Methods

Study design

This study used repeated cross-sectional study from five waves of the Special Eurobarometer surveys on tobacco (waves 77.1 [2012], 82.4 [2014], 87.1 [2017], 93.2 [2020], and 99.3 [2023]), conducted by the European Commission and designed to be nationally representative of individuals aged ≥15 years in each EU MS.19 Sample sizes were approximately 1000 respondents per country, with smaller samples of around 500 respondents, in countries such as Cyprus, Luxembourg, and Malta. Twenty-six EU MS were included; Croatia and the United Kingdom were excluded to ensure consistency across waves, as they were not covered in the first and last survey years, respectively.

The Eurobarometer surveys use a standardised multistage random sampling method. Primary Sampling Units (PSUs) are selected with probability proportional to population size and stratified by region and degree of urbanisation, ensuring coverage of all Eurostat NUTS II regions as well as metropolitan, urban, and rural areas. Within each PSU, addresses are selected using a combination of random starting points and systematic sampling procedures. One individual per household is then selected using the nearest birthday rule and interviewed in the national language. Post-stratification weights provided by Eurostat were applied to align samples with national and EU population distributions by age, sex, and area of residence. Further methodological details are available elsewhere.19

The 2020 wave was conducted under COVID-19 pandemic conditions, which led to deviations in fieldwork procedures, including disruptions to standard face-to-face data collection and related adaptations in fieldwork timing and logistics, with several countries, such as Belgium, Denmark, Estonia, Finland, Ireland, Luxembourg, Netherlands, Spain, and Sweden, switching partially or entirely to online computer-assisted web interviewing.

We defined young people as individuals between 15 and 30 year of age. The analytic sample of this study with complete cases included a total of 20,466 respondents across 26 EU MS (Supplementary Fig. S1).

Outcome measures

Ever use

Ever use of waterpipe and e-cigarettes was used as an indicator of lifetime experimentation or exposure and early adoption of these products. All participants were asked if they had ever tried waterpipe. In 2012 and 2014, responses included “Yes, you use or used it regularly”; “Yes, you use or used it occasionally”; “Yes, you tried it once or twice; “No”; and “Don’t Know”. Responses starting with “Yes” were considered as ever use of waterpipe. From 2017, responses were frequency-based, including “Every day”; “Every week”; “Every month”; “Less than monthly”; “You used to use it regularly, but you have stopped”; “You have tried it only once or twice”; “Never”; “Refusal”; and “Don’t know”. Individuals selecting any of the first six responses were classified as ever users of waterpipe.

All participants were asked if they had tried e-cigarettes. In 2012, the responses included “Yes, you use or used it regularly”; “Yes, you use or used it occasionally”; “Yes, you tried it once or twice”; “No”; and “Don’t know”. The first three responses were considered as ever use of e-cigarettes. From 2014 onwards, responses (with slight variations in wording) were “You currently use electronic cigarettes or similar electronic devices (e.g., e-shisha, e-pipe)”; “You used them in the past but no longer use them”; “You tried them in the past but no longer use them”; “You have never used them”; and “Don’t know”. Individuals selecting any of the first three responses were classified as ever users of e-cigarettes.

The survey questions, response options, and definitions used to classify waterpipe and e-cigarette ever use across Eurobarometer waves are summarised in Supplementary Table S1.

Current use

We defined current use of waterpipe or e-cigarettes as use at least monthly at the time of data collection, as an indicator of recent, ongoing and socially patterned behaviour rather than established or dependent use. Analyses were limited to waves 2017–2023 due to non-comparable responses in earlier waves.

Current waterpipe use was derived from the frequency-based responses used to assess ever use: individuals reporting use at least monthly were classified as current users. For e-cigarettes, respondents who reported that they “currently use” the product were asked about frequency of use, with response options including “Every day”; “Every week”; “Every month”; “Less than monthly”; “You have tried only once or twice”; “Never”; and “Refusal”. Those reporting e-cigarettes use at least monthly were classified as current users.

Covariates

Individual-level variables

Sociodemographic characteristics

Data were collected on participants’ age, gender, area of residence and financial circumstance. We categorised age as 15–17, 18–24 and 25–30 years. Gender was recorded as man or woman in 2012–2017; from 2020 onwards, participants could also select “None of the above/Non-binary/Do not recognise yourself in the above categories”. To ensure comparability across waves, analyses were restricted to respondents identifying as man and woman (0.09% excluded). Area of residence was classified as rural if “rural area or village”, and urban if “town” or “city” was selected. Responses on difficulty paying bills in the last 12 months were used as a proxy for experiencing financial difficulty and grouped into two categories: almost never/never and most of the time/from time to time. Descriptive statistics for these variables are presented in Supplementary Tables S2 and S3.

Smoking status

Smoking status was derived from responses to the question: “Regarding smoking cigarettes, cigars, cigarillos, or a pipe, which of the following applies to you?”. Respondents reporting “You currently smoke” were classified as current smokers, those reporting “You used to smoke but you have stopped” as former smokers, and those reporting “You have never smoked” as never smokers. The question wording and response options were consistent across all survey waves.

Country-level variables

Actual Individual Consumption (AIC) per capita

AIC per capita measures the value of all goods and services consumed per person in a country, including household purchases and services. It was used to adjust for underlying differences in living standard and purchasing power across EU. Data for each country and survey year were obtained from Eurostat.20

Tobacco Control Scale (TCS)

TCS scores indicate the implementation level of comprehensive national tobacco control policies based on six cost-effective measures that should be prioritised according to the World Bank, including price, smoke-free laws, public spending on tobacco control, advertising bans, health warning, and treatment. This score increases with the strength of tobacco control policies up to a possible maximum of 100 points, indicating full implementation.21 For each EU MS, the most recent score available before each survey wave was used as an indicator of tobacco control policy implementation, accounting for lagged policy effects.

Statistical analysis

Weights provided in the Eurobarometer datasets were applied to account for the sampling design when estimating prevalence of product use across the EU. Descriptive results are presented as percentages with 95% confidence intervals (CIs). Observations with missing values or responses of “Don’t know” or “Refusal” were excluded (n = 1,024; 4.8% of observations) (Supplementary Fig. S1).

We used the survey wave (year; hereafter wave) as a proxy for time to examine trends over time. Modified Poisson regression was used to estimate adjusted prevalence ratios (aPRs). Two multilevel Poisson regression models (mepoisson) with robust standard errors were used to examine associations between wave and ever and current use of waterpipe and e-cigarettes. Given the limited number of survey waves, wave was modelled as a categorical variable rather than assuming a linear trend or using joinpoint regression. These regression analyses were fitted without survey weights, because the variables used to construct the weights were included as covariates in the models, an approach supported in the survey methodology literature.22 Given the repeated cross-sectional design, multilevel models were specified with countries set as the higher-level unit, and relevant covariates were included to account for between-country and between-wave heterogeneity.

Initial models included age and gender as covariates based on established literature.10,23 Additional individual-level candidate confounders were assessed comparing alternative model specifications using Akaike Information Criterion and Bayesian Information Criterion to determine their contribution to model fit and parsimony. After individual-level covariates had been finalised, country-level variables were incorporated. Final models were adjusted for area of residence, financial circumstance, smoking status, AIC per capita and TCS score. Analyses of current use were restricted to data from 2017 to 2023. To examine changes over time, we treated wave as a categorical variable with 2017 as the reference year, representing the midpoint of the study period and the first year with comparable data available on current use for waterpipe.

To explore whether trends differed between sociodemographic characteristics, interaction terms between each sociodemographic variable and wave were tested for statistical significance. To account for multiple testing across these interaction terms, p-values were adjusted using the Benjamini–Hochberg false discovery rate procedure.24 Statistically significant interactions after correction were used to define primary stratified analysis, with models stratified by the relevant sociodemographic factor while adjusting for other covariates. For sociodemographic variables whose interaction with wave did not remain statistically significant after correction in a given model, stratified results are presented as exploratory. This allowed examination of temporal changes in product use across subgroups.

Sensitivity analyses were conducted using alternative definitions of ever and current use of waterpipe and e-cigarette, including ever regular use (≥occasionally; respondents reporting having tried it once or twice were classified as never users), ever use (experimentation; respondents reporting ever regular use were set to missing), current use including less-than-monthly use, and current daily use, each compared with never use. To assess the potential impact of methodological deviations in the 2020 survey wave, all primary regression analyses for ever and current use of waterpipe and e-cigarettes were repeated after excluding the 2020 wave, and estimates were compared with the main analyses to evaluate the robustness of observed temporal trends.

Results are presented as aPRs with 95% CI, with statistical significance set at p-value < 0.05. All statistical analyses were performed using Stata V19.5.

Role of the funding source

The funders did not influence the study design, data collection, data analysis, interpretation, or writing of the manuscript.

Ethics statement

This study involved a secondary analysis of publicly available, fully anonymised data and therefore did not require ethical approval. The requirement for written informed consent was waived because the analysis used anonymised data with no possibility of participant identification.