Family Structure and Peer Influences as Correlates of Lifetime Smokeless Tobacco Use Among Undergraduate Students at The University For Development Studies, Ghana: An Analytical Cross-Sectional Study

Abstract

Background

Smokeless tobacco (SLT) use is a growing public health concern among young adults in sub-Saharan Africa. Evidence on family-level factors associated with lifetime SLT use among university students in Ghana is lacking. This study estimated the lifetime prevalence of SLT use and identified associated factors, focusing on family structure, among undergraduates at the University for Development Studies (UDS), Tamale, Ghana.

Methods

An analytical cross-sectional study was conducted (January–March 2023) among 600 undergraduate students at UDS (response rate 85.7%). Data were collected using a locally adapted WHO student drug-use survey. Subgroup prevalence was calculated with 95% Wilson score confidence intervals. Between-group differences were assessed using the chi-square test or Fisher’s exact test where the corresponding expected cell count was fewer than five. Factors associated with lifetime SLT use were identified using Firth penalised logistic regression, with standard logistic and modified Poisson regression as sensitivity analyses.

Results

Lifetime SLT prevalence was 9.67% (95% CI: 7.54–12.31%). In the primary multivariable analysis, having friends who used drugs during upbringing (aOR = 1.84; 95% CI: 1.02–3.32; p = 0.043) and having relatives who used drugs during upbringing (aOR = 1.82; 95% CI: 1.01–3.27; p = 0.045) were independently associated with lifetime SLT use. Associations were consistent across all three regression models.

Conclusions

Early-life exposure to drug use within peer and family networks is independently associated with lifetime SLT use among university students at UDS, Ghana. Family-based and peer-focused prevention strategies may be useful in reducing SLT use in this context.

1. Introduction

Tobacco use remains a leading preventable cause of morbidity and mortality worldwide. Smokeless tobacco (SLT), including chewing tobacco, oral snuff, and nasal snuff administered without combustion, accounts for more than 650,000 deaths annually, predominantly from oral and pharyngeal cancers and cardiovascular disease. ¹ SLT products are carcinogenic ² and contain high concentrations of nicotine and tobacco-specific nitrosamines ³ ; their discreet, odour-free nature facilitates concealment and continued use, especially in contexts where combustible tobacco is more visibly regulated. ⁴

Globally, approximately 4.4% of adolescents aged 12–16 years report current SLT use, with higher prevalence among males (5.7%) than females (3.1%). ⁵ In sub-Saharan Africa, a multi-country analysis across 53 nations reported a mean SLT prevalence of 6.4% among school-going adolescents aged 11–17 years. ⁶ In Ghana, the 2017 Global Youth Tobacco Survey (GYTS) indicated that 3.1% of students reported current SLT use and 9.2% reported ever-use of SLT. ⁷ These figures likely underestimate the true burden, given the cultural normalisation of SLT in some communities and inadequate regulatory enforcement. ⁸

University students occupy a uniquely vulnerable position: the transition away from parental supervision, exposure to diverse peer networks, academic stress, and increased product availability may collectively heighten susceptibility to SLT initiation. ¹ Family structure has consistently been identified as a critical determinant of adolescent and young adult risk behaviours in low- and middle-income country settings,^8,9 with adolescents from households characterised by parental absence, limited supervision, or substance-using family members exhibiting higher susceptibility to tobacco and other substance use.^8-12 These observations are broadly consistent with social learning theory, which posits that behaviours are acquired through observation and reinforcement from significant individuals in one’s social environment. ¹³ Cross-sectional evidence, however, cannot establish that such family-level exposures causally precede SLT use; the associations documented may instead reflect shared unmeasured factors.

Despite this evidence base, research specifically examining how family structure relates to lifetime SLT use among university students in Ghana is lacking. Addressing this gap is important for informing prevention strategies aligned with Sustainable Development Goal 3, which aims to reduce preventable risk factors including tobacco use. ¹⁴ This study therefore sought to: (1) estimate the lifetime prevalence of SLT use among undergraduate students at UDS, Tamale, Ghana; and (2) identify factors associated with lifetime SLT use, with a particular focus on family structure variables.

2. Methods

2.1. Study Design and Setting

An analytical cross-sectional design was used. The study was conducted between January and March 2023 at the University for Development Studies (UDS), Tamale: the first public university in northern Ghana. UDS enrols more than 10,000 undergraduate students across health-based and non-health-based schools, faculties, and departments.

2.2. Study Population and Eligibility

The target population was all undergraduate students enrolled at UDS during the study period. Eligible participants were students present on campus who provided informed consent. Students who declined to participate or were severely unwell at the time of data collection were excluded.

2.3. Sample Size, Sampling Procedure, and Participant Flow

This study forms part of a larger cross-sectional survey on substance use among university students; the methodology and primary findings for the broader study have been published elsewhere. ¹⁵ The sample size for the parent study was calculated using the Cochran formula ¹⁶ :

$$\mathrm{N}\ge {\mathrm{Z}}^2\text{pq}\hspace{0.17em}/\hspace{0.17em}{\mathrm{d}}^2$$

where Z = 1.96 (95% confidence level), p = 0.5 (lifetime prevalence of substance use among comparable student populations), q = 1 − p, and d = 0.05 (5% precision error). Adjusting for a 10% non-response rate yielded a minimum sample of 384. A final sample of 600 participants was recruited to improve precision of subgroup estimates across multiple substances of interest. ¹⁷ For this SLT-specific analysis, 600 participants provide a margin of error of ±2.3 percentage points around the observed 9.67% prevalence at 95% confidence.

Figure 1 presents the participant flow. A total of 700 students were invited to participate. Of these, 680 returned questionnaires (97.1% initial return). Of the 680 returned questionnaires, 80 were subsequently excluded: 65 for incomplete responses (missing one or more key items), 10 for failure to meet age-eligibility requirements in the absence of the required parental/guardian consent form, and 5 for submission outside the stipulated two-week timeframe. A final sample of 600 participants with complete data were included in the analysis (overall response rate: 85.7%). No item-level missing data were identified after exclusion of incomplete responses, so listwise deletion resulted in no further attrition.

Figure 1.

Participant flow diagram

A quota sampling approach proportional to population size was used, incorporating school enrolment, departmental and faculty size, class size, and sex distribution. Within each quota cell, simple random sampling was applied using students’ registration records. Questionnaires were distributed electronically via students’ WhatsApp handles and institutional email addresses, accompanied by an embedded consent form. Students aged under 18 years were required to submit both respondent and parental/guardian consent forms. A two-week response window was provided, with reminders issued every two days by text message, telephone call, and email.

2.4. Data Collection and Instrument

The WHO Student Drug-Use Survey instrument ¹⁸ was adapted for the Ghanaian university context, informed by relevant literature^17,19-22 and expert consultation. The instrument was structured, anonymously self-administered, and captured: (1) sociodemographic characteristics; (2) lifetime substance use history; (3) family structure and residential characteristics; and (4) parenting and supervision characteristics during upbringing. Data were entered into the Google Forms electronic platform.

The questionnaire was pre-tested among 30 students at a nearby higher education institution; these data were excluded from the final analysis. Three teams of research assistants, each comprising three individuals with Master of Public Health (MPH) degrees and fluent in English and relevant local dialects (Dagbani, Twi), were recruited from UDS and trained for three days. Questionnaires were assessed for completeness daily.

2.5. Variables

2.5.1. Outcome Variable

Lifetime smokeless tobacco use was defined as use of any non-combusted tobacco product (administered through nasal inhalation or oral chewing) at any point in the participant’s lifetime (binary: yes/no). Locally, this encompasses “snuff” (finely ground tobacco powder inhaled through the nose or placed in the mouth).

2.5.2. Primary Exposure (Family Structure)

Six variables were measured: residential status (living in a family house; renting/living alone; living in a friend’s house); parents living together during upbringing (yes/no); nature of parental/guardian supervision during upbringing (liberal/strict/moderate); frequency of family quarrels during upbringing (sometimes/always/never); relatives using any psychoactive substance (alcohol, tobacco, or illicit drugs) when growing up (yes/no); and friends using any psychoactive substance when growing up (yes/no). For this study, ‘drugs’ refers broadly to psychoactive substances including alcohol, tobacco, and illicit substances, consistent with the WHO student drug-use survey instrument. ¹⁸ ‘Upbringing’ refers to the period of childhood and adolescence prior to university enrolment.

2.5.2. Covariates

Sociodemographic covariates included age, sex, study year, religion, and marital status, selected based on prior literature documenting their association with tobacco or substance use in similar student populations,^8,9,19-22 theoretical relevance to social learning theory, ¹³ and feasibility of measurement. Substance use covariates were lifetime alcohol use and lifetime cannabis use, included given evidence of polysubstance use clustering among young adults. ²² All variables were measured at the same time point; no temporality is implied by their designation as ‘covariates'.

2.6. Statistical Analysis

Data were exported from Kobo Collect, cleaned in Microsoft Excel, and analysed in STATA/MP version 17.0. Descriptive statistics were generated as frequencies and percentages for all variables. Subgroup prevalence of lifetime SLT use was computed as the number of SLT users in each stratum divided by the total number of participants in that stratum (SLT users/stratum N × 100), with 95% Wilson score confidence intervals. For sociodemographic variables where individual-level stratum denominators were available in the dataset, subgroup prevalence was similarly computed. Between-group differences were assessed using the chi-square (χ²) test; Fisher’s exact test (Freeman–Halton extension for comparisons with more than two categories) was substituted wherever an expected cell count was fewer than five, which applied to the religion and residential-status comparisons in Table 2.

Table 2.

Subgroup Lifetime Prevalence of Smokeless Tobacco Use by Participant Characteristics

Variable	Category	SLT users (n)	Stratum N	Subgroup prevalence % (95% CI)	Test statistic (df)	p-value
Sociodemographic characteristics
Age	<25 years	39	420	9.29 (6.87–12.44)	0.36 (1)	0.547
	≥25 years	19	180	10.56 (6.92–15.74)	​	​
Sex	Female	44	368	11.96 (9.00–15.70)	3.71 (1)	0.054
	Male	14	232	6.03 (3.64–9.77)	​	​
Study year	Year 1	7	98	7.14 (3.49–13.89)	11.44 (5)	0.043*
	Year 2	9	112	8.04 (4.23–14.69)	​	​
	Year 3	10	104	9.62 (5.30–16.79)	​	​
	Year 4	12	96	12.50 (7.26–20.60)	​	​
	Year 5	3	90	3.33 (1.07–9.73)	​	​
	Year 6	17	100	17.00 (10.87–25.49)	​	​
Religion	Christianity	31	324	9.57 (6.82–13.25)	0.04 (2)‡	0.980
	Islam	25	252	9.92 (6.78–14.28)	​	​
	Other	2	24	8.33 (2.01–28.73)	​	​
Marital status	Married/cohabiting	8	72	11.11 (5.73–20.39)	0.24 (1)	0.626
	Never married/single	50	528	9.47 (7.21–12.33)	​	​
Family structure variables
Friends using drugs when growing up	No	26	374	6.95 (4.80–10.00)	8.18 (1)	0.004**
	Yes	32	226	14.16 (10.20–19.30)	​	​
Relatives using drugs when growing up	No	21	331	6.34 (4.20–9.50)	9.72 (1)	0.002**
	Yes	37	269	13.75 (10.10–18.40)	​	​
Family quarrels during upbringing	Never	12	252	4.76 (2.70–8.10)	6.93 (2)	0.035*
	Sometimes	27	138	19.57 (13.80–27.00)	​	​
	Always	19	210	9.05 (5.90–13.70)	​	​
Residential status	Family house	23	248	9.27 (6.30–13.50)	2.84 (2)‡	0.242
	Renting/alone	30	325	9.23 (6.50–12.90)	​	​
	Friend’s house	5	26	19.23 (8.50–37.90)	​	​
Parents together during upbringing	No	11	130	8.46 (4.80–14.50)	0.27 (1)	0.599
	Yes	47	470	10.00 (7.60–13.00)	​	​
Parental/guardian supervision	Liberal	7	66	10.61 (5.20–20.30)	0.52 (2)	0.769
	Strict	29	327	8.87 (6.20–12.40)	​	​
	Moderate	22	207	10.63 (7.10–15.60)	​	​
Substance use covariates
Lifetime alcohol use	No	42	494	8.50 (6.40–11.30)	4.36 (1)	0.037*
	Yes	16	106	15.09 (9.50–23.10)	​	​
Lifetime cannabis use	No	52	500	10.40 (8.00–13.40)	1.85 (1)	0.174
	Yes	6	100	6.00 (2.80–12.50)	​	​

Notes. All values represent subgroup lifetime prevalence computed as SLT users/stratum total × 100. 95% CIs are Wilson score intervals. The χ² test was used for all comparisons except those marked ‡ (religion; residential status), where the expected count in at least one cell was <5 and Fisher’s exact test (Freeman–Halton extension for the 3-category comparisons) was used instead; the reported statistic and df for these two comparisons are shown for descriptive reference only, and the associated p-value is the Fisher’s exact p-value. *p < 0.05, **p < 0.01.

The primary multivariable estimator was Firth penalised maximum likelihood logistic regression (Model 1), selected a priori because of its robustness to quasi-complete separation and small-sample maximum likelihood bias (outcome events = 58). Results are expressed as adjusted odds ratios (aOR) with 95% confidence intervals (CIs) and p-values. Two pre-specified sensitivity analyses were conducted: (1) standard binary logistic regression with robust standard error estimators, reporting aORs; and (2) modified Poisson regression with robust variance estimation, reporting adjusted prevalence ratios (aPRs). All three models included the same set of covariates. Because Firth logistic regression, standard logistic regression, and modified Poisson regression use different estimation frameworks, model fit diagnostics (AIC, BIC, log pseudo-likelihood, and McFadden’s pseudo-R² for the two logistic models) are not directly comparable across models. These diagnostics are therefore reported as within-model reference values in Supplemental Table S1 rather than in the main results table, together with an explanatory note on their non-comparability.

Multicollinearity was assessed using the variance inflation factor (VIF); the overall VIF was <5, indicating no problematic collinearity. No item-level missing data were present in the analytical dataset; listwise deletion was applied only for exclusion of incomplete questionnaires at the data-cleaning stage, as described in Section 2.3. All statistical tests were two-tailed; the significance threshold was set at α = 0.05.

2.7. Ethical Considerations

Ethical approval was granted by the Institutional Ethics Review Board of the University for Development Studies (Approval No.: UDS/RB/01/01/23). Additional institutional approval was obtained from Deans and Heads of Department. Written informed consent was provided by all participants aged ≥18 years. For participants aged <18 years, both written respondent assent and written parental/guardian consent were required; verbal consent was not accepted for minors. The parental/guardian consent form was embedded within the electronic questionnaire package; parents or guardians provided a physical signature on the printed form, which the student then scanned or photographed and uploaded together with the completed questionnaire, so that a documentary record of consent was retained for each minor participant. Written, rather than verbal, consent was required because the fully remote mode of data collection did not permit reliable independent verification of parental identity and assent, and a verifiable documentary record was considered necessary given the sensitivity of substance-use data collected from minors. Ten prospective participants aged <18 years who could not produce a valid signed parental/guardian consent form were excluded prior to analysis (Figure 1).

Participants were assured of strict confidentiality; responses were numerically coded and personal identifiers excluded from the analysis dataset. The study was conducted in accordance with the principles of the Declaration of Helsinki (2013 revision). The STROBE guidelines for cross-sectional studies were followed in reporting (STROBE checklist provided as Supplemental File 1).

3. Results

3.1. Participant Flow and Characteristics

Of the 700 students invited, 600 provided complete, eligible responses, yielding an overall response rate of 85.7% (Figure 1). The prevalence of lifetime SLT use was 9.67% (58/600; 95% CI: 7.54–12.31%). The sample was predominantly younger, with 70.0% aged below 25 years, and had a higher proportion of females (61.3%) compared to males (38.7%). Participants were relatively evenly distributed across academic years (each year: 15–19%). Christianity (54.0%) and Islam (42.0%) were the dominant religious affiliations. Over one-third (37.7%) reported having friends who used drugs, and nearly half (44.8%) had relatives who used drugs. About 58.0% reported either occasional or frequent family quarrels during upbringing. More than half the students (54.2%) lived independently. Parental supervision was most often described as strict (54.5%), and 78.3% of participants reported that their parents lived together during upbringing. Regarding co-occurring substance use, 16.7% reported lifetime cannabis use and 17.7% reported lifetime alcohol use. Table 1 presents the frequency distribution for the full sample.

Table 1.

Frequency Distribution of Study Variables (N = 600)

Variable	Category	n	%
Lifetime SLT use	No	542	90.33
	Yes	58	9.67
Sociodemographic characteristics
Age	<25 years	420	70.00
	≥25 years	180	30.00
Sex	Female	368	61.33
	Male	232	38.67
Study year	Year 1	98	16.33
	Year 2	112	18.67
	Year 3	104	17.33
	Year 4	96	16.00
	Year 5	90	15.00
	Year 6	100	16.67
Religion	Christianity	324	54.00
	Islam	252	42.00
	Other	24	4.00
Marital status	Married/cohabiting	72	12.00
	Never married/single	528	88.00
Family structure variables
Friends using drugs when growing up	No	374	62.33
	Yes	226	37.67
Relatives using drugs when growing up	No	331	55.17
	Yes	269	44.83
Family quarrels during upbringing	Never	252	42.00
	Sometimes	138	23.00
	Always	210	35.00
Residential status	Living in family house	248	41.33
	Renting/living alone	325	54.17
	Living in friend’s house	26	4.33
Parents lived together during upbringing	No	130	21.67
	Yes	470	78.33
Parental/guardian supervision	Liberal	66	11.00
	Strict	327	54.50
	Moderate	207	34.50
Substance use covariates
Lifetime cannabis use	No	500	83.33
	Yes	100	16.67
Lifetime alcohol use	No	494	82.33
	Yes	106	17.67

3.2. Bivariate Associations With Lifetime SLT Use

Table 2 presents subgroup lifetime prevalence of SLT use for all study variables, computed as the number of SLT users in each stratum divided by the total number of participants in that stratum. For sociodemographic variables, stratum denominators from Table 1 were used to derive these prevalence estimates. Proportions among SLT users (composition) are not reported in Table 2 to avoid conflation with prevalence estimates, as these are statistically distinct quantities.

Subgroup prevalence was higher among students whose friends used drugs during upbringing (14.16% vs 6.95%; χ²(1) = 8.18, p = 0.004) and whose relatives used drugs during upbringing (13.75% vs 6.34%; χ²(1) = 9.72, p = 0.002). Family quarrel frequency was associated with SLT use (χ²(2) = 6.93, p = 0.035), with the highest subgroup prevalence among those reporting quarrels ‘sometimes’ (19.57%). Lifetime alcohol use was associated with higher SLT prevalence (15.09% vs 8.50%; χ²(1) = 4.36, p = 0.037). Sex, age, religion, marital status, residential status, parental co-residence, parental supervision, and lifetime cannabis use were not statistically associated with lifetime SLT use at the α = 0.05 threshold.

3.3. Multivariable Factors Associated With Lifetime SLT Use

Table 3 presents results from all three regression models. In the primary Firth penalised logistic regression model, two factors were independently associated with lifetime SLT use. Students whose friends used drugs during upbringing had 84% greater adjusted odds of lifetime SLT use (aOR = 1.84; 95% CI: 1.02–3.32; p = 0.043), and those whose relatives used drugs during upbringing had 82% greater adjusted odds (aOR = 1.82; 95% CI: 1.01–3.27; p = 0.045). These associations were consistent across the two sensitivity analyses. No other covariate reached statistical significance at p < 0.05 in the primary model. Model fit diagnostics for all three models are within-model reference values only and are reported in Supplemental Table S1, together with an explanation of why they cannot be compared across estimation frameworks.

Table 3.

Factors Associated With Lifetime Smokeless Tobacco Use: Results From Three Regression Models (N = 600)

Variable (reference)	Firth logit aOR (95% CI)	p	Standard logit aOR (95% CI)	p	Modified poisson aPR (95% CI)	p
Family structure
Friends using drugs – growing up (yes vs no)	1.84 (1.02–3.32)*	0.043	1.92 (1.05–3.53)*	0.036	1.72 (1.03–2.89)*	0.038
Relatives using drugs – growing up (yes vs no)	1.82 (1.01–3.27)*	0.045	1.90 (1.02–3.53)*	0.043	1.73 (1.01–2.97)*	0.047
Family quarrels: always vs sometimes	1.26 (0.65–2.46)	0.493	1.28 (0.63–2.59)	0.496	1.23 (0.68–2.21)	0.490
Family quarrels: never vs sometimes	0.60 (0.28–1.26)	0.178	0.57 (0.26–1.28)	0.174	0.60 (0.29–1.24)	0.170
Residential status: renting/alone vs family house	0.93 (0.49–1.76)	0.816	0.93 (0.50–1.75)	0.829	0.94 (0.55–1.62)	0.833
Residential status: friend’s house vs family house	1.49 (0.41–5.49)	0.546	1.47 (0.34–6.43)	0.607	1.40 (0.42–4.61)	0.572
Parents together during upbringing (yes vs no)	1.50 (0.66–3.44)	0.337	1.59 (0.62–4.08)	0.335	1.51 (0.68–3.32)	0.311
Parental supervision: strict vs liberal	0.80 (0.28–2.30)	0.674	0.82 (0.24–2.74)	0.742	0.86 (0.31–2.36)	0.773
Parental supervision: moderate vs liberal	0.91 (0.32–2.64)	0.865	0.94 (0.29–3.06)	0.921	0.97 (0.37–2.57)	0.953
Sociodemographic covariates
Age ≥25 years (vs <25)	1.42 (0.55–3.67)	0.465	1.47 (0.47–4.57)	0.502	1.41 (0.54–3.71)	0.477
Study year 2 (vs 1)	1.40 (0.50–3.90)	0.524	1.42 (0.48–4.21)	0.527	1.37 (0.51–3.66)	0.527
Study year 3 (vs 1)	1.42 (0.52–3.88)	0.496	1.43 (0.51–3.97)	0.499	1.38 (0.55–3.48)	0.491
Study year 4 (vs 1)	1.41 (0.51–3.87)	0.508	1.41 (0.47–4.19)	0.537	1.38 (0.52–3.65)	0.516
Study year 5 (vs 1)	0.66 (0.15–2.92)	0.581	0.57 (0.10–3.25)	0.526	0.62 (0.13–2.93)	0.547
Study year 6 (vs 1)	2.13 (0.74–6.16)	0.161	2.23 (0.70–7.10)	0.176	1.94 (0.70–5.39)	0.200
Sex: male vs female	0.59 (0.29–1.21)	0.152	0.56 (0.26–1.20)	0.138	0.60 (0.31–1.16)	0.131
Religion: Islam vs Christianity	0.80 (0.45–1.43)	0.455	0.79 (0.43–1.45)	0.447	0.82 (0.49–1.38)	0.459
Religion: other vs Christianity	3.57 (0.62–20.71)	0.154	3.66 (0.61–22.01)	0.153	2.52 (0.78–8.12)	0.120
Marital status: married/cohabiting vs never married	1.28 (0.49–3.36)	0.613	1.37 (0.45–4.13)	0.574	1.27 (0.51–3.18)	0.607
Substance use covariates
Lifetime alcohol use (yes vs no)	1.39 (0.68–2.86)	0.366	1.41 (0.67–2.95)	0.363	1.31 (0.71–2.42)	0.387
Lifetime cannabis use (yes vs no)	0.51 (0.21–1.25)	0.143	0.45 (0.17–1.17)	0.100	0.51 (0.22–1.17)	0.113

Notes. Firth logit = Firth penalised maximum likelihood logistic regression (primary model); standard logit = binary logistic regression with robust standard errors; modified Poisson = modified Poisson regression with robust variance. aOR = adjusted odds ratio; aPR = adjusted prevalence ratio; CI = confidence interval. *p < 0.05. Model fit diagnostics (AIC, BIC, log pseudo-likelihood, McFadden’s pseudo-R²) are reported in Supplemental Table S1 as within-model reference values only; because the three models use different estimation frameworks, these diagnostics are not comparable across models and are not used to select among them.

4. Discussion

4.1. Summary of Key Findings

This study documented a lifetime SLT prevalence of 9.67% among undergraduate students at UDS in northern Ghana. Having friends who used drugs during upbringing and having relatives who used drugs during upbringing were the only factors independently associated with lifetime SLT use in the primary multivariable model, and these associations were consistent across three pre-specified regression models.

4.2. Lifetime Prevalence of Smokeless Tobacco Use

Approximately one in ten students reported ever having used SLT. This figure is higher than the 3.1% current SLT use and 9.2% ever-use of SLT reported in the 2017 GYTS among Ghanaian junior high school students ⁷ ; direct comparisons should be made with caution, as the two studies differ in population (junior high school vs university students), sampling frame, study period, and SLT definition (current vs lifetime use). The present lifetime prevalence also exceeds the 6.4% average reported across sub-Saharan African adolescent populations, ⁶ though it remains lower than rates in South and South-East Asia, where cultural acceptance and regulatory gaps can drive prevalence above 20%. ²³

Several contextual factors may be relevant to the relatively higher prevalence observed in this student sample, including limited enforcement of Ghana’s Tobacco Control Act with respect to SLT products, poor public awareness of SLT-specific health harms, and the discreet nature of SLT use, which may facilitate use in campus settings. ⁴ Misconceptions about the relative safety of SLT compared with combustible tobacco ²⁴ may further reduce perceived risk among young adults. These are plausible contextual considerations rather than established explanations, since the cross-sectional design does not permit conclusions about their causal role.

4.3. Family and Peer Influences

The finding that friends’ and relatives’ drug use during upbringing were independently associated with lifetime SLT use is consistent with a broader literature identifying early social exposures as relevant correlates of tobacco initiation.^25,26 Students reporting friends who used drugs had approximately 84% higher adjusted odds of lifetime SLT use, broadly consistent with the social learning framework, ¹³ in which substance use within peer networks may be reinforced through modelling and increased access. The cross-sectional design, however, does not permit conclusions about the directionality of this association, and residual confounding by unmeasured social or genetic factors cannot be excluded.

The independent association with relatives’ drug use suggests an additional pathway beyond peer effects. Early family exposure to substance use may be associated with household behavioural norms, parental monitoring capacity, and emotional regulation strategies, all of which have been linked to substance use susceptibility in other contexts.^10,11 That both associations remained significant after mutual adjustment is consistent with the possibility that they operate through partially distinct pathways, although the present design cannot establish this. These findings are broadly concordant with evidence that family-strengthening interventions may reduce substance use initiation^12,27 and that peer education can be relevant to tobacco prevention.^25,26

4.4. Non-significant Covariates

Several characteristics that differed in bivariate analysis, including sex, study year, and lifetime alcohol use, were not statistically significant in the multivariable models. This attenuation after adjustment may reflect partial confounding by family and peer exposure variables or may reflect limited power to detect modest individual effects given 58 outcome events. McFadden pseudo-R² values of 0.09–0.10 for the two logistic models (Supplemental Table S1) indicate that the included covariates explain a modest proportion of variance in the outcome, consistent with other cross-sectional behavioural studies where unmeasured psychosocial, biological, and contextual factors account for substantial unexplained variation.^28,29

4.5. Strengths and Limitations

Strengths of this study include the a priori use of three regression models yielding consistent findings, the pre-specified selection of Firth penalised regression to address small-sample bias, a WHO-adapted and locally validated instrument with comprehensive quality assurance.

This study also has several limitations. First, the cross-sectional design means the temporal sequence between early social exposures and SLT initiation cannot be determined from these data, and the reported associations do not establish causation. Second, self-reported lifetime SLT use is susceptible to recall and social desirability bias; the electronic distribution of the questionnaire may have introduced selection bias if students with limited internet access or lower engagement with digital platforms were systematically less likely to participate, potentially over-representing digitally active students and limiting representativeness within UDS. Third, the study was conducted at a single university in northern Ghana, limiting generalisability to other Ghanaian universities or to younger or non-student populations. Fourth, with 58 SLT users, the study may have been underpowered to detect modest effects for some covariates. Finally, quota sampling, while ensuring proportional representation across strata, does not guarantee the same inferential properties as probability sampling, which may affect external validity.

4.6. Implications for Policy and Practice

Prevention strategies may benefit from combining family-based and peer-focused components, although the evidence from this single cross-sectional study should be interpreted with appropriate caution. Educational campaigns could target parents and extended family members on the potential downstream risks of household substance use exposure, alongside community- and campus-based peer education promoting substance-free norms. Universities could consider periodic, confidential substance use screening with referral pathways for early intervention, particularly for students with a history of early-life substance use exposure. Ghana’s tobacco regulatory framework could be strengthened to explicitly cover SLT products consistent with WHO Framework Convention on Tobacco Control guidance, ¹ enforce campus restrictions, and address public misconceptions about SLT safety.

4.7. Future Research

Longitudinal cohort studies are needed to examine the temporal relationships between early social exposures and SLT initiation. Multi-institutional studies across Ghanaian universities incorporating mental health assessments, tobacco-related knowledge and attitude measures, and social network analyses would strengthen the evidence base for national policy and intervention development. Future research should also examine mechanisms through which family and peer influences may operate, including perceived substance use norms, self-efficacy, and access to SLT products.

5. Conclusions

This study found that early-life exposure to drug use within peer and family networks is associated with lifetime SLT use among undergraduate students at UDS, Ghana. With a prevalence of 9.67%, SLT use represents a meaningful public health concern in this university setting, and this study provides, to our knowledge, the first multivariable evidence on family structure as a factor associated with lifetime SLT use among students at this institution. Prevention efforts that address the social environments in which young people are raised and educated, including family-based and peer-focused approaches, may be relevant to reducing SLT use in similar university settings. Longitudinal research is warranted to clarify the temporal and mechanistic pathways underlying these associations.

Appendix.

List of Abbreviations

aOR

adjusted odds ratio

aPR

adjusted prevalence ratio

AIC

Akaike Information Criterion

BIC

Bayesian Information Criterion

CI

Confidence interval

FCTC

Framework Convention on Tobacco Control

GYTS

Global Youth Tobacco Survey

IRB

Institutional Review Board

SLT

Smokeless tobacco

UDS

University for Development Studies

VIF

Variance inflation factor

WHO

World Health Organization

ORCID iD

Yula Salifu https://orcid.org/0000-0003-1223-1765

Ethical Considerations

Ethical approval was granted by the Institutional Ethics Review Board of the University for Development Studies (Approval No.: UDS/RB/01/01/23).

Consent to Participate

Written informed consent was provided by all participants ≥18 years; for participants <18 years, both written respondent assent and written parental/guardian consent were required, recorded via a signed consent form as described in Section 2.7. The study was conducted in accordance with the Declaration of Helsinki (2013 revision).

Data Availability Statement

The datasets used and analysed during this study are available from the corresponding author upon reasonable request.*