DATA AND METHODS

The data for the study came from the household questionnaire fielded under NFHS-2—a nationally representative, cross sectional, household sample survey conducted in 1998–99 covering 99% of India’s population living in 26 states (the administrative divisions in India). The survey did not cover 1% of the population living in union territories (the administrative divisions directly under the control of central government) of India. A uniform sample design was adopted in all the states, which is described elsewhere in detail.¹⁵ Briefly, in each state the sample was selected in two stages: the selection of villages and urban blocks (primary sampling units) with probability proportional to size in the first stage, followed by random selection of households within each primary sampling unit in the second stage. The sample was stratified according to a number of variables in each state. These included regions, sub-regions, village size, percentage of males in the non-agricultural sector, percentage of scheduled castes and scheduled tribes, and female literacy.¹⁵

The sample size was large enough to provide reliable estimates of prevalence of tobacco consumption by sex and for different socioeconomic population groups for each of the 26 states. A sample of 91 196 households yielded 334 553 individuals—160 871 men and 154 726 women—15 years and older, which constituted the study population for estimating the prevalence of tobacco consumption.

The questionnaire was administered face-to-face to the head of the household or to any other competent adult member of the household. Fifty nine per cent of the household respondents were female. Majority of the respondents were in age groups 25–39 (42.6%) and 40–59 years (30.9%); 14.5% of the respondents were 15–24 years old while the rest (12.0%) of respondents were 60 years and older. Only 27.8% of the respondents reported themselves to be the head of household. The respondent to the household questionnaire was first asked to list all the usual residents and visitors who stayed in the household the night before the interview. Data for tobacco consumption for each household member 15 years and older including the household respondent were elicited with the help of two questions: does “he or she chews pan masala or tobacco?” or “smokes tobacco?”. The individuals were classified as “chewing tobacco” if the household respondent answered “yes” to the question on chewing tobacco. Similarly, individuals were categorised as “smokers” if household respondent answered “yes” to the question on “smoking tobacco”. No questions were asked on the amount or frequency of tobacco consumption and on the type of tobacco smoked—biris or cigarettes or other forms of smoking prevalent in India (chutta, hukka, etc), which is an important data limitation of the study.

Socioeconomic and demographic information were sought from the household informant at both household level (ownership of different assets, caste, religion) and at individual level (for example, age, sex, education, marital status) for all the household members. Socioeconomic differentials of tobacco prevalence were assessed with respect to three measures of socioeconomic status—caste, education, and household wealth. The caste is the basis of social hierarchical organisation of the Hindu religion—the predominant religion followed in India. The government of India has identified, in a schedule of the Constitution of India, the castes occupying the lowest rung of social hierarchy as the most socially disadvantaged, and classified them as scheduled castes (SCs) and the scheduled tribes (STs). In addition, the government has identified some occupational castes as socially backward and classified them as “other backward castes” (OBCs).

To measure the economic status, we created a “household wealth index” using principal component analysis of different household assets and dwelling characteristics (type of floor, roof, etc) based on methods previously described.¹⁶ The household wealth index was then used to divide the population into five quintiles.

STATA 7.0 was used to carry out statistical analyses.¹⁷ Univariate analysis was done to assess the distribution of the sample and to compute overall prevalence of chewing of tobacco/pan masala and smoking of tobacco. Both point estimates and robust 95% confidence intervals (based on robust standard errors after adjusting for strata and clustering at primary sampling unit (PSU) level are presented. The sample based prevalence estimates were converted into population based estimates taking into account the population of India as per the Census 2001. Bivariate and multivariate logistic regression models were estimated to assess the unadjusted and adjusted association, respectively, of different socioeconomic (household wealth, years of schooling, religion, and caste) and demographic characteristics (urban/rural residence, sex and age) with tobacco consumption. Both bivariate and multivariate regression models were estimated after applying for sampling weights and adjusting for multi-stage clustered sampling designs using “svy” command in STATA 7.0. Collinearity between explanatory variables was tested using variance inflation factor (VIF) command in STATA, which was found to be insignificant. MAPINFO software was used to produce thematic maps of male tobacco smoking and chewing prevalence.

RESULTS

Overall prevalence

Sixteen per cent of the study population (29.3% men and 2.3% women) smoked tobacco; 20% of the study population (28.1% men and 12.0% women) chewed tobacco/pan masala; and 30% of the study population (46.5% men and 13.8% women) either smoked or chewed tobacco. Table 2 shows the percentage and approximate number of people who consumed some form of tobacco. Approximately, 194 million people aged 15 years and older (150 million men and 44 million women) consumed some form of tobacco. Almost 79% of tobacco consumers lived in rural areas, slightly more than the share of rural population in the total population (73%).

State level variation

The prevalence of both smoking and chewing tobacco/pan masala varied significantly among different states in India (table 3, fig 1). Some regional patterns were observed for chewing tobacco/pan masala. Chewing of tobacco was relatively more common in the Central, Eastern, Western (except Goa) and Northeastern states (except Meghalaya) compared to in the Northern and Southern states. However, in the Northern states, where chewing is relatively less common, smoking of tobacco is relatively higher (except in Punjab where tobacco prevalence is one of the lowest as majority of its population (58%) practice Sikh religion, which prohibits tobacco consumption).

Demographic and socioeconomic predictors

Unadjusted relationships

Columns 2–4 and columns 6–8 of tables 4 and 5, respectively, summarise the prevalence and unadjusted odds of smoking and chewing in different socioeconomic and demographic groups by sex. The change in prevalence of tobacco consumption with age is also shown in fig 2. The prevalence of both smoking and chewing tobacco increased prominently with age until 50 years and then remains constant or declined. Compared to chewing, the smoking prevalence declined more steeply after 50 years of age among men.

The prevalence of both chewing tobacco/pan masala and smoking tobacco was significantly higher in rural, poorer, and uneducated populations compared to urban, wealthier, and more educated populations, respectively, both in men and women, though the differentials for chewing tobacco were smaller. The socioeconomic gradients (by household wealth as well as by education) were steeper for women than for men for both chewing tobacco/pan masala and smoking tobacco. Men in the poorest quintile have 3.5 times higher unadjusted odds for smoking and 3.7 times higher unadjusted odds for chewing tobacco than among men from the richest quintile, while the similar figures among women were 9.9 times for smoking and 4.8 times for chewing tobacco. Men with no schooling had 5.6 times higher unadjusted odds of smoking and 3.1 times higher unadjusted odds of chewing tobacco than men with more than 11 years of schooling, while in women the unadjusted odds were 41 times higher for smoking and 13 times higher for chewing tobacco.

Prevalence of chewing tobacco/pan masala was the highest among the scheduled tribe (ST) population and lowest among forward castes and other backward castes (OBC). The forward castes were also least likely to smoke tobacco. While no significant differences were seen in the prevalence of smoking and chewing tobacco/pan masala among Hindu, Muslim, and Christian populations, the Sikh population had a significantly lower prevalence of tobacco consumption.

Multivariate analysis: adjusted relationships

Column 5 and column 9 in table 4 (for smoking) and table 5 (for chewing tobacco/pan masala) present the adjusted odds ratios from multivariate logistic regression models to tease out the adjusted association between different socioeconomic and demographic characteristics and adjusted odds of chewing and smoking tobacco. Though the relations observed in bivariate analysis persisted in multivariate analysis, the effect of certain sociodemographic variables got diluted after controlling for other confounding socioeconomic characteristics.

Compared to the younger population (15–24 years), the older population (25+ years) had a greater likelihood of both chewing and smoking tobacco. In the multivariate analysis, the number of school years emerged as one of the strongest predictors—in terms of odds ratio—for both smoking and chewing among both men and women followed by household wealth. Scheduled castes and scheduled tribes were more likely to smoke and chew tobacco compared to the forward castes (except for smoking among men where no significant differentials were observed among ST, OBC, and forward castes), even after controlling for household wealth and education. Compared to Hindus, Muslim men were more likely to smoke, though no differentials were observed among women. However, Muslim women were more likely to chew tobacco than Hindu women. Sikh religion emerged as one of the strongest predictord among both men and women for not chewing or smoking tobacco. The differentials by state of residence also persisted in the multivariate analysis (not shown in tables 4 and 5).

No significant association was observed between urban/rural residence and smoking or chewing of tobacco/pan masala among men after controlling for other characteristics. However, rural women were less likely to chew tobacco than urban women (odds ratio 0.87, p < 0.05).

DISCUSSION

Prevalence

This is the first study to provide nationally representative aggregate prevalence estimates of tobacco consumption by different socioeconomic and demographic characteristics. Though superior to the existing studies in terms of study design and representativeness, this study also suffers from several data limitations, which may potentially affect the prevalence estimates. The first step in validating the estimates from the NFHS-2 survey may be to compare the estimates from earlier studies while noting the methodological differences that may be responsible for the observed differences. Table 1 summarises prevalence estimates of tobacco consumption arrived in previous studies, and comparable estimates from the NFHS-2 survey. The prevalence rates estimated from NFHS-2 are slightly lower for smoking, but higher for chewing than the estimates arrived in the 52nd NSS, which used the same sampling methodology and household informants (table 1). However, the questions were phrased differently in the two surveys that may partly account for the differences. For example, while the question on chewing tobacco was very ambiguous in the 52nd NSS as discussed earlier in the first section, NFHS-2 explicitly asked about chewing tobacco as well as about chewing paan masala. Similarly, while the 52nd NSS explicitly probed for all forms of smoking tobacco (biri/cigar/cigarette/hukka), NFHS-2 simply asked whether any household member “smokes tobacco”, leaving the interpretation of “tobacco” to the respondents themselves, which may explain the lower estimates of smoking prevalence in NFHS-2.

However, the estimates from both the NFHS-2 and the 52nd NSS using proxy household informants are lower than the prevalence rates of tobacco consumption estimated by WHO,¹ based on small studies conducted in India in different time periods, and also from those of other small studies based on special population groups in small geographical areas (table 1). Use of household informants in NFHS-2 who may not be aware of the use of tobacco by other household members rather than the individual self reports used in most of the earlier small scale studies (table 1) may be partly responsible for the differences observed, besides the differences in the reference population groups in these studies. To estimate the potential degree of under-reporting in NFHS-2 due to use of household informants, the prevalence of tobacco consumption was estimated separately among household respondents themselves and among other household members (table 6). The age adjusted prevalence rates in the household respondents’ population were higher among men and women, respectively, by almost 10.6% and 1.5% for chewing tobacco and by 5.4% and 0.5% for smoking tobacco. When examined by age group (table 6, fig 2), the under-reporting is consistently higher in the younger age groups among both men and women. Though a recent study conducted in Delhi¹⁸ found a high degree of agreement between individual self reports and household informant reports for tobacco use, results from this study show that prevalence may be underestimated when it is elicited through a household informant. Taking into account all the limitations of the study and the previous literature on prevalence of tobacco consumption in India, it is safe to conclude that the study provides robust lower bound estimates for the prevalence of tobacco consumption in India. The real estimates may be almost 11% and 1.5% higher for chewing among men and women, respectively. Similarly the prevalence of smoking may be underestimated by almost 5% in men and 0.5% in women.

State level variation

The odds of tobacco consumption varied significantly across different Indian states even after controlling for individual-level socio-economic and demographic characteristics. For example, the chewing prevalence was distinctly higher in Central and Eastern India, and in the Northeastern states. The state level variation may reflect distinct regional socio-cultural patterns or effect of different public policies on tobacco in different states. Future studies should explore the reasons for inter-state differentials as this can provide important insights into effect of different public policies and their interaction with local socio-cultural patterns on use of tobacco.

Poverty and illiteracy

Tobacco consumption was the highest in the least educated, poorest, and scheduled castes and scheduled tribes. Though systematic under-reporting of tobacco use in some social groups due to differentials in social stigma attached with tobacco use in different sociodemographic groups may have contributed to this apparent trend, it is unlikely to account completely for the trends observed, considering the very wide differentials. The socioeconomic differentials in tobacco consumption from this study also compare well with the findings from previous studies in India^10,11,19 and elsewhere.²⁰

Why poor or less educated people consume more tobacco remains an open empirical question for further investigation. Education emerged as a relatively stronger predictor than household wealth, both among men and women. It is likely that poor and less educated people are: less aware of the health hazards of tobacco consumption; more likely to find themselves in conditions predisposing them to initiation of smoking and chewing of tobacco; and more likely to have higher degree of fatalism or higher overall risk taking behaviour.²⁰ The findings of the study highlight that an agenda to improve health outcomes for the poor and other disadvantaged groups in India must also include effective interventions to control tobacco, as these groups may suffer from disproportionate burden of tobacco induced morbidity and mortality. In addition, each intervention should be evaluated for its effectiveness separately in different socioeconomic and cultural groups, since access and effectiveness of different programmatic strategies may vary across these groups.

Demographics of tobacco consumption

The cross sectional nature of the data did not allow us to assess the trends in tobacco consumption over time or with age. The observed increase in prevalence of tobacco consumption with age can be due to a cohort effect (declining prevalence over time with younger cohorts having lower prevalence) or an age effect (younger people having lower prevalence, with more people initiated into tobacco consumption as they get older) or simply due to under reporting of tobacco use among young people. Previous literature suggests no declining trends in tobacco consumption over time in India. As indicated earlier, the differences in the prevalence of tobacco use between respondent and non-respondent population is much higher in the 15–24 year age group (fig 2, table 6). However, as is very clear in fig 2, trends in prevalence rates with age are very similar in both respondent and non-respondent sub-samples (table 6, fig 2). Hence the observed increase in tobacco prevalence is the partly the result of under-reporting in the younger age groups and party due to an actual increase in prevalence of tobacco consumption with age up to mid 50s (fig 2). This has important policy and programme implications—the initiation into tobacco use may occur at any age and not just among young people. This implies that programmes to control tobacco have to focus on almost all age groups up to the age of 50.

Conclusion

In India, as in most low income countries, death in middle age is increasing in relative importance due to an increase in smoking related deaths.²¹ The disease burden, health care costs as well as other fiscal losses resulting from premature deaths attributable to tobacco consumption will rapidly increase.^8,22–24 Earlier WHO estimates suggest that deaths and disability adjusted life years (DALYs) attributable to tobacco use in India will increase from 129 000 deaths and 1719 DALYs in 1990 to over 1.5 million deaths and 24 024 DALYs by 2002.⁸ The high levels of tobacco consumption among disadvantaged population groups may lead to a doubling of the disease burden in these social groups from chronic illnesses related to tobacco consumption as well as from communicable and nutrition related diseases, which still account for a large share of total disease burden in the disadvantaged social groups in India. In addition to the differences in prevalence of tobacco consumption between disadvantaged and better off groups, the type and amount of tobacco consumption, which are shown to be directly associated with incidence of tobacco related diseases and mortality, may also vary between the two, further aggravating the differences in disease burden attributable to tobacco between the two groups.¹⁹ Though no data were collected in the NFHS-2 on the type and amount of tobacco consumed, which remains one of the weaknesses of survey, other studies suggest that the poor are 8–10 times more likely to smoke bidis^22,25 that may lead to greater risk of oropharyngeal and lung cancer in this group than among the better off.^7,25

The public policy implications of our study are twofold. Research on tobacco use needs to be considerably systematised with use of more consistent definitions of tobacco consumption and study methodologies. More rigorous comparable prevalence studies over time are needed to establish the trends in prevalence and evaluate the effect of different public policies pursued to control tobacco use. Though studies involving use of household informants are much more economical and easier to administer than those based on individual self reports, the prevalence may be underestimated and need to be adjusted upwards to provide real estimates. In addition, future studies should investigate the prevalence rates of different tobacco products (both smoking and chewing tobacco) separately, as the economic and health effects of different products may vary considerably, and because of the potential differences in prevalence of use of different tobacco products across different sociodemographic groups.