Association of alcohol consumption with colorectal cancer risk:a dose-response meta-analysis

Yan Li,Zhi-Li Xu,Yi-Yi Cui,Yong Guo＊

1Departmet of Integrative Oncology,The First Affiliated Hospital of Zhejiang Chinese Medicine University,54 Youdian Road,Hangzhou 310006,China.

2Department of First Clinical Medicine,Zhejiang Chinese Medicine University,548Bingwen Road,Hangzhou 310053,China.

3Departmet of Integrative Oncology,The Third Affiliated Hospital of Zhejiang Chinese Medicine University,219Moganshan Road,Hangzhou 310013,China.

Abstract Background:Currently,there are many meta-analyses on the correlation between alcohol conception and the risk of colorectal cancer,but the findings are inconsistent.We conducted a dose-response meta-analysis to investigate the association between alcohol intake and the risk of colorectal cancer.Methods:A literature search was performed on PubMed and EMBASE databases to identify relevant articles published before December 09,2019.Summarized relative risks(RRs)and 95% confidence intervals(CIs)were calculated using a fixed or random-effects model for the dose-response meta-analysis.The Cochran Q and I2 statistics were used to assess statistical heterogeneity among studies.Results: Five cohort and 19 case-control studies were included in the meta-analysis.An increased intake of 15-39.9g of daily alcohol was related to a 21% growth risk of CRC(RR=1.21,95% CI:1.05-1.38).Our dose-response analysis indicated that for drinkers of light,moderate and heavy level of alcohol drinking,the estimated RRs of CRC were 0.94(95% CI:0.69-1.27),1.08(95% CI:0.93-1.26)and 1.90(95% CI:1.29-2.97)respectively and for drinkers of moderate beer,heavy beer,moderate wine and heavy wine intake,the estimated RRs of CRC were 0.98(95% CI:0.72-1.34),1.48(95% CI:1.05-2.07),0.86(95% CI:0.56-1.34)and 1.04(95% CI:0.83-1.29)respectively.The risks with no significant association were consistent in the subgroup analyses of general alcohol consumption status and frequency.Conclusions:Heavy level of alcohol consumption,especially heavy beer intake seems to be associated with an increased the risk of CRC.

Keywords:Alcohol consumption,Colorectal cancer risk,Dose-response,Meta-analysis

Introduction

According to Globocan2018,colorectal cancer(including colon and rectal cancer,CRC)is the third commonly diagnosed cancer and the fourth leading cause of cancer death in both male and female[1].Approximately 60-65% of CRC cases occur in individuals without inherited genetic mutations arising CRC risk and family history of CRC[2].Of note,not just genetic variations known to be associated with CRC risk,environmental factors and changes in diet and lifestyle also contribute to the rapid rises in incidence of carcinogenesis undergoing economic development.Recognition of possible risk factors during the development and progression of colorectal cancer might help with its prevention through reducing modifiable risk factors[3].

Alcohol consumption has been identified as a major risk factor for human cancer,which plays a causal role in liver,colon,rectal,stomach,pancreatic and female breast cancers by the epidemiological evidence from the International Agency for Research on Cancer(IARC)[4].The 2017 extensive summary from the World Cancer Research Fund(WCRF)and American Institute of Cancer Research(AICR),inferred that alcohol increase CRC risk[5].Conversely,alcohol may prevent the occurrence and development of cancer through increasing insulin sensitivity or inhibiting nuclear factor kappa B(NF-kB)and activator protein-1(AP-1)transcription[6,7].Furthermore,there exists a dose-response association and alcohol types evidence between alcohol use and CRC risk[8-18].Previous studies found that moderate drinking was associated with a lower risk of colorectal cancer than heavy drinking[8,16].Other studies found that red wine and light alcohol consumption may have a protective effect on the development of CRC,although this effect is minimal[7,12].

Taking into account all the diverse results of studies along with the promotion of worldwide alcohol consumption,the association between alcohol drinking and colorectal cancer risk seems worth meta-analyzing.Thus,we performed this dose-response meta-analysis to estimate the correlation between alcohol drinking and the odds ratio(ORs)or relative risk(RRs)of CRC,following the Meta-analysis Of Observational Studies in Epidemiology(MOOSE)[19].

Methods

Data sources and searches

We carried out this meta-analysis in accordance with the PRISMA Statement guidelines[20].A systematic search was performed in PubMed and EMBASE databases.All eligible articles were screened out until December 09,2019.The retrieval strategy including Medical subject headings(MeSH)and Emtree headings were as follows:(colorectal OR colorectum OR colon OR rectum OR bowel)AND(adenomatous OR adenoma OR neoplasm OR tumor OR carcinoma OR cancer)AND(ethanol OR alcohol OR drink)AND(risk OR incidence OR prevalence).In terms of language limitations,we only selected articles published in English.The complete retrieval and filtering process is presented in Figure 1.

Assessment of study quality

Two of the authors(Y L and ZLX)identified and extracted each study independently following the MOOSE statement and the Newcastle-Ottawa Scale(NOS)[21].Two different evaluation tools in NOS were used to evaluate the case-control study and cohort study,respectively.The Newcastle-Ottawa Scale included three parameters:selection(highest score equates to 4),comparability(highest score equates to 2),and exposure/outcome assessment(highest score equates to 3).Each study with a score equals or more than 7 points presents high quality.NOS scores less than or equal to 6 points were considered as low-quality literature.Studies with insufficient information that could not be evaluated by NOS were also considered as low-quality literature.

Study selection

The inclusion criteria for this meta-analysis were as follows:(1)Colorectal cancer patients with definite pathological diagnosis;(2)original observational studies with control group(including case-control study design and cohort study design);(3)reported a correlation of the parameters between alcohol consumption and colorectal cancer prevalence;(4)outcome indicators were RRs(relative risk)or ORs(odds ratio)with 95% confidence intervals;(5)provided roughly stratified levels of alcohol consumption.

Studies were exclude if they meet the following criteria:(1)conference abstracts,cross-sectional studies,review papers,case reports,studies at a cellular level or in animals;(2)haven’t provided correlation parameter,such as ORs or RRs with 95%confidence interval(CI);(3)provided a HR(hazard ratio)with 95% confidence interval(CI);(4)no associated with colorectal cancer risk;(5)no associated with alcohol consumption;(6)the selected literature was not published in English;(7)the selected literature with duplicated data or analysis.

Data abstraction

Two of the investigators(YL and ZLX)independently extracted data from the 24 studies that were eventually screened out.The dissenting parts were checked by a third investigator(YYC).The full data information extracted from each final included study were as follows:the name of the first author,geographical location,year of publication,type of clinical trial(case-control studies or cohort studies),characteristics of study(gender,site of colorectal cancer,stage),characteristics of case-control studies(the number of cases and controls,respectively),characteristics of cohort studies(the patients number of cancer and all participants,duration of follow-up),variables adjusted parameters,alcohol stratification parameters,and assessment of clinical outcome(OR,RR,95%CI).

Statistical methods

To calculate the summarized relative risks(RRs)with 95% confidence intervals(CIs)of this meta-analysis,we used both the random-effects model and the fixed-effects model.In the selection of statistical analysis method,the Der-Simonian-Laird method in random effects model was performed.Based on the data from the selected articles,each study used their different unit to measure the alcohol consumption level,therefore,we used g/day as a standard units of ethanol level by the equations as follows:1mL=0.8g,1standard=1.33g,1 tertile=5.6g and 1 drink=12.5g.Nondrinkers and occasional alcohol drinkers were the reference category.The light alcohol drinking consumption was defined as≤15 g/day of ethanol,moderate as 15-39.9 g/day of ethanol,heavy as consumption of≥40 g/day of ethanol.The overall analysis between ever/or current alcohol drinking and CRC risk was performed.We also extracted the corresponding RR or OR values for each drinking stratification group and combined these corresponding values to obtain the median data of different hierarchical analysis.

The Cochran Q test and I2statistics were performed to conduct heterogeneity across these selected studies[22].For I2test,if the detected value is more than 50%,the heterogeneity of the study was considered to be significant.We used a stratified analysis based on the random effects model and meta-regression to further analyzing the sources of the heterogeneity.Subgroup analyses were performed by subtype of sample size(≤500;500-1000;≥1000),type of analyses(univariate analysis;multivariate analysis),follow up time(m)(≤24;＞24),type of studies(case-control;cohort),type of study design(retrospective study;prospective study),stage(Ne stage;all stage),general alcohol consumption status(ever;current),levels of alcohol drinking(the light;moderate;or heavy drinkers),types of drinking(beer or wine),and frequency of alcohol drinking(<4/month;1-4/week;≥4/week).Additionally,we conducted sensitivity analysis by eliminating each study one by one to determine whether the merger effect is stable and whether a study is the major source of heterogeneity.Begger’s funnel polt test was also carried out to evaluate publication bias between alcohol drinking and colorectal cancer risk[23].All the statistical results were performed with STATA,version 12(Stata,College Station,Texas,USA)and P-value<0.05 was determined to be statically significant.

Results

Search results

As shown in Figure 1,a total of 4423 citations meeting the requirements of inclusion were retrieved through Pubmed and Embase database.After removing 1741 duplicated references,a total of 2682 citations were used for the following title/abstracts screening.A total of 85 citations were included in the full-text assessed after removing 2597 references through the exclusion criteria.Additional studies without available ORs or RRs with 95% CI(n=23)nor with overlapping patients(n=4)or including a population affected with a particular disease(n=30)or cross-sectional study(n=4)were also removed.Therefore,a total of 24 articles with 19 case-control studies and 5 cohort studies involving 2122740 participants were enrolled in this meta-analysis.

Characteristics of the studies included

The specific clinical characteristic information of these 19 case-control studies included in the meta-analysis were shown in Table 1.These 19 studies were published from 2010 to 2019.Six of these 19 studies were initiated in Korea,2 in Australia,2 in China,the rest of studies in North Carolina,Oman,India,Greece,UK,Sri Lankan,Canada,Italy,as well as Siberia,involving 12755 cases and 23405 controls.The specific clinical characteristic information of the five cohort studies included in the meta-analysis were shown in Table 2.These 5 studies were published from 2011 to 2016.Five of these studies were initiated in Korea,US,UK,Iowa and North Bengal,involving 2086580 participants.We evaluated the quality of the included literature with NOS and found that 15 literatures(62.5%)were classified as high-quality literatures and 9 literatures(37.5%)were classified as quality literatures within the acceptable range.

In general

Meta-analysis of the 24 studies with a random-effects model demonstrated that an increased alcohol intake about 15g to 39.9g daily was related to a 35%additional risk of CRC(RR=1.35,95% CI:1.00-1.81),especially in the subgroup of cohort studies with a 21%excess risk of CRC(RR=1.21,95% CI:1.05-1.38).As shown in Figure 2,Significant heterogeneity was found through combing the results of these studies(Pheterogeneity=0.000,I2=97.8%)

Dose-response meta-analysis

The results of the subgroup analysis were performed in Table 3,between general alcohol consumption status(ever drinking or current drinking),levels of alcohol drinking(light drinkers;moderate drinkers;or heavy drinkers),types of alcohol(beer or wine),and frequency of alcohol drinking(<4/month;1-4/week;≥4/week).

In general alcohol consumption status subgroup,about ever or current drinking,eighteen eligible literatures provided the drinking data;Ever or current consumption of alcohol was related to no significantly increased risk of CRC(RR=1.18,95% CI:1.00-1.40 for general alcohol consumption status;RR=1.15,95%CI:0.91-1.45 for ever;RR=1.21,95% CI:0.97-1.51 for current).In levels of alcohol consumption subgroup,we found that an increase in heavy level of alcohol consumption was associated with a significantly increased risk of CRC(RR=1.90,95%CI:1.29-2.97),but the same association was not present for moderate level of alcohol consumption(RR=1.08,95% CI:0.93-1.26)and light level of alcohol consumption(RR=0.94,95% CI:0.69-1.27).In type of alcoholic beverage,a significantly increased risk of CRC was found in heavy beer subgroup(RR=1.48,95% CI:1.05-2.07);however,there was no significant association between alcohol consumption and the risk of colorectal cancer in moderate beer subgroup(RR=0.98,95% CI:0.72-1.34),moderate wine subgroup(RR=0.86,95% CI:0.56-1.34)and heavy wine subgroup(RR=1.04,95% CI:0.83-1.29).In frequency of alcohol drinking subgroup,six studies showed that there was no significant association between the frequency of alcohol drinking and the risk of CRC(RR=1.03,95% CI:0.92-1.14 for frequency of alcohol consuption;RR=0.92,95% CI:0.76-1.12 for <4/month;RR=1.14,95% CI:0.89-1.47 for 1-4/week;RR=1.04,95% CI:0.87-1.25 for≥4/week).

Table 3.Summary of the dose-response meta-analysis

Table 4 shows the results of subtype of sample size,type of analyses,follow up time(m),type of studies,type of study design,and stage.Through subgroup analysis,we found that the combined data did not indicate a significant correlation in the following subgroups:sample size(≤500;500-1000;≥1000),variable type(univariate analysis;multivariate analysis),follow up time(≤24;＞24),type of study(case-control;Cohort),study design(retrospective study;prospective study),and stage(Ne stage;all stage).Stratified analysis was carried out on the levels of alcohol consumption or type of alcoholic beverage and the design characteristics of the clinical study,we found that the stratified data of the former factor indicated more statistically different,whereas the factors of sample size,variable type,follow up time,type of study,study design and stage did not shown a statistically different.

Table 4.Summary of the meta-analysis results in subgroups

Meta-regression analysis

We also carried out the meta-regression analysis regarding alcohol consumption and CRC risk to find the impact of the eligible study feature on RRs of CRC.Taken sample size,variable type,follow up time,type of study,study design and stage as covariates,none of these covariates could not account for the observed heterogeneity in multivariable meta-regression analyses(type of study:p=0.978;sample size:p=0.262;variable type:p=0.277;follow up time:p=0.942;study design:p=0.84;stag:p=0.629).

Sensitivity analyses

We conducted a sensitivity analysis by eliminating each study one by one to determine whether the merger effect is stable and whether a study is the major source of heterogeneity.As shown in Figure 3,we did not found that any of the studies was a major source of heterogeneity affecting the stability of the synthetic effectors.The sensitivity analyses further confirmed that the results of the correlation between colorectal cancer risk and alcohol consumption obtained before was stable.

Publication bias

Begger’s funnel polt test was performed to evaluate publication bias between alcohol drinking and colorectal cancer risk.Publication bias was detected as shown in Figure 4(Pr >/z/=0.046 by Bgger’s test).

Discussion

A large number of studies and meta-analysis have demonstrated the correlation between alcohol consumption and the colorectal cancer risk while these correlations were not consistent and conclusive.Therefore,on the basis of reviewing the relevant literature published,we carried out a meta-analysis including more recent studies and conducting a more detailed stratification of alcohol consumption to estimate the association.Our comprehensive meta-analysis combined the outcomes of 2122740 participants from 24 individual studies,indicating that our results support a causal relationship between heavy alcohol consumption(consumption of≥40 g/day of ethanol)and the colorectal cancer risk(RR=1.90,95% CI:1.29-2.97,p=0.00).An increased alcohol intake of 15g to 39.9g daily was related to a 21%additional risk of CRC(RR=1.21,95% CI:1.05-1.38,p=0.043)in cohort studies.

With the development of economics and the improvement of people’s living conditions,the requirement of alcohol consumption become further increased.The exact mechanisms involved in the correlation between alcohol drinking and the occurrence,development and progression of colorectal cancer were still not fully researched.From the viewpoints of physiology and pathology,the molecular mechanisms of different pathways under different phenotypes can demonstrate these correlations.Heavy level of alcohol drinking reported with high-level IGF2 DMR0 hypomethylation involved in an early step during the process of colonic carcinogenesis[37].Alcohol may also interfere with anticarcinogenic related nutrents to suppress tumor immune surveillance and activate other procarcinogens.However,alcohol may also prevent the occurrence and development of cancer based NF-kB and activator AP-1 transcriptional regulation[6,7].Furthermore,long-term sustained alcohol intake may alter the expression levels of functional proteins whose functions including protein trafficking,inflammation,metabolism and regulating the development and progression of colorectal cancer[38].Therefore,based on the diversity of alcohol drinking and CRC related regulation mechanism,different levels of alcohol drinking may also related to appropriate levels of colorectal cancer risk.

During our current meta-analysis,we found that the correlation of colorectal cancer risk to heavy levels of drinking was opposite to light alcohol consumption,the heavy levels of drinking group had an additional higher risk(RR in heavy alcohol consumption vs.light alcohol consumption=1.90 vs.0.88,respectively).In a subgroup analysis of types of alcohol,we also found a positive correlation between beer consumption and colorectal cancer risk rather than wine consumption,but only the heavy beer subgroup had a statistically significant difference(RR=1.48,95% CI:1.05-2.07).Our data further confirmed the role of drinking consumption with different dose-linearly dependent in the development of colorectal cancer.And yet,there existed not only a limited number of stratified analysis of alcohol consumption but also a varies number of stratified characteristics,we should be cautious of the detected publication bias.Moreover,a published data showed that light drinking was related to a lower risk of CRC and very heavy drinking was related to a higher risk[39],which was similar to our findings in the light drinking and heavy drinking group.The main difference is that a J-shaped correlation of alcohol drinking to the risk of CRC by regression models was conducted in moderate levels of alcohol consumption.

Nevertheless,there still existed several limitations of our meta-analysis as follows:(1)the most important limitation of this study was the problem of the significant heterogeneity((Pheterogeneity=0.000,I2=97.8%for all eligible literatures with random-effects model;Pheterogeneity=0.043,I2=59.3% for the cohort studies in a random-effects model).Because of the limited number of studies on drinking stratification and the inconsistency of stratification characteristics of drinking stratification,we still could not find the major source of heterogeneity and demonstrated more accurate merging data even after sensitivity analysis and meta-regression.(2)Original data of alcohol intake was assessed only once,therefore,we could not completely avoid the measurement error.In addition,the inconsistency of stratification characteristics of drinking stratification may also generate measurement error,such as participants with different frequency units or different intake units,or different hierarchical definition.(3)The quality of the published data is relatively lower,all of the included studies which include the dose-stratified analysis of alcohol consumption were relatively small.(4)We did not conduct a supplementary search and there may have missed potential published data.(5)We did not included studies published in other language,so the publication bias may be unavoidable.In fact,our meta-analysis did detect a certain degree of publication bias.

In summary,the results from this meta-analysis revealed no remarkable effect on the correlation between alcohol drinking and CRC risk from the overall analysis;Heavy level of alcohol consumption,especially heavy beer intake seems to be associated with an increased the risk of CRC from the dose-response subgroup analysis.Lastly,it is clear that only uniformed stratified analysis and more and more original stratified studies can shed the light on how alcohol drinking affects the risk of CRC.