Study and population characteristics
The searches retrieved 6,619 studies (6,510 papers from the initial searches, plus 109 papers from supplemental searches). Two researchers selected 21 papers for inclusion in the study (Appendix 2); comprising 14 unique longitudinal prospective studies (Appendices 4 ‒ 8). The data in the primary studies were collected between 2013 and 2016 and their longitudinal follow-up periods ranged from 4 months to 2.5 years. Only one study had two follow-up time points [33]. Fifteen primary papers were completed using North American populations [33-47] and six primary papers were based on European populations [48-53]. The studies’ populations ages ranged from 13 to 19 years at baseline. The included studies had a range of research questions related to e-cigarette use; 17 enquired about ever use of e-cigarettes [33, 36-40, 42, 44-53], and 4 queried current e-cigarette use in the past 30 days [35, 40, 43, 44]. Data on e-cigarette type, generation or liquid were not asked about in any study. The papers assessed tobacco cigarette smoking during follow-up as an outcome variable: 18 papers studied ever use of tobacco cigarettes between baseline and follow-up [33, 35, 37-52], and 4 queried past 30-day use of tobacco cigarettes [39, 40, 42, 44]. The publications completing regression analysis included potential confounding variables as covariates in their regression model, ranging from the inclusion of 3 variables to the inclusion of 17. Based on previous research, we grouped the covariates into three groups: demographic (e.g. age, gender, ethnicity, family affluence), interpersonal (e.g. number of friends/family members that smoke) or intrapersonal (e.g. such impulsivity, sensation seeking, rebellion) [17]. One paper [36] collected data on variables from a single domain, while eight papers [35, 38, 43, 46, 47, 50, 51, 53] had variables representing two domains, and ten papers [33, 34, 39, 40, 42, 44, 45, 48, 49, 52] had variables from all three domains.
Quality assessment
Overall, using the NHLBI quality assessment tool [22], we considered four studies to be high-quality[39, 44, 47, 52] as they had a representative and clearly defined sample with a participation rate of more than 50%, a loss to follow-up rate of 20% or less, and a sample size justification or variance calculation for the main outcomes (Appendix 3). The remaining studies in the meta-analysis were judged to be moderate quality and tended to have higher loss to follow up or lower participation rates.
Feasibility assessment
In order to ascertain whether a meta-analysis was appropriate and which studies should be included, a feasibility analysis was conducted, assessing if the primary study authors employed the same method of analysis considering outcome, exposure, unit of measurement, and length of time to follow-up (Appendix 9). Based on these criteria, nine studies with 16,808 participants were considered eligible for meta-analysis of ever using e-cigarette at baseline and smoked tobacco cigarettes at any time during the follow-up period (Figure 1), while three studies of past-30-day e-cigarette use at baseline and smoked tobacco cigarettes at any time during the follow-up period were eligible.
Ever e-cigarette use at baseline and subsequent cigarette smoking at follow-up
Our ever e-cigarette-use meta-analysis was based on the primary study adjusted odds ratios and calculated a combined 4.06 (95%CI: 3.00-5.48, I2 68%) times higher odds of initiating smoking tobacco cigarettes for those who had ever used e-cigarettes at baseline, although this combined odds ratio decreased marginally to 3.71 times (95%CI: 2.83-4. 86, I2 35%) when only the four high-quality studies [39, 44, 47, 52] were analysed in a sensitivity analysis(Appendix 10). The initial meta-analysis model had a high statistical heterogeneity, sensitivity analysis restricted to high-quality studies had moderate statistical heterogeneity.
Figure 1 here
One additional sensitivity analysis was completed on six studies [33, 39, 44, 48, 49, 52] that controlled for the three domains of covariates – that is, demographic, interpersonal, and intrapersonal factors. The results of this sensitivity analysis (OR: 3.82; 95% CI: 2.66–5.48; I2 69%) were like the results for the high-quality studies sensitivity analysis, but the level of heterogeneity remained high (Appendix 10).
Three subgroup analysis were done (Appendix 10). The first compared studies which collected data pre-2014 [33, 39, 47] with those which collected their initial data from 2014 onwards [38, 44, 48, 49, 51, 52], due to the rise in e-cigarette use that was noticed around this time [54]. The combined OR for studies which collected data from 2014 onwards increased substantially (pre-2014 AOR: 2.81, 95%CI: 2.45–3.72 I2 78%); compared to 2014 onwards (5.16, 95%CI: 3.69–7.21 I2 38%). However, the confidence intervals overlap indicating that they are not statistically significantly different.
The second subgroup analysis compared the length of time to follow-up, as studies included in the analyses had follow-up periods which ranged from 4 months to 2 years. However, as only two studies had follow-up periods of less than 1 year [49, 51] including one study which had a very small sample size, the meta-analysis for this subgroup did not provide useable results.
Finally, considering the significance of the geographical, regulatory, and cultural context of these studies, we compared European studies with those from the USA. The combined OR was higher in the European studies (OR: 6.22, 95%CI: 3.73–10.38 I2 54%) [48, 49, 51, 52] compared with the USA studies (OR: 3.18, 95%CI: 2.26–4.47 I2 65%) [33, 38, 39, 44, 47]. However, the confidence intervals overlap indicating that they are not statistically significantly different.
Past-30-day e-cigarette use at baseline and subsequent cigarette smoking at follow-up
Four studies examined the impact of past-30-day e-cigarette use at baseline and subsequent cigarette smoking at follow-up [35, 40, 43, 44]. One of the four studies was excluded following feasibility analysis prior to meta-analysis. A meta-analysis was completed using the remaining three primary studies adjusted odds ratios (Appendix 10) and included 30,018 participants. The meta-analysis identified a significant positive association between past-30-day e-cigarette use at baseline and subsequent cigarette smoking initiation at follow-up (OR: 2.14; 95% CI: 1.75–2.62; I2 0%) (Figure 2) [35, 40, 44].
Figure 2 here
Level and certainty of evidence
We assigned a level of evidence of 3 using British Medical Journal guidelines [30], as this is a systematic review of cohort studies, some of which had high loss to follow-up and/or very small sample sizes. With respect to certainty of evidence [31] and taking account of study design and statistical heterogeneity, we have moderate confidence that the true effect is likely to be close to the estimated effect for initiating smoking at follow-up for those who had ever used e-cigarettes at baseline.