The present meta-analysis was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (16). A comprehensive literature search was carried out in the PubMed, EMBASE, and Web of Science from their inception to February 2020. The following keywords and their combinations were used as search strategy: (“alcohol” or “alcohols” or “alcoholic” or “ethanol” or “wine” or “beer” or “liquor”) and (“VTE” or “thromboembolism” or “venous thrombosis” or “pulmonary embolism”) and (“cohort” or “prospective” or “nested case–control” or “trial”). The search strategy in PubMed is provided in Supplementary Text S1. Similar search strategy was used in EMBASE and Web of Science. The reference lists of retrieved articles and reviews were examined for additional relevant studies. There were no restrictions for language or publication date.

Eligibility criteria were reported using the PICOS strategy (Supplementary Table S1). Included studies must meet all the following criteria: (1) cohort or nested case–control study or randomized trial; (2) evaluated the association between alcohol intake and VTE risk; and (3) reported relative risk (RR) and its 95% confidence interval (CI). There were no restrictions on types of VTE, which included fatal or nonfatal, symptomatic or asymptomatic, and provoked or unprovoked VTE. For articles based on same study population, the article reporting the largest sample size or longest follow-up was included. Study selection was independently performed by two reviewers (MC and MJ). Any divergences were resolved by consensus.

The following information was recorded from each included study with a predesigned data collection form: first author's surname, study country, study design, study population/source, duration of follow-up, number of cases and sample size, method of exposure assessment, and risk estimates with their 95% CIs. Data extraction was performed by the same two independent reviewers (MC and MJ).

The quality evaluation was performed with an eight-item instrument named Newcastle–Ottawa Scale (NOS, http://www.ohri.ca/programs/clinical_epidemiology/oxford.asp). Each item represents 1 point, except for Comparability (2 points). A study with ≤6 points was considered low quality, and >6 points was considered high quality. The quality assessment was carried out by two independent reviewers (MC and MJ). Any disagreements were resolved through discussion and consensus.

The summary RR with its 95% CI was calculated using a DerSimonian random effect model that accounted for both within- and between-study heterogeneity (17). Heterogeneity across included studies was assessed using the Cochran chi-square test (the level of significance was set at 0.1) and the I² statistic (18). Publication bias was assessed using a visual plot, Begg's test (rank correlation method) (19), and Egger's test (linear regression method) (20). Sensitivity analysis was performed by removing each included study in turn and then recalculating the summary risk estimate. Subgroup analyses were also carried out based on the following factors: geographical region, gender, number of cases, and publication year.

Next, studies that reported at least three levels of alcohol intake and corresponding VTE risk estimates were included in the dose–response analysis. For studies that only provided a range for exposures, the mid-point in each category was assigned to the corresponding RR. When the highest category was open ended, the width of the interval was assumed to be the same as in the closest category. Generalized least squares trend (GLST) regression model (21, 22) was used to explore the potential dose–response relation between alcohol intake and VTE risk. Finally, a potential nonlinear dose–response association was examined by modeling alcohol intake using restricted cubic splines with three knots at percentiles 25, 50, and 75% of the distribution (23). Statistical analyses were performed with STATA 10.0 (StataCorp, College Station, TX) with two-sided P values (set at 0.05).

The detailed process of study searching and selection is outlined in Figure 1. We identified 1,237 potentially eligible studies. Most of them were excluded after screening the titles and/or abstracts. Fourteen studies were further evaluated by full-text reading. Finally, 10 prospective studies (10–15, 24–27) were included in our meta-analysis.

All of the included studies were published in English. A total of 441,128 individuals with 10,221 VTE cases were included in these 10 studies. Six studies were from North America and four from Europe. Articles were published between 1996 and 2019. The ascertainment of alcohol intake was based on self-administrated questionnaires (eight studies) or interview (two studies). The quality of the included studies was assessed by NOS, which was summarized in Supplementary Table S2, and the mean NOS score was 7.5 (range, 6–9). Detailed study characteristics are presented in Table 1.

Four studies (12, 14, 15, 25) reported the results separately by gender and thus were regarded as independent cohorts. Therefore, 14 cohorts were included in the final summary analysis. Overall, the highest consumption of alcohol was not associated with the VTE risk compared with the lowest group [Figure 2; RR, 0.96 (95% CI, 0.89–1.04), P = 0.293]. No obvious heterogeneity of RRs was observed across these studies (P = 0.249 for heterogeneity, I² = 18.8%).

The impact of each included study on the summary RR was determined by repeating the meta-analysis after omitting each study in turn. The results indicated that the pooled risk estimate became statistically significant when the study by Johansson et al. (25) was removed [Figure 3; RR, 0.92 (95% CI, 0.86–0.99)]. Funnel plot (Figure 4) and Egger's and Begg's tests all showed no evidence of publication bias (Egger, P = 0.443; Begg, P = 0.730). Finally, various stratified analyses were performed. Although no significant association was observed in any subgroups, a borderline significant association between alcohol consumption and VTE risk was observed in women [RR, 0.91 (95% CI, 0.82–1.00)] (Table 2).

Seven studies (nine cohorts) (10–13, 25–27) were available for the dose–response analysis. Alcohol intake for the different categories, together with RRs and CIs, was provided in Supplementary Table S3. After evaluating dose–response patterns for alcohol intake per day, we observed a linear decrease in VTE risk with increasing alcohol intake (P = 0.634 for nonlinearity). However, the reduced risk was not statistically significant (Figure 5).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.