Cigarette smoking and colorectal cancer incidence and mortality: Systematic review and meta-analysis
Article first published online: 26 NOV 2008
Copyright © 2009 Wiley-Liss, Inc.
International Journal of Cancer
Volume 124, Issue 10, pages 2406–2415, 15 May 2009
How to Cite
Liang, P. S., Chen, T.-Y. and Giovannucci, E. (2009), Cigarette smoking and colorectal cancer incidence and mortality: Systematic review and meta-analysis. Int. J. Cancer, 124: 2406–2415. doi: 10.1002/ijc.24191
- Issue published online: 16 MAR 2009
- Article first published online: 26 NOV 2008
- Accepted manuscript online: 26 NOV 2008 12:00AM EST
- Manuscript Accepted: 14 NOV 2008
- Manuscript Received: 28 AUG 2008
- Harvard School of Public Health
- colorectal cancer;
The association between cigarette smoking and colorectal cancer (CRC) has been controversial. To synthesize the available data, we conducted a comprehensive meta-analysis of all prospective studies. A total of 36 studies were included in our meta-analysis. We examined the association between smoking and CRC, colon cancer and rectal cancer in terms of incidence and mortality. Separate analyses were conducted for smoking status, daily cigarette consumption, duration, pack-years and age of initiation. Relative to nonsmokers, current and former smokers had a significantly increased risk of CRC incidence and mortality, respectively. When CRC data were combined with colon/rectal cancer data, current smokers had a significantly increased risk of CRC incidence. All 4 dose–response variables examined—daily cigarette consumption (RR = 1.38 for an increase of 40 cigarettes/day), duration (RR = 1.20 for an increase of 40 years of duration), pack-years (RR = 1.51 for an increase of 60 pack-years) and age of initiation (RR = 0.96 for a delay of 10 years in smoking initiation)—were significantly associated with CRC incidence (all p-values < 0.0001). The relationship between duration of smoking and rectal cancer incidence was also significant. Among the subset of studies that distinguished cancer by site, a higher risk was seen for rectal cancer than for colon cancer for all analyses. Among prospective studies, a consistent association exists between smoking and CRC. The association is stronger for rectal cancer than for colon cancer in the subset of studies that differentiated cancer by site. © 2008 Wiley-Liss, Inc.
Colorectal cancer (CRC) is the third most common cancer among both men and women in the United States, with 154,000 new cases and 52,000 deaths anticipated in 2007.1 Worldwide, it is the fourth most commonly diagnosed malignancy, with an estimated 1,023,000 new cases and 529,000 deaths each year.2 Genetic risk factors for CRC are well-established, and modifiable risk factors such as physical activity, weight and diet have also been implicated in the development of CRC.3
Cigarette smoking has been shown to cause many nonpulmonary cancers, including those of the oral cavity, pancreas and kidney.4 However, the association between smoking and CRC is controversial. Most of the early studies examining this relationship did not find an association.5–15 In contrast, the majority of studies with follow-up after 1970 have supported the association.3, 16–29 It has been proposed that the reason for this discrepancy may be a 35- to 40-year lag time between exposure and disease, which would not be captured by earlier studies and studies with shorter follow-up.3 Nevertheless, smoking is currently not recognized as a risk factor for CRC by the International Agency for Research on Cancer (IARC) or the U.S. Surgeon General.
Recent meta-analysis studies suggest a positive association between smoking and colorectal adenomas and CRC. Botteri et al. conducted a comprehensive systematic review and found a significant association between smoking and the incidence of adenomatous polyps, which are recognized precursor lesions of CRC.30 The only published meta-analysis examining the relationship between smoking and CRC was conducted by Chen et al., which examined 14 case-control studies published in China and concluded that cigarette smoking was a significant risk factor for CRC.31 To date, a comprehensive meta-analysis of the published literature has not been conducted.
To further investigate the controversial relationship between cigarette smoking and CRC, we conducted a comprehensive systematic review and meta-analysis of all published prospective studies.
Material and methods
Two investigators (PL, TC) independently searched for relevant studies in the PubMed and EMbase databases. The search criteria used on Pubmed was (“Colorectal Neoplasms”[Mesh] OR “Colorectal Neoplasms, Hereditary Nonpolyposis”[Mesh]) AND (“Smoking”[Mesh] OR “Tobacco”[Mesh] OR “Tobacco Use Disorder”[Mesh]). The search criteria used on EMBase was (“colorectal cancer”/exp) AND (“tobacco”/exp OR “cigarette smoking”/exp). No limits were placed on the searches. References of select articles were also checked for additional relevant studies. The search results were compared and discrepancies were resolved by consensus. A total of 879 abstracts were screened, 175 papers were selected for further evaluation, 53 papers were chosen for data extraction and a final 36 papers were used in the meta-analysis (Fig. 1).
We included only prospective studies that examined the relationship between cigarette smoking and CRC. Case-control studies nested within prospective cohorts were included. For studies that published data from the same cohort, we chose only the most recent and complete report for analysis.
Studies on adenomatous polyps, carcinoma in situ, hereditary syndromes and inflammatory bowel disease were excluded. We also excluded review articles and studies without risk estimates or a variance measure such as confidence interval (CI) or standard error. Data on other forms of tobacco use, such as cigar and pipe, were not analyzed.
Two investigators (PL, TC) independently extracted the following information from all articles: author, year of publication, country of study, type of study, study year, mean follow-up, study population, number of subjects, adjustments used in analysis, smoking classification, risk estimate and variance. For studies that presented several risk estimates, only the most adjusted risk estimate was recorded. Adjusted variables for each study are shown in Table I. The extracted data were cross-checked and discrepancies were resolved by consensus.
|Ref.||Nation||Study time||Sex||Subjects||I/M||CR/C/R||Smoke status||Cigs/d||Smoke dur||Pk-yr||Init age||Adjusted|
|21||JPN||1990–97||M||21,695||I||CR||X||X||X||X||Age, alc, BMI, diet, edu, exercise, fam hx|
|50||JPN||1966–81||B||265,118||M||C/R||X||Age, occup, region, other|
|49||JPN||1968–87||B||61,505||I||C/R||X||Age, region, sex, other|
|16||USA||1982–96||B||781,351||M||CR||X||Age, alc, BMI, diet, edu, exercise, race|
|41||USA||1967–97||B||39,299||M||CR||X||X||Age, BMI, edu, race, other|
|3||USA||1986–92||M||47,935||I||CR||X||X||Age, alc, BMI, diet, fam hx|
|24||USA||1954–80||M||248,046||M||C/R||X||Age, SES, other|
|25||USA||1966–86||M||17,633||M||CR/C||X||X||X||Age, alc, region|
|43||USA||1980–04||F||104,519||M||CR||X||X||X||Age, alc, BMI, exercise, fam hx, other|
|51||KOR||1993–98||B||14,103||I||CR||X||X||X||X||X||Age, alc, BMI, sex, other|
|44||USA||1978–84||B||106,203||I||C/R||X||Age, alc, BMI, edu, race, sex, other|
|57||FIN||1966–94||B||56,973||I||CR/C/R||X||X||Age, BMI, occup, region, sex, other|
|45||USA||1986–99||F||34,467||I/M||CR||X||X||X||X||X||Age, alc, BMI, diet, exercise, other|
|58||NLD||1986–93||B||2,948||I||CR||X||X||X||X||Age, BMI, fam hx, sex|
|53||JPN||1990–99||B||90,004||I||CR/C/R||X||X||Age, alc, BMI, exercise, fam hx, region|
|46||USA||1991–05||F||146,877||I||CR/C/R||X||X1||X1||X1||Age, alc, exercise, diet, fam hx, race, other|
|20||GBR||1980–99||B||10,998||I||CR||X||Age, alc, sex|
|54||JPN||1992–00||B||29,051||I||C/R||X||Age, alc, BMI, edu|
|28||USA||1982–95||M||22,011||I||CR||X||X||X||Age, alc, BMI, diet, exercise, other|
|60||SWE||1961–97||B||17,118||I||CR/C/R||X||X||Age, BMI, exercise, sex|
|47||CAN||1980–93||F||89,835||I||CR/C/R||X||X1||X1||Age, alc, edu, exercise, other|
|22||NLD||1987–98||B||639||I||CR||X||X||X||Age, alc, BMI, region, sex, other|
|48||SGP||1993–04||B||61,321||I||CR/C/R||X||X||X||X||Age, alc, BMI, edu, exercise, fam hx, sex, other|
|55||JPN||1988–97||B||59,879||I||C/R||X||X||X||X||X||Age, alc, BMI, diet, edu, exercise, fam hx, region|
We separately analyzed the incidence and mortality of CRC, colon cancer and rectal cancer with respect to the following variables: smoking status (former vs. never smokers and current vs. never smokers), daily cigarette consumption among current smokers, duration of smoking, pack-years and age of initiation. For CRC, pooled risk estimates were calculated using data on CRC results only as well as CRC data combined with colon and rectal cancer data. In the latter analysis, colon and rectal cancer data were incorporated only if CRC data were not presented for the study. The effect measures of interest were relative risk (RR) and the associated 95% CI. All results were reported as RR, and odds ratios of nested case-control studies were taken as unbiased estimates of the RR. We used random-effects models32 to calculate all summary estimates.
Conventional methods were used to analyze the binary relationship between smoking status and CRC.33, 34 For studies that presented separate risk estimates by sex, both risk estimates were included in the pooled analysis. We tested for heterogeneity using both the H and I2 statistics, as proposed by Higgins and Thompson.35 Publication bias was assessed with the funnel plot, Begg's rank correlation test,36 and Egger's linear regression test.37
To analyze the relationship between smoking and dose–response variables (daily cigarette consumption, duration, pack-years, age of initiation), we used the generalized least squares for trend estimation method as described by Greenland and Longnecker.38 For analysis of daily cigarette consumption, duration and pack-years, never smokers were used as the referent group since they had the lowest exposure to cigarettes. Dose categories used in the pooled analysis were derived from the original dose intervals of each study. For dose intervals in which both the upper and lower bounds were given, the median value was used, e.g., 10–20 cigarettes/day was entered as 15 cigarettes/day. For dose intervals in which only the lower bound was given, the range was assumed to be equal to that of next lowest interval, and the median value was used. For example, if a study used 1–19 cigarettes/day and >20 cigarettes/day as dose intervals, the higher interval was entered as 30 cigarettes/day. For the analysis of age of initiation data, the exposure was the number of years before smoking initiation and the referent group with the lowest exposure was those who began smoking at the youngest age. Never smokers had the highest “exposure” since they have never smoked. Since all groups must be assigned a value for age of initiation, we used the study-specific baseline age as the age of initiation for never smokers. The RRs of all age groups were recalculated relative to the referent group in the age of initiation analysis.
Statistical significance was defined as a 2-sided p value of <0.05. Stata version 10.0 (StataCorp, College Station, TX) was used for all analysis.
Our meta-analysis includes 36 prospective studies with data from a total of 3,007,011 subjects. Of the 36 studies, 2 were nested case-control studies22, 39 and the other 34 were cohort studies. Sixteen studies were conducted in North America,3, 16, 17, 23–25, 28, 29, 40–47 9 in Asia21, 48–55 and 11 in Europe,18–20, 22, 39, 56–61 Four studies were published before 1980,17, 29, 44, 52 16 studies were published between 1990 and 2000,3, 16, 18, 19, 23–25, 28, 40, 42, 49, 50, 56, 57, 59, 61 and 16 studies were published after 2000.20–22, 39, 41, 43, 45–48, 51, 53–55, 58, 60 Twenty studies included both sexes, 8 studies included only female subjects18, 39, 40, 42, 43, 45–47 and 8 included only male subjects.3, 21, 23–25, 28, 52, 59 For outcome measures, 27 studies collected data on incidence, 8 collected data on mortality,16, 19, 24, 25, 41, 43, 50, 52 and 1 included both incidence and mortality.45 Table I shows the demographic variables and outcome characteristics of the included studies.
Colorectal cancer (CRC)
For current smokers, 18 risk estimates were pooled from 16 studies for incidence3, 18, 20–22, 28, 29, 42, 45–48, 51, 53, 57, 58 and 5 estimates were pooled from 4 studies for mortality16, 25, 43, 45 (Table II). Compared to never smokers, current smokers had a 17% (95% CI: 0.97–1.40) higher risk of developing CRC and a 40% (95% CI: 1.06–1.84) higher risk of CRC mortality. For former smokers, 18 risk estimates were pooled from 16 studies for incidence18, 20–22, 28, 29, 45–48, 51, 53, 57, 58, 60, 61 and 6 estimates were pooled from 5 studies for mortality.16, 25, 41, 43, 45 The RR for CRC incidence was 1.25 (95% CI: 1.04–1.51) and the RR for CRC mortality was 1.15 (95% CI: 0.90–1.48) in comparison to never smokers. The Forest plots in Figure 2 show overlap between the CIs of all studies from each smoking status category, which is a graphical demonstration of the absence of heterogeneity. All H statistics had associated p-values greater than 0.05 and all I2 statistics were 0, further providing evidence that there was no heterogeneity among CRC studies. There was no evidence of publication bias in any analysis by Begg's and Egger's tests (p > 0.05), however a nonsignificant bias (p = 0.096) toward positive studies was noted in the funnel plot for the analysis of CRC incidence among current smokers (Fig. 3). Single study sensitivity analysis did not change the statistical significance level of RRs except for the analysis of CRC mortality among current smokers. Excluding Chao et al.'s16 risk estimate for female current smokers from the analysis resulted in a decreased RR from 1.40 to 1.39 that was reduced to marginal significance (95% CI: 0.99–1.95).
|Smoking status||Outcome||Studies*||RR (95%CI)||H statistic p-value|
|Current smokers||Incidence||18 (16)||1.17 (0.97–1.40)||Non-significant|
|Incidence (combined1)||34 (22)||1.15 (1.00–1.32)||Non-significant|
|Mortality||5 (4)||1.40 (1.06–1.84)||Non-significant|
|Mortality (combined1)||15 (7)||1.27 (1.05–1.53)||Non-significant|
|Former smokers||Incidence||18 (16)||1.25 (1.04–1.51)||Non-significant|
|Incidence (combined1)||36 (22)||1.20 (1.04–1.38)||Non-significant|
|Mortality||6 (5)||1.15 (0.90–1.48)||Non-significant|
|Mortality (combined1)||11 (7)||1.20 (0.98–1.47)||Non-significant|
|Current smokers||Incidence||14 (11)||1.10 (0.89–1.36)||Non-significant|
|Mortality||6 (4)||1.13 (0.82–1.57)||Non-significant|
|Former smokers||Incidence||15 (13)||1.10 (0.90–1.35)||Non-significant|
|Mortality||4 (3)||1.27 (0.86–1.89)||Non-significant|
|Current smokers||Incidence||14 (11)||1.19 (0.94–1.52)||Non-significant|
|Mortality||5 (3)||1.23 (0.84–1.81)||Non-significant|
|Former smokers||Incidence||15 (13)||1.20 (0.94–1.53)||Non-significant|
We also examined the effect of combining data on colon cancer or rectal cancer only with the CRC data (Table II). Among current smokers, the combined RR for CRC incidence was 1.15 (95% CI: 1.00–1.32) and the combined RR for CRC mortality was 1.27 (95% CI: 1.05–1.53). Among former smokers, the combined RR was 1.20 (95% CI: 1.04–1.38) and 1.20 (95% CI: 0.98–1.47) for CRC incidence and mortality, respectively.
For current smokers, 14 risk estimates were pooled from 11 studies for incidence23, 40, 46–49, 53, 55–57, 59 and 6 estimates were pooled from 4 studies for mortality.19, 24, 25, 50 Compared to never smokers, current smokers had a nonsignificant 10% (95% CI: 0.89–1.36) higher risk of developing colon cancer and a 13% (95% CI: 0.82–1.57) higher risk of colon cancer mortality. For former smokers, 15 risk estimates were pooled from 13 studies for incidence23, 40, 44, 46–49, 53, 55–57, 59, 60 and 4 estimates were pooled from 3 studies for mortality.19, 24, 25 The RR for colon cancer incidence was 1.10 (95% CI: 0.90–1.35) and the RR for colon cancer mortality was 1.27 (95% CI: 0.86–1.89) in comparison to never smokers. There was no evidence of heterogeneity or publication bias in any of the analyses. Single study sensitivity analysis did not alter the significance level of any results.
For current smokers, 14 risk estimates were pooled from 11 studies for incidence23, 40, 46–49, 53, 55–57, 59 and 5 estimates were pooled from 3 studies for mortality.19, 24, 50 Compared to never smokers, current smokers have a nonsignificant 19% (95% CI: 0.94–1.52) higher risk of developing rectal cancer and a nonsignificant 23% (95% CI: 0.84–1.81) higher risk of rectal cancer mortality. For former smokers, 15 risk estimates were pooled from 13 studies in the analysis of rectal cancer incidence.23, 40, 44, 46–49, 53, 55–57, 59, 60 Analysis of mortality among former smokers was not performed, because data were only available from 2 studies. The RR for rectal cancer incidence was 1.20 (95% CI: 0.94–1.53) in comparison to never smokers. There was no evidence of heterogeneity or publication bias and no single study changed the significance of the pooled RR on sensitivity analysis.
Daily cigarette consumption
Eleven studies were included in the dose–response analysis of daily cigarette consumption and CRC incidence,21, 22, 28, 45, 46, 48, 51, 57, 58, 60 4 studies were included in the analysis of CRC mortality25, 41, 43, 45 and the same 6 studies each were included in the analysis of both colon cancer incidence and rectal cancer incidence.40, 48, 55, 57, 59, 60 There was a significant increased risk of CRC incidence and mortality (p < 0.0001) with increased daily cigarette consumption (Table III). An increase of 20 cigarettes/day (1 pack/day) led to a 17.5% increase in RR for CRC incidence and a 40.7% increase in RR for CRC mortality. An increase of 40 cigarettes/day (2 packs/day) led to a 38.0% increase in RR for CRC incidence and a 98.0% increase in RR for CRC mortality. An increase in daily cigarette consumption of 20 cigarettes was associated with a 3.0% increase in RR (p = 0.611) for colon cancer and a 13.1% increase in RR (p = 0.068) for rectal cancer, but these results were not significant. When data on colon or rectal cancer only were combined with CRC data, the increase in RR associated with an additional 20 cigarettes per day was 13.7% for CRC incidence and 41.0% for CRC mortality (both p < 0.0001).
|Variable||Outcome||Studies||RR of moderate exposure1||RR of high exposure2||p value|
|Incidence||6||0.978 (0.022)||0.956 (0.044)||<0.0001|
|Incidence (combined)*||7||0.979 (0.021)||0.958 (0.042)||<0.0001|
Duration of smoking
Eight studies were included in the dose–response analysis for duration of smoking and CRC incidences21, 22, 45–48, 51, 58 and the same 3 studies were included in both the analysis of colon and rectal cancer incidence.48, 55, 59 For CRC incidence, a 20-year increase in duration of smoking was associated with a 9.4% increase in RR and a 40-year increase in duration was associated with a 19.7% increase in RR (Table III). These results were highly significant (p < 0.0001). Smoking duration was also significantly associated with rectal cancer incidence, with a 20-year increase in duration being associated with a 13.5% increase in RR (p = 0.0004). A statistically nonsignificant 2.9% reduction in RR for colon cancer incidence was observed for every 20 years of smoking duration (p = 0.458). The combined CRC and colon/rectal cancer analysis showed an 8.7% increase in RR for CRC incidence with a 20-cigarette per day increase in consumption (p < 0.0001).
Five studies were included in the dose–response analysis of pack-years of cigarette smoking and CRC incidence3, 28, 45, 51, 53 and the same 4 studies were included in the analysis of colon and rectal cancer incidence.23, 53–55 With additional pack-years of smoking, a statistically significant increase in RR was observed for CRC incidence but not for colon cancer incidence or rectal cancer incidence (Table III). The associated increases in RR for CRC incidence were 26.9% for an increase of 35 pack-years and 50.5% for an increase of 60 pack-years (p < 0.0001). A 35 pack-year increase was associated with an increase in RR of 2.2% for colon cancer incidence (p = 0.99) and 18.4% for rectal cancer incidence (p = 0.07). Results of the combined CRC and colon/rectal cancer data showed that an increase in 35 pack-years of exposure was associated with a 23.9% increase in RR for CRC incidence (p < 0.0001).
Age of initiation
Six studies were included in the dose–response analysis of age of initiation of smoking and CRC incidence.21, 45, 46, 48, 51, 58 A 5-year delay in smoking initiation was associated with a 2.2% reduction in RR for CRC incidence, and a 10-year delay was associated with a 4.4% reduction in RR. These associations were highly significant (p < 0.0001, Table III). Analysis of colon cancer and rectal cancer incidence data alone was not performed due to the small number of available studies. When the colon/rectal cancer only data were combined with CRC data, a 5-year delay in smoking initiation was associated with a 2.1% reduction in RR for CRC incidence (p < 0.0001).
Smoking and CRC incidence and mortality
Our results indicate that both past and current smokers have an increased risk of CRC incidence and mortality. Significantly increased risk was found for current smokers in terms of mortality (RR = 1.40), former smokers in terms of incidence (RR = 1.25) and current smokers in terms of incidence when CRC data were combined with colon/rectal cancer data (RR = 1.15). The RR of CRC mortality in former smokers was not significant with CRC data alone, but when combined with colon/rectal cancer data the RR increased from 1.15 to 1.20, which was marginally significant (95% CI: 0.98–1.47). Adding colon/rectal data to CRC data resulted in narrower CIs for all smoking status categories and made the RR for incidence among current smokers significant.
The risk of CRC mortality was higher for current smokers than for former smokers. Continuous smoking increased cancer mortality, which can be explained by biological causes, potential confounders or a combination thereof. Cigarette smoking has been shown to facilitate tumor growth by induction of angiogenesis and/or suppression of cell-mediated immunity.62 Smoking has also been shown to be an independent risk factor for increased mortality among operable CRC patients.63 Another proposed theory is poorer response among smokers to cancer treatment.64 Nonetheless, one should be cautious about possible confounders, including a difference in health behavior between smokers and nonsmokers and competing illnesses in smokers. Smokers may be less likely to receive preventive and curative care, and may have a higher incidence of comorbid conditions such as cardiovascular disease.
In our study, current smokers also had a higher risk of CRC mortality relative to incidence. This can be explained by a more advanced stage of CRC at diagnosis among smokers,65–67 which could be due to more rapid progression of CRC and less health-conscious behavior among smokers. Although the introduction of CRC screening leads to earlier detection and decreased mortality from CRC,68 from a public health perspective it is still important to further decrease cancer mortality among CRC patients by life-style modification.
In addition to finding an association between binary smoking status and CRC, we also found an association between smoking and CRC in a dose-dependent manner. All 4 dose–response variables we examined—daily cigarette consumption (RR = 1.38 for an increase of 40 cigarettes/day), duration (RR = 1.20 for an increase of 40 years of duration), pack-years (RR = 1.51 for an increase of 60 pack-years) and age of initiation (RR = 0.96 for a delay of 10 years in smoking initiation)—were significantly associated with CRC incidence (all p-values < 0.0001). The association between daily cigarette consumption and CRC mortality, which was the only dose–response analysis of CRC mortality, was also significant (RR = 1.98 for an increase of 40 cigarettes/day, p < 0.0001). When colon/rectal cancer only data were combined with CRC data, all associations remained significant (p < 0.0001). Associations between CRC mortality and other dose–response variables were not analyzed due to an insufficient number of studies.
Smoking and colon and rectal cancer
Our primary analyses are for CRC rather than subsite-specific for several reasons.
CRC is still conceptualized as a single disease by many individuals and organizations, and because organizations such as IARC still do not recognize a causative relationship between smoking and CRC, establishing a relationship between smoking and the broader category of CRC is an important step. In addition, because only a relatively small portion of studies differentiated colon and rectal cancer, CRC is the category that is both most comprehensive and least likely to be influenced by publication bias. Nonetheless, since colon and rectal cancer are likely to have some etiologic differences, we additionally considered them separately. Only about half the studies presented information separately for these subsites, and only 7 studies provided data on CRC, colon cancer and rectal cancer.39, 46–48, 53, 57, 60 For current and past smoking, based on 14 to 15 studies of the 36 total, results for neither rectal nor colon cancer were significant, although the magnitude of the RR's were about twice as strong for rectal cancer as for colon cancer (Table II). Colon cancer incidence was not significantly associated with any of the 3 dose–response variables analyzed, whereas rectal cancer incidence was significantly associated with longer duration of smoking alone although the number of studies in these comparisons was small (Table III).
Since the same studies were used for both colon and rectal cancer incidence analyses, this suggests a genuinely higher RR for rectal cancer incidence than for colon cancer incidence. However, to investigate the possibility that the discrepancy in CRC and colon/rectal cancer results arises from differences in individual study characteristics, we compared the following characteristics of studies that were not included in both CRC and colon/rectal cancer analyses: publication year, mean follow-up, sex, number of adjusted factors, country of study and baseline age. It did not appear that we can attribute the discrepancy in results to any consistent underlying differences in the characteristics of the individual studies. Our results are consistent with the finding that smoking has been associated with a higher RR for rectal cancer than colon cancer in most studies in the literature,16, 19, 23, 24, 42, 69, 70 and in some studies the association was significant only for rectal cancer but not for colon cancer.6, 40, 59, 71 One recent study that examined possible differences between colon cancer and rectal cancer found clinical and prognostic differences, but did not discover differences in tumor marker expression or histological characteristics.72
Although our data suggest that a stronger association is likely for rectal cancer than for colon cancer, insufficient evidence exists to declare a lack of association between smoking and colon cancer. Although we did not find significant trends for any of the dose variables, only 3 to 6 of the 36 studies provided information in dose–response variables by subsite. Our analysis shows a similar effect of smoking on CRC and rectal cancer, which in of itself strongly suggests the presence of publication bias among studies that presented subsite-specific data. Rectal cancers comprise only a minority of total CRC (∼30%), so if the association between smoking and CRC were entirely due to rectal cancer, one might expect much stronger associations for smoking and risk of rectal cancer. This suggests that the increase in risk of CRC is at least partly driven by an increase in colon cancer. Moreover, smoking has been associated with both colon and rectal adenoma.30 Although the data do support a stronger association between smoking and rectal cancer than between smoking and colon cancer, it would be premature to conclude that the increased risk for CRC can be attributed entirely to rectal cancer.
A major limitation of our study, and of all meta-analysis studies, was our limited access to the primary data when calculating summary risk estimates. This issue was most apparent for the data on colon and rectal cancer separately, and on dose–response variables (see Table I). Further, we were unable to correct for the potential confounders present in individual studies. Confounding was assessed variably across the studies, though most of the studies and in particular the more recent studies assessed and evaluated potential confounding by diet and alcohol, and we used the most fully adjusted RR in our analyses. No study reported confounding by any individual factor that was strong enough to eliminate any apparent association between smoking and CRC. In many studies, smoking status was recorded at baseline and did not account for changes in smoking habits over the course of follow-up, and therefore misclassification and information bias are likely to exist in the pooled results. Studies did not generally separate distal and proximal colon cancers, which could have different etiologies.
An additional limitation of our meta-analysis was our inability to include studies that did not present data on both risk estimates and measures of variance or those that presented data using unconventional classifications, e.g., ever smokers instead of current and former smokers for smoking status. Consequently, our meta-analysis excluded some large studies, notably Doll's British physicians studies,5, 73 Hammond and Horn's 9-state study9 and Hammond's study of 1 million US men and women.74 The studies by Hammond and Hammond and Horn are among the first to be published on this topic, and although they both amassed a substantial number of person-years of follow-up, this was due to large cohorts with relatively short follow-up times of 4 and 3.7 years, respectively. Given the proposed lag time of at least 35 years between smoking and CRC, negative results in these early publications with short follow-up time is not surprising. In the most recent update of Doll's study of male British physicians in 2005, the authors found a significant association between smoking and CRC that was driven by a highly significant association with rectal cancer despite a null association with colon cancer. The finding was a reversal of earlier nonsignificant results from the same cohort,6, 69 however, it continued a trend of progressively more significant risk estimates for rectal cancer over time. This provides further support for the lag-time hypothesis for smoking and CRC.
Another important limitation is that besides stratifying by colon and rectal cancer, we could not characterize the CRCs into biologically relevant subtypes. Recent evidence has supported that smoking history may be a particularly important risk factor for the subtype of CRC characterized by microsatellite instability,75, 76 and by CpG island methylator phenoptype (CIMP) status and BRAF mutations.77 Also, smoking has been associated with an increased risk of hyperplastic polyps,78 particularly when they coexist with adenomas.79–81 Interestingly, hyperplastic polyps have been linked to BRAF mutations and CIMP positivity.82 These data suggest that smoking may be associated primarily with a subtype of CRC, characterized by a pathway involving hyperplastic (serrated) polyps, BRAF mutations and CIMP positive status. Future studies focused on such subtypes of CRC may yield much stronger associations with smoking compared to studies pooling all subtypes of CRC.
We conducted a comprehensive systematic review and meta-analysis of prospective studies examining the relationship between cigarette smoking and CRC incidence and mortality. Our findings provide strong evidence that smoking is associated with an increased risk of CRC, which was observed consistently in analysis of both current and past smokers, daily cigarette consumption, duration, pack-years and age of initiation. We also found a consistently higher risk for rectal cancer than for colon cancer across all smoking variables, which suggests a different qualitative or quantitative relationship between smoking and risk of cancer development between the 2 anatomic sites. The studies encompassed diverse populations in North America, Europe and Asia, and no significant heterogeneity was observed.
The relationship between smoking and CRC has been controversial probably due to initial studies that did not show an association. The reason for the initial null results had been postulated to an effect of smoking primarily early in carcinogenesis, which required a sufficiently long follow-up time to demonstrate an association. In this meta-analysis, an increased risk was slightly stronger for former smokers than for current smokers, suggesting that the effect of smoking persists irreversibly or at least for many years, even after cessation. Our results also show that both a relatively high intensity of smoking and a long duration, also reflected in total pack-years, is required to demonstrate appreciably elevated RRs. Of note, the magnitude of the elevated RRs are still weaker than those observed for adenomatous and hyperplastic polyps. For example, we estimated that a 60 pack-year history of CRC was associated with a RR of 1.45 for CRC; in contrast, based on the meta-analysis of colorectal adenomas by Botteri et al.,30 a 60 pack-year history would be associated with a RR of 2.08. The stronger association for adenomas than for cancer may be contributed in part by the longer induction period required for cancer, and perhaps because smoking appears to influence a subtype of adenoma, and different subtypes may have varying rates and likelihoods of progressing to cancer. Our results, including other evidence such as a consistent association between smoking and colorectal adenomas, the emerging evidence of a stronger association with a unique subtype of CRC, and the strong biological plausibility given that tobacco provides numerous proven carcinogens through the circulatory system or direct ingestion, provide strong support of a causal association between cigarette smoking and CRC.
The authors thank Dr. Eric Ding and Dr. Rob van Dam of the Harvard School of Public Health for their assistance with the methodology. This work was supported by Dr. Giovannucci's research and teaching funding at the Harvard School of Public Health. The authors have no potential conflicts of interest to report.
- 2GLOBOCAN 2002: cancer incidence, mortality and prevalence worldwideed. Lyon: IARC Press, 2004., , , .
- 4IARC Working Group on the Evaluation of Carcinogenic Risks to Humans. Tobacco smoke and involuntary smoking. IARC Monogr Eval Carcinog Risks Hum 2004; 83: 1–1438.
- 81Hyperplastic polyposis in the New Zealand population: a condition associated with increased colorectal cancer risk and European ancestry. N Z Med J 2007; 120: U2827., , , , .