Introduction

Colorectal cancer has been rapidly increasing since 1975 and is now the third most common cancer worldwide, which has an impact on approximately 1.23 million people [1, 2]. Many studies suggest that most colorectal cancers originate from precursor benign polyps [3, 4]. Besides the well-known colorectal adenomas (CRAs) which may cause tumorigenesis via the traditional adenoma-carcinoma pathway, the World Health Organization recently defined a new group of polyps called serrated polyps.

Serrated polyps (SPs), generally considered as “hyperplastic polyps” in the past, have been now classified into hyperplastic polyp (HP), sessile serrated adenoma (SSA), and traditional serrated adenoma (TSA) [5]. These polyps show different colonoscopic appearances, histology, and molecular alteration, and risk of progression to CRC. HPs are the most common category and account for 75–95 % of total SPs, while SSAs take up less than a quarter. TSAs are the rarest kind, comprising only approximate 1 % of SPs [6–11].

Rosty et al. [5] found that 30 % of CRC developed via the serrated neoplasia pathway, which was characterized by the mutation of the BRAF oncogene and development of CpG island methylator phenotype (CIMP). It is likely that different risk factors act at each step of the serrated pathway and exacerbate the SP into CRC. SPs (including HPs) share a number of common risk factors with conventional adenomas and carcinomas, such as cigarette smoking and overeating [12–14]. Some studies demonstrate smoking is an independent risk factor for developing SPs via DNA hypermethylation and even has greater association with colorectal SP than CRA [15]. However, some epidemiologic evidence suggested that serrated polyps may have many different risk factors. For instance, aging was thought to be a strong risk factor of traditional adenomas [16], yet not many studies find any prominent relationship between age and risk for serrated (or hyperplastic) polyps [17–19].

In some studies, alcohol consumption was regarded as one of the most important risk factors in the development of SPs (including HPs), as shown by Martinez et al. [20] (>9.4 g/day vs. non-drinkers: RR 2.01, 95 % CI 1.05, 3.91). However, the mechanism by which alcohol increased the risk of colorectal SPs or HPs is not yet clearly known. Alcohol may not be a direct-acting carcinogen, but the various compounds including phenols, higher alcohols, aldehydes and nitrosamines in different types of drinking may be involved in the serrated pathway [21]. Nevertheless, not many studies have proven that alcohol intake is a serious risk factor for SPs. [18]. Meanwhile, the relative risks may vary due to different dose of alcohol consumption.

Therefore, the relationship between alcohol intake and colorectal SPs of different categories is still under debate, and no comprehensive meta-analysis is conducted. Thus, this systematic review with meta-analysis is to estimate the summarized relative risk of colorectal SPs associated with alcohol intake in different categories as well as with the dose–response analysis.

Methods

Data Sources and Searches

A comprehensive, computerized literature search was carried out on PubMed from the beginning of indexing to October 2014, following the Preferred Reporting Items for Systemic Reviews and Meta-Analyses (PRISMA) guidelines [22]. Two independent investigators (Y.M. Wang and J.Z. Zhu) used the terms (colorectal OR colorectum OR colon OR rectum OR bowel) AND (polyps OR adenoma OR adenomatous) AND (ethanol OR alcohol OR drink) to search for relevant studies. Abstracts or unpublished reports were not considered for inclusion in the meta-analysis.

Study Selection

Two authors (Y.M. Wang, K.F. Zhu) independently accessed all the potentially eligible studies retrieved according to prespecified selection criteria. Discrepancies between the two reviewers were decided by another author (J.Z. Zhu). Studies were included if they met the following criteria: (1) published as an original article using a case–control, cohort or cross-sectional design; (2) reported relative risk (RR) estimates with corresponding 95 % CIs for at least three categories of alcohol intake levels (including a category of nonalcohol intake or occasional alcohol intake) and the risk of colorectal SP. Studies were excluded if the cases had both colorectal SP and adenoma. Besides, neither the nonpeer-reviewed articles, experimental animal studies nor the mechanistic studies were included.

Data Abstraction and Study Quality Assessment

Two authors (Q.Y. Zhou and K.F. Zhu) independently extracted data from the included full-text citations. The following information was abstracted: study design (case–control, cohort, or cross-sectional study), the first author’s last name, country of origin, gender, age of cases, site of SPs, publication year, number of cases and controls with the definition for each level of alcohol consumption, race/ethnicity, examination method (endoscopy or/and CT) and case ascertainment.

Two authors (J.Z. Zhu and K.F. Zhu) independently assessed the methodological quality of every included study according to the Newcastle–Ottawa Scale (NOS) [22, 23]. The NOS contains nine items (1 point for each) for the observational studies: selection (4 items), comparability (2 items), and exposure/outcome assessment (3 items). A study with no less than 7 points was regarded as high quality and a study with insufficient information was considered to be of low quality.

Statistical Methods

Different studies used different ways to calculate the alcohol consumption. Thus, the alcohol consumption was converted into grams of ethanol per day using the following equations: 1 mL = 0.8 g and 1 drink = 11.0 g. Nondrinkers or occasional drinkers were defined as the reference category. The standard deviation, quartiles, mean, and median of the lower boundaries of heavy dose of the selected studies were calculated to set up the boundary of light, moderate and heavy doses. The upper boundary of light dose was set according to the first quartile of the lower boundaries of heavy doses of all included studies. And the upper boundary of moderate dose was set by the difference of the maximum and standard deviation of the lower boundaries of heavy doses of the studies. Therefore, the alcohol consumption categories were defined using the following values: light (≤8 g/day), moderate (8–36 g/day), and heavy (≥36 g/day).

Abstracted data from the included studies were entered into Stata Statistical Software (version 12.0; StataCorp LP, College Station, TX, USA). Fixed or random effects models were used to calculate the summarized relative risks of colorectal SPs with 95 % CIs. Forest plots were made for any, light, moderate, and heavy versus non-/occasional alcohol consumption. Heterogeneity across studies was assessed with the chi-square statistic, with P < 0.10 considered significant, and with the I² test [24]. An I² value greater than 50 % was regarded as substantial heterogeneity. Furthermore, the sources of heterogeneity were investigated by subgroup analyses and meta-regression analyses according to study design, race, geographic location, subtype of SPs, and published year. Additionally, sensitivity analysis was carried out to estimate the influence of each individual study on the summary estimates by conducting the random effects meta-analysis again after omitting one of the included studies in each turn. Publication bias was assessed using the Egger linear regression test [25], the Begger–Mazumdar test and visual inspection of funnel plots. Potential publication bias would be adjusted by trim-and-fill method if applicable (P < 0.05).

A dose–response analysis was conducted using both linear and nonlinear random effects models on the natural logarithm of the RR. This method is able to explain the correlation between reported risks estimates within one study, heterogeneity between the studies and nonlinear dose–risk association. Linear random effect models and ten second-order fractional polynomial random effects models were tested to find out the best-fitting model. Finally, the best-fitting model that had the lowest Akaike’s information criterion was selected as the final dose–risk relation model. The statistics of the chosen model showed that the model was appropriate [26, 27].

Results

Literature Search

Figure 1 shows the study selection details in the meta-analysis. A total of 467 citations were identified, of which 250 citations were excluded due to irrelevant topics after screening of the titles and abstracts. A total of 207 of the remaining articles were excluded for various reasons after evaluating the full texts. Finally, ten articles (7 case–control studies [12, 18–20, 28–30], 2 cross-sectional studies [31, 32], and 1 cohort study [13]) were included in this meta-analysis.

Fig. 1
figure 1

Flowchart of the study selection process

Full size image

Study Characteristics

A total of ten studies concerning alcohol consumption and colorectal SP incidence were published between 1995 and 2014. All studies inquired about the alcohol intake of the participants, whether they ever had symptoms, and the histological diagnosis of the colorectal polyps after undergoing the examination of colonoscopy or sigmoidoscopy. Among the ten qualified studies, seven were conducted in the United States, two in Europe, and one study in Asia, as shown in Table 1. In addition, eight studies reported RRs for both colon and rectum SPs, while two did for distal colon and rectum. In terms of subtype of the SPs, only two recent articles summarized the data [30, 32], from which we can only abstract the RRs of HPs and SSAs. Due to the limited data, the analysis of the gender and the size of polyps cannot be performed.

Table 1 Characteristics of the included case–control, cohort and cross-sectional studies
Full size table

In General

Meta-analysis of the ten studies in a random-effects model showed that alcohol intake was associated with a 24 % increased risk of colorectal SP (RR 1.24; 95 % CI 1.07–1.44). There heterogeneity existed among studies (P < 0.001; I² = 70.2 %), so subgroup meta-analyses and meta-regression were used to explore the sources of heterogeneity. Moreover, RRs of different levels of alcohol drinkers were as follows: light drinkers, 1.05 (95 % CI 0.93–1.18); moderate drinkers, 1.19 (95 % CI 1.02–1.40); heavy drinkers, 1.60 (95 % CI 1.35–1.91) (Fig. 2; Table 2).

Fig. 2
figure 2

Pooled risk estimates for colorectal SP incidence for all drinkers versus non-/occasional drinkers, light drinkers versus non-/occasional drinkers, moderate drinkers versus non-/occasional drinkers and heavy drinkers versus non-/occasional drinkers, with respect to types of study (cohort, case–control and cross-sectional studies)

Full size image
Table 2 Stratified meta-analyses of alcohol consumption and risk of colorectal serrated polyp
Full size table

Type of the Studies

In the stratified analysis by study type, the summary RRs of colorectal SP were 1.24 (95 % CI 1.02–1.52) in case–control studies, 1.16 (95 % CI 0.94–1.42) in cross-sectional studies and 1.41 (95 % CI 1.10–1.80) in the cohort study, comparing all drinkers and non-/occasional drinkers (Fig. 2; Table 2).

Geographical Region

In terms of different regions, the pooled estimate showed a significant increased risk of colorectal SP in the United States [RR 1.29 (95 % CI 1.06–1.57)] when compared to the non-/occasional alcohol intake controls, but with a substantial heterogeneity between trials. However, there was no statistical difference between all drinkers and non-/occasional drinkers in Europe [RR = 1.12 (95 % CI 0.90–1.41)] and those in Asia [RR = 1.16 (95 % CI 0.88–1.52)]. Yet, further analysis suggested that heavy alcohol consumption was associated with a significant increased risk of colorectal SP in Europe [1.70 (95 % CI 1.15–2.51)] but not in Asia [1.43 (95 % CI 0.90–2.25)] (Fig. 3; Table 2).

Fig. 3
figure 3

Pooled risk estimates for colorectal SP incidence for all drinkers versus non-/occasional drinkers, light drinkers versus non-/occasional drinkers, moderate drinkers versus non-/occasional drinkers and heavy drinkers versus non-/occasional drinkers, with respect to geographical region of studies (USA, Europe and Asia)

Full size image

Subgroup of SP

Only data from the two recent studies were extracted for subgroup meta-analysis. Figure 4 demonstrated that only the heavy alcohol consumption drinkers had an increased risk of colorectal SP, with the RR of 1.24 (95 % CI 1.01–1.52). Furthermore, the limited analysis showed that alcohol intake appeared to have a slight association with SSA and HP, with the RR of 1.14 (95 % CI 0.84–1.56) and 0.99 (95 % CI 0.86–1.14), respectively, when compared to all drinkers with non-/occasional drinkers (Table 2).

Fig. 4
figure 4

Pooled risk estimates for colorectal SP incidence for all drinkers versus non-/occasional drinkers, light drinkers versus non-/occasional drinkers, moderate drinkers versus non-/occasional drinkers and heavy drinkers vs. non-/occasional drinkers with respect to subgroup of SPs [hyperplastic ployps (HPs) and sessile serrated adenomas (SSAs)]

Full size image

Sub-analysis of Smokers with SP

The risk factor of cigarette smoking can also be evaluated in the included studies, from which eight were associated with former/past smokers and nine were about the current/present smokers with colorectal SP. The studies on both past and current smokers demonstrated statistically increased risks of colorectal SP in contrast with the non-smokers, with RR of 1.58 (95 % CI 1.31–1.89) and 2.25 (95 % CI 1.78–2.86) in past smokers and current smokers, respectively (Fig. 5).

Fig. 5
figure 5

Pooled risk estimates for colorectal SP incidence for former/past smokers and current/present smokers versus non-smokers

Full size image

Meta-regression Analysis

We conducted a meta-regression analysis to investigate the impact of these study characteristics on the study estimates of RR. No variables such as study design, published year and geographical region, showed statistically significant associations with the sources of heterogeneity in univariate or multivariate meta-regression analyses.

Sensitivity Analysis

The sensitivity analysis was conducted by omitting one study at a time and calculating the pooled RRs for the remainder of the studies. There were no changes in the tendency of the risk effect when any one study was excluded. For instance, when the study by Burnett et al. [30] (the study that carried the most weight) was excluded from the analysis, the summarized RR of all drinkers remained significant (SRR 1.33; 95 % CI 1.21–1.45), compared with non-/occasional drinkers. Meanwhile, the RR for light drinkers (1.08, 95 % CI 0.94–1.24), moderate drinkers (1.26, 95 % CI 1.13–1.40) and heavy drinkers (1.72, 95 % CI 1.52–1.94) also share the same effect. Therefore, this analysis confirmed the stability of the risk tendency between alcohol consumption and colorectal SP (Fig. 6a–d).

Fig. 6
figure 6

Sensitivity analysis by omitting the study of Burnett-Hartman et al. [30] (the study that carried the most weight) and calculating the pooled RRs of the remaining studies when comparing all drinkers and non-/occasional drinkers (a), light drinkers and non-/occasional drinkers (b), moderate drinkers and non-/occasional drinkers (c), heavy drinkers and non-/occasional drinkers (d)

Full size image

Publication Bias

The shape of the funnel plots for studies on the association between alcohol consumption and colorectal SP seemed to be symmetrical. What’s more, the P value of Begg’s adjusted rank correlation test was >0.05, indicating no statistical evidence of publication bias (Fig. 7a–d).

Fig. 7
figure 7

a Funnel plot of studies evaluating the association between all drinkers versus non-/occasional drinkers (Begg’s test, P = 0.030) and colorectal SP risk. b Funnel plot of studies evaluating the association between light drinkers versus non-/occasional drinkers (Begg’s test, P = 0.154) and colorectal SP risk. c Funnel plot of studies evaluating the association between moderate drinkers versus non-/occasional drinkers (Begg’s test, P = 0.144) and colorectal SP risk. d Funnel plot of studies evaluating the association between heavy drinkers versus non-/occasional drinkers (Begg’s test, P = 0.144) and colorectal SP risk

Full size image

Dose–Response Meta-Analysis

Among the second-order fractional polynomial random effects models, the best explanatory variable is dose. The relationship between alcohol intake and CRA risk was ln (RR) = −0.09679 + 0.00569178 × dose. Compared with non-/occasional alcohol consumption, the fractional polynomial model estimates of RR were 0.96 (95 % CI 0.88–1.05), 1.05 (95 % CI 0.96–1.14), 1.21 (95 % CI 1.11–1.32) and 1.60 (95 % CI 1.47–1.75) for 10, 25, 50 and 100 g/day of alcohol intake, respectively (Fig. 8).

Fig. 8
figure 8

Relative risk function and the corresponding 95 % confidence interval, describing the best-fitting dose–response association of alcohol drinking (in g/day) and colorectal adenoma incidence

Full size image

Discussion

The comprehensive meta-analysis attempted to reveal the association between alcohol consumption and colorectal SP by pooling the data from ten observational studies. The systematic review with meta-analysis and dose–response analysis of the alcohol and colorectal SP had never been reported before.

In general, the categorical meta-analysis demonstrated that all drinkers were associated with a 24 % increased risk of colorectal SP (RR 1.24; 95 % CI 1.07–1.44), and the RR of moderate drinkers (8–36 g/day) and heavy drinkers (≥36 g/day) were 1.19 (95 % CI 1.02–1.40) and 1.60 (95 % CI 1.35–1.91), respectively, when compared with the non-/occasional drinkers. Nevertheless, the light drinkers (≤8 g/day) might not have statistically significant influence on the colorectal SP, with an RR of 1.05 (95 % CI 0.93–1.18). Notably, the further dose–response analysis produced consistent results. For drinkers of 10, 25, 35, 50, 75, and 100 g/day alcohol intake, the estimated RRs of colorectal SP were 0.96 (95 % CI 0.88–1.05), 1.05 (95 % CI 0.96–1.14), 1.11 (95 % CI 1.02–1.21), 1.21 (95 % CI 1.11–1.32), 1.39 (95 % CI 1.28–1.52) and 1.60 (95 % CI 1.47–1.75) in the fractional polynomial model. Moreover, the correlation of alcohol consumption with increased risk of colorectal SP in the United States and Europe was stronger than in Asia, especially in comparison with the heavy dose in the subgroup analysis of geographical region, as well as in the subgroup analysis of race. In terms of sub-classification of colorectal SP, alcohol consumption had a greater influence on SSA than HP, with an RR of 1.14 (95 % CI 0.84–1.56) versus 0.99 (95 % CI 0.86–1.14) for all drinkers. Cigarette smoking both in the past and at present is associated closely with serrated polyps compared to non-smokers. The estimated RR of past smokers was 1.58 (95 % CI 1.31–1.89) and current smokers was 2.25 (95 % CI 1.78–2.86).

Although the mechanisms underlying the association between alcohol intake and the increased risk of colorectal SP were not clear, several possibilities have been hypothesized. Serrated neoplasia is a relatively new concept in the field of CRC pathogenesis. Recent studies have further indicated that some SPs may develop into serrated carcinomas via serrated pathways which account for approximately 35 % of the CRCs [33–36]. Unlike the traditional microsatellite-stabled adenoma-carcinoma pathways, these pathways are characterized by microsatellite instability through the nucleotide mismatch repair defect, methylation or loss of hMLH1 or methylguanine DNA methyltransferase (MGMT), exaggerated crypt serration, excess mucin expression, and evidence of architectural dysplasia [37–41]. Epidemiologic studies on CRC have suggested that risk factor profiles may have different effects between the microsatellite instability tumors and microsatellite stable tumors [42–44]. Alcohol may activate the BRAF gene mutation, which inhibits normal apoptosis of colonic epithelial cells. Furthermore, alcohol induces CpG island methylation (CIMP) that can lead the normal mucin to become serrated lesions which are mainly mirocvesicular hyperplastic polyps (MVHPs) and SSAs [37]. Additionally, intestinal bacteria can utilize the alcohol dehydrogenase to transfer excessive ethanol to acetaldehyde, which can change the status of the intestinal flora microecology [45]. Knowledge of risk factors and their effects in the early stage of tumorigenesis is the pivotal issue in the prevention of CRC [46]. Based on the results of the present study, the screening guideline of precancerous polyps could be adjusted to consider individuals who are accustomed to high dose of alcohol (≥36 g/day) as the susceptible population.

This meta-analysis has the following strengths: (1) studies were identified after a comprehensive exclusion criteria which includes at least three categories of alcohol consumption by a multidisciplinary work team having specialists in gastroenterology, gastrointestinal endoscopy and statistics, (2) study participants in the included studies were nearly homogeneous, for they all went through the endoscopy examination and had histological evidence if polyps were found in the colon or rectum, (3) in addition, the value of relative risk was extracted from the primary data source which could avoid confounders including cigarette smoking, nonsteroidal anti-inflammatory drug use, BMI level and other factors in the selected studies, (4) finally, subgroup analyses, sensitivity and meta-regression analyses were made to reduce or remove the sources of heterogeneity and to make some better understanding of the SP-CRC sequence.

However, the systematic review with the meta-analysis and dose–response analysis also has some limitations that might affect the results. First, nine of the selected studies used a case–control or cross-sectional design which may have the selection and recall bias more or less. Second, the results were influenced by the alcohol consumption misclassification resulting from the different classifications of the alcoholic levels in each included studies although we used the accurate statistics methods to integrate them. Furthermore, the significant heterogeneity was observed across the studies when pooling the estimate RRs of all drinkers, which to some extent might cast doubt on the reliability. This significant heterogeneity might come from the study design, characteristics of the population, duration of the follow-ups and unawareness of the accurate subgroup of serrated polyps or neoplasia. We were unable to clarify these confounders, despite the random-effect models. In addition, the distributions of some other lifestyle factors may have a co-effect of alcohol consumption in the progression of the disease that we could not clearly differentiate from our analysis.

In summary, this meta-analysis suggested that moderate or heavy alcohol intake was related to a significant increased risk of colorectal SP. To be specific, the risk was stronger in SSA than in HP, and significantly higher in white people than Asian. This systematic review shed light on the fact that alcohol consumption may to some extent be involved in the etiology and pathogenesis of colorectal SP to CRC and provided additional evidence that limiting heavy dose drinking is important for early prevention of colorectal SP and CRC.