A systematic review comparing the health effects of dual use of electronic and conventional cigarettes with health effects of exclusive smoking of conventional cigarettes

Background: A high prevalence of dual use (DU) of e-cigarettes and conventional cigarettes has been reported across the world. In some countries most users of e-cigarettes are dual users (DUs). We wanted to investigate the health effects of DU and compare with the health effects of exclusive smoking of conventional cigarettes (ESCC). Methods: A systematic search was carried out in PubMed, EMBASE, CINAHL, and Cochrane library. The last search was conducted on April 26, 2021. We included original articles on any topic relevant to health, in all languages. Reviewers independently assessed the main risks of bias without the use of automated tools. We followed the PRISMA guidelines. Both reviewers independently screened and read all publications. Results: Fifty-ve publications (52 studies) were included, 12 of the studies were prospective. There was great heterogeneity across studies both in methodology and outcome. Several studies, especially experimental studies with short-term outcome, found higher levels of harmful substances in ESCC than in DUs, however, the two largest population-based studies, with low risk of selection-bias, found higher levels of harmful substances in DUs than in ESCC. Most studies investigating symptoms or risk of disease were large population-based surveys. One study found that DUs reported a signicantly better health than ESCC, while fteen found a higher risk of e.g., pulmonary, cardiovascular or metabolic risk factors/symptoms, self-reported general health or cancer in DUs than in ESCC. The study with the longest follow-up, six years, found that DUs had an adjusted odds ratio of 1.48 (95% condence interval 0.81–2.70) of a possibly smoking-related disease (conrmed by hospital discharge abstracts) compared with ESCC. Many methodological weaknesses were identied, such as risk of reverse causality. We found a correlation between high tobacco consumption in DUs and ndings of negative health outcomes. Conclusion: Due to many methodological weaknesses, it is dicult to draw any strong conclusions, but the results indicate that DU might be as or even more harmful than ESCC. Well-designed longitudinal studies are needed. Before recommending EC for smoking cessation health authorities should consider the high risk of DU and its potential consequences. of the frequency/dose of EC use by DUs. A survey found that the of CVD was signicantly higher in DUs with a daily use of ECs than in those with occasional use of ECs (23). Another survey found increased odds of COPD with increasing frequency of EC use among people who had never indicating a stepped harm of EC use (57). The impact of frequency/dose of EC use and potential biological interactions of ECs and CCs need further investigation.


Introduction
The number of electronic cigarette/e-cigarette (EC) users has been increasing rapidly, and it is estimated that the number of adults who vape will reach almost 55 million by 2021 (1). Available evidence on the bene ts and risks of EC use are mixed and interpreted differently. Some believe that ECs have the potential to reduce the include/exclude a title. After reading the title and abstract, 623 titles were clearly not about the health effects of DU of ECs and CCs, leaving 88 papers eligible for full-text review. All were in English. Full documents were obtained for a thorough reading and a further 42 papers were excluded due to the following criteria: wrong outcome (e.g. smoking cessation or level of nicotine dependence), comparison between DU and ESCC was not possible (no analyses performed), article retracted, not EC-CC dual use (Figure 1), leaving 46 papers. An overview table of all excluded titles including the reason for exclusion, can be found in Appendix 2. Additionally, references from the screened full text papers were carefully examined for missed papers, and our own reference data base was hand-searched for possible overlooked titles. Nine relevant studies were identi ed and included, leaving a total of 55 papers in the nal analysis.

Data collection process
During the data collection process, each reviewer independently read and extracted data from each paper to a prede ned table framework. Results from the independent data collection were then compared, discussed and merged into one detailed table (Appendix 3).

Study risk of bias assessment
Both reviewers independently assessed the main risks of bias without the use of automated tools. Due to the heterogeneity of study methods, it was not possible to use a universal assessment method. When assessing selection bias, we considered sampling, volunteer bias and attrition bias. We also looked to see whether data were weighted for non-participation and if the study had taken confounding into account.

Effect measures
Effect measures varied depending on the outcome. Most papers on symptoms or disease risk presented unadjusted and adjusted odds ratios with a 95% con dence interval (CI). We present adjusted odds ratios if available (aOR, Appendix 3). Papers on toxic effects typically presented geometric means and 95% CI and/or ranges and interquartile intervals.

Data items
Papers were included if any outcome data comparing DU with ESCC were presented, even if signi cance levels between ESCC and DU were not shown. We extracted the same prede ned information from all papers (Appendix 3). If data on any variable were missing, we searched in supplementary material and/or in study protocols.

Synthesis methods
There was great heterogeneity both in methods and outcomes across papers, so merging of results in a metaanalysis was not possible. After completing Appendix 3, which gave us an overview of all studies, we distributed papers according to four categories: toxicity/carcinogenicity, pulmonary effects, cardiovascular and metabolic effects, and other health effects (Tables 1-4). Some studies reported on more than one type of health outcome (e.g. both pulmonary and cardiovascular disease) and are therefore to be found in more than one of the tables. Results were synthesized into ve overall categories, marked by signs (##, #, ¤, *, **) (Tables 1-4).

Results
Appendix 4 shows the characteristics of studies, study design, de nition of use and prevalence of use in detail.
De nitions of use (Appendix 4 for details) There was great variation both in duration and in frequency of use of products for details. The de nitions of ESCC, EC users and DUs were based on self-reports and varied a lot across studies.

Con ict of interest
Sixteen (29%) of the studies had a con ict of interest (COI): two with an EC manufacturer (50,51), two had received nancial support from anonymous contributors (26, 40), and the remaining had a COI with pharmaceutical companies. One study was tobacco-industry-sponsored and found a bene cial effect of replacing ESCC with the ECs they manufactured, also in DUs (41).
Toxicity and carcinogenicity (Table 1) Fourteen studies investigated the levels of harmful and potentially harmful substances in urine, blood, hair and saliva of DUs of ECs and CCs and in ESCC. Some studies measured only one or two harmful substances (50,53,65) while others tested up to 50 (11). Six studies found that levels of harmful substances were higher in ESCC (37,41,48,(52)(53)(54). One study found that DUs had higher values for most substances tested, but it did not test for signi cance (51). Three studies found signi cantly higher levels of harmful biomarkers in DUs (11,55,56), and nine studies found no signi cant difference. Results of the largest population-based studies, with low risk of selection bias, indicate that real-world DUs' might be exposed to higher levels of several harmful substances than ESCC.
Three studies had an experimental design with short-term follow-up. In one study, DUs should switch to ESCC for one week. The same or higher levels of carcinogen metabolites were measured in the ESCC period compared with the DU period (37). In the second study, smokers who wanted to stop smoking should switch from ESCC to ECs. After one month, 52% used both ECs and CCs (DUs) but had reduced their tobacco consumption signi cantly (approx. half of them had smoked less than ve CCs in the last week). DUs had signi cant reductions in carbon monoxide and a major metabolite of acrolein (48), a bladder carcinogen. One study was tobacco-industry-sponsored: ESCC were randomized to DU for ve days. DUs' who had halved the number of CPD had signi cant reductions in most of the harmful biomarkers assessed (41).
Two studies included long-term users who had used ECs and CCs for at least six months. One study included almost 500 persons (54) and found that DUs and ESCC had similar levels of toxic and carcinogenic substances, but ESCC had signi cantly higher levels of three tobacco-speci c nitrosamines and acrylonitrile than DUs. The other long-term study included almost 200 persons and found that DUs and ESCC had similar levels of toxic and carcinogenic substances, but DUs had a signi cantly higher level of benzene, a carcinogenic substance (56).
Three studies measured carcinogen and toxin exposure in large samples of the general population, re ecting the levels of real-world use (11,55,66). The largest study measured 50 biomarkers of toxicity/carcinogenicity in urine in more than 5,000 adults and found that DUs had signi cantly higher concentrations of most biomarkers than ESCC (11). The second largest study included almost 3,000 adults, measured over 40 biomarkers of toxicity/carcinogenicity in urine and blood and found that DUs had signi cantly higher levels of some toxic and carcinogenic biomarkers compared to ESCC (55). A third large population-based study included more than 1,100 persons and found the same levels of metals in DUs and ESCC (66). All three studies had low risk of selection bias, had weighted data and adjusted for relevant confounding (11).

Studies investigating health outcomes (Table 2-4)
One study found that ESCC had signi cantly worse health outcome than DUs, while 15 studies found the opposite.
Two studies in adolescents, based on the same survey in 2018, found that odds of asthma were higher in DUs than in ESCC (68, 69), but signi cance tests were not presented. One of the studies found higher odds of allergic rhinitis (68) in DUs than in ESCC, whereas the other study found comparable odds (69). The same odds of asthma were found in one study, but in analyses of complete data of current users, DUs had a signi cantly higher risk of asthma than ESCC (67). Three other studies found higher odds of asthma in adolescent (70) and adult DUs (59, 71) than in ESCC, but signi cance levels were not tested.
One study found the same odds of respiratory symptoms (72), and another study found higher odds in DUs than in ESCC (22) without, however, testing the signi cance level. Two studies found signi cantly higher odds of COPD, emphysema and chronic bronchitis (21,57) and a third of breathing di culties (58) in DUs rather than in ESCC. Further, one study found higher odds of COPD (71) and one of respiratory disease (46) in a cohort study, but signi cance levels were not tested.
In a cohort of military personnel incident cases of acute respiratory infections (in-and outpatient diagnoses) in the last 9 months were investigated, and the study found that DUs had higher incident rate of acute respiratory infections than ESCC but signi cance level was not tested (47). Finally, different pulmonary responses were found in mice exposed to both aerosol from EC and smoke from CC than in mice exposed to smoke from CC only. Dual exposure increased airway resistance compared with mice exposed to smoke from CC only but signi cance level was not stated (38).
Cardiovascular and metabolic outcomes (Table 3) Eleven studies investigated the cardiovascular and metabolic outcomes in DUs and in ESCC. All, except two studies (48, 73), were large population-based surveys with self-reported symptoms/diagnoses, including from almost 3,500 (74) to almost 450,000 (23) adults.
The largest study from the general population found signi cantly higher odds of self-reported CVD (stroke, myocardial infarction or coronary heart disease) (23) and signi cantly higher odds of self-reported premature CVD in DUs than in ESCC. The second largest nationally representative study found signi cantly higher odds of self-reported stroke (61) in DUs compared with ESCC, and another large survey found a signi cantly greater proportion of DUs reporting past/current arrhythmia than ESCC (58). Further, a large survey found that DUs had higher odds of myocardial infarction and stroke than CCU, but signi cance levels were not tested (75).
Three surveys investigated cardiovascular risk factors. One found that DUs had a signi cantly higher prevalence odds ratio of cardiovascular risk factors and diagnosis of metabolic syndrome than ESCC (17). One found that DUs had signi cantly higher odds of elevated human c-reactive protein, a biomarker of in ammation and predictor of cardiovascular disease, than ESCC (60). The last found that DUs had signi cantly higher odds of abdominal obesity than ESCC, but otherwise found no signi cant differences, even though there was a tendency to higher odds in DUs than in ESCC, except for blood pressure, for which ESCC had higher odds (18). This contrasts with a survey nding that DUs had higher odds of hypertension than ESCC, but a signi cant difference was not reached (24).
Three population-based surveys investigated diabetes-related measures. One study found that DUs had similar fasting glucose as ESCC (17), and another found that DUs had higher glycosylated hemoglobin levels than ESCC, but the signi cance level was not tested (76). A third study found the same levels of insulin resistance (74). In a clinical study, non-invasive vascular function testing was performed in almost 500 young persons, and the study found that DUs had similar arterial stiffness as ESCC (73).
In the prospective study with the longest follow-up, almost 1,400 persons were included at baseline (26). Adherence was high for six years. After one year, DUs had the same self-reported health as ESCC (45). Two years after baseline, DUs still had the same self-reported health as ESCC and a signi cantly higher probability of serious adverse events (25). However, six out of ten DUs stopped using ECs and continued to smoke, those who still were DUs at the 24-month follow-up had signi cant improvement in self-rated health. After four years, there was still no signi cant difference in self-reported health score and possible smoking-related disease between the DU group and CC users, but the study found generally worse outcomes in DUs (40). After six years, a possibly smoking-related disease was recorded in 10% of the participants, with no signi cant differences from the baseline group. Moreover, self-reported health showed a very small change over time in all groups. DUs had an adjusted odds ratio (OR) 1.48 (0.81-2.70 95% CI) of a possibly smoking-related disease and an adjusted coe cient 0.16 (-0.08, 0.39 95% CI) of self-rated health compared with ESCC. The results did not differ substantially when the sample was restricted to those who did not switch smoking/vaping group or to those who were visited or had their outcomes con rmed through a linkage with hospital discharge abstracts (26). Advanced analyses and adjustment for many confounders was performed.

Pregnancy outcomes
Four studies included pregnant women. Two found higher odds of giving birth to a small-for-gestational-age child among DUs than ESCC, but signi cance level was not tested (42,43) whereas one found that DUs and ESCC had the same risk of small-for-gestational-age (49). This study also found that DUs had lower odds of preterm birth than ESCC, but signi cance level was not tested (49). The last study found that offspring of DUs had same birthweight, Apgar score and mean gestation at delivery as ESCC and that offspring of DUs had a higher rate of admission to neonatal intensive care unit and higher incidence of birthweight <10th centile than ESCC, but signi cance level was not tested (39). Further, a prospective study found lower fecundability ratio in DUs than in ESCC, but not signi cantly different (44).
Two large surveys looked at sleep in adolescents. One study found that high-school students with DU were signi cantly more likely to report insu cient sleep compared with ESCC (64). The other study also found that DUs had a higher risk of sleep-related complaints than ESCC, but it was not signi cant (77). The third study including adolescents found higher odds of dental problems than ESCC, but the signi cance level was not tested (78).
Two large population-based surveys investigated SARS-CoV-2(COVID-19)-related issues. Both found that DUs had higher odds of symptoms and higher odds of con rmed/suspected COVID-19 diagnosis than ESCC, but signi cance levels were not tested (20,79).
Further, population-based surveys found signi cantly worse median general health scores (58) and signi cantly higher levels of uric acid and prevalence of hyperuricemia (19) in DUs than in ESCC and the same prevalence of poor physical health (72). An interview survey in more than 4,000 homeless adults found that DUs had signi cantly higher rates of cancer compared to ESCC (63). A large survey in male soldiers found that DUs had signi cantly worse tness than ESCC (62). Finally, a clinical study found that DUs had higher levels of most biomarkers of systemic in ammation than ESCC, but the difference was not signi cant (80).

Discussion
This is the rst systematic review comparing the (general) health effects of DU with ESCC. We identi ed 55 papers/52 studies. There was great heterogeneity across studies, both in the de nition of use, in methodology and in outcome measurement. Only 12 studies had a prospective design, and many studies had methodological weaknesses, so it is di cult to draw strong conclusions.
Several studies, especially experimental studies with short-term outcome, found higher levels of toxic/carcinogenic substances in ESCC than in DUs, however, the largest population-based studies with low risk of selection bias indicate that real-world DUs' might be exposed to higher levels of harmful substances than ESCC. Most studies investigating symptoms or risk of disease were large population-based survey. The longest follow-up of a cohort was six years. One study found that DUs reported a signi cantly better health than ESCC while fteen found a higher risk of e.g., pulmonary, cardiovascular or metabolic risk factors/self-reported symptoms, self-reported general health or cancer in DUs than in ESCC.
If smokers replaced most of the CCs with ECs, there might be a bene cial effect of DU (81). The included studies found that DUs smoked the same number of CPD as ESCC (11,17,26,41,45,55,62), one study actually found a signi cantly higher number of CPD in DUs than in ESCC (58). This is in agreement with other studies, not included in this review (32-34, 82, 83). Thus, known and unknown (84) harmful compounds and transformation products formed during the vaping process of ECs (85) (81) are added to the harmful substances in tobacco smoke.
The experimental studies con rm a correlation between the number of CPD and harm in DUs, DUs who had substantially reduced the number of CPD with ECs exhibited reduced biomarker levels (41,48) proportional to the reduced numbers of CPD (41). Also, a large population-based study showed that the frequency of CC use was positively correlated with toxicant concentration (11). Most of the studies that reported the same or higher tobacco consumption in DUs as in ESCC found signi cantly worse health outcomes in DUs than in ESCC (11,17,26,55,58,62). None of the studies where DUs reported smoking a lower number of CPD than ESCC found signi cantly worse outcome in DUs (50,51,54), in fact, one of them found that ESCC had a worse outcome (54).
Only two studies investigated the potential impact of the frequency/dose of EC use by DUs. A survey found that the risk of premature CVD was signi cantly higher in DUs with a daily use of ECs than in those with occasional use of ECs (23). Another survey found increased odds of COPD with increasing frequency of EC use among people who had never smoked, indicating a stepped harm of EC use (57). The impact of frequency/dose of EC use and potential biological interactions of ECs and CCs need further investigation.
It is important to note that the majority of studies were cross-sectional (conclusions on causality cannot be drawn) and most DUs have been ESCC for decades and might have persistent smoking related disease. Further, some of the studies investigating disease asked if participants ever had been diagnosed with, for example, heart disease (23) or stroke (61), without taking into account if this occurred before or after they started using ECs. The study by Bhatta found p<0.001 for reverse causality (46).
Confounding factors could also be a reason for the tendency towards worse health outcomes in DUs. Almost all studies adjusted for sociodemographic factors and many adjusted for other relevant factors, such as exposure to second-hand smoke (68). However, few studies adjusted for differences in previous tobacco consumption, years smoked or age at smoking debut. The Italian cohort study, however, had adjusted for years of tobacco smoking and still found that the risk of possible smoking-related disease tended to be slightly worse in persistent DUs after six years (26).
Few of the included studies were designed to compare DUs with ESCC. Also, the de nition of DU varied a lot across studies and disease was mostly self-reported, which imposes the risk of both recall and misclassi cation bias. However, a large retrospective survey using hospital records also found a tendency towards a higher risk of respiratory infections in DUs (47). Misclassi cation might also be a problem when it comes to smoking/vaping status (86).
On the other hand, the included studies had also several strengths. Some were very large. Most surveys weighted data, and almost all studies took confounding into account. Regarding the misclassi cation of disease diagnoses, the Italian cohort study also performed analyses on participants who had their outcomes con rmed through a linkage with hospital discharge abstracts (26), and reached the same conclusion.
Well-designed longitudinal studies on health effects of DU are needed.

Limitations
Due to the large heterogeneity of the studies, we were unable to assess the quality of studies in a systematic way. Due to the large heterogeneity in outcome measurement we were not able to perform a meta-analysis.

Strengths
Both authors independently read all abstracts and full text of papers and extracted data. Eventual disagreements were discussed. Potential COIs that might in uence the ndings were described.

Conclusion
A high prevalence of DU of ECs and CCs has been reported across the world. In some countries a majority of users of ECs are DUs. Our ndings indicate that DU might be as or maybe even more harmful than ESCC, but strong conclusions cannot be drawn due to many methodological weaknesses. Well-designed longitudinal studies on health effects of DU are needed. Before recommending e-cigarettes for smoking cessation or implementing a legislation that favours use of EC health professionals and decision makers should consider the high risk of DU and its consequences on public health.    The PRISMA ow-chart of the search and inclusion of papers in the systematic review

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Appendix1Searchstrategy.pdf