Age of initiation, intensity and cessation of tobacco smoking among Bangladeshi adults: evidence from a cross-sectional nationally representative survey

Abstract

Background

Tobacco smoking is a major global health issue, linked to nearly 8 million deaths annually. This study explores the smoking behaviors of adults in Bangladesh, focusing on the age of initiation, intensity, and cessation attempts.

Methods

The study is based on the data from the Global Adult Tobacco Survey, 2017-18, which comprised 12,774 respondents from Bangladesh aged 15 years and older. Statistical analyses included descriptive statistics, weighted t-tests, weighted ANOVA, Rao-Scott Chi-square tests, and regression models (Survey-weighted linear and negative binomial, Firth’s penalized logistic) to assess determinants of smoking behaviors.

Results

Among the respondents, 18.13% were tobacco smokers. Among smokers, 43.85% initiated tobacco use at or before age 18, 35.20% had a successful cessation attempt, and 10.78% were heavy smokers. Geographical division and education level were significant differentials for early initiation, smoking intensity, and successful cessation attempt. Lower paid working people including agriculture workers, daily laborers, and unemployed individuals able to work were associated with high smoking intensity. Gender and age of respondents were significantly associated with age of initiation but not with smoking intensity or cessation attempt. Residence and wealth index were not significant for any outcome.

Conclusion

These findings emphasize the need for targeted tobacco control strategies in Bangladesh to address early initiation, high smoking intensity, and low cessation attempt rates. Analytical results highlight the need for tailored intervention programs and laws. Prioritize tobacco as an addictive substance and target lower paid working and less educated populations, along with high-burden regions such as Mymensingh, Chattogram, and Sylhet.

Introduction

Tobacco smoking is one of the largest global public health threats, recognized as the leading behavioral risk factor for non-communicable diseases (NCDs), accounting for nearly 8 million fatalities annually worldwide [1]. The distribution and intensity of tobacco use are crucial, particularly for low and middle-income countries, with Bangladesh being one of the world’s leading tobacco users and a low-income country [2].

The widespread prevalence of cigarette smoking found in literature highlighting the urgent need for comprehensive tobacco control measures [3]. The age at which individuals begin smoking and the intensity of their smoking, along with the cessation process, are closely associated with long-term negative consequences, such as a higher chance of developing chronic illnesses like lung cancer, COPD, heart disease, stroke, and respiratory infections; difficulties in school; substance abuse; and an increased risk of developing diseases like diabetes, high blood pressure, and chronic respiratory disorders as a result of extended exposure to harmful substances and unsuccessful or delayed attempts to quit [4–10] Initiating smoking at a young age significantly increases the likelihood of developing lung cancer, leading to a reduction in life expectancy by at least ten years compared to nonsmokers [11, 12]. Addressing tobacco use requires global strategies like the WHO’s Framework Convention on Tobacco Control (FCTC) and the MPOWER measures, which aim to reduce tobacco use through monitoring, smoke-free laws, tobacco cessation support, health warnings, advertising bans, and tax increases [13, 14]. The FCTC, adopted by over 182 countries, promotes tools such as raising taxes, smoke-free laws, graphic health warnings, and advertising bans. However, the implementation and enforcement of these guidelines vary significantly, particularly in Asian countries like Bangladesh, India, and China [13, 14].

Despite the global push for stringent tobacco control, Bangladesh faces persistent and complex challenges in reducing tobacco consumption among its population. World Health Organization (WHO) reported that about 19.2 million adults were current tobacco smoker in Bangladesh [15, 16]. This highlights the critical need for an in-depth understanding of smoking behaviors, particularly the age of initiation, intensity, and cessation, as a fundamental element for effective public health strategies. Even slight variations in the age of smoking initiation, intensity, and cessation in Bangladesh can substantially influence mortality risk and the burden of tobacco-related diseases later in life, with WHO highlighting that continued smoking into adulthood increases mortality risk and emphasizing the importance of early cessation to reduce health burdens [15, 17–19]. The age of smoking initiation, smoking intensity patterns, and cessation may be influenced by the strength and nature of anti-smoking efforts, as well as socio-demographic factors such as gender (higher in males), education level (lower education correlates with higher smoking), socio-economic status (higher income may increase initiation risk but lower wealth associated with continuation), and occupation (day laborers show higher prevalence) [20–22]. In England, Australia, Canada, and the United States, lower socioeconomic status, including factors such as lower education levels and unemployment, has been linked to higher smoking rates, highlighting significant socio-economic disparities in tobacco use, while in low and middle-income countries, individuals with lower educational attainment are more likely to smoke and less likely to quit [23–25]. Several studies have been conducted on tobacco smoking behavior within country and in other countries [1, 17, 26–34]. However, some of these studies were carried out in developed countries (e.g., the USA, UK, and Scotland) [31, 34]. While others provide only partial information related to the topic [17, 26, 29, 33, 35]. No prior study has simultaneously modeled three smoking behavior outcomes: age of initiation, smoking intensity, and cessation in Bangladesh. Therefore, main objective of this study is to examine the age of smoking initiation, intensity and cessation among Bangladeshi adults and to identify key socio-demographic factors associated with smoking behaviors. Using data from the Global Adult Tobacco Survey 2017-18 in Bangladesh, this study further aims to address the cracks in policies in the country.

Methods

Study overview and design

For this study, data from the Global Adult Tobacco Survey (GATS), conducted in Bangladesh during 2017–2018 has been used. The Global Adult Tobacco Survey (GATS) is a standardized and unified monitoring tool established by the World Health Organization (WHO) to collect, assess and distribute critical data on tobacco use from participating countries [36]. Bangladesh, located in South Asia on the northern land of the Bay of Bengal, lies between latitudes 20°34’ and 26°38’ North and longitudes 88°01’ and 92°41’ East [37]. The country covers an area of approximately 148,460 square kilometers (57,320 square miles) [38]. Known for its diverse landscapes and rich cultural heritage, Bangladesh has a unique identity.

The survey was carried out in collaboration with the National Institute of Preventive and Social Medicine (NIPSOM), the National Institute of Population Research and Training (NIPORT), and the Bangladesh Bureau of Statistics (BBS), following WHO guidelines. The survey collected data through standardized questionnaires administered via face-to-face interviews. A multi-stage, stratified cluster sampling design was employed to produce nationally representative estimates covering both urban and rural populations. Trained data collectors adhered to standardized protocols to ensure data quality and reliability, and rigorous quality control measures were implemented throughout data collection and processing to maintain accuracy. Further details on the survey methodology, questionnaires and data gathering methods for GATS Bangladesh 2017 are accessible in their respective survey publications [39, 40].

Study population

The study population included Bangladeshi individuals aged 15 years and older residing in Bangladesh at the time of data collection. The 2017–2018 GATS Bangladesh survey employed a multistage stratified cluster sampling design. Its sampling frame was based on 293,533 Enumeration Areas (EAs) defined by the national census, each representing a cluster referred to as Primary Sampling Units (PSUs). To enhance representativeness these EAs were first stratified according to the eight administrative divisions as well as rural areas and urban areas, ensuring relatively homogenous subgroups within each stratum. From this stratified frame, 496 EAs were randomly selected and treated as clusters (PSUs), meaning that individuals within the same EA could share similar socioeconomic and demographic characteristics. In each selected EA, 30 households were systematically sampled, and one eligible respondent aged 15 years or older was chosen from each household for interview. In total, 14,880 households were initially selected, and 12,783 individuals completed the survey, yielding a response rate of 90.8%. After applying the predefined inclusion criteria, the final analytic sample included 12,774 participants.

To account for the complex survey design, sampling weights were computed for each respondent based on the inverse probability of selection at the PSU, household, and individual levels. These weights were additionally adjusted for non-response to ensure that the final estimates remain nationally representative of the Bangladeshi population.

Outcome measurement

In this study, the outcome variables are age of initiation, smoking intensity, and cessation. Age of initiation is defined as the age at which individuals first began using tobacco smoking. Specifically, it was determined through the question: “How old (in years) were you when you first started smoking tobacco daily?” A cutoff age of 18 was used to distinguish between early and later initiation. This cutoff has been used only for the descriptive report. The model has been fitted using scale data.

Smoking intensity is defined as the number of smoking products (e.g., bidi, cigarette, churut, etc.) a respondent consumes per day. Respondents were classified into three categories based on their daily consumption: Light smokers (smoking ≤ 10 tobacco products consumed per day), Moderate smokers (smoking 11–20 tobacco products consumed per day), Heavy smokers (smoking > 20 tobacco products consumed per day) [41]. This classification has again been used only for descriptive purposes; the model has been fitted to rate.

Cessation has been determined based on the question: “Thinking about the last time you tried to quit, how long were you away from smoking?” If the respondent reported being abstinent for 28 days or more, the attempt was classified as successful attempt; if the period of abstinence was shorter than 28 days, the attempt was classified as unsuccessful [6].

Tobacco smokers were identified based on the response to the question: “Do you currently smoke tobacco daily, less than daily, or not at all?” Respondents who reported smoking daily or less than daily were categorized as smokers, while those who reported not smoking at all were categorized as nonsmokers.

Explanatory variables

The selected socio-demographic and economic variables, which are available in the survey data, are considered as exploratory variables. These variables include gender, age, urban/rural residence, geographical region, level of education, occupation, and economic status. In line with methodological guidelines from previous research, some variables were reclassified to enhance the precision of the analysis. Age was categorized into five groups: 15–24, 25–34, 35–44, 45–54, and ≥ 55 years. This segmentation helps to understand patterns of smoking and cessation across different life stages. Economic status is measured by the Wealth Index, which categorizes respondents as poorest, poorer, poor, rich, or richest. Occupation varies widely, spanning government employee, non-government employee, small business, large business, farming (land owner and farmer), agricultural worker, industrial worker, daily laborer, other self-employed, student, homemaker or housewife or housework, paid domestic worker, retired, unemployed (able to work), unemployed (unable to work), and other. The Place of residence is noted as urban or rural. At the same time, the division Barisal, Chattogram, Dhaka, Khulna, Mymensingh, Rajshahi, Rangpur, and Sylhet represents the respondent’s geographical location. Education level (no formal schooling, less than primary school, primary school, less than secondary school, and secondary school and above) is also included as socio-demographic variables.

Statistical analysis

The data have undergone a comprehensive statistical analysis using appropriate techniques. Respondents’ characteristics have been described with descriptive statistics: categorical variables as observed frequencies and weighted percentages, and continuous variables as weighted means and weighted standard errors. Associations between socio-demographic and economic characteristics and the outcome variables were assessed using Rao-Scott Chi-square tests, independent weighted t-tests, and weighted ANOVA. Smoking cessation has been evaluated with Rao-Scott Chi-square tests to compare the weighted percentage of successful cessation attempts across categories. Weighted t-tests has been used for bivariate comparisons, and weighted ANOVA for comparisons involving more than two groups. Statistical computations has been performed using Stata version 14.2.

Multivariable modeling has been carried out in R using the R survey package, which includes the complex survey design features (primary sampling units (PSUs), sampling weights, and strata). First, a survey-weighted linear regression model has been used to estimate the log-transformed age of smoking initiation, and the results are presented as regression coefficients and p-values [42]. Next, the significant correlates of the intensity of smoking has been identified using survey-weighted negative binomial regression and the results are presented as incidence rate ratios (IRRs) with 95% confidence intervals (CIs) [43]. Finally, Firth penalized logistic regression has been used to identify the significant correlates of success smoking cessation attempt (coded as 1 (successful) and 0 (unsuccessful)), and penalized likelihood estimation has been employed in the logistf package, which is appropriate for sparse data and rare-event bias [44].

Sensitivity analyses has been conducted for age of initiation and intensity by comparing the primary survey-weighted model with weight-trimmed estimates at the 99th percentile and the unweighted model [45]. For smoking cessation, a sensitivity analysis is performed by comparing estimates from the Firth-penalized logistic model with those from the survey-weighted and unweighted logistic regression models [46, 47]. The stability of estimates across all models has been assessed using a 10% or less percentage-change threshold [48]. The two occupational groups: Homemaker/Housewife and Paid Domestic Worker are excluded from all three models and sensitivity analyses because they had very few cessation events (n = 1 and n = 0, respectively), which precluded stable estimation.

Model predictive performance has been assessed using root mean squared error (RMSE) and mean absolute error (MAE) for the survey-weighted linear and negative binomial regression models, and the area under the receiver operating characteristic curve (AUC) for the Firth-penalized logistic regression model. These metrics evaluate predictive precision for continuous and count outcomes, and discriminatory ability for the binary cessation outcome, respectively.

Results

The characteristics of the respondents are presented in Table 1. The study analyzed data from 12,774 participants, comprising 49.01% males and 50.99% females. The mean age of the respondents was 39.39 years (Standard Error, SE = 17.49).

Table 1.

Socio-demographic and economic characteristics of study subjects from Bangladesh GATS, 2017-18 (N = 12774)

Characteristics variable	n (Weighted %)
Gender
Male	6079 (49.01)
Female	6704 (50.99)
Age (years)	39.39(17.49)*
15–25	2345 (28.22)
25–35	3363 (23.57)
35–45	3034 (20.46)
45–55	2088 (12.56)
≥55	1953 (15.19)
Residence
Urban	6356 (25.11)
Rural	6427 (74.89)
Division
Barisal	1609 (5.23)
Chattogram	1577 (18.95)
Dhaka	1506 (26.92)
Khulna	1621 (10.97)
Mymensingh	1591 (7.46)
Rajshahi	1705 (13.17)
Rangpur	1653 (10.59)
Sylhet	1521 (6.70)
Educational Level
No Formal Schooling	3581 (27.79)
Less Than Primary School	2057 (16.27)
Primary School Completed	1573 (12.17)
Less Than Secondary	2710 (22.03)
School Completed	1187 (10.27)
High School Completed	871 (6.80)
Graduation Completed	483 (2.59)
Postgraduate Degree	320 (2.09)
Occupation
Government Employee	281 (1.37)
Non-Government Employee	873 (6.13)
Business (Small)	1172 (8.19)
Business (Large)	246 (1.49)
Farming (Land Owner & Farmer)	850 (8.00)
Agricultural Worker	326 (2.53)
Industrial Worker	210 (2.19)
Daily Laborer	1068 (7.66)
Other Self-Employed	520 (3.53)
Student	869 (10.68)
Homemaker / Housewife	5338 (39.64)
Paid Domestic Worker	153 (0.84)
Retired	164 (1.03)
Unemployed, Able to Work	191 (2.33)
Unemployed, Unable to Work	251 (2.15)
Other (Specify)	271 (2.24)
Wealth Index
Poorest	2558 (20.00)
Poorer	2564 (19.59)
Poor	2603 (20.77)
Rich	2721 (21.14)
Richest	2337 (18.49)

*weighted mean (weighted standard error) has been reported

It has been found that 18.13% of the respondents were smokers (not reported in any table). Table 2 shows that, in 2017–18, 43.85% of smokers began smoking before age 18, and 64.80% reported unsuccessful smoking cessation attempt. Regarding smoking intensity, 10.78% were heavy smokers, 23.17% medium smokers, and 66.05% light smokers.

Table 2.

Distribution of Outcome Variable - Smoking Behavior in Bangladesh, GATS 2017-18 (N = 2492)

Outcome variable	n (Weighted %)
Age of initiation
Early initiation (≤ 18)	992 (43.85)
Later initiation (> 18)	1231 (56.15)
Cessation
Greater than 28 days	345 (35.20)
28 days or less	717 (64.80)
Intensity*
Light smoker	1492 (66.05)
Medium smoker	539 (23.17)
Heavy smoker	228 (10.78)

*Light smoker is defined as a person smoking ≤ 10 tobacco products per day, medium smoker is defined as a person smoking 11–20 tobacco products per day, and Heavy smoker is defined as a person smoking > 20 tobacco products per day. Missing values are not reported in the table and have been excluded from the analysis

Mean age of smoking initiation, smoking intensity, and cessation attempt rates across demographic and socioeconomic characteristics are reported in Table 3. The mean age of smoking initiation for males was 20.05 years (SE: 0.18), while for females it was 27.97 years (SE: 7.16). Among different age groups, those aged 15–25 years initiated smoking at the youngest age with a mean age of 16.79 years (SE: 0.28), followed by those aged 25–35 years, with a mean age of 19.79 years (SE: 0.25). Mymensingh had the lowest mean age of smoking initiation at 18.53 years (SE: 0.54) among all regions. Education levels were also related to smoking initiation age, and those with no formal schooling (mean: 19.75 years, SE: 0.46) or less than primary education (mean: 19.72 years, SE: 0.29) initiated smoking earlier than those with higher education levels. Students (mean: 18.11 years, SE: 1.47) and unemployed individuals unable to work (mean: 17.39 years, SE: 0.82) began smoking at an earlier age.

Table 3.

Comparing socio-demographic and economic characteristics to smoking behavior in Bangladesh (GATS 2017-18) (N = 2492)

Variable	Age of initiation		Smoking Intensity		Smoking cessation
	Weighted mean (SE)		Weighted mean (SE)		n (Weighted %)
Gender
Male	20.05 (0.18)		11.43 (0.35)		344 (35.37)
Female	26.97 (7.16)		7.65 (1.32)		1 (17.61)
p-value	0.333		0.006		0.382
Age
15–25	16.79 (0.28)		8.74 (1.25)		25 (41.34)
25–35	19.79 (0.25)		10.24 (0.41)		72 (29.73)
35–45	21.39 (0.36)		12.44 (0.62)		95 (30.55)
45–55	20.20 (0.36)		12.75 (0.75)		75 (41.39)
≥55	20.69 (0.81)		11.22 (0.69)		78 (37.4)
p-value	< 0.001		< 0.001		0.339
Residence
Urban	20.35 (0.33)		10.35 (0.73)		156 (36.99)
Rural	20.12 (0.28)		11.69 (0.40)		189 (34.61)
p-value	0.596		0.108		0.631
Division
Barisal	20.05 (0.44)		8.05 (0.50)		48 (53.62)
Chattogram	19.80 (0.42)		11.77 (0.70)		28 (16.02)
Dhaka	20.21 (0.38)		11.07 (0.85)		55 (45.41)
Khulna	20.96 (0.48)		11.14 (0.72)		35 (40.03)
Mymensingh	18.53 (0.54)		11.65 (0.67)		35 (34.95)
Rajshahi	19.96 (0.48)		10.64 (0.67)		46 (33.68)
Rangpur	22.69 (1.43)		13.41 (1.39)		63 (41.67)
Sylhet	19.06 (0.52)		11.74 (0.74)		35 (28.62)
p-value	0.011		< 0.001		< 0.001
Educational Level
No Formal School	19.75 (0.46)		12.19 (0.51)		110 (32.9)
Less Than Primary	19.72 (0.29)		10.71 (0.58)		75 (34.16)
Primary School Completed	20.58 (0.40)		10.74 (0.64)		34 (33.74)
Less Than Secondary	19.97 (0.43)		10.55 (0.66)		62 (40.48)
School Completed	21.45 (0.55)		12.01 (2.36)		21 (29.02)
High School Completed	23.15 (0.55)		13.12 (2.56)		17 (47.89)
Graduation Completed	22.62 (1.01)		9.21 (1.60)		17 (33.24)
Postgraduate Degree	22.48 (1.36)		9.29 (2.38)		9 (59.48)
p-value	< 0.001		0.177		0.620
Occupation
Government Employee	20.30 (1.17)		7.50 (1.41)		5 (14.00)
Non-Government Employee	20.68 (0.44)		9.86 (0.90)		25 (40.27)
Business (Small)	20.77 (0.41)		11.69 (0.71)		69 (28.79)
Business (Large)	23.73 (1.32)		14.70 (2.23)		12 (34.09)
Farming (Land Owner & Farmer)	20.06 (0.38)		12.06 (0.71)		72 (46.46)
Agricultural Worker	20.54 (0.47)		12.27 (1.19)		26 (41.55)
Industrial Worker	19.84 (0.71)		9.67 (0.94)		3 (33.52)
Daily Laborer	19.08 (0.27)		11.09 (0.51)		54 (25.39)
Other Self-Employed	19.26 (0.42)		10.85 (0.69)		32 (31.87)
Student	18.11 (1.47)		9.47 (3.09)		6 (39.73)
Homemaker / Housewife	31.36 (8.93)		8.11 (1.53)		1 (29.36)
Paid Domestic Worker	25.00 (…)		3.00 (…)		0 (0.00)
Retired	20.21 (0.79)		9.89 (1.05)		10 (58.46)
Unemployed, Able to Work	18.98 (0.87)		17.40 (4.37)		6 (19.72)
Unemployed, Unable to Work	17.39 (0.82)		8.47 (1.09)		7 (43.56)
Other (Specify)	19.89 (0.66)		10.25 (1.12)		17 (44.62)
p-value	< 0.001		< 0.001		0.085
Wealth Index
Poorest	19.85 (0.36)		10.60 (0.57)		55 (35.59)
Poorer	19.79 (0.33)		11.32 (0.71)		54 (30.58)
Poor	19.54 (0.37)		11.55 (0.67)		54 (28.17)
Rich	20.24 (0.29)		10.39 (0.55)		86 (42.65)
Richest	20.87 (0.60)		12.45 (0.70)		96 (36.44)
p-value	0.285		0.166		0.241

(…) indicates that the standard error (SE) is not available. P-values have been obtained using the Rao–Scott adjusted chi-square test for categorical variables, and -weighted t-test and ANOVA for continuous variables

Average smoking intensity was 11.43 smoked tobacco products per day (SE: 0.35) for males and 7.65 smoked tobacco products per day (SE: 1.32) for females. The highest smoking intensity was observed in the 45–54 years age group, with an average of 12.75 smoked tobacco products per day (SE: 0.75). Among divisional regions, Rangpur had the highest smoking intensity, with an average of 13.41 smoked tobacco products per day (SE: 1.39). Smoking intensity was higher among individuals with lower levels of education. Those with no formal schooling had an average of 12.19 smoked tobacco products per day (SE: 0.52), and individuals with a high school diploma reported an average of 13.12 smoked tobacco products per day (SE: 2.56). Additionally, individuals with large businesses (average of 14.70 smoked tobacco products per day, SE: 2.23) and unemployed individuals able to work (average of 17.40 smoked tobacco products per day, SE: 4.37) exhibited higher smoking intensities than their counterparts.

Table 3 also shows that successful smoking cessation attempt was 35.37% among males and 17.61% among females. Successful smoking cessation rates were lower in Chattagram (16.02%) and Sylhet (28.62%) than in other divisions. Additionally, unemployed individuals who were able to work had the lowest rate of successful cessation attempts (19.72%).

To identify the factors associated with age of smoking initiation, smoking intensity, and successful smoking cessation attempt, regression models have been used, and the results are summarized in Table 4. A survey-weighted linear regression model [42] has been used to identify factors associated with the age of initiation (log transformed). The normality assumption has been evaluated using a Q–Q plot (Fig. 1). The residual plot confirms the linearity assumption and the absence of heteroscedasticity (Fig. 2). The analytical results indicate that the respondent’s gender, respondent’s age, geographical division, and educational level were significantly associated with the age of smoking initiation.

Table 4.

Socio-demographic and economic correlates to smoking behavior in Bangladesh (GATS 2017-18)

Variable	Log Age of Initiation*	Smoking Intensity**	Smoking Cessation***
	Coeff (p-value)	IRR (95%CI)	OR (95%CI)
Gender
Male (rf)		1	1
Female	-0.24 (0.003)	0.64 (0.29, 1.41)	0.64 (0.10, 4.21)
Age
15–24 (rf)		1	1
25–34	0.16 (< 0.001)	1.03 (0.69, 1.52)	0.85 (0.47, 1.53)
35–44	0.23 (< 0.001)	1.51 (0.99, 2.31)	1.00 (0.56, 1.80)
45–54	0.19 (< 0.001)	1.22 (0.83, 1.80)	0.96 (0.53, 1.76)
≥55	0.19 (< 0.001)	1.31 (0.78, 2.19)	1.20 (0.64, 2.23)
Residence
Urban (rf)		1	1
Rural	-0.02 (0.479)	0.84 (0.71, 1.01)	0.90 (0.67, 1.21)
Division
Barisal (rf)		1	1
Chattogram	0.01 (0.998)	1.00 (0.47, 2.11)	0.30 (0.17, 0.53)
Dhaka	0.01 (0.777)	0.96 (0.46, 2.01)	1.12 (0.67, 1.89)
Khulna	0.05 (0.165)	2.32 (0.93, 5.81)	0.80 (0.46, 1.42)
Mymensingh	-0.10 (0.003)	4.72 (1.86, 11.97)	0.60 (0.35, 1.05)
Rajshahi	0.01 (0.787)	0.87 (0.40, 1.88)	0.81 (0.48, 1.39)
Rangpur	0.07 (0.019)	1.10 (0.51, 2.38)	0.82 (0.50, 1.35)
Sylhet	-0.03 (0.349)	0.93 (0.45, 1.96)	0.50 (0.29, 0.87)
Educational Level
No Formal School (rf)		1	1
Less Than Primary	0.05 (0.014)	0.83 (0.64, 1.09)	1.33 (0.91, 1.95)
Primary School Completed	0.08 (0.001)	1.00 (0.74, 1.34)	0.98 (0.61, 1.58)
Less Than Secondary	0.06 (0.015)	0.80 (0.63, 1.02)	1.10 (0.72, 1.67)
School Completed	0.13 (< 0.001)	1.27 (0.70, 2.32)	0.98 (0.53, 1.79)
High School Completed	0.18 (< 0.001)	1.11 (0.78, 1.57)	1.62 (0.76, 3.48)
Graduation Completed	0.17 (0.001)	0.94 (0.58, 1.54)	2.23 (1.02, 4.88)
Postgraduate Degree	0.16 (0.006)	0.82 (0.47, 1.43)	3.13 (1.13, 8.70)
Occupation
Government Employee (rf)		1	1
Non-Government Employee	0.11 (0.037)	1.31 (0.83, 2.05)	1.32 (0.44, 3.96)
Business (Small)	0.17 (0.007)	1.57 (0.94, 2.63)	2.03 (0.69, 5.91)
Business (Large)	0.09 (0.127)	1.53 (0.89, 2.63)	1.86 (0.55, 6.31)
Farming (Land Owner & Farmer)	0.12 (0.032)	1.63 (0.91, 2.90)	3.03 (1.00, 9.17)
Agricultural Worker	0.11 (0.088)	1.90 (1.03, 3.51)	2.49 (0.76, 8.17)
Industrial Worker	0.09 (0.111)	1.28 (0.75, 2.21)	1.26 (0.25, 6.35)
Daily Laborer	0.07 (0.193)	1.95 (1.14, 3.35)	1.49 (0.49, 4.51)
Other Self-Employed	0.06 (0.463)	1.82 (1.02, 3.24)	2.05 (0.66, 6.39)
Student	0.36 (0.021)	0.88 (0.40, 1.93)	2.87 (0.61, 13.43)
Retired	0.10 (0.14)	1.51 (0.58, 3.97)	3.39 (0.87, 13.24)
Unemployed, Able to Work	0.08 (0.215)	2.22 (1.10, 4.5)	1.49 (0.38, 5.86)
Unemployed, Unable to Work	-0.02 (0.851)	0.97 (0.54, 1.77)	1.97 (0.50, 7.80)
Other (Specify)	0.08 (0.195)	1.28 (0.75, 2.18)	2.88 (0.86, 9.63)
Wealth Index
Poorest (rf)		1	1
Poorer	0.02 (0.496)	0.97 (0.64, 1.46)	0.58 (0.29, 1.18)
Poor	-0.02 (0.635)	0.83 (0.56, 1.22)	0.63 (0.31, 1.28)
Rich	0.03 (0.33)	0.95 (0.64, 1.42)	1.15 (0.65, 2.02)
Richest	0.03 (0.286)	0.98 (0.66, 1.45)	0.82 (0.48, 1.40)
Constant	2.66 (< 0.001)	7.27 (2.43, 21.81)	0.27 (0.04, 1.73)

OR Odds Ratio, IRR Incidence Rate Ratio, Ref  Reference category (IRR = 1, OR = 1)

*A survey-weighted linear model has been fitted to the log-transformed age of initiation

**A survey-weighted negative binomial regression model has been applied to smoking intensity

***Firth’s penalized logistic regression model has been employed to cessation (successful cessation attempt is coded as 1 and unsuccessful cessation attempt is coded as 0)

Fig. 1.

Q–Q Plot of residuals: model of log age at smoking initiation

Fig. 2.

Residual plot to check homoscedasticity and linearity

For smoking intensity, a survey-weighted negative binomial regression model [43] has been used to identify associated factors. The approximately symmetric distribution of Pearson residuals around zero with no systematic trend supports the appropriateness of the model for analyzing smoking intensity (Fig. 3).

Fig. 3.

Pearson residuals versus fitted values from the negative binomial regression model for smoking intensity

From the results (Table 4) it has been found that geographical divisions are significantly associated with smoking intensity. Smokers residing in Mymensingh division exhibited 4.72 times higher smoking intensity than those in Barisal division (IRR = 4.72, 95% CI: 1.86, 11.97). Regarding occupation, smoking intensity was 90% higher among agricultural workers (IRR = 1.90, 95% CI: 1.03, 3.51) than government employees. It has been also found that smoking intensity was 95% higher among daily labors (IRR = 1.95, 95% CI: 1.14, 3.35), 82% higher among other self-employed workers (IRR = 1.82, 95% CI: 1.02, 3.24), and 2.22 time higher among unemployed individuals who were able to work (IRR = 2.22, 95% CI: 1.10, 4.50) than government employees.

The Firth’s penalized logistic regression model [44] has been fitted to data on successful smoking cessation attempt to identify significant cofactors. Results (Table 4) revealed that respondents who completed graduation were 2.23 times more likely (OR = 2.23, 95% CI: 1.02, 4.88) to have successful cessation attempt than respondents with no formal schooling. It has been also found that respondents with postgraduate degrees were 3.13 times more likely (OR = 3.13, 95% CI: 1.13, 8.70) to have successful cessation attempt than respondents with no formal schooling. Respondents from Chattragram division had 70% less likely to have successful cessation attempt (OR = 0.30, 95% CI: 0.17, 0.53) and respondents from Sylhet division had 50% less likely to have successful cessation attempt (OR = 0.50, 95% CI: 0.29, 0.87) than respondents from Barisal division.

The performance of the three models was examined using observed-versus-predicted comparisons and discriminative performance metrics, as shown in Figs. 4, 5 and 6. The Survey-Weighted Linear Model for age of smoking initiation demonstrates moderate predictive accuracy, with a correlation of 0.41, an RMSE of 0.25, and an MAE of 0.19 (Fig. 4), suggesting reasonable predictive ability on the log scale. The Survey-Weighted Negative Binomial Model for smoking intensity yields a correlation of 0.32, an RMSE of 7.42, and an MAE of 5.26 (Fig. 5), reflecting inherent over dispersion and wide variability in individual-level smoking intensity. The Firth’s Penalized Logistic Regression Model for smoking cessation achieves an AUC of 0.72 (Fig. 6), indicating acceptable discriminative ability in classifying successful and unsuccessful cessation attempt.

Fig. 4.

Model performance evaluation for age of smoking initiation

Fig. 5.

Model performance evaluation for smoking intensity

Fig. 6.

Model performance evaluation for smoking cessation

To assess the robustness of the survey-weighted linear regression model, a sensitivity analysis has been conducted [45] by comparing it with a weight-trimmed model at the 99th percentile and an unweighted model; the results are presented in Table 5. The survey-weighted negative binomial regression model showed high consistency with the trimmed-weight model, with 24 variables remaining stable, confirming that the main model is robust. In contrast to the unweighted linear regression model, only 10 variables remaine stable.

Table 5.

Sensitivity analysis comparing primary survey-weighted linear regression with trimmed-weighted and unweighted models

Variable	Coeff (SE)			% Change from M1		Verdict	Verdict
	M1: Primary (weighted)	M2: Trimmed weighted(P99)	M3: Unweighted	M1→M2	M1→M3	(M1→M2)	(M1→M3)
Gender
Male (rf)
Female	-0.24 (0.08)	-0.24 (0.08)	-0.25.(0.09)	0	4.17	Stable	Stable
Age
15–24 (rf)
25–34	0.16 (0.02)	0.16 (0.02)	0.12 (0.02)	0	25	Stable	Unstable
35–44	0.22 (0.02)	0.22 (0.02)	0.18 (0.02)	0	18.19	Stable	Unstable
45–54	0.19 (0.03)	0.19 (0.03)	0.18 (0.02)	0	5.27	Stable	Stable
≥55	0.19 (0.03)	0.19 (0.03)	0.18 (0.03)	0	5.27	Stable	Stable
Residence
Urban (rf)
Rural	-0.02 (0.01)	-0.02 (0.01)	-0.03 (0.01)	0	50	Stable	Unstable
Division
Barisal (rf)
Chattogram	0.01 (0.03)	0.00 (0.03)	-0.01 (0.02)	100	200	Unstable	Unstable
Dhaka	0.01 (0.03)	0.00 (0.03)	0.01 (0.02)	100	0	Unstable	Stable
Khulna	0.05 (0.03)	0.04 (0.03)	0.02 (0.02)	20	60	Unstable	Unstable
Mymensingh	-0.1 (0.03)	-0.1 (0.03)	-0.13 (0.02)	0	30	Stable	Unstable
Rajshahi	0.01 (0.03)	0.01 (0.03)	0.01 (0.02)	0	0	Stable	Stable
Rangpur	0.07 (0.03)	0.07 (0.03)	0.05 (0.02)	0	28.58	Stable	Unstable
Sylhet	-0.03 (0.03)	-0.03 (0.03)	-0.03 (0.02)	0	0	Stable	Stable
Educational Level
No Formal School (rf)
Less Than Primary	0.05 (0.02)	0.05 (0.02)	0.03 (0.01)	0	40	Stable	Unstable
Primary School Completed	0.08 (0.02)	0.08 (0.02)	0.08 (0.02)	0	0	Stable	Stable
Less Than Secondary	0.06 (0.02)	0.06 (0.02)	0.07 (0.02)	0	16.67	Stable	Unstable
School Completed	0.13 (0.03)	0.13 (0.03)	0.11 (0.02)	0	15.39	Stable	Unstable
High School Completed	0.18 (0.03)	0.15 (0.03)	0.16 (0.03)	16.67	11.12	Unstable	Unstable
Graduation Completed	0.17 (0.04)	0.16 (0.04)	0.15 (0.04)	5.89	11.77	Stable	Unstable
Postgraduate Degree	0.16 (0.06)	0.16 (0.05)	0.18 (0.05)	0	12.5	Stable	Unstable
Occupation
Government Employee
Non-Government Employee	0.11 (0.05)	0.11 (0.05)	0.07 (0.04)	0	36.37	Stable	Unstable
Business (Small)	0.17 (0.05)	0.1 (0.05)	0.09 (0.04)	41.18	47.06	Unstable	Unstable
Business (Large)	0.09 (0.06)	0.14 (0.06)	0.1 (0.05)	55.56	11.12	Unstable	Unstable
Farming (Land Owner & Farmer)	0.12 (0.05)	0.08 (0.05)	0.06 (0.04)	33.34	50	Unstable	Unstable
Agricultural Worker	0.11 (0.05)	0.11 (0.05)	0.11 (0.05)	0	0	Stable	Stable
Industrial Worker	0.11 (0.06)	0.11 (0.06)	0.04 (0.05)	0	63.64	Stable	Unstable
Daily Laborer	0.09 (0.05)	0.07 (0.05)	0.06 (0.04)	22.23	33.34	Unstable	Unstable
Other Self-Employed	0.07 (0.05)	0.06 (0.05)	0.06 (0.05)	14.29	14.29	Unstable	Unstable
Student	0.36 (0.08)	0.06 (0.08)	0 (0.07)	83.34	100	Unstable	Unstable
Retired	0.1 (0.06)	0.1 (0.06)	0.06 (0.06)	0	40	Stable	Unstable
Unemployed, Able to Work	0.08 (0.06)	0.08 (0.06)	0.03 (0.06)	0	62.5	Stable	Unstable
Unemployed, Unable to Work	-0.02 (0.07)	-0.02 (0.07)	-0.06 (0.06)	0	200	Stable	Unstable
Other (Specify)	0.08 (0.06)	0.07 (0.05)	0.06 (0.05)	12.5	25	Unstable	Unstable
Wealth Index
Poorest
Poorer	0.02 (0.02)	0.02 (0.02)	0.02 (0.02)	0	0	Stable	Stable
Poor	-0.02 (0.02)	-0.01 (0.02)	0.01 (0.02)	50	150	Unstable	Unstable
Rich	0.03 (0.02)	0.02 (0.02)	0.04 (0.02)	33.34	33.34	Unstable	Unstable
Richest	0.032 (0.02)	0.02 (0.02)	0.02 (0.02)	37.5	37.5	Unstable	Unstable
Constant	2.66 (0.06)	2.66 (0.06)	2.7 (0.05)	0	1.51	Stable	Stable

M1 Primary model survey-weighted linear regression. M2 Survey-weighted model with weights trimmed at the 99th percentile, M3 Unweighted ordinary least squares regression model and β Unstandardized regression coefficient (log scale), SE Standard error. Percentage change has been calculated as |βM1 − βMx| / |βM1| × 100, where βMx denotes βM2 or βM3. Results have been deemed stable if the percentage change has been < 10%. — indicates |βM1| < 0.01; the percentage change is not reported because the denominator is near zero and the calculation is unreliable

A sensitivity analysis of smoking intensity also has been conducted by comparing the survey-weighted negative binomial regression model with a 99th-percentile weight-trimmed model and an unweighted model [45]. The results are presented in Table 6. The survey-weighted negative binomial regression model is highly consistent with the trimmed-weight model, with 27 variables stable, indicating robustness to extreme weights. In contrast, only 8 variables are stable in the unweighted model, underscoring the need to incorporate survey weights.

Table 6.

Sensitivity analysis comparing primary survey-weighted negative binomial regression with trimmed-weighted and unweighted models

Variable	IRR (95%CI)			% Change		Verdict	Verdict
	M1	M2	M3	M1→M2	M1→M3	M1→M2	M1→M3
Gender
Male (rf)
Female	0.64 (0.29, 1.41)	0.61 (-0.14, 0.14)	0.59 (-0.14, 1.31)	4.69	7.82	Stable	Stable
Age
15–24 (rf)
25–34	1.03 (0.69, 1.52)	0.98 (0.58, 0.58)	0.9 (0.72, 1.09)	4.86	12.63	Stable	Unstable
35–44	1.51 (0.99, 2.31)	1.45 (1.01, 1.01)	1.27 (1.08, 1.45)	3.98	15.9	Stable	Unstable
45–54	1.22 (0.83, 1.8)	1.18 (0.79, 0.79)	1.03 (0.83, 1.22)	3.28	15.58	Stable	Unstable
≥55	1.31 (0.78, 2.19)	1.26 (0.76, 0.76)	1.11 (0.90, 1.31)	3.82	15.27	Stable	Unstable
Residence
Urban (rf)
Rural	0.84 (0.71, 1.01)	0.9 (0.70, 0.70)	0.75 (0.65, 0.84)	7.15	10.72	Stable	Unstable
Division
Barisal (rf)
Chattogram	1 (0.47, 2.11)	1.05 (0.30, 0.30)	1.19 (0.99, 1.38)	5.00	19	Stable	Unstable
Dhaka	0.96 (0.46, 2.01)	0.95 (0.21, 0.21)	0.99 (0.79, 1.17)	1.05	3.13	Stable	Stable
Khulna	2.32 (0.93, 5.81)	2.32 (1.40, 1.40)	2.07 (1.87, 2.25)	0.00	10.78	Stable	Unstable
Mymensingh	4.72 (1.86, 11.97)	4.77 (3.85, 3.85)	2.91 (2.73, 3.08)	1.06	38.35	Stable	Unstable
Rajshahi	0.87 (0.4, 1.88)	0.87 (0.10, 0.10)	1.13 (0.94, 1.31)	0.00	29.89	Stable	Unstable
Rangpur	1.1 (0.51, 2.38)	1.09 (0.34, 0.34)	1.15 (0.96, 1.34)	0.91	4.55	Stable	Stable
Sylhet	0.93 (0.45, 1.96)	0.92 (0.18, 0.18)	0.98 (0.79, 1.17)	1.08	5.38	Stable	Stable
Educational Level
No Formal School (rf)
Less Than Primary	0.83 (0.64, 1.09)	0.85 (0.58, 0.58)	0.95 (0.82, 1.06)	2.41	14.46	Stable	Unstable
Primary School Completed	1 (0.74, 1.34)	0.98 (0.67, 0.67)	1.09 (0.93, 1.24)	2.00	9.01	Stable	Stable
Less Than Secondary	0.8 (0.63, 1.02)	0.8 (0.53, 0.53)	0.77 (0.63, 0.90)	0.00	3.75	Stable	Stable
School Completed	1.27 (0.7, 2.32)	1.21 (0.56, 0.56)	0.9 (0.69, 1.10)	4.73	29.14	Stable	Unstable
High School Completed	1.11 (0.78, 1.57)	0.86 (0.51, 0.51)	0.7 (0.42, 0.97)	22.53	36.94	Unstable	Unstable
Graduation Completed	0.94 (0.58, 1.54)	0.83 (0.39, 0.39)	0.68 (0.38, 0.98)	11.71	27.66	Unstable	Unstable
Postgraduate Degree	0.82 (0.47, 1.43)	0.92 (0.44.1.91)	0.98 (0.81,1.19)	12.20	19.52	Unstable	Unstable
Occupation
Government Employee
Non-Government Employee	1.31 (0.83, 2.05)	0.74 (0.19, 0.19)	0.62 (0.21, 1.01)	43.52	52.68	Unstable	Unstable
Business (Small)	1.57 (0.94, 2.63)	1.31 (0.91, 0.91)	1.04 (0.67, 1.40)	16.57	33.76	Unstable	Unstable
Business (Large)	1.53 (0.89, 2.63)	1.54 (1.12, 1.12)	1.37 (1.00, 1.72)	0.66	10.46	Stable	Unstable
Farming (Land Owner & Farmer)	1.63 (0.91, 2.9)	1.28 (0.74, 0.74)	0.95 (0.53, 1.36)	21.48	41.72	Unstable	Unstable
Agricultural Worker	1.9 (1.03, 3.51)	1.63 (1.11, 1.11)	1.22 (0.85, 1.59)	14.22	35.79	Unstable	Unstable
Industrial Worker	1.28 (0.75, 2.21)	1.87 (1.30, 1.30)	1.44 (1.04, 1.83)	46.1	12.5	Unstable	Unstable
Daily Laborer	1.95 (1.14, 3.35)	1.21 (0.76, 0.76)	1 (0.54, 1.44)	37.95	48.72	Unstable	Unstable
Other Self-Employed	1.82 (1.02, 3.24)	1.91 (1.44, 1.44)	1.32 (0.94, 1.68)	4.95	27.48	Stable	Unstable
Student	0.88 (0.4, 1.93)	1.77 (1.26, 1.26)	1.5 (1.12, 1.88)	101.14	70.46	Unstable	Unstable
Retired	1.51 (0.58, 3.97)	2.91 (1.59, 1.59)	3.23 (2.72, 3.72)	92.72	113.91	Unstable	Unstable
Unemployed, Able to Work	2.22 (1.1, 4.5)	2.24 (1.59, 1.59)	1.3 (0.83, 1.76)	0.91	41.45	Stable	Unstable
Unemployed, Unable to Work	0.97 (0.54, 1.77)	0.94 (0.42, 0.42)	0.71 (0.21, 1.19)	3.10	26.81	Stable	Unstable
Other (Specify)	1.28 (0.75, 2.18)	1.22 (0.76, 0.76)	0.9 (0.48, 1.31)	4.69	29.69	Stable	Unstable
Wealth Index
Poorest (rf)
Poorer	0.97 (0.64, 1.46)	1.02 (0.59, 0.59)	1.14 (0.98, 1.29)	5.16	17.53	Stable	Unstable
Poor	0.83 (0.56, 1.22)	0.82 (0.41, 0.41)	0.86 (0.70, 1.00)	1.21	3.62	Stable	Stable
Rich	0.95 (0.64, 1.42)	0.97 (0.57, 0.57)	1.18 (1.03, 1.32)	2.11	24.22	Stable	Unstable
Richest	0.98 (0.66, 1.45)	0.96 (0.56, 0.56)	1.06 (0.92, 1.20)	2.05	8.17	Stable	Stable
Constant	7.27 (2.43, 21.81)	7.41 (6.37, 6.37)	10.5 (10.05 10.93)	1.93	44.43	Stable	Unstable

M1 Primary survey-weighted negative binomial regression model with complex sampling design, M2 Survey-weighted model with weights trimmed at the 99th percentile, M3 Unweighted negative binomial regression model, IRR Incidence rate ratio, 95% CI 95% confidence interval. Percentage change has been calculated as |IRR_M1 − IRR_Mx| / |IRR_M1| × 100, where IRR_Mx denotes IRR_M2 or IRR_M3. Results have been deemed stable if the percentage change has been < 10% and unstable if it has been ≥ 10%

To evaluate the robustness of the Firth-penalized logistic regression model, a sensitivity analysis further has been conducted by comparing it with survey-weighted and unweighted logistic regression models [46, 47]. The results are shown in Table 7. The Firth-penalized logistic regression model remaines stable compared with the unweighted model, with 28 variables remaining stable, indicating the robustness of the primary findings. In contrast, comparison with the survey-weighted logistic regression model reveals instability for several covariates.

Table 7.

Sensitivity analysis comparing Firth penalized logistic regression with survey-weighted and unweighted models

Variable	OR (95%CI)			% Change from M1		Verdict	Verdict
	M1	M2	M3	M1→M2	M1→M3	M1→M2	M1→M3
Gender
Male (rf)
Female	0.64(0.10, 4.21)	0.00 (0.00, 0.00)	0.00	100	100	Unstable	Unstable
Age
15–24 (rf)
25–34	0.85(0.47,1.53)	0.66 (0.28, 1.57)	0.86 (0.47, 1.56)	22.36	1.18	Unstable	Stable
35–44	1.00(0.56,1.80)	0.67 (0.28, 1.59)	1.02 (0.56, 1.86)	33	2.00	Unstable	Stable
45–54	0.96(0.53,1.76)	1.07 (0.47, 2.42)	0.97 (0.52, 1.80)	11.46	1.05	Unstable	Stable
≥55	1.20(0.64,2.23)	1.00 (0.42, 2.38)	1.25 (0.66, 2.36)	16.67	4.17	Unstable	Stable
Residence
Urban (rf)
Rural	0.90(0.67,1.21)	1.18 (0.75, 1.86)	0.90 (0.67, 1.22)	31.12	0.00	Unstable	Stable
Division
Barisal (rf)
Chattogram	0.30(0.17,0.53)	0.15 (0.07, 0.34)	0.27 (0.15, 0.49)	50	10	Unstable	Unstable
Dhaka	1.12(0.67,1.89)	0.72 (0.38, 1.35)	1.13 (0.66, 1.93)	35.72	0.90	Unstable	Stable
Khulna	0.80(0.46,1.42)	0.53 (0.23, 1.21)	0.79 (0.44, 1.40)	33.75	1.25	Unstable	Stable
Mymensingh	0.60(0.35,1.05)	0.40 (0.18, 0.86)	0.58 (0.33, 1.03)	33.34	3.34	Unstable	Stable
Rajshahi	0.81(0.48,1.39)	0.42 (0.20, 0.91)	0.80 (0.46, 1.38)	48.15	1.24	Unstable	Stable
Rangpur	0.82(0.50,1.35)	0.63 (0.31, 1.30)	0.80 (0.48, 1.34)	23.18	2.44	Unstable	Stable
Sylhet	0.50(0.29,0.87)	0.35 (0.15, 0.79)	0.49 (0.28, 0.86)	30	2.00	Unstable	Stable
Educational Level
No Formal School (rf)
Less Than Primary	1.33(0.91,1.95)	1.33 (0.81, 2.18)	1.36 (0.92, 2.01)	0.00	2.26	Stable	Stable
Primary School Completed	0.98(0.61,1.58)	1.15 (0.59, 2.27)	0.98 (0.61, 1.60)	17.35	0.00	Unstable	Stable
Less Than Secondary	1.10(0.72,1.67)	1.44 (0.79, 2.64)	1.10 (0.72, 1.69)	30.91	0.00	Unstable	Stable
School Completed	0.98(0.53,1.79)	1.01 (0.38, 2.72)	0.98 (0.52, 1.82)	3.07	0.00	Stable	Stable
High School Completed	1.62(0.76,3.48)	4.16 (1.36, 12.70)	1.65 (0.75, 3.63)	156.8	1.86	Unstable	Stable
Graduation Completed	2.23(1.02,4.88)	1.09 (0.21, 5.58)	2.31 (1.03, 5.17)	51.13	3.59	Unstable	Stable
Postgraduate Degree	3.13(1.13,8.70)	3.59 (0.67, 19.18)	3.30 (1.14, 9.53)	14.7	5.44	Unstable	Stable
Occupation
Government Employee (rf)
Non-Government Employee	1.32(0.44,3.96)	4.90 (1.27, 18.92)	1.40 (0.44, 4.45)	271.22	6.07	Unstable	Stable
Business (Small)	2.03(0.69,5.91)	3.67 (0.80, 16.83)	2.22 (0.72, 6.84)	80.79	9.36	Unstable	Stable
Business (Large)	1.86(0.55,6.31)	2.86 (0.50, 16.41)	1.99 (0.55, 7.17)	53.77	6.99	Unstable	Stable
Farming (Land Owner)	3.03(1.00,9.17)	7.32 (1.53, 35.10)	3.35 (1.05, 10.70)	141.59	10.57	Unstable	Unstable
Agricultural Worker	2.49(0.76,8.17)	7.03 (1.53, 32.24)	2.75 (0.80, 9.50)	182.33	10.45	Unstable	Unstable
Industrial Worker	1.26(0.25,6.35)	4.96 (0.57, 42.88)	1.24 (0.22, 6.94)	293.66	1.59	Unstable	Stable
Daily Laborer	1.49(0.49,4.51)	3.58 (0.75, 17.21)	1.63 (0.51, 5.21)	140.27	9.40	Unstable	Stable
Other Self-Employed	2.05(0.66,6.39)	3.98 (0.81, 19.46)	2.23 (0.68, 7.33)	94.15	8.79	Unstable	Stable
Student	2.87(0.61,13.43)	2.03 (0.18, 22.63)	3.18 (0.63, 16.11)	29.27	10.81	Unstable	Unstable
Retired	3.39(0.87,13.24)	19.85 (3.11, 126.60)	3.69 (0.89, 15.29)	485.55	8.85	Unstable	Stable
Unemployed, Able to Work	1.41(0.36, 0.56)	3.40 (0.51, 22.86)	1.55 (0.37, 6.57)	141.14	9.93	Unstable	Stable
Unemployed, Unable to Work	1.97(0.50,7.80)	5.28 (0.63, 44.64)	1.88 (0.43, 8.10)	168.03	4.57	Unstable	Stable
Other (Specify)	2.88(0.86,9.63)	10.05 (1.74, 58.18)	3.20 (0.91, 11.28)	248.96	11.12	Unstable	Unstable
Wealth Index
Poorest (rf)
Poorer	0.58 (0.29, 1.18)	0.56 (0.28, 1.11)	0.83 (0.51, 1.34)	3.45	43.11	Stable	Unstable
Poor	0.63 (0.31, 1.28)	0.58 (0.29, 1.16)	0.71 (0.44, 1.15)	7.94	12.7	Stable	Unstable
Rich	1.15 (0.65, 2.02)	1.10 (0.62, 1.93)	1.10 (0.70, 1.72)	4.35	4.35	Stable	Stable
Richest	0.82 (0.48, 1.4)	0.79 (0.46, 1.35)	0.96 (0.62, 1.48)	3.66	17.08	Stable	Unstable
Constant	0.27 (0.04, 1.73)	0.28 (0.04, 1.84)	0.32(0.08, 1.27)	3.71	18.52	Stable	Unstable

M1 refers to the Firth penalized logistic regression model, M2 denotes the survey-weighted logistic regression model, and M3 represents the unweighted logistic regression model. OR indicates odds ratio, and 95% CI refers to the 95% confidence interval. Percentage change has been calculated as |OR_M1 − OR_Mx| / |OR_M1| × 100, where OR_Mx corresponds to either OR_M2 or OR_M3. Results have been deemed stable if the percentage change has been < 10% and unstable if it has been ≥ 10%

Discussion

This study investigated the age at smoking initiation, smoking intensity, and successful tobacco cessation attempts in Bangladesh using a nationally representative sample. The average age of male onset for tobacco smoking was 20.05 years, and a significant percentage initiated smoking before 18 years old. Overall, the successful cessation attempt rate was 35.20%. Among the smokers, 10.78% were heavy smokers.

Model-based analytical results reveal that people in early adulthood initiated tobacco smoking earlier, though it is hoped that smoking intensity has been found to be lower in that group. However, age has not been found to be significant to a successful cessation attempt. Literature shows that initiating smoking at a younger age and higher smoking intensity are consistently associated with a greater risk of subsequent regular and heavy smoking later in life and is highly associated with smoking-related diseases, such as lung cancer [11].

No significant differences have been found in age of smoking initiation, smoking intensity, or successful cessation attempt rates between rural and urban residents. Geographical divisions have been found to be another significant differential in the age of smoking initiation, smoking intensity, and successful cessation attempt. It has been found that respondents from Mymensingh and Sylhet divisions initiate tobacco smoking earlier than Barisal division. Further respondents from Mymensingh division exhibit the most smoking intensity, while respondents from Chattogram and Sylhet exhibit the least successful cessation attempt. This might be driven by a combination of socio-economic factors, peer influence, easy access, occupational exposure (lack of regulations), cultural norms, and limited awareness of tobacco’s health dangers [49–51]. However, it is evident that early smoking initiation also contributes to higher consumption prevalence in those divisions [52].

Furthermore, education has been proven to be a significant protective factor as higher education levels are associated with late initiation, reduced smoking intensity, and higher cessation attempt rates. Education always plays as a crucial public health intervention [50] and also plays as a protective factor for smoking rate [52]. Agricultural workers, daily laborers, and unemployed individuals who are able to work have been significantly associated with high smoking intensity. This can be explained by heightened stress at work, physically strenuous work environments, lower socioeconomic status, less access to health information, and more permissive norms around smoking in the workplace [53]. However, occupation is not found to be significantly associated with age of smoking initiation or smoking cessation. The wealth index has not been found to be statistically significant to any of the three smoking behavior in Bangladesh.

The strength of the study is substantial: (i) Use of a large and nationally representative dataset, (ii) High relevance for public health policy, (iii) Comprehensive analysis of multiple smoking-related outcomes, and (iv) Inclusion of key socio-economic determinants.

However, there are some limitations to consider. First, the study relies on self-reported data, introducing potential biases, such as recall and social desirability biases, which could affect the accuracy of smoking behavior reports. Second, the definition of successful smoking cessation was based on a 28-day abstinence threshold, which merely reflects an attempt to quit rather than sustained long-term cessation. Additionally, the exclusion of key psychological and behavioral factors related to tobacco use hinders a more comprehensive understanding of the root causes of smoking behaviors. Lastly, accurately assessing wealth and socioeconomic status is challenging, as relying on indirect indicators may misrepresent participants’ financial circumstances.

Conclusion

This study highlights prevalence and significant correlates of three key smoking behaviors - age of initiating smoking, smoking intensity and successful cessation attempt in Bangladesh. Despite many tobacco control efforts in Bangladesh in the last few decades, challenges persist, particularly concerning high smoking intensity, early initiation and low cessation rates. Moreover, these three key behaviors vary substantially across less education levels and geographical location, highlighting the need for tailored intervention programs and laws. Prioritizing tobacco as addiction substance interventions should be taken targeting lower working status and less educated populations, high-burden regions such as Mymensingh, Rangpur, Chattogram and Sylhet.

Abbreviations

NCD

Noncommunicable disease

WHO

World health organization

CDC

Centers for disease control and prevention

CI

Confidence interval

OR

Odds ratio

IRR

Incidence rate ratios

Authors’ contributions

Conception, access of data and material collection: PS. Statistical analysis and interpretation: PS and ZS. Drafting the manuscript: PS and ZS. Finalizing the manuscript: PS, ZS and MF.

Funding

The authors received no specific funding for this study.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

This study used secondary, de-identified data from GATS Bangladesh 2017-18, conducted by the Bangladesh Bureau of Statistics with support from WHO and CDC. The survey was granted ethical approval by Bangladesh Medical Research Council (BMRC) National Research Ethics Committee (NREC) [Ref: BMRC/NREC/2016-2019/344].

Competing interests

The authors declare no competing interests.