of tobacco use is higher 43,44 ; access and availability of tobacco products is greater; the implementation of state policies has been less effective, and levels of adherence to these policies has been lower 45 ; and female employment in the tobacco sector has increased 46 .
A further explanation for the relatively high tobacco consumption levels currently observed in the northeastern states is related to the tobacco regulations that were implemented in the 1990s and 2000s, which were less effective in targeting bidi smoking and smokeless tobacco use 43,45 . Indeed, additional analysis revealed that smoking prevalence is still relatively high in states with bidi smoking traditions (Table S1).
In these states, women tend to smoke more bidis and use more smokeless tobacco, while smoking fewer cigarettes (Table S1). This pattern may be partially explained by the comparatively high levels of employment of women in the bidi industry 47 , and by women's preferences due their economic constraints to consume more bidis and smokeless tobacco than cigarettes 37 .
Despite the overall progress India made in reducing tobacco consumption, it remains a public health priority, not only because of the situation in the north-eastern regions, but also because in India as a whole, the levels of tobacco use in 2015-2016 were largely comparable to the levels in 1998-1999 (Table S2), and smokeless tobacco and bidi consumption still play a major role in many parts of the country (Table S1). The complex tax structure in India has kept taxes on cigarettes relatively low, and taxes on bidis and smokeless tobacco products very low, compared to other countries 48 , 36 . Thus, bidis and smokeless tobacco products remain quite affordable. Whereas taxes on cigarettes account for approximately 38% of the retail price, taxes on bidis account only 9% of the retail price. These taxes are well below the tax rates on tobacco products recommended by the World Bank of 65-80% of the retail price 49 . In 2011 in India, the price of a single cigarette was US$0.026, whereas the price of a single bidi was US$0.0033 50

Evaluation of data and methods
Our study is the rst to provide insights into the progression of the tobacco epidemic by gender across Indian states. For our study, we used several rounds of the NFHS to estimate the age-standardized smoking and smokeless tobacco prevalence. Speci cally, we used information on tobacco consumption for men and women between 15 to 49 years of age, as the cumulative tobacco consumption levels between these ages are a strong predictor of adverse health consequences later in life 51,52 . The NFHS has a huge sample size and uses the same sampling procedure across its different rounds to provide relatively accurate representation of the population at national and state levels 30  considerable under-sampling in the 10 states with highest levels of tobacco consumption. In addition, discrepancies in reporting of the procedures used for data collection and quality assurance in GATS may have signi cantly affected its estimates of tobacco use prevalence 54 . Similarly, the NHSDAA covered only men from 25 states 55 , and the GYTS was designed to collect information on tobacco use among young people between ages 13-15 only 56 . Moreover, the NHSDAA was conducted at only one point in time, while the GYTS was conducted at just two points in time. Using the NFHS for the present study was advantageous, as it is a repeated cross-sectional survey representing 99% of the Indian population. Prevalence of tobacco consumption was captured in the NFHS rounds using self-administered questionnaires, which meets the global standardized guidelines 57 .
However, some limitations of our study should be taken into account when interpreting our results. First, we provide information only on current tobacco use, and not on the severity (duration/amount) of the consumption. This latter information is, unfortunately, not provided by the NFHS. Still, we believe that for studying the tobacco epidemic, the information provided by NFHS is reliable. Second, because the NFHS survey rounds were conducted with substantial time lags, the exact timing of the peak of the smoking epidemic cannot be assessed using this data, especially among females. Third, the overall prevalence of tobacco use in the Indian population may be higher 39 , as we focus only on 15-49 age group. Finally, the present study considered only the prevalence of tobacco use, while the classical smoking epidemic model also includes smoking-attributable mortality. This is certainly an important eld for future research that will provide further insights into the overall health burden of tobacco smoking in India.

Conclusion
Our results suggest that India and the majority of its states have experienced a 'compressed tobacco epidemic' which is quite distinct from the experiences of western countries. Despite the overall progress India made in reducing tobacco consumption, lowering it further remains a public health priority, as the prevalence of tobacco use in 2015-16 was highly comparable to the prevalence of 1998-99, and remains high especially in the north-eastern regions. Consumption of smokeless tobacco and bidis still plays a major role in many parts of the country, which can be linked to the very low taxes on bidis and smokeless tobacco products. We therefore recommend that the taxes and/or the prices be raised for these products in particular, but also for cigarettes. Strict implementation of policies, especially in the north-eastern part of the country, would substantially reduce the overall prevalence of tobacco use in India, and, in turn, the burden of tobacco-related morbidity and associated premature mortality in the country.

Availability of Data and Materials
The datasets generated and/or analysed during the current study are available from the Demographic and Health Survey (DHS) website (Link: The DHS Program -Available Datasets).