Study design and sample size

The study was designed as a cross-sectional survey. Considering the area, population, and geographical diversity of Bangladesh, four divisions (Dhaka, Chittagong, Khulna, and Rangpur) were purposely selected for the survey. Since the ST prevalence is high among women, adult females aged ≥15 years³ were the participants. Applying division-wise ST prevalence³, the representative sample size was estimated as 2190 households (50% with female ST users and 50% with ST non-users) (Supplementary file Table 1A). Equal allocation was considered for the power and robustness of comparison. Considering a non-response of around 1%, 2213 households were surveyed.

Sampling design and survey approach

A multi-stage sampling design was used. A total of 12 districts (three from each of the four divisions) were selected by simple random sampling. These were followed by subdistricts (called upazilla) and lower administrative units (called unions) under each subdistrict, which were selected randomly. The lowest administrative component, i.e. village or ward in each union, was the primary sampling unit (PSU). The households were randomly recruited in all the areas with urban–rural stratification. Estimated samples were equally distributed for all districts under each division (Supplementary file Table 2A).

Household recruitment, characterization, and survey scope

In a selected PSU, the sampling frame included all households with at least one adult female (self-reported) daily ST user (at least for the last year) as ‘user households’ and those with none of their members as users of any kind of tobacco as ‘non-user households’. In the user households, all the adult female ST users who were present during the survey were approached for participation. In contrast, in the non-user households, all available adult females were approached to take part. Since ST prevalence is predominant among women³ and smoking prevalence among them is negligible, the male respondents and smoking attributes were out of the survey scope.

Survey tool, data collection, and storage

The survey was conducted simultaneously between October and November 2021 in four divisions. Written consent was obtained from the participants. The data were collected with a pre-tested semi-structured questionnaire, which the trained enumerators performed electronically with Survey CTO¹⁹. The questionnaire was designed following the Global Adult Tobacco Survey³. It had four sections: household characteristics and individual sociodemographic characteristics, ST use behavior and practices, onset of disease and treatment cost, and quality of life information. The quality-of-life segment was adopted from the EuroQol¹⁷. The translated Bengali version of the EQ-5D-5L questions was read out, and the enumerators recorded the responses in the Survey CTO application. For the EQ-VAS (Visual Analogue Scale), enumerators showed them the scale on a data collection device. They instructed the participants to assess their health state by selecting a point on the scale.

Variables and data analysis

Household and individual characteristics, ST use practice and disease onset

The household characteristics such as age and gender of the household head, family size, location, and income and expenditure of the households, were collected. Sociodemographic characteristics of the individuals, such as age, gender, religion, marital status, occupation, and education level, were considered. Regarding ST use behavior and practice, at the individual level, daily used ST products, frequency of use, initiation age, influential person for initiation, quit attempt and quit duration, average expenditure behind ST, and source of the purchase, were enquired. The onset of diseases (cancer, asthma/COPD/breathing complications, heart disease, delivery complications, and oral diseases) for individuals was also considered. In practice, the women were asked to self-report the above disease onset and whether they were treated or under treatment.

Quality of life measurement

The quality of life was measured using EQ-5D instrument. This widely accepted, precise, and generic instrument for self-reported physical and mental health measures was introduced by EuroQol^17,20. For descriptive health profiling, the EQ-5D-5L version was adopted, considering its less ceiling effects^20,21. Besides, participants’ assessment of their general health state was coded through EQ-VAS¹⁷. Covering five health dimensions (mobility, usual activities, self-care, pain and discomfort, and anxiety and depression), quality of life was enquired with five levels centering around the extent of a problem: no, mild, moderate, severe, and extreme²⁰. The level responses were recorded on a scale of 1 to 5 (1 = ‘no problem’ and 5 = ‘extreme problem’) while the EQ-VAS recorded the health state assessment on a scale from 0 (worst health one can think of) to 100 (best health one can aspire). Since EQ-5D value sets are unavailable for Bangladesh, Thailand’s value set (with the score ranging from -0.42 to 1.0) was referred for deriving the preference-based index value or HRQoL scores (also called EQ-5D-5L utility scores)²².

Summary statistics and descriptive comparison

Considering their sociodemographic characteristics, the EQ-5D-5L health state profiles of female ST users were compared with non-users. For the user group, the initiation age, influential person for ST uptake, product and daily frequency of use, quit attempt, and quit duration were analyzed. The difference in the proportion reporting ‘any sort of problem’ in health dimensions among female ST users and non-users, was tested statistically. Average HRQoL score and EQ-VAS values for the two groups were estimated concerning socioeconomic and sociodemographic factors, disease onset (cancer, oral disease, heart disease, delivery complications, asthma/COPD/breathing problem), ST use behavior and practice. The statistical significance of the difference in averages between the two groups, which were conditioning the mentioned factors, was tested. The correlation coefficient among them was measured, as a strong correlation between HRQoL score and EQ-VAS values is normally expected. Three different correlation (r) statuses were considered, defining them as: weak, r<0.3; moderate, 0.3≤r≤0.5; and strong, r>0.5²³.

Econometric modelling

The impact of ST use on the outcome of HRQoL scores and EQ-VAS values was estimated using two different multiple linear regression models. In Model 1, the outcome was HRQoL scores, while in Model 2, it was EQ-VAS values. Both models had sociodemographic and ST-related exposures. The sociodemographic exposures were listed considering the significance of the difference in averages of HRQoL scores and EQ-VAS values for user and non-user females. A utility decrement or disutility score (1-HRQoL Score) was modeled to avoid negative HRQoL scores^23,24. Thus, while exposure with a positive coefficient would increase the disutility score, it would decrease the HRQoL scores and worsen the health state²³.

Since EQ-5D-5L data commonly have clustering effects, which leads to compromised consistency and biased estimates for ordinary least squares (OLS)²⁰, a more robust estimation method, the Generalized Linear Model (GLM), was applied²⁵. These are robust against non-normality and provide more unbiased and efficient estimates^23,26. The GLM application method requires the identification of the ‘family of distribution’ and ‘link function for fitted values’²⁵. An appropriate family of distribution (Gaussian, Gamma, Inverse Gamma, and Poisson) was chosen using the lowest χ² value. Pregibon’s link test was used to decide the link function^23,27. It was found that the disutility score followed a Poisson distribution, while EQ-VAS values followed the Gaussian distribution. The best link function for the fitted values of both the dependent variables was tested to be Log. Standard errors of the regression estimates were bootstrapped. STATA package eq5dds provided the descriptive profiling of HRQoL scores²⁸. The value set was generated using the package named eq5d ²⁹. All statistical calculations were operationalized using STATA 17³⁰.

Socioeconomic statistics, disease onset, and ST use pattern

Around 86% of the ST users and 91% of non-user households had a male as their head. A higher number of female ST users resided in higher age groups (43% of female ST users aged ≥55 years). The female ST users mainly were married (71%), and slightly over a quarter were widows (27%). They also had low education level (63% had no schooling). The vast majority were homemakers (86%) (Supplementary file Table 3A). Most of the female ST users started at the age of 25–34 years (31%) or 15–24 years (30%). They were mainly influenced by in-laws (47%). It was found that around 6 to 7 users out of 10 used zarda or zarda with paan (betel leaf). A quarter used sada pata (sun-dried tobacco leaf) with paan, and gul (tobacco powder mixed with ingredients) was used by 13%, and 6% chewed sada pata directly. Around 4% used ST 6 to 10 times daily, while 8% used it more intensively (>15 times a day). Only 16% of users attempted to quit; among them, only 17% quit for a month or more (Supplementary file Table 4A).

The female ST users had higher disease (directly or indirectly attributed to ST use) onset than the non-users. Predominantly oral disease (teeth and/or gum problems) was more than twice as prevalent (27%) among the users than among the non-users (around 12%). Asthma/chronic obstructive pulmonary disease (COPD)/breathing problems, heart disease, and cancer (oral, laryngeal/esophageal/throat, lungs/chest, pancreas/stomach) were also more than twice as prevalent among the users (around 7.8%, 7.9%, and 1.3 %, respectively) (Supplementary file Figure 1A).

Health outcome: application of EQ-5D-5L utility score

Regardless of health dimensions, a more significant proportion of female ST non-users reported no problems with their health state (mobility: 56.6% vs 80.4%, self-care: 70.3% vs 80.0%, usual activities: 51.2% vs 78.1%, pain or discomfort: 30.2% vs 59.3%, and anxiety or depression: 38.6% vs 62.4%) (Supplementary file Figure 2A). Table 1 contains the percentage of female ST users and non-users who had any type of problem (ranging from ‘slight’ to ‘extreme’) in all dimensions. Female users remained consistently higher in this regard than non-users. More specifically, considering movement, personal care, and usual activities, the percentage of women with ST use attributes facing any problem was more than double that of their non-user counterparts. For anxiety and depression, the percentages were 61.3% (95% CI: 58.5–64.1) and 37.5% (95% CI: 35.1–40.0) among users and non-users, respectively. A similar finding was observed for pain and discomfort [users: 69.8% (95% CI: 67.0–72.3) vs non-users: 40.6% (95% CI: 38.1–43.1)]. The differences in proportions were statistically significant.

The overall minimum HRQoL score was -0.42. The average score for all women was 0.86, with a standard deviation (SD) of 0.19. The non-users had an average of 0.90 (95% CI: 0.89–0.90), while that for users was 0.79 (95% CI: 0.78–0.81). This difference was statistically significant at the 1% level (p<0.01). Also, for other socioeconomic and sociodemographic subgroups, the average HRQoL score for ST-using females was lower. Except for a few exceptions, the differences in average scores were statistically significant on all occasions (Table 2). Although the difference is small, the average utility score decreased for more intensive ST-using females. Those who used raw ST (e.g. chewing sada pata or ST with paan masala) had lower HRQoL scores.

The overall average of EQ-VAS was 76.14 (SD=17.83) with a median value of 80. The female ST non-users assessed their health status as being relatively well compared to the users [EQ-VAS averages were 80.76 (95% CI: 79.9–81.6) and 70.27 (95% CI: 69.2–71.2), for non-users and users, respectively]. Regardless of the sociodemographic characteristics, such a trend was persistent. Among the users who started using ST at a relatively younger age and used it more intensively had lower EQ-VAS average. The score was 71.24 for those who used ST less than 10 times daily. Conversely, it decreased to 59 for those who used it more than 30 times daily. Additionally, the EQ-VAS average was lower for those who used the raw form of ST (Table 2).

The distributions of HRQoL scores and EQ-VAS values had longer left wings and were more peaked than a normal distribution (Supplementary file Figure 3A). Additionally, the clustering effect of the scores was clearly visible. The participants were clustered around the value of 1 (perfect health) under the utility score. Although the clustering effect was not as severe as the utility score for the EQ-VAS values, participants were clustered around the scale range of 80–100.

The relationship between HRQoL scores and self-assessed EQ-VAS values was measured using a correlation coefficient along with its significance (Supplementary file Table 5A). In the overall sample, it was 0.63, while for the user and non-user subsets, they were 0.60 and 0.61, respectively. The association was statistically significant and categorized as strong. Except for a few occasions, this association for every subset of the data was strong and significant. Thus, irrespective of the sociodemographic characteristics and ST use status, females with higher HRQoL scores also had self-evaluated better health status in general.

Econometric modelling: association of ST use with HRQoL scores and EQ-VAS values

Considering the non-normality and clustering of HRQoL scores, a GLM was estimated for disutility score with Poisson as the distribution family and Log as the link function. Results in Table 3 show that controlling all the sociodemographic factors, the average HRQoL score of female ST users was significantly lower by 0.15 points than that of the non-users. Furthermore, although the duration of use had no significant effect, intensive use of ST reduced scores significantly, with a coefficient value of 0.03 points. Area of residence, family size, and marital status were significant factors for the HRQoL score of females. In the test for appropriate distribution family, the χ² for the Poisson distribution was found to be the lowest (0.06). The insignificance of the coefficient of the square of the predicted value of the base regression, established the appropriateness of the Log link in Pregibon’s link test (Supplementary file Table 6A).

The GLM regression with EQ-VAS values as a response was estimated with a Gaussian distribution and Log as the link function (Table 4). It reinforced the findings from the HRQoL score regression. Controlling all the sociodemographics, the attribute of ST use among females decreased their EQ-VAS values. The coefficient was -0.04 (p<0.05). Although negative, the magnitude of the effect of duration of ST use was very small. Marital and education level and area of residence were significant contributing factors for EQ-VAS values. The value of the distribution family test was 14.94 for Gaussian distribution and the lowest compared to the others; additionally, the model did not fail Pregibon’s link test for Log as the link function (Supplementary file Table 7A).

Limitations

The study has some limitations. The non-users were identified through self-report. No biomarker was used to verify their ST use status. However, as ST use is socially and culturally acceptable, we believe that there was less incentive for the user to hide their consumption status. Although the study sample is representative at the divisional level, the generalizability of the findings is limited as it used only four divisions. The external validity of the findings is also limited for other countries. Additionally, since the findings used cross-sectional data instead of cohort analysis, the effect measure might not be a true causal impact. The residual confounding was not controlled for in the multivariate models. Despite the shortcomings, the study fills the research gap by underlining the evidence that ST use has severe implications for the quality functioning of life of women in Bangladesh.