Current and Former Smoking and Risk for Venous Thromboembolism: A Systematic Review and Meta-Analysis

Summary

Background:
Smoking is a well-established risk factor for atherosclerotic disease, but its role as an independent risk factor for venous thromboembolism (VTE) remains controversial. We conducted a meta-analysis to summarize all published prospective studies and case-control studies to update the risk for VTE in smokers and determine whether a dose–response relationship exists.

Methods and Findings:
We performed a literature search using MEDLINE (source PubMed, January 1, 1966 to June 15, 2013) and EMBASE (January 1, 1980 to June 15, 2013) with no restrictions. Pooled effect estimates were obtained by using random-effects meta-analysis. Thirty-two observational studies involving 3,966,184 participants and 35,151 VTE events were identified. Compared with never smokers, the overall combined relative risks (RRs) for developing VTE were 1.17 (95% CI 1.09–1.25) for ever smokers, 1.23 (95% CI 1.14–1.33) for current smokers, and 1.10 (95% CI 1.03–1.17) for former smokers, respectively. The risk increased by 10.2% (95% CI 8.6%–11.8%) for every additional ten cigarettes per day smoked or by 6.1% (95% CI 3.8%–8.5%) for every additional ten pack-years. Analysis of 13 studies adjusted for body mass index (BMI) yielded a relatively higher RR (1.30; 95% CI 1.24–1.37) for current smokers. The population attributable fractions of VTE were 8.7% (95% CI 4.8%–12.3%) for ever smoking, 5.8% (95% CI 3.6%–8.2%) for current smoking, and 2.7% (95% CI 0.8%–4.5%) for former smoking. Smoking was associated with an absolute risk increase of 24.3 (95% CI 15.4–26.7) cases per 100,000 person-years.

Conclusions:
Cigarette smoking is associated with a slightly increased risk for VTE. BMI appears to be a confounding factor in the risk estimates. The relationship between VTE and smoking has clinical relevance with respect to individual screening, risk factor modification, and the primary and secondary prevention of VTE.

Please see later in the article for the Editors' Summary

Introduction

Although cigarette smoking has been responsible for approximately 5 million deaths every year, there are still an estimated 1.1 billion smokers worldwide [1],[2]. The magnitude of this public health challenge is growing, and estimates suggest that as many as 8 million people may die from smoking-related diseases by 2030 [2]. Venous thromboembolism (VTE) is a serious medical event and associated with a substantial risk of mortality [3]. In ambulatory population-based cohorts, the estimated 28-d mortality for the first episode of VTE is 11% [4]. Autopsy studies have found that VTE exists in about one-third of deaths in hospitals and 13% of all autopsies showed signs of fatal pulmonary embolism (PE) [5],[6].

Smoking is a well-established risk factor for atherosclerotic disease, but its role as an independent risk factor or effect modifier for VTE remains controversial. Several prospective studies reported smoking to be an independent risk factor [7],[8], whereas others failed to detect a significant relationship between smoking and VTE [9],[10]. A recent meta-analysis showed a statistically nonsignificant odds ratio (OR) for VTE of 1.18 (95% CI 0.95–1.46) for smokers compared with non-smokers [11]. However, the meta-analysis (involving a total of ten studies) included only about one-third of the data currently available. In addition, six of the ten studies included were clinical trials of oral contraceptives, in which the samples may not be representative of the general population. Furthermore, the VTE risk may be underestimated due to lack of distinction between former and current smokers and no adjustment for cardiovascular risk factors.

Smoking can be potentially reduced by individual and population-related measures; therefore, demonstrating the link between smoking and the risk of VTE may help reduce the burden of this disease. Therefore, we conducted a meta-analysis with the following aims: (1) to estimate the link between smoking and risk of VTE in the general population; (2) to measure the smoking-VTE relationship according to different degrees of adjustment for confounding factors, study designs, study populations, sex category, and type of VTE; and (3) to study dose-response patterns of tobacco exposure on the risk of VTE.

Methods

Search Strategy

This meta-analysis follows PRISMA guidelines (Text S1). We searched the publications listed in the electronic databases MEDLINE (source PubMed, January 1, 1966 to June 15, 2013) and EMBASE (January 1, 1980 to June 15, 2013) using the following text and key words in combination both as MeSH terms and text word “thromboembolism”, “venous thrombosis”, “pulmonary embolism”, deep-vein thrombosis”, “risk factors”, “smoke”, “cigarette”, “tobacco” or “smoking”. We searched articles published in any language and scrutinized references from these studies to identify other relevant studies.

Study Selection

To minimize differences between studies, we imposed the following methodological restrictions for the inclusion criteria: (1) Studies that contained the minimum information necessary to estimate the relative risk (RR) associated with smoking, including case-control and cohort studies published as original articles; (2) Studies in which populations were representative of the general population and not those with selected participants on the basis of risk factor for VTE, such as tumor, surgery, or use of oral contraceptives. In instances of multiple publications, the most up-to-date or comprehensive information was used.

Data Abstraction

Articles were reviewed and cross-checked independently by two authors (YJC and ZHL). Because there is no standardized quality scoring system for observational studies, we selected a priori several important design characteristics that may affect study quality, including method of case confirmation, percentage of patients completing planned follow-up, smoking as the primary analysis of interest, selection criteria for control participants, matching criteria, and control for confounding. Percentage agreement between the two authors on the quality review ranged from 88% to 100%. Any disagreements were resolved by consensus. Data on the following characteristics were independently extracted: study size, number of patients who developed VTE, total person-years of follow-up, study population, publication year, study design, sampling framework, study location (defined as Europe, North America, or Asia); gender category, site of VTE studied (deep vein thrombosis [DVT] or PE), type of VTE studied (unprovoked or provoked), ascertainment of VTE (validated or not validated), smoking category (ever, current, or former), and reported adjustment for potential confounders. When available, we used the most comprehensively adjusted risk estimates.

Data Analysis

RR was used as a measure of the relationship between smoking and the risk of VTE. For case-control studies, the OR was used as a surrogate measure of the corresponding RR. Because the absolute risk of VTE is low, the OR approximates the RR [12].

Summary RRs (95% CI) were calculated by pooling the study-specific estimates using a random-effects model that included between-study heterogeneity (parallel analyses used fixed-effects models), because significant heterogeneity was anticipated among studies. Pooled RRs were expressed with 95% CIs. We calculated the I² (95% CI) statistic to assess heterogeneity across studies, applying the following interpretation for I²: <50% = low heterogeneity; 50%–75% = moderate heterogeneity; >75% = high heterogeneity [13].

We calculated the population attributable fraction (PAF) as {prevalence of smoking×(RR−1)/[prevalence of smoking×(RR−1)+1]}, where RR indicates pooled RRs [14]. On the basis of population-based cohort studies, the average prevalence of three categories of smoking was estimated by weighting by the sample size of each study.

Subgroup analyses and meta-regression models were carried out to investigate potential sources of between-study heterogeneity. When several risk estimates were present in a single study (i.e., separate estimates for current and former smokers), we adjusted the pooled estimates for intra-study or within-study correlation [15].

In the dose-response analysis, we considered cigarettes per day and pack-years as explanatory variables. Because for many studies continuous exposures were reported as categorical data with a range, we assigned the mid-point in each category to the corresponding RR for each study. When the highest category was open ended, we considered 60 cigarettes per day and 60 pack-years as the maximum (for example, one study reported >20 cigarettes per day as an open range; we considered 40 cigarettes per day as the mid-point in this category). We used generalized least squares (GLST) regression models to assess the pooled dose-response relation between smoking and risk of VTE across studies that had heterogeneous categorizations of smoking [16]. Linear models were fitted and evaluated on the logarithm of the RRs.

To enable the total person-years of observation to be calculated, we included data from reports that specified one or more of the following: total person-time of follow-up; sample size and mean (or median) follow-up per patient; or sample size and cumulative incidence rate. The principal summary measure was event rate expressed per 100,000 patient-years of follow-up. Weighted meta-analytic prevalence estimates for outcomes were calculated with the variance-stabilizing Freeman-Tukey double-arcsine transformation with an inverse-variance random-effects model [17].

Small study bias, consistent with publication bias, was assessed with funnel plot, by Begg's adjusted rank correlation test and by Egger's regression asymmetry test [18]. We used STATA, version 11.0 (Stata Corp) for all analyses. Statistical tests were two sided and used a significance level of p<0.05.

Results

Study Selection

With the search strategy, 1,531 unique citations were initially retrieved. Of these, 231 articles were considered of interest and full text was retrieved for detailed evaluation. One hundred ninety-nine of these 231 articles were subsequently excluded and finally 32 articles were included in the meta-analysis (Figure 1).

Study Characteristics

Thirty-two independent observational studies reporting 3,966,184 individuals and 35,151 incident cases were identified [7],[8],[19]–[48]. Fifteen studies were based in Europe, eight in North America, and nine in Asia. No studies were based in Africa or South America. Studies were published between January 1980 and March 2013. Thirteen studies were prospective cohort studies and 19 were case-control studies. 15 studies recruited participants from population registers and 15 were hospital-based.

The methodological quality of the included studies was generally good. Of the primary studies, 100% had described independent, consecutive sampling of their cohort. Average follow-up duration ranged from 5.0 to 20.1 y. Patients were followed up for an average of over 10 y in a majority of studies (84.6%). The proportion of patients with complete follow-up to the end of the study was given for 11 studies and ranged from 70.5% to >99%. The sizes of the cohorts ranged from 855 to 2,314,701 (in total 3,926,048), with the two largest studies recruiting participants over 1 million (Table 1) [26],[29]. Nineteen case-control studies were designed to evaluate risk factors for VTE, and eight of them used either hospital discharge data or data from registries. In 12 of the 19 incident case-control studies, controls were matched for age and/or sex only (Table 2).

Of all the studies, two included only patients with DVT [36],[44] and four investigated only patients with PE [19],[31],[35],[40]. Four cohort studies [8],[22],[27],[28] and four case-control studies [33],[38],[39],[48] compared the prevalence of smoking between patients with unprovoked VTE and provoked VTE. Eight studies investigated only women [19],[23],[25],[29],[30],[32],[34],[48] and three studies included only men [20],[22],[24]. The association between smoking and VTE was the primary outcome of interest for 20 studies, whereas it was a secondary question in 12 studies. The ascertainment of VTE varied across studies; 24 studies based on medical record, radiology or autopsy (validated), and eight confirmed by questionnaire or patient registry (not validated) (Tables 1 and 2).

Adjusted RRs could be determined for all cohort studies and nine of the case-control studies. Most risk estimates were adjusted for age (19 studies) and sex (11 studies). Eighteen studies (56.3%) reported an adjusted estimate for at least one of the cardiovascular risk factors: BMI (11 cohort and seven case-control studies), cholesterol (three cohort and one case-control studies), diabetes (three cohort and three case-control studies), hypertension (four cohort and four case-control studies), alcohol consumption (four cohort studies), or physical activity (four cohort studies). Detailed information on adjustments is reported in Tables 1 and 2.

Smoking and Risk of VTE

Figures 2, S1, and S2 showed the results from the random-effects model (parallel analysis with fixed-effects model) combining the RRs for VTE. Overall, the ever smokers compared with the reference group experienced a significantly increased risk for developing VTE (RR: 1.17 [95% CI 1.09–1.25, p<0.001]). The pooled RRs for current versus never smokers and former versus never smokers were 1.23 (95% CI 1.14–1.33, p<0.001) and 1.10 (95% CI 1.03–1.17, p = 0.002), respectively.

There was evidence of moderate heterogeneity of RRs across these studies. The findings from the sensitivity analyses based on different inclusion and exclusion criteria were presented in Table 3. Risk estimates changed little after analyses with fixed effects models, inclusion of the studies with adjusted RRs, or exclusion of the two largest and the outlier studies, yet moderate heterogeneity was still present. However, when the analysis was confined to those large prospective cohort studies (high quality), the overall combined RR did not materially change, but heterogeneity was decreased to 34.68% for ever smokers, 10.61% for current smokers, and 0% for former smokers.

Because the study by Ray et al. [32] only reported risk estimates for current but not for former smokers, in order to allow an unbiased comparison between the two smoking classes, we also computed the RR for current smokers (RR: 1.25 [95% CI 1.17–1.35]) from the remaining 14 studies reporting both estimates. Compared with former smokers, current smokers experienced a significant higher risk for developing VTE (p = 0.02). Neither funnel plots nor Egger and Begg tests showed evidence of publication bias for ever smokers (Egger, p = 0.88; Begg, p = 0.21), current smokers (Egger, p = 0.06; Begg, p = 0.11), and former smokers (Egger, p = 0.41; Begg, p = 0.83) (Figure 3).

Stratified Analyses

To explore study heterogeneity, we performed stratified analyses across a number of key study characteristics and clinical factors (Table 4). The finding of increased VTE risk in smokers was consistently observed in most of the stratified analyses. Study design, geographical area, or publication year were not significant sources of heterogeneity. In addition, the RRs in studies in which VTE cases were validated with imaging examination or medical record were not systematically different from studies in which they were not (Table 4; Figures S3, S4, S5). Level of adjustment in the primary studies seemed to be associated with the results (p = 0.03 for ever smokers and p<0.001 for current smokers). Studies with no adjustments for cardiovascular risk factors found no significant association between smoking and risk of VTE, while the analysis of studies adjusted for cardiovascular risk factors, especially for BMI, yielded relatively higher RRs for ever and current smokers. There was evidence of moderate heterogeneity for former smokers (I²: 57.10% [95% CI 20.40%–76.88%], p = 0.01), but not for ever smokers (I²: 30.44% [95% CI 0%–60.69%], p = 0.11) or current smokers (I²: 19.73% [95% CI 0%–57.71%]). In a case-control study by Zhu et al. [43], the adjusted risk estimate for former smokers (RR: 3.25 [95% CI 1.92–5.49]) was much higher than the pooled risk estimate. After excluding this single study, there was no evidence of heterogeneity (I²: 0% [95% CI 0%–58.32%], p = 0.44) and the pooled risk estimate still reached statistical significance (RR: 1.09 [95% CI 1.05–1.12]) (Figure 4). The risk for developing VTE was significantly higher in current smokers than former smokers after adjustment for BMI (p<0.001).

We undertook meta-regression to further identify the relationship between BMI and smoking-VTE risk. Although baseline BMI did not seem to be significantly correlated with the smoking-VTE risk for ever smokers, BMI-adjusted risk estimates were significantly higher than unadjusted ones (p = 0.02) (Figure S6).

We also evaluated whether a difference existed between men and women, DVT and PE, and unprovoked and provoked VTE in the smoking-VTE relationship. The stratified analyses shown in Table 4 suggest no modification of the relationship by these characteristics. To allow an unbiased comparison, we also calculated the RRs from studies reporting both estimates for men and women, DVT and PE, and unprovoked and provoked VTE. Similar pooled risks were again observed in both sexes (p = 0.95) [7],[8],[26],[37], sites of VTE (p = 0.31) [29],[33],[37], and types of VTE (p = 0.38) [8],[22],[27],[28],[33],[38],[39],[48].

Dose-Response Relationship and Incidence of VTE

After evaluating dose-response patterns for cigarettes per day and pack-years for ever versus never smokers, we observed a linear increase in VTE risk with increasing smoking consumption. The risk increased by 10.2% (95% CI 8.6%–11.8%) for every additional ten cigarettes per day or by 6.1% (95% CI 3.8%–8.5%) for every additional ten pack-years (for example, an individual who smoked one pack of cigarettes per day for 40 y or two packs per day for 20 y has a relative increased risk of 26.7% [95% CI 16.0%–38.4%] for developing VTE compared with someone who never smoked) (Figure 5).

From eight population-based studies that reported information on person-years in smokers and nonsmokers, we could calculate absolute annual rates of VTE cases from the general population: 176.3 cases per 100,000 person-years in smokers and 152.0 cases in nonsmokers, corresponding to an absolute risk increase of 24.3 (95% CI 15.4–26.7) cases per 100,000 person-years.

PAF Calculations

Using the average prevalence of smoking from included cohort studies and the summary estimates obtained from all studies combined, the PAF of VTE due to smoking were 8.7% (95% CI 4.8%–12.3%) for ever smoking, 5.8% (95% CI 3.6%–8.2%) for current smoking, and 2.7% (95% CI 0.8%–4.5%) for former smoking. If cardiovascular risk factor-adjusted risk estimates were used, then the proportions of VTE explained by three categories of smoking increased to 10.6% (95% CI 7.8%–12.8%), 7.7% (95% CI 6.1%–9.1%), and 2.8% (95% CI 1.1%–4.5%), respectively.

Discussion

The present meta-analysis, involving approximately 4 million participants and more than 35,000 patients with VTE from 32 observational studies, found a slightly increased risk of VTE for smokers compared with non-smokers. The risk was higher in studies adjusted for conventional cardiovascular risk factors, especially for BMI. The risk of developing VTE was greater for current smokers than for former smokers, and a dose-response relationship was found for daily smoking and pack-years smoked.

Recent studies have suggested that patients with obesity, hypertension, diabetes, or dyslipidemia were at risk of developing VTE, whereas conflicting results were reported for smoking [10],[11],[49]–[53]. This meta-analysis is the first to our knowledge to confirm smoking to be an independent risk factor for VTE. The risk magnitude appears to be less robust than those reported for well-established major risk factors such as cancer, surgery, pregnancy, use of estrogens, or mutation of factor V Leiden and prothrombin [54]–[57]. However, smoking is more common and its coexistence is associated with an additive causative effect. For example, there was a synergistic effect on VTE risk for smoking and oral contraceptive use. Pomp et al. reported the OR of developing VTE for oral contraceptive users was 3.90, but increased to 8.79 when current smoking was added [37]. One prospective cohort study also identified a hazard ratio of 3.75 for the association of the combination of current smoking and the prothrombin mutation with the risk of VTE, significantly higher than that of the prothrombin mutation only [58]. Thus, given the multi-factorial nature of VTE, it is highly likely that the concomitant action of smoking may be responsible for a proportion of VTE in the general population.

A causal relationship between VTE and smoking may be mediated by different mechanisms. Our results suggest that the association of smoking with VTE risk may be largely mediated by an acute mechanism, supported by a dose-response relationship for the amount of current smoking and the higher risk in current compared to former smokers. In addition, the association was not solely due to smoking-related secondary diseases, because we found a positive association between current smoking and both unprovoked and provoked VTE. Furthermore, there is biological plausibility for the relationship. A procoagulant state, reduced fibrinolysis, inflammation, and increased blood viscosity may underlie the association between smoking and VTE risk [59]–[61]. Smoking is associated with a higher level of plasma fibrinogen, hence the increase of factor VIII, which has been reported to be associated with VTE [62]. It has been shown that the fibrinogen concentration decreased quickly after cessation of smoking and the fibrinogen concentration was nearly equal in former smokers and never smokers [63],[64]. Yarnell et al. detected a positive relationship between the amount of current tobacco consumption and plasminogen activator inhibitor-1 concentration, which may also be related to VTE [65]–[67]. These findings might suggest an acute causal association and the dose-response relationship between VTE and smoking. However, a relatively weak association between former smoking and the risk of VTE was also observed. We suppose this association may be mediated by secondary smoking-related diseases. Former smoking is related to cardiovascular diseases, diabetes, and certain types of cancer [68]–[70], which may be associated with risk of VTE [54],[71]. It is also possible that chronic inhalation of tobacco smoke, causing progressive lung destruction, chronic obstructive pulmonary disease, and emphysema [72], may also result in a hypercoagulable state and thus contribute to an increased risk of VTE [73].

It is of note that lack of adjustment for BMI tends to deflate the pooled risk estimate, indicating that BMI is an important confounding factor when assessing the smoking-VTE association. Limiting studies to those adjusted for BMI identified no significant heterogeneity for ever and current smokers, suggesting that BMI may be a source of heterogeneity. Current smokers tend to be thinner than nonsmokers or former smokers [74]–[76], and several studies have shown that smokers' BMI is lower [77]. However, previous studies also identified obesity or weight gain to be an independent risk factor for developing VTE [11],[20],[78]. Thus, given that body leanness of some smokers might partly reduce the risk, the true magnitude of association between smoking and VTE may be greater. This may be an explanation for the non-significant association observed in previous observational studies and meta-analysis that did not control for BMI.

Strengths of this meta-analysis include the strict inclusion criteria, the large number of patients analyzed, the robustness of the findings in sensitivity analyses, the dose-response relationship, and the fact that all subgroup analyses were prespecified a priori. The absence of important publication bias supports the robustness of the study findings. A possible limitation of our study is the heterogeneity of the studies with regard to adjustment of the estimates for potential confounders. Although differences in levels of adjustment seems, at least in part, to explain this finding, heterogeneity still exists in former smokers, even after we confined the analysis to studies that adjusted for BMI. This suggests that apart from BMI, there are other factors that potentially may confound the risk estimates. Furthermore, baseline BMI was not significantly associated with the smoking-VTE risk, indicating that the relation between BMI and risk of VTE in smokers needs to be further elucidated. Inclusion of different types of studies into one meta-analysis may also introduce heterogeneity into the results. Despite this, the consistency of the finding of an increased risk of VTE among smokers in both case-control and cohort studies suggests that the association is valid. In addition, the results from a given study for the three comparisons (ever versus never, former versus never, current versus never) are not independent and there is a possibility of a type I error. However, the results continued to be statistically significant after adjusting for multiple comparisons by setting α = 0.05/3. Like all meta-analyses, our study has the limitation of being a retrospective analysis. Another limitation was the lack of individual participant data, which precluded determining the independent associations of individual variables with study outcomes. Instead, we used between-study meta-regressions, when possible.

In conclusion, the results from this meta-analysis suggest that smoking slightly increases the risk of VTE, independent of conventional cardiovascular risk factors. BMI may be a potential confounding factor in the risk estimates. The association between smoking and VTE has clinical relevance with respect to individual screening, risk factor modification, and the primary and secondary prevention of VTE. Future prospective studies are needed to elucidate the specific pathogenic mechanisms.

Supporting Information

Attachment 1

Attachment 2

Attachment 3

Attachment 4

Attachment 5

Attachment 6

Attachment 7