Trends and determinants of tobacco use among women of reproductive age in East Africa: a trend and multilevel analysis using demographic and health survey data

Abstract

Background

Tobacco use remains a leading preventable risk factor for morbidity and mortality worldwide. Although global evidence indicates a general decline in tobacco use among women, it is unclear to what extent these trends and associated determinants apply to populations in East Africa. Therefore, this study aimed to examine the trends in tobacco use and its associated factors among women aged 15–49 across East African countries.

Objective

To examine trends in tobacco use and identify individual- and community-level factors associated with current tobacco use among women of reproductive age (15–49 years) in East Africa.

Methods

We analysed data from Demographic and Health Surveys (DHS) across multiple East African countries. Prevalence trends were assessed over four survey periods (1999–2005, 2006–2010, 2011–2015 and 2016–2022). A cross-sectional analysis of the most recent survey wave was used for the multilevel logistic regression models to identify individual-level and community-level factors associated with current tobacco use, including sociodemographic and socioeconomic characteristics (such as age, education, wealth status and marital status) and community-level factors (such as place of residence and contextual community characteristics). Country-specific trends were also assessed.

Results

A total of 543 013 women of reproductive age from 11 East African countries were included in this analysis. In each DHS survey, most respondents resided in rural areas. Of the pooled sample years, the overall prevalence of current tobacco use was 2.84% (95% CI 2.7% to 2.9%). After a slight increase from 3.74% (1999–2005) to 4.04% (2006–2010), the prevalence declined markedly to 1.35% (2011–2015) and then rose to 2.74% (2016–2022). Compared with cigarette smoking, the use of other tobacco products has consistently greater fluctuations. Country-specific trends showed a declining prevalence in most countries, although increases or fluctuations were observed in Madagascar and Zambia. In the multilevel analysis, age and education showed the strongest association with tobacco use. Women aged 35–49 years had significantly higher odds of tobacco use (adjusted OR (AOR) 2.86, 95% CI 2.44 to 3.36), and those aged 25–34 years (AOR 1.86, 95% CI 1.62 to 2.15) compared with women aged 15–24 years. Women with no formal education were substantially more likely to use tobacco (AOR 4.30, 95% CI 3.01 to 6.15) than those with higher education levels. Additional factors associated with increased odds included experience of child death (AOR 1.39, 95% CI 1.27 to 1.52) and abortion history (AOR 1.21, 95% CI 1.09 to 1.33). At the community level, urban residence (AOR 1.60, 95% CI 1.40 to 1.84) and lower altitudes of <1000 m (AOR 1.40, 95% CI 1.03 to 1.91) were also significantly associated with tobacco use.

Conclusion

Tobacco use among women of reproductive age in East Africa remains relatively low but varies substantially across countries and over time. Older age, lack of formal education, reproductive factors and contextual community characteristics were important determinants. Targeted and context-specific tobacco control interventions are therefore required to reduce tobacco use among vulnerable populations in the region.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study used nationally representative Demographic and Health Surveys (DHS) data from multiple survey years, enabling cross-country comparisons and trend analyses.

• The application of multilevel modelling accounted for clustering at the community level, thereby improving the accuracy of individual and contextual estimates.

• Tobacco use was self-reported, which may have led to underestimation due to social desirability bias in some societies.

• The cross-sectional design of the DHS limits the ability to infer causal relationships between variables, and the dataset lacks detailed contextual variables such as tobacco industry marketing exposure, peer influence and access to cessation services, which could influence the observed patterns.

• The study included only East African countries with available DHS data on tobacco use, which may restrict generalisability. Due to model non-convergence, a two-level model with community-level random effects was used while treating country as a fixed effect, which may not fully capture country-level variation.

Introduction

Tobacco use continues to be a major preventable cause of morbidity and mortality, greatly contributing to the burden of non-communicable diseases, including cancer, respiratory conditions and cardiovascular disorders.^{1 2} According to a WHO report, 80% of tobacco users worldwide reside in low and middle-income nations, and tobacco kills over 7 million people annually, including an estimated 1.6 million non-smokers who are exposed to secondhand smoke.³ Even though the prevalence of tobacco use has decreased globally over the past 20 years, low and middle-income countries (LMICs) currently bear a disproportionate amount of the burden, with the tobacco industry aggressively targeting sub-Saharan Africa (SSA) as a growing market.⁴

Although women have historically smoked at much lower rates than men, they represent a critical demographic in global tobacco control. Projections suggest that unless effective interventions are sustained, the number of women who smoke in LMICs could rise due to demographic changes, aggressive marketing and evolving gender norms.⁵ Given the region’s growing urbanisation, changing sociocultural norms and differing degrees of tobacco control policy implementation, East Africa is of particular interest.

Global programmes, such as the WHO Framework Convention on Tobacco Control (WHO FCTC), have effectively reduced tobacco consumption in several high-income countries.⁶ However, the epidemic is spreading to low-income and middle-income nations, perhaps as a result of lax enforcement, laxer legislation and aggressive tobacco industry marketing.⁷ East Africa is a crucial region for examining these issues because of its rapid population growth, urbanisation and shifting socioeconomic landscape.⁸ Despite the gradual implementation of tobacco control measures, such as pricing controls, advertising bans and public smoking bans, it is yet unknown how these policies will impact women.⁹ Attempts to quantify and solve this public health concern have been hindered by social stigma and conventional norms surrounding women’s tobacco use.

The health risks of tobacco for women are broad and specific, threatening maternal and child health, especially for those of reproductive age.¹⁰ In addition to common issues such as heart disease, cancer and chronic respiratory conditions, tobacco use is linked to adverse reproductive outcomes. It increases the risk of pregnancy problems such as placenta previa, placental abruption and pre-eclampsia and reduces fertility.¹¹ Additionally, it is a strong predictor of low birth weight, stillbirth and neonatal death, which can have long-term consequences for the development and health of the child.^{11 12} These effects put further pressure on the already overburdened healthcare systems of East African nations.¹⁰

Most research on tobacco in Africa has traditionally focused on men and smoked products; however, there is an increasing need for detailed data on women’s tobacco use, especially among those of reproductive age. Information on the prevalence and trends of various forms of tobacco, such as cigarettes, smokeless tobacco and waterpipes, among women in East Africa, is limited but crucial for developing effective public health strategies. Concerns are also growing about the expanding diversity and availability of tobacco products and whether their use among women is increasing or stabilising. Understanding these patterns and their determinants is vital because they inform public health policies, reveal social and structural influences such as education, wealth, urbanisation and media exposure, and help identify higher risk groups for targeted interventions. In this context, this study aimed to comprehensively analyse tobacco use among women of reproductive age in East Africa using DHS data. It described prevalence trends and identifies the factors associated with tobacco use. The findings will aid policymakers, health officials and civil society organisations in the region in developing targeted, evidence-based prevention and cessation programmes that focus on women’s reproductive health and overall well-being of women.

Methods and materials

Data source and settings

This study utilised publicly available cross-sectional survey data from the DHS for selected East African countries conducted between 1999 and 2022. The DHS programme uses standardised questionnaires to collect nationally representative data on population and health indicators to enable cross-country comparisons to be made. A rigorous, multistage, stratified cluster sampling design was used in the surveys to ensure that the data were comparable across nations and survey waves.

Country classification was based on the United Nations Geoscheme for Africa, which organises countries into regional groupings for statistical and analytical purposes.¹³ Under this classification, East Africa comprises 19 countries: Burundi, Comoros, Djibouti, Ethiopia, Eritrea, Kenya, Madagascar, Malawi, Mauritius, Mozambique, Réunion, Rwanda, Seychelles, Somalia, Somaliland, Tanzania, Uganda, Zambia and Zimbabwe.

DHS data were available for 13 of the 19 East African countries. The remaining six countries (Djibouti, Somalia, Somaliland, Seychelles, Mauritius and Réunion) were excluded because of a lack of DHS data. Among the 13 countries with DHS data, Comoros and Eritrea had two survey rounds; however, only one Comoros survey (DHS 2012) contained tobacco-related variables, while earlier datasets (DHS 1996) did not include tobacco use data. Similarly, both Eritrean surveys (DHS 1995 and DHS 2002) did not include tobacco use data and were excluded from the analysis.

In this study, East African countries with at least two DHS survey rounds containing comparable data on tobacco use among women aged 15–49 years were included. Based on this criterion, 11 countries were retained for analysis, ensuring the availability of sufficient repeated measures to assess trends over time (table 1). The DHS Individual Record (IR) dataset was used to extract the required variables.

Table 1. Study participants by country and corresponding DHS survey years for trends and determinants of tobacco use among women of reproductive age in East Africa.

Serial number	Country	Survey year	Weighted frequency (n)
1.	Burundi	2010	9388
		2016/17	17 269
2.	Ethiopia	2005	14 059
		2011	16 511
		2016	15 683
3	Kenya	2003	8192
		2009	8438
		2014	31 076
		2022	32 156
4.	Madagascar	2003/04	7948
		2008/09	17 366
		2021	18 869
5.	Malawi	2000	13 219
		2004	11 661
		2010	23 011
		2015/16	24 562
6.	Mozambique	2003	12 406
		2011	13 745
		2022	13 183
7.	Rwanda	2000	10 418
		2005	11 308
		2010	13 669
		2015	13 496
		2020	14 634
8.	Tanzania	2005	10 325
		2010	10 135
		2016	13 266
		2022	15 254
9.	Uganda	2001	7242
		2006	8531
		2011	8666
		2016	18 506
10.	Zambia	2002	7656
		2007	7140
		2014	16 410
		2018	13 683
11.	Zimbabwe	1999	5903
		2006	8902
		2011	9171
		2015	9955

DHS, Demographic and Health Surveys.

Study population and sampling

Women of reproductive age (15–49 years) who participated in the selected surveys comprised the study population. DHS employs a two-stage stratified cluster sampling procedure, with enumeration areas selected in the first stage and households systematically selected in the second stage. All eligible women residing in the selected households were invited to participate in the study. Sampling weights provided by the DHS were applied to adjust for survey design, unequal probability of selection and non-response, ensuring nationally representative estimates.

Patient and public involvement

Patients or members of the public were not involved in the design, conduct, reporting or dissemination of this research, as it was based on a secondary analysis of anonymised DHS data. No patients or members of the public were involved in the reporting or dissemination of this study.

Variables

All variables included in this study were selected based on peer-reviewed literature and DHS datasets, ensuring their relevance to the research objectives. These variables were categorised and coded according to the guidelines in the DHS Recode Manual.¹⁴

Outcome variable

The primary outcome variable was the current tobacco use status among women of reproductive age (15–49 years). This was defined based on self-reported responses to questions in the DHS questionnaire regarding the current use of any tobacco products. The following were ideally included:

• Current cigarette smoking was defined by ‘Do you currently smoke cigarettes?’ (yes/no).

• Current smokeless tobacco use was defined as ‘Do you currently use smokeless tobacco (eg, chewing tobacco, snuff)?’ (yes/no).

• Other smoked tobacco forms (if available): questions about pipes, cigars or waterpipe (shisha/hookah) use were included if they were consistently available across relevant survey rounds.

A composite ‘current any tobacco use’ variable was created (yes/no) if a respondent reported using any of the available tobacco products. Separate analyses were conducted for specific tobacco products.

Independent variables

Independent variables were categorised into individual-level and community-level factors based on the prior literature and data availability in the DHS datasets.

Individual-level factors included age (15–24, 25–34 and 35–49 years); educational attainment (no education, primary, secondary, higher), marital status (never married, married, widowed/divorced/separated), employment status (currently working vs not working), media exposure (exposed vs not exposed to at least one of radio, television, newspapers or the internet), parity (number of children ever born), household size (<5 vs ≥5 members), history of abortion (yes/no), history of child death (yes/no), religion (categorised as reported in the DHS) and household wealth class (lower class, middle class and higher class).

Community-level factors

Included place of residence (urban/rural), country (categorised by survey country) and cluster altitude in metres (was categorised based on DHS geospatial data into high (2001–3500 m), middle (1000–2000 m) and low (<1000 m above sea level)).

All variables were coded according to the DHS standard definitions and recoded as necessary for the analysis. Detailed descriptions of the variable measurements and the original survey questions are provided in online supplemental file 1.

Data management and statistical analysis

The data set for each country was extracted and cleaned to remove missing and inconsistent data observations. Missing data were assessed across all study variables prior to analysis, in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines. Most variables had little or no missing data. However, body mass index, health insurance, current alcohol consumption and partner-related characteristics (husband’s occupation and education) had substantial levels of missingness, largely due to subsampling procedures inherent to the DHS, and were, therefore, excluded from the multivariable regression analysis. Variables with low levels of missingness (occupation and altitude) were managed using Multiple Imputation by Chained Equations (MICE) under the assumption of missing at random (MAR), incorporating all study variables and country identifiers. However, the variable ‘religion of the respondent’ was not reported in the Tanzania DHS dataset, leading to structural missingness that violated the MAR assumption; therefore, it was excluded from the final multivariable analysis to ensure comparability across countries. Observations with missing values for the outcome variable were excluded from the analysis. The extent of missing data for all variables is reported in online supplemental file 2.

The cleaned datasets of the selected countries were appended into a single dataset for further analysis after the variable names and coding were harmonised. The data were checked for completeness, and sampling weights from the DHS were applied to account for unequal selection probabilities and complex survey designs.

All prevalence estimates were calculated using survey-weighted analyses (svy procedures) that accounted for clustering and stratification. For pooled analyses across countries and survey waves, sampling weights were applied to reflect the relative contribution of each survey, based on its sample size and design. Period-specific prevalence estimates were derived using weighted analyses.

Descriptive statistics, including percentages and frequencies, were used to summarise the background characteristics of the study participants. The results are presented in tables and figures. Trends over time in tobacco use were displayed graphically using line plots, stratified by country. The prevalence of current tobacco use was estimated by survey period: 1999–2005, 2006–2010, 2011–2015 and 2016–2022. Trends were examined overall, by product type (cigarettes vs other tobacco products) and by country. The pooled prevalence across the study period was also calculated. All analyses were conducted using STATA V.17 and R V.4.5.1.

The geographic distribution of tobacco use was also assessed using cluster-level prevalence estimates from the most recent DHS (2016–2022). The geographic coordinates of the survey clusters were mapped using ArcGIS software. Cluster-level prevalence values were linked to their corresponding geographic locations to visualise the spatial variation in tobacco use across East Africa. To enhance interpretability, the prevalence values were categorised into five classes using a data-driven classification approach based on natural breaks (Jenks optimisation) within ArcGIS. This method group values into intervals that minimise within-class variance and maximise between-class differences, allowing for a clearer visualisation of spatial heterogeneity.¹⁵

In addition, a supplementary meta-analysis was conducted to assess the robustness of the estimates across countries and survey years. Country-specific and year-specific prevalence estimates were pooled using a random-effects model with inverse variance weighting to account for heterogeneity. The results are presented as forest plots in the supplementary materials, showing individual and pooled estimates with corresponding CIs.

Given the hierarchical structure of DHS data (individuals nested within communities), a two-level multilevel logistic regression model was fitted to identify the determinants of tobacco use. The model accounts for clustering at the community level and estimates both fixed effects (ORs for predictors) and random effects (variances attributable to community-level factors). A two-level mixed-effects logistic regression model was employed, with individuals (level 1) nested within clusters (level 2). Because cluster identifiers (v001) were not unique across countries, a unique cluster identifier was created by combining the country and cluster variables. A random intercept was specified at the cluster level to account for the intracluster correlation. Although a three-level model was considered, it failed to converge; therefore, the two-level specification was used.

The regression analysis involved fitting four consecutive models.^{16 17} The empty model (model I) contained only the random intercept and was used to estimate the proportion of variance in tobacco use attributable to differences between communities, expressed as the intraclass correlation coefficient (ICC). The second model (model II) was fitted using individual-level explanatory variables. In this model, the significance and contribution of each individual-level explanatory variable, as well as any changes that occurred in the first-level and second-level variance terms, were evaluated. The third model (model III) was fitted by including community-level predictors. We examined whether the explanatory variables at the community level accounted for the between-group variation in the outcome variable. The last model (model IV) was fitted to examine the effect of both individual-level and community-level explanatory variables simultaneously.

Parameter estimation methods

The fixed effect’s outcome was presented as an adjusted OR (AOR) with a 95% CI. Statistical significance was determined for all variables with p values <0.05. These models quantify the relationship between individual and community-level characteristics and their likelihood of occurrence. The selection of variables for inclusion in the multivariable multilevel model was based on a comprehensive review of the peer-reviewed literature and the structure and availability of variables within the DHS datasets. The most recent DHS survey wave was used for hierarchical analysis due to inconsistencies in the measurement and availability of key covariates across earlier waves. This approach ensured the internal consistency and comparability of variables while maintaining the policy relevance of the findings.

The random effects were measures of variation in the prevalence of tobacco use across clusters. The proportional change in variance (PCV), ICC, and median OR (MOR) were calculated to measure the variation.

The ICC measures the degree of heterogeneity of tobacco use between clusters (ie, the percentage of the overall observed variation in tobacco use that is attributable to between-cluster variations), and the following formula was used to calculate it¹⁸:

ICC=σ²u0/σ²u0+σ²e, where σ²u0=variance of the level 2 residual errors (cluster level variance), cluster level variance residual, and σ²e=the variance of the level 1 (individual level) residual errors.

The MOR quantifies the variation or heterogeneity in tobacco use between clusters in terms of the OR scale and is defined as the median value of the OR between the cluster at a high likelihood of tobacco use (ie, the median increase in odds of tobacco use if a woman moved to another cluster with a higher risk).¹⁹

MOR=exp (0.6745*√(2*σ²u0))^, σ²u0=cluster level variance for respective model

The total variation attributed to individual and/or community-level factors in each model was measured by the PCV, which was calculated as follows:

PCV=VN−VM/VA, where VN=variance of the null model and VM=variance of the model with more terms.¹⁶

Model fit statistics

The sequential multilevel models were compared using deviance and Akaike’s Information Criterion (AIC). These statistics were used descriptively to assess the model fit across models 1–4. The final model for interpretation was selected based on the lowest deviance and AIC values.²⁰

Results

Background characteristics of the study participants

This analysis included 543 013 women of reproductive age from 11 East African countries. In each DHS survey, most of the respondents resided in rural areas: about 91 888 (76.36%) in DHS 1999–2005, 83 227 (78.09%) in 2006–2010, 78 320 (65.80%) in 2011–2015 and 131 478 (71.5%) in 2016–2022. The survey-weighted distribution (weighted using DHS sampling weights) showed that most of the respondents were in the age group 15–24 in all survey years: 51 286 (42.6%) for DHS 1999–2005, 44 390 (41.6%) for DHS 2006–2010, 52 605 (39.8%) for DHS 2011–2015 and 74 643 (40.6%) for DHS 2016–2022. There was a reduction in the proportion of non-educated women, from 28.67% in the DHS 1999–2005 to 17.2% in the DHS 2016–2022 (table 2).

Table 2. Survey-weighted background characteristics of women aged 15–49 for trends and determinants of tobacco use in East Africa (1999–2022, n=542 528).

Variables		DHS 1999–2005, N (%)	DHS 2006–2010, N (%)	DHS 2011–2015, N (%)	DHS 2016–2022 n (%)	DHS 1999–2022, N (%)
Age group	15–24	51 286 (42.6)	44 390 (41.6)	52 605 (39.8)	74 643 (40.6)	222 924 (41.1)
	25–34	36 748 (30.5)	33 361 (31.3)	42 606 (32.2)	56 288 (30.6)	169 003 (31.1)
	35–49	32 302 (26.8)	28 831 (27.1)	37 086 (28.0)	52 867 (28.8)	151 086 (27.8)
Residence	Urban	28 448 (23.6)	23 354 (21.9)	40 711 (34.2)	52 321 (28.5)	149 645 (27.6)
	Rural	91 888 (76.4)	83 227 (78.1)	78 320 (65.8)	131 478 (71.5)	393 368 (72.4)
Educational level	No education	34 498 (28.7)	18 512 (17.4)	19 358 (16.3)	31 671 (17.2)	105 985 (19.5)
	Primary	65 241 (54.2)	60 062 (56.4)	55 448 (46.6)	86 182 (46.9)	275 144 (50.7)
	Secondary	18 689 (15.5)	25 196 (23.6)	37 025 (31.1)	53 764 (29.3)	137 599 (25.3)
	Higher	1907 (1.6)	2811 (2.6)	7188 (6.0)	12 182 (6.6)	24 271 (4.5)
Wealth class	Low	28 608 (37.7)	39 223 (36.8)	46 533 (35.2)	65 672 (35.7)	180 036 (36.1)
	Middle	14 448 (19.0)	20 432 (19.2)	24 674 (18.7)	34 762 (18.9)	94 316 (18.9)
	Higher	32 842 (43.3)	46 926 (44.0)	61 090 (46.2)	83 365 (45.4)	224 223 (45.0)
Media exposure	No exposure	33 502 (27.9)	21 917 (20.6)	24 457 (20.6)	59 860 (32.6)	141 717 (26.1)
	Have exposure	86 583 (72.1)	84 616 (79.4)	94 388 (79.4)	123 939 (67.4)	400 811 (73.9)

DHS, Demographic and Health Surveys.

Prevalence and trends of tobacco use

The overall prevalence of current tobacco use among women of reproductive age in East Africa varies across countries and over time. In the earliest survey rounds (1999–2005), the survey-weighted prevalence of tobacco use was estimated at 3.74%. This proportion increased slightly to 4.04% between 2006 and 2010, reaching its highest value during the observation period. However, in subsequent years, the prevalence dropped significantly to 1.35% in 2011–2015 and then rose slightly to 2.74% in 2016–2022. Overall, the pooled prevalence across the entire study period was 2.84% (figure 1) (online supplemental file 3).

Figure 1. The survey-weighted prevalence of tobacco use among reproductive-age women in East Africa by DHS survey year. DHS, Demographic and Health Surveys.

The country-specific prevalence also varied over time. Burundi experienced a significant decline from 10.17% in 2010 to 5.45% in 2017, whereas Ethiopia maintained consistently low levels, dropping from 1.4% in 2005 to 0.5% in 2016. Kenya showed a notable decrease from 2.5% in 2003 to 0.7% in 2014, with a slight increase to 1.3% in 2022. Madagascar displayed an upward trend, increasing from 7.8% in 2004 to 12.12% in 2021. Rwanda steadily decreased from 8.3% in 2000 to 1.5% in 2020, while Tanzania’s numbers declined from 1.5% in 2005 to 0.6% in 2022. Zambia decreased from 2.7% in 2002 to 1.6% in 2014 and then rose again to 2.7% in 2018. Overall, most countries showed declining tobacco use, with Madagascar and Zambia experiencing increases or fluctuations, highlighting regional and country-specific differences in tobacco control trends (figures 2 and 3, online supplemental file 3).

Figure 2. Survey-weighted trends in tobacco use among women aged 15–49 in East Africa by country, 1999–2022. DHS, Demographic and Health Surveys.

Figure 3. Country-specific survey-weighted trends in tobacco use among women aged 15–49 in East Africa, 1999–2022 (countries shown separately). DHS, Demographic and Health Surveys.

The distribution of any tobacco uses among women aged 15–49 in East Africa, based on recent DHS data (2016–2022), revealed notable variability at the cluster level. The highest prevalence (ranging from 23.65% to 39.23%) was observed in the eastern and southern regions of Madagascar, southeastern Uganda, western Ethiopia and the northwestern Kenya. Clusters in these areas exhibited substantially higher tobacco use than the surrounding regions. In contrast, clusters in most other East African countries consistently showed a low prevalence of 0%–1.01% (figure 4).

Figure 4. Spatial distribution of any tobacco use among women aged 15–49 in East Africa, using recent DHS (2016–2022). DHS, Demographic and Health Surveys.

Factors associated with tobacco use among reproductive age women in East Africa

A multilevel logistic regression analysis was conducted using recent DHS data (2016–2022) to identify individual and community-level factors. Four models were fitted: model 1 (null model), model 2 (individual-level variables), model 3 (community-level variables) and model 4 (full model with both levels). The null model revealed a cluster-level variance of 3.26 and an ICC of 47.79%, indicating that 47.79% of the total variation in tobacco use was attributable to differences between clusters. Model 2, which included age, education, marital status, employment status, parity, abortion history, current pregnancy, current breastfeeding, current amenorrhea, child death experience, media exposure, wealth class, family size, household head sex and accessing healthcare status, reduced the cluster variance to 3.15 (PCV=3.37%), showing that individual characteristics explained part of the between-cluster variation. Model 3, which included residence altitude and region, yielded a cluster variance of 1.35 (PCV=58.59%), demonstrating that these community-level factors also contributed to explaining the cluster-level differences.

The full model (model 4) combined individual-level and community-level predictors, further reducing the cluster variance to 1.22 (PCV=62.57%) with an ICC of 27.09%, indicating that a substantial portion of the cluster-level variation was explained. The MOR decreased across the models, highlighting the reduction in unexplained cluster heterogeneity. Model comparison based on deviance and AIC showed that the fourth model provided the best fit for the data (table 3).

Table 3. Random effect parameters and model comparison for the multilevel analysis.

Parameters	Null model	Model II	Model III	Model IV
Cluster-level variance	3.26	3.15	1.35	1.22
ICC	47.78%	48.91%	29.11%	27.09%
PCV	–	3.37%	58.59%	57%
MOR	5.592	5.435	3.029	2.868
Model fitness
Deviance	38 577.122	35 025.316	36 409.203	32 825.771
AIC	38 581.12	35 073.32	36 437.2	32 897.77

AIC, Akaike’s Information Criterion; ICC, intraclass correlation coefficient; MOR, median OR; PCV, proportional change in variance.

According to the multilevel logistic regression analysis, both individual characteristics and community context significantly influence tobacco use (table 4). From the final model analysis, individual factors such as being in middle or late reproductive age, having lower educational attainment, marital status, multiple parity, abortion history, child death experience and belonging to a lower wealth class were found to be significantly associated with tobacco use among women of reproductive age.

Table 4. Multilevel multivariable logistic regression analysis of factors associated with odds of tobacco use among women of reproductive age in East Africa, DHS 2016–2022 (N=543 013).

Variables	Category	Model II AOR (95% CI)	Model III AOR (95% CI)	Model IV AOR (95% CI)
Age group	15–24	1	–	1
	25–34	1.84 (1.60 to 2.13)*	–	1.86 (1.62 to 2.15)*
	35–49	2.85 (2.43 to 3.35)*	–	2.86 (2.44 to 3.36)*
Educational status	No education	4.57 (3.22 to 6.49)*	–	4.30 (3.01 to 6.15)*
	Primary	2.46 (1.74 to 3.49)*	–	2.32 (1.63 to 3.31)*
	Secondary	1.48 (1.04 to 2.10)*	–	1.25 (0.87 to 1.78)
	Higher	1	–	1
Marital status	Never in union	1.08 (0.86 to 1.35)	–	0.98 (0.78 to 1.22)
	Married	1	–	1
	Widowed/divorced/separated	1.35 (1.19 to 1.52)*	–	1.25 (1.10 to 1.41)*
Employment status	Not employed	0.69 (0.62 to 0.78)*	–	0.91 (0.82 to 1.02)
	Employed/not currently	0.98 (0.81 to 1.17)	–	1.17 (0.98 to 1.41)
	Currently employed	1	–	1
Parity	Nulliparous	1	–	1
	Multiparous	1.44 (1.15 to 1.81)*	–	1.42 (1.14 to 1.79)*
	Grand multiparous	1.87 (1.45 to 2.40)*	–	1.88 (1.46 to 2.41)*
Abortion history	No	1	–	1
	Yes	1.20 (1.08 to 1.32)*	–	1.21 (1.09 to 1.33)*
Currently pregnant	No/unsure	1.16 (0.98 to 1.38)	–	1.17 (0.99 to 1.38)
	Yes	1	–	1
Currently breastfeeding	No	1	–	1
	Yes	0.93 (0.83 to 1.05)	–	0.92 (0.82 to 1.03)
Currently amenorrheic	No	1	–	1
	Yes	0.94 (0.82 to 1.07)	–	0.93 (0.82 to 1.06)
Child death experience	No	1	–	1
	Yes	1.37 (1.25 to 1.50)*	–	1.39 (1.27 to 1.52)*
Media exposure	No exposure	1.13 (1.02 to 1.24)*	–	1.05 (0.95 to 1.16)
	Exposure	1	–	1
Wealth class	Lower class	2.07 (1.84 to 2.33)*	–	2.37 (2.09 to 2.68)*
	Middle class	1.37 (1.20 to 1.56)*	–	1.50 (1.31 to 1.72)*
	Higher class	1	–	1
Family size	<5	1.18 (1.07 to 1.29)*	–	1.15 (1.05 to 1.26)*
	≥5	1	–	1
Household head sex	Male	1.06 (0.94 to 1.18)	–	0.93 (0.83 to 1.04)
	Female	1	–	1
Healthcare access	Not big problem	1	–	1
	Big problem	1.31 (1.20 to 1.44)*	–	1.16 (1.06 to 1.27)*
Residence	Urban	–	0.63 (0.56 to 0.71)*	1.60 (1.40 to 1.84)*
	Rural	–	1	1
Altitude in metres	<1000 m	–	1.49 (1.10 to 2.01)*	1.40 (1.03 to 1.91)*
	1000–2000 m	–	1.33 (0.95 to 1.86)	0.98 (0.70 to 1.38)
	2001–3500 m	–	1	1
Country of residence	Burundi	–	10.63 (8.42 to 13.41)*	7.20 (5.63 to 9.22)*
	Ethiopia	–	0.78 (0.50 to 1.19)	0.44 (0.28 to 0.69)*
	Madagascar	–	32.70 (25.29 to 42.27)*	35.34 (27.32 to 45.71)*
	Malawi	–	1.09 (0.79 to 1.49)	1.00 (0.72 to 1.37)
	Mozambique	–	3.36 (2.42 to 4.65)*	3.04 (2.22 to 4.17)*
	Rwanda	–	2.64 (2.02 to 3.46)*	2.24 (1.70 to 2.94)*
	Tanzania	–	0.96 (0.68 to 1.36)	0.75 (0.54 to 1.06)
	Uganda	–	2.80 (2.10 to 3.73)*	2.17 (1.63 to 2.88)*
	Zambia	–	5.77 (4.46 to 7.47)*	5.01 (3.87 to 6.48)*
	Kenya	–	1	1
Cluster level variance		3.15	1.35	1.22

^*Statistically significant.

1, reference group; AOR, adjusted OR; DHS, Demographic and Health Surveys.

The odds of women aged 25–34 and 35–49 being tobacco users were 1.86 times (AOR=1.86; 95% CI 1.62 to 2.15) and 2.86 times (AOR 2.86; 95% CI 2.44 to 3.36), respectively, compared with women aged 15–24, after controlling for other confounders. Women with no education and primary education had 4.30 times (AOR=4.30; 95% CI 3.01 to 6.15) and 2.32 times (AOR 2.32, 95% CI 1.63 to 3.31) higher odds of using tobacco than women with higher education. The odds of tobacco use among women with an abortion history were 1.21 times higher (AOR=1.21; 95% CI 1.09 to 1.33) than among those without. Additionally, widowed and/or divorced women had 1.25 times higher odds (AOR=1.25; 95% CI 1.10 to 1.41) of tobacco use compared with married women.

Community-level factors such as rural residence and lower altitude were also significantly associated with tobacco use among women of reproductive age. After adjusting for other variables, women living in urban areas had 1.60 times higher odds (AOR (95% CI) 1.60 (1.40 to 1.84)) of tobacco use than women in rural areas. Similarly, women living at altitudes of less than 1000 m had 1.40 times the odds (AOR (95% CI) 1.40 (1.03 to 1.91)) of being tobacco users compared with women living at higher altitudes, controlling for community media exposure.

Discussion

This study examined the trends and determinants of tobacco use among women of reproductive age in East Africa, utilising pooled data from multiple DHS conducted between 1999 and 2022. Overall, the prevalence of current tobacco use among women in this region remained relatively low compared with global estimates, although temporal and country-specific variations were evident. The combined prevalence across the study period was 2.84% (95% CI 2.7% to 2.9%), with a slight increase in the early 2000s, followed by a marked decline after 2010 and a modest uptick in recent years. This finding was consistent with previous DHS-based findings from other SSA regions, showing that female smoking prevalence rarely exceeds 5% except in some contexts.²¹ This is much lower than the global female average report of tobacco use²² and substantially below female smoking rates in Europe, where almost one in five women smoke.²³

The initial increase from 3.74% in 1999–2005 to 4.04% in 2006–2010, followed by a sharp drop to 1.35% in 2011–2015, and a subsequent rise to 2.74% in 2016–2022, highlights the dynamic nature of tobacco use epidemiology in the region. The significant decrease observed in the 2011–2015 period could be attributed to the impact of robust global and regional tobacco control efforts, such as the ratification and implementation of the WHO FCTC, which gained momentum in many African countries during the mid-2000s.²⁴ This decline suggests that public health campaigns and policy changes, such as graphic warning labels and public smoking bans, may have been effective in reducing tobacco initiation and promoting cessation.²⁵ The slight increase in the most recent period may be a cause for concern, possibly indicating a need to re-evaluate existing strategies or counter new promotional tactics by the tobacco industry that specifically target women.

When disaggregating by product type, other tobacco (smokeless tobacco) forms showed an initial high at 2.95%, followed by declines to 2.05% and 0.76%, before a rebound to 1.90%. Cigarette smoking remained low and fluctuated between 0.45% and 0.76%. These findings spotlight the predominance of smokeless or traditional tobacco forms among women in East Africa, echoing global findings that smokeless tobacco prevalence among reproductive-age women in some LMICs can be around 0.56%–0.78%,¹⁰ while globally, rates are much higher; for instance, pooled smokeless tobacco prevalence among reproductive-age women exceeds 9%.¹⁰ This underlines not only the diversity of product choice in the East African region but also the need for tobacco control policies to comprehensively target both smoked and smokeless forms.

The country-specific trends may further illuminate the varied success of tobacco control measures. The consistent decline in countries like Ethiopia, Kenya and Rwanda, where prevalence rates dropped significantly, could demonstrate the efficacy of sustained, multifaceted public health programmes.^26–28 In contrast, the upward trend in Madagascar and the fluctuations in Zambia suggest that these nations may face unique challenges in implementing and sustaining effective tobacco control. These challenges may include socioeconomic factors, weaker policy enforcement or a greater cultural acceptance of tobacco use among women.^{29 30} The data for ‘other tobacco’ use, which consistently constitutes a higher proportion of tobacco use than cigarettes, is particularly important. This indicates that interventions must not solely focus on cigarette smoking but also address traditional and other forms of tobacco, which may be more prevalent and culturally embedded in these communities.

The multilevel regression analyses’ findings revealed that several individual and community-level factors were significantly associated with higher odds of tobacco use, many of which align with findings from global studies on the social determinants of health. The association of tobacco use with age demonstrated that women in their middle and late reproductive years were more likely to use tobacco compared with younger women. This finding is consistent with earlier reports from SSA, which noted that smoking and smokeless tobacco use increased with age.²¹ Globally, similar patterns have been documented in South Asia and other LMICs, where tobacco initiation in younger cohorts has slowed, but cumulative exposure and persistent use drive higher prevalence in older women.¹⁰

Educational attainment showed a strong protective effect, with women who had no or low education being significantly more likely to use tobacco compared with those with secondary or higher education. This association is widely recognised across Africa and globally. Evidence from previous studies confirms that uneducated women are more likely to consume tobacco, partly due to a lack of awareness of health consequences and vulnerability to cultural acceptance or peer pressure.³¹ Similar findings from South and Southeast Asia indicate that low literacy contributes to a higher prevalence of smokeless tobacco use, especially among women.³²

Concerning marital status, widowed and/or divorced women exhibited significantly higher odds of tobacco use compared with their married or never-married counterparts. This pattern reflects psychosocial stress, social isolation and the weakening of normative constraints, which are common explanations in other studies.³³ Similar associations have been observed in Nigeria, Kenya and Iran, where formerly married women reported greater tobacco use. These findings highlight how social disruption and loss of economic support can contribute to tobacco adoption as a coping mechanism. Reproductive history also played a significant role. Women with higher parity, a history of abortion and child death were more likely to use tobacco. These results are consistent with the notion that tobacco use may be adopted as a coping strategy for psychosocial distress.³⁴ Women experiencing reproductive adversity or maternal loss have shown higher engagement in health-risk behaviours, including tobacco consumption.³⁵ These associations suggest that cumulative life stressors, such as raising many children or coping with infant loss, may increase smoking behaviour. In support, a South African study found that multiparity independently raised the odds of tobacco exposure among young women.³⁶ In turn, tobacco use itself is associated with poor reproductive outcomes such as infertility, miscarriage and stillbirth.³⁷

The relationship between family size and tobacco use further reinforced the role of household stress, as women living in larger families were more likely to consume tobacco. This is in line with evidence from Ethiopia and other SSA countries, where women in resource-constrained, high-burden households were more susceptible to tobacco use.^{21 38} Stress related to large family responsibilities may function as a driver, and the diversion of limited household resources to tobacco has also been observed in poor communities.²¹

Women who reported that accessing healthcare was a big problem had higher odds of tobacco use than those who did not face such barriers, consistent with evidence that limited or inequitable access to health services is associated with higher tobacco use and lower engagement in cessation support.^{39 40} This finding suggests that limited access to healthcare may reduce women’s exposure to preventive counselling, tobacco screening and cessation support, thereby increasing vulnerability to continued tobacco use.^{41 42} In settings where healthcare access is constrained by distance, cost, time or poor service availability, women may also have fewer opportunities to receive health education on the risks of tobacco consumption and the benefits of quitting.^{39 43} The association further implies that tobacco use should be addressed not only through population-level control policies but also through improvements in women’s access to primary healthcare services, where early identification and counselling can play an important preventive role.^{41 44 45}

As in many other LMIC settings, wealth gradients were strongly evident. Women in lower wealth quintiles were significantly more likely to consume tobacco than those in higher quintiles. This finding is consistent with previous African studies and global reports showing that tobacco use disproportionately affects the poor, where poorer populations may have less access to cessation resources and are more susceptible to the pressures of the tobacco industry marketing.^{10 46} Disadvantaged women in low-income and middle-income settings disproportionately bear the burden of tobacco use, further supporting the critical intersection between inequality and health risks.²³ In addition, a study from South Asian countries reported that poor women were more likely to consume smokeless forms, often to suppress hunger or due to aggressive marketing strategies that target marginalised populations.³²

This study finding also showed women in urban areas were more likely to use tobacco than those in rural settings, a pattern that contrasts with prior studies in SSA, where higher rural prevalence has been linked to the use of traditional smokeless products and weaker antitobacco messaging.^{21 47} The higher urban prevalence may be driven by greater exposure to tobacco marketing, peer influence and shifting gender norms, with the tobacco industry actively portraying smoking as a marker of empowerment and modernity among urban women.^{5 48} Cultural and religious taboos that discourage female smoking in rural areas may further explain this difference. Similar findings from Ethiopia and other LMICs support the association between urban residence and higher smoking risk. These results underscore the need for urban-focused interventions, including stricter advertising regulations and tailored campaigns to counter industry narratives and prevent uptake among women.

Importantly, country of residence remained a strong determinant of tobacco use compared with Kenya, women in Burundi, Madagascar, Mozambique, Rwanda, Uganda and Zambia had higher odds of tobacco use, while women in Ethiopia had lower odds. This captures macrolevel influences such as national tobacco control efforts, media campaigns, industry presence and price/tax policies. For example, Ethiopia has enforced strict tobacco laws (FCTC implementation, high taxes), likely contributing to its very low female smoking rates. By contrast, Madagascar, despite some controls, continues to struggle with high smoking; it is one of the few SSA countries where women’s tobacco use is in double digits.

Strengths and limitations of the study

This study’s strengths include using nationally representative DHS data, which allows for trend analysis and cross-country comparisons across multiple survey years. Because the multilevel modelling approach accounts for community clustering, it provides a reliable estimate of both contextual and individual-level factors. However, there are several limitations to acknowledge. The actual prevalence of tobacco use may be underestimated since it was based on self-reported data, which can be affected by the social stigma associated with smoking in some societies. The cross-sectional design of the DHS limits the ability to establish causal relationships from the observed associations, and the lack of detailed information on contextual factors such as tobacco industry marketing, peer influence and access to cessation services may also affect the observed pattern. Furthermore, the study included only East African countries with available DHS data on tobacco use; therefore, the findings are generalisable to DHS-surveyed countries in the region rather than the entire East African subregion. Although the data are clustered at both community and country levels, a three-level hierarchical model could not be fitted due to non-convergence, so country was included as a fixed effect, which adjusts for between-country differences but does not fully account for country-level random variation, potentially limiting broader generalisability.

Implications of the study

The study findings emphasise the need for targeted, context-specific tobacco control strategies in East Africa. Although overall tobacco use among women remains low, the persistent or rising rates in countries like Madagascar and Zambia highlight gaps in policy implementation and enforcement; therefore, governments should focus on adopting and enforcing WHO FCTC-compliant measures, including taxation, advertising restrictions and mass communication campaigns. Interventions should target vulnerable groups, such as older, less-educated, lower-wealth and previously married women, as well as those with high parity or adverse reproductive histories. Programmes should address both smoked and smokeless tobacco by incorporating culturally sensitive health education, counselling within reproductive health services and community-based initiatives. Involving community and religious leaders can enhance the reach and acceptance of these efforts, while strengthened national policies and surveillance systems are essential to prevent further increases in female tobacco use.

Conclusion

In conclusion, although the prevalence of tobacco use among women of reproductive age in East Africa remains relatively low, increasing or fluctuating trends in some countries highlight the need for targeted public health interventions. Key individual and community-level factors, including education, wealth status, reproductive health characteristics, residence and altitude, should be addressed through sustained, culturally appropriate strategies. Comprehensive and gender-sensitive approaches that integrate tobacco control into reproductive health services, strengthen WHO FCTC-aligned policies and promote community-based interventions are also essential to reduce tobacco use and its associated health risks among women. Importantly, these findings are generalisable to the East African countries included in the DHS datasets used in this analysis, rather than the entire East African subregion. Future research across all East African countries should include qualitative studies to better understand the sociocultural determinants of tobacco use among women across the region as well as longitudinal or cohort studies to explore causal pathways further.

Data availability statement

Data are available in a public, open access repository.