Cross-sectional studies investigating the association between current smoking and binge drinking among adults (18 years and over) were included despite gender, nationality, race, religion, or publication date. Papers published in non-English languages were excluded. Studies that were conducted in specific populations such as pregnant or breastfeeding women, or people with various diseases such as AIDS, metabolic disorders, etc., were also excluded. In addition, studies in which the effect of age was not adjusted were excluded from the study.

Binge drinking is defined as “a pattern of drinking that brings a person’s blood alcohol concentration to 0.08 g/dl or above. This typically happens when men consume 5 or more drinks or women consume 4 or more drinks in about 2 h” (2). Current smoking is defined as use of any type of smoked tobacco product on a daily or occasional basis (14). All studies that assessed the association between the past month and the past year binge drinking with current smoking were included in this study.

This study is part of a larger systematic review on the relationship between all types of smoking and alcohol consumption. The search strategy was a combination of the selected keyword sets within the titles, abstracts and keywords. All major electronic databases including Web of Knowledge; PubMed; Scopus; Embase and Ovid (Ovid MEDLINE, Ovid EMBASE, Ovid PsysARTICLES, Ovid PsyscINFO, Ovid-PsycEXTRA, and Ovid PsysTests) were searched from database inception to 27 May 2020. Boolean operators and truncations were different according to the databases. Relevant MeSH and -Emtree terms were included when searching the PubMed and Embase databases, respectively. Details of all searches can be found in the supporting information section (Supplementary material 2).

To manage the search results of all databases, protocols for study screening, data extraction and quality assessment were first developed by the research team. All protocols were clear and concise and were refined during the pilot study. A training workshops was held for the research team members and screening of 20 studies was done on a trial basis. During study, meetings with members of the research team were repeated as necessary. All search results of databases were combined using Excel version 2016, and duplicates were deleted. Each of the retrieved studies were screened by two independent reviewers based on the title and abstract and were classified into three groups: related, unrelated and undetectable. Undetectable studies were re-reviewed by a senior reviewer. The full text of all selected studies was searched from databases, and when necessary, from other sources such as searching ResearchGate and emailing corresponding authors. The screened positive studies were assessed based on their full text to figure out eligible cross-sectional studies on the association current smoking and binge drinking. Five reviewers [LM (50%), ZA (50%), MG (40%), HK (40%), and EJ (20%)] were involved in finding eligible studies. Any discrepancies between reviewers at each step were resolved by discussion until consensus was reached, otherwise a senior reviewer judged the case.

Data extraction was carried out by 8 authors [LM (20%), ZA (30%), MG (30%), HK (20%), EJ (25%), NG (25%), FG (25%), and SS (25%)] using a previously piloted excel sheet. The following key data were extracted: first author’s name, publication year, country, participant characteristics (age, gender, and nationality), sample size, type, and method of smoking and drinking measurement, smoking and alcohol definitions, and adjusted odds ratio (AOR) with 95% confidence interval (CI).

The methodological quality of the included studies was explored by two independent authors. Five authors [LM (10%), SS (20%), NG (25%), MS (20%), and VM (25%)] did the quality assessment using the Newcastle-Ottawa Scale adapted for cross-sectional studies (15). This scale assigns up to ten stars for selection (maximum 5 stars), comparability (maximum 2 stars), and outcome (maximum 3 stars) to the cross-sectional studies. The studies with seven or more stars were considered to have a low-risk bias, and those with six or less stars were considered as high risk. Any disagreement between reviewers at each step was resolved by discussion.

Adjusted measures of effect were extracted from each study. Effect measures were extracted as adjusted odds ratios (AOR), with 95% confidence intervals (CI). Random-effects meta-analysis and the inverse variance weighting method were adopted to estimate the pooled measures of effect across studies. Heterogeneity was investigated using Q-test (16) and I² statistic (17). Sensitivity analysis was employed to investigate the heterogeneity sources between studies. Also, a random-effects meta-regression was runed in the studies reporting the association in both genders to explore whether effect size varied by study sample size, data collection year (2007–2013, 2014–2020), and nationality (American and European, Asians and Africans). The Egger’s (18), and Begg’s (19) tests, and a funnel plot were applied to assess publication bias. Stata software (version 14.1, StataCorp, College Station, TX, USA) was used to perform analyses.

The systematic search of electronic databases retrieved 190,040 references until 27 May 2020, and after removing duplicate references, 56,718 studies remained. After reviewing the titles and abstracts, 53,547 studies that were not related to the association of smoking and alcohol drinking were excluded. For the 3,171 remained studies, full texts were reviewed and 2,692 studies were excluded for the following reasons: 2,243 studies were not related to the association between smoking and drinking, 205 studies were not in English, 123 studies did not report the effect size, and 120 studies’ full texts were not found. In general, 480 studies related to smoking and alcohol drinking were identified. Of these, 365 studies were excluded for the following reasons: in 220 studies smoking was outcome, in 127 studies, study population aged less than 18 years, and 18 studies were focused on patients (HIV/AIDS, ADHD, epilepsy, etc.) or individuals with specific conditions like pregnancy or breastfeeding.

Among the remaining 115 studies, studies related to current smoking and binge drinking were selected, and 106 studies (complete list is available at supporting information section Supplementary material 3) were excluded for the following reasons: in 100 studies the relationship between other types of smoking and alcohol was investigated, in 3 studies the study design was not cross-sectional (two cohorts and one case-control study), and in 3 study the adjusted odds ratios were not reported. Eventually, 9 studies were eligible for inclusion in the current systematic review (Figure 1).

Nine cross-sectional studies reported data on the association between current smoking and binge drinking among adults, published between 2007 and 2018. Five studies were done by researchers from the United States (5, 9, 12, 13, 20), one study was from the United States and northern Europe (21), the other three studies were from Hong Kong (10), South Korea (22), and South Africa (11). The summary of included studies, the odds ratios and 95% confidence intervals for crude and adjusted estimates, and description confounders are presented in Tables 1, 2.

Table 1 summarizes the study characteristics of the included studies. The total number of participants in the studies was 64,516 (each study sample size ranged from 563 to 21,953). In one of the studies, the association was reported for six Asian-American subpopulations, separately (Chinese, Filipino, South Asian, Japanese, Korean, and Vietnamese) (13). Five studies reported the association in all subjects without separating by gender (11, 13, 20–22), two studies reported association separately for male, female, and total subjects (5, 10), one study reported association separately for male and female subjects without reporting for total sample (9), finally one study was female only (12).

Overall AOR for both genders, based on pooling the results of 7 studies, was 2.97 (95% CI = 1.98 to 4.45). Heterogeneity was assessed using the Q-test and the I² statistic. Figure 2 shows a significant heterogeneity (Q-test = 115.88, P < 0.001; I² = 90.5%) among studies addressing association between current smoking and binge drinking. The result of the subgroup analysis of gender found that gender explains to some extent the high level of heterogeneity [Female: 3.68 (95% CI = 1.03 to 13.18; I² = 98.9%) vs. (Male: 2.53 (95% CI = 1.87 to 3.42; I² = 73.1%)] (Figure 3).

Sub-group analysis in the studies reporting the association in both genders by nationality showed that the adjusted effect size among American and European population was 1.36 (95% CI: 1.01–1.83, I² = 47.4%), while it was 3.93 (95% CI: 2.99–5.17, I² = 61.3%) among Asians and Africans. So, nationality is another characteristic that could explain heterogeneity (Figure 4).

Table 3 shows the random effect meta-regression results. The nationality influenced the magnitude of the effect size between current smoking and binge drinking. Effect size was larger among samples from Asians and Africans versus American and European. Nationality explained alone about 78% of between-studies variance.

Additionally, to examine heterogeneity between studies, the Metaninf commands were used to show the effect of removing each study on the overall result, in both sexes. According to the results, the exclusion of none of the studies had a significant role in changing the AOR. Among them, the exclusion of the study by Bartoli et al. (21) had the largest effect on the estimated AOR (3.28) (Figure 5 and Supplementary material 4).

The risk of bias assessment of the studies listed in Table 1 are based on the Newcastle-Ottawa scale (15). All 9 studies had a low risk of bias. The possibility of publication bias was explored using a funnel plot, Egger and, Begg statistical tests. Based on Begg’s funnel plot results (Figure 6), the studies are symmetrically scattered on both sides of the horizontal line, and there is no evidence of publication bias in the studies. Also, the results of the Egger and Begg statistical test confirmed this finding. The Egger’s test (P = 0.273) and Begg’s test (P = 0.457) were statistically non-significant.