Cigarette Smoking Associated with Colorectal Cancer Survival: A Nationwide, Population-Based Cohort Study

We investigate whether cigarette smoking is associated with survival in patients with colorectal cancer (CRC) through a nationwide population-based cohort study in Taiwan. The Taiwan Cancer Registry and National Health Insurance Research Database were used to identify data from patients with CRC from 2011 to 2017. Tobacco use was evaluated based on the smoking status, intensity, and duration before cancer diagnosis. A total of 18,816 patients was included. A Kaplan–Meier survival analysis indicated smoking to be significantly associated with the CRC mortality risk (log-rank p = 0.0001). A multivariable Cox model indicated that smoking patients had a 1.11-fold higher mortality risk (HR = 1.11, 95% CI = 1.05–1.19) than nonsmoking patients did. This increased risk was also present in patients with CRC who smoked 11–20 cigarettes per day (HR = 1.16; 95% CI = 1.07–1.26) or smoked for >30 years (HR = 1.14; 95% CI = 1.04–1.25). Stratified analyses of sex and cancer subsites indicated that the effects of smoking were higher in male patients and in those with colon cancer. Our results indicate that cigarette smoking is significantly associated with poor survival in patients with CRC. An integrated smoking cessation campaign is warranted to prevent CRC mortality.


Introduction
Colorectal cancer (CRC) is among the most common cancers and among the leading causes of cancer deaths worldwide [1][2][3][4]. Although it has historically been more prevalent in the West, the incidence rates of CRC have been increasing in East Asian countries [5]. In Taiwan, CRC is one of the most commonly diagnosed cancers [6]. Despite the progress that has been achieved in its diagnosis and treatment, approximately half of patients with CRC die within 5 years of diagnosis [4]. Therefore, further efforts to identify and obviate the risk factors of CRC mortality are required to improve the prognosis of this cancer.
Cigarette smoking is a serious public health concern; it is annually responsible for millions of deaths around the world [7]. Smoking is estimated to be responsible for more than 30% of cancer deaths in the United States each year. Smoking has also been observed to increase the risk of mortality in CRC [8]. The association between smoking and CRC Data in this study were collected from the Taiwan Cancer Registry (TCR) and Taiwan's National Health Insurance Research Database (NHIRD). Both of these databases are managed by the Health and Welfare Data Science Center (HWDC) of the Ministry of Health and Welfare. The TCR was established to gather information on individual demographics, cancer stages (AJCC 7th edition), primary sites, histology, and treatment types in patients with cancer to understand the incidence and mortality rates of cancer in Taiwan. The NHIRD was established for research purposes; it contains data from Taiwan's single-payer insurance system, in which more than 99% of Taiwan's 23 million citizens are registered. The NHIRD contains registration files and original inpatient and outpatient reimbursement claim data from 1996 to 2017. The datasets of the HWDC are all de-identified forms. This study was conducted in compliance with the Declaration of Helsinki of 1964 and was approved by the Research Ethics Committee of Chi Mei Hospital (IRB no. 10702-E04). The requirement for informed consent was waived by the Research Ethics Committee of Chi Mei Hospital.

Study Population
The TCR was used to identify patients with CRC based on the International Classification of Diseases for Oncology, third edition (ICD-O-3); in this study, colon (ICD-O-3: C18), rectosigmoid junction (ICD-O-3: C19), and rectum (ICD-O-3: C20) cancers were included. Because the TCR began recording smoking and drinking behavioral information in 2011, patient data from 2011 to 2017 were selected. Patients with a history of CRC before 2011 were excluded to reduce omitted variable bias. In addition, because the aim of this study was to estimate the association between cigarette smoking and risk of mortality in patients with CRC, included patients were categorized as those with and without a history of prediagnostic smoking. Those with a smoking history included both current and ever smokers, for whom the duration of smoking in years and smoking count per day were included in the analysis. To reduce the potential confounding factors of mortality, including age, gender, clinical stage, grade, and cancer subsite, between patients with smoking and those without, we randomly selected two patients without smoking to match each patient with smoking using propensity score approach. A propensity score matching approach with the nearest-neighbor matching algorithm was used in this study according to SAS macro "%OneToManyMTCH". The flowchart of the study population selection is presented in Figure 1. smokers, for whom the duration of smoking in years and smoking count per day were included in the analysis. To reduce the potential confounding factors of mortality, including age, gender, clinical stage, grade, and cancer subsite, between patients with smoking and those without, we randomly selected two patients without smoking to match each patient with smoking using propensity score approach. A propensity score matching approach with the nearest-neighbor matching algorithm was used in this study according to SAS macro "%OneToManyMTCH". The flowchart of the study population selection is presented in Figure 1.

Measurements
The major outcome of this study was overall mortality. Mortality was defined using Taiwan's cause-of-death data. All patients were right censored to date of death or 31 December 2017, whichever came first. The study variables, namely, age at diagnosis, sex, clinical stage, histological grade, and alcohol drinking habit, were all collected from the TCR. Age was divided into groups of <40, 40-49, 50-59, 60-69, and ≥70 years. Charlson comorbidity index (CCI) scores were calculated using patients' diagnosis records from the NHIRD to represent patients' comorbidities, which were defined before the date of diagnosis of CRC. To generate the index score, each of the 19 identified medical conditions was scored from 1 to 6.

Statistical Analysis
The frequency was presented as a percentage for categorical variables among the study population. The distribution difference between smoking and nonsmoking groups was compared using Pearson's chi-square test. The trend of mortality during the study

Measurements
The major outcome of this study was overall mortality. Mortality was defined using Taiwan's cause-of-death data. All patients were right censored to date of death or 31 December 2017, whichever came first. The study variables, namely, age at diagnosis, sex, clinical stage, histological grade, and alcohol drinking habit, were all collected from the TCR. Age was divided into groups of <40, 40-49, 50-59, 60-69, and ≥70 years. Charlson comorbidity index (CCI) scores were calculated using patients' diagnosis records from the NHIRD to represent patients' comorbidities, which were defined before the date of diagnosis of CRC. To generate the index score, each of the 19 identified medical conditions was scored from 1 to 6.

Statistical Analysis
The frequency was presented as a percentage for categorical variables among the study population. The distribution difference between smoking and nonsmoking groups was compared using Pearson's chi-square test. The trend of mortality during the study period was plotted using the Kaplan-Meier approach, with a log-rank test for estimating the statistical difference between smoking and nonsmoking patients with CRC. Multivariable Cox proportional regression was constructed to estimate the mortality risk and control for potential confounders by adjusting for age, sex, drinking habit, residence in a remote area, cancer site, cancer clinical stage, cancer grade, and CCI group. Stratified analyses of age, sex, and CRC subsites were also presented. To observe the progress of mortality risk on smoking counts per day and smoking years, the linear trend test was used to estimate the potential trends. All analyses were conducted using SAS statistical software version 9.4 (SAS Institute, Cary, NC, USA). Significance was set at p < 0.05. Kaplan-Meier curves were plotted using STATA (version 12; Stata, College Station, TX, USA).

Characteristics of Study Population
The baseline characteristics of the matched cohort are presented in Table 1. Of the 18,816 patients with CRC included in this study, 6272 were smokers and 12,544 were not. The smoking group comprised more patients with drinking habits (52.1% vs. 13.8%, p < 0.0001). In addition, the mortality rate was significantly higher in the smoking group (30.1% vs. 27.9%, p = 0.0012). Otherwise, the two groups were balanced with regard to age, sex, residence in remote area, cancer subsite, clinical stage, tumor grade, and CCI grouping.

Cigarette Smoking and Mortality Risk Stratified by Sex
The data presented in Table 3 revealed the risk of mortality associated with smoking in patients with CRC stratified by sex. The proportions of male and female smokers were the same (33.3%), but the male smoking population was 9.47 times of the female. Smoking men had a 1.09-fold higher mortality risk (95% CI = 1.02-1.18; p = 0.0156) than nonsmoking men did. A further analysis revealed a significantly higher risk of mortality in men who smoked 11-20 cigarettes per day (HR = 1.16; 95% CI = 1.06-1.26; p = 0.0011) and who smoked for more than 10 years (HR = 1.11; 95% CI = 1.01-1.24; p = 0.0390 for men smoking for 11-30 years; HR = 1.14; 95% CI = 1.04-1.25; p = 0.0076 for those smoking for >30 years) than in those who were nonsmokers. For women, a 1.59-fold higher mortality risk was observed only in those who smoked for 1-10 years (95% CI = 1.03-2.45; p = 0.0367) compared with nonsmoking women. However, the increase in risk was not significant in the other levels of smoking intensity.

Cigarette Smoking and Mortality Risk Stratified by Age
The effects of smoking on the mortality risk were assessed with the stratification of the patients' ages (Table 5). For patients younger than 60, an increased CRC mortality risk was not associated with smoking, with the exception of those who smoked for more than 30 years (HR = 1.43; 95% CI = 1.10-1.87; p = 0.0074). By contrast, smoking significantly increased the risk of mortality in patients with CRC who were older than 60 (HR = 1.12; 95% CI = 1.03-1.21; p = 0.0061). Higher risks were observed in such patients who smoked 11-20 cigarettes per day (HR = 1.16; 95% CI = 1.05-1.27; p = 0.0034) and who smoked for more than 10 years (HR = 1.15; 95% CI = 1.02-1.30; p = 0.0190 for patients who smoked for 11-30 years; HR = 1.12; 95% CI = 1.02-1.24; p = 0.0209 for those who smoked for more than 30 years).

Discussion
Through the combined analysis of data from nationwide health insurance and cancer registries, we demonstrated that cigarette smoking was associated with a significantly increased risk of mortality in patients with CRC. The increased risk was more prominent in patients with higher levels of smoking intensity and duration. This pattern was especially present in men, patients with colon cancer, and patients older than 60. A dose-response effect on the risk of mortality was also observed with smoking duration in the whole population and with smoking amount in patients with colon cancer. Although the increase in the mortality risk was moderate in most of the analyzed categories, its significance merits further consideration to improve the prognosis of CRC.
Despite widespread skepticism towards an association between cigarette smoking and CRC, accumulating evidence has suggested an increased risk of incidence incurred with smoking [5]. In the Iowa Women's Health Study (IWHS), ever smokers had a moderately increased CRC risk (RR of approximately 1.20) compared with never smokers [11,27]. In the Cancer Prevention Study II Nutrition Cohort, the incidence of CRC was approximately 30% higher in current smokers than in never smokers [18]. Two meta-analyses demonstrated that the pooled RR increased from 15% to 20% in ever smokers compared with never smokers [1,16].
Furthermore, cigarette smoking was associated with an increased risk of mortality in patients with CRC. In the IWHS, ever smokers had an increased risk of overall mortality (RR = 1.31) compared with never smokers, which was similar to observations of CRC incidence [11]. In a previous meta-analysis, current smokers exhibited a significantly higher risk of CRC mortality (RR = 1.58) compared with nonsmokers [16]. In addition to allcause mortality, disease-specific mortality was affected by current smoking [23,24]. Further evidence of this was presented in a meta-analysis that demonstrated that smokers had a 26% higher risk of all-cause mortality than never smokers did. Notably, 30-day mortality was reported to be higher by between 49% and 100% [4]. Compared with these previous studies, the mortality risk increased only moderately, though significantly, in smokers in our study. Factors such as components in cigarettes or differences in the study population might be implicated.
Our results indicated an increased risk of mortality in patients who smoked more than 10 cigarettes per day and who smoked for more than 10 years. This amount and the duration were much lower than previously reported. In the Chicago Heart Association cohort, a significant association between smoking and an increased CRC mortality was observed only in patients who smoked more than 20 cigarettes per day [19]. Smoking more than 15 cigarettes per day and having a 20-pack-year history were reported by Walter et al. to affect CRC survival [4]. In addition to differences in the characteristics of cigarettes and study populations, the lower threshold of smoking intensity for an increasing CRC mortality risk in our study may be attributable to the numerous events that may increase the detectability of differences in risks.
The mechanisms underlying the association between smoking and CRC mortality are multifold and incompletely understood. Cigarette smoke contains more than 60 carcinogens [7,17,29]. Of them, nicotine and 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone may enhance metastasis, which is the leading cause of death in patients with CRC, by enhancing cell migration and epithelial-mesenchymal transformation [6,30,31]. Nicotine may also interfere with the antiproliferative and proapoptotic effects of chemotherapeutic agents [4,32,33]. Tobacco smoking may cause a mutation of the GSTM1 gene, resulting in the impaired detoxification of tobacco carcinogens and enhancement of carcinogens' tumorigenic actions. Smoking may also induce aberrant promoter DNA methylation and silence regulatory genes in tumor progression. Consequent genetic alterations, such as a high microsatellite instability (MSI), the CpG island methylator phenotype, and the BRAF V600E mutation, may impair patient survival [13].
In previous studies, risk factors associated with an increased CRC mortality in smokers included an active smoking status, increased smoking amount or duration, and younger age at initiation [4,9,13,18]. The effects of smoking were also more significant in patients younger than 50 [19,24]. Notably, in our study, smoking patients younger than 60 had a lower risk of mortality. An exception to this was the increased risk in those who smoked for more than 30 years, indicating effects of prolonged smoking duration and younger age at initiation. By contrast, a generally significant association was noted between smoking and an increased risk of mortality in patients older than 60. This contradicted the observations by Colangelo et al. that the association between CRC and smoking mortality was significant only in those younger than 50 [19]. Whether this variance can be explained by other unanalyzed factors, such as genetic alterations, requires further elucidation.
For the effect of sex, Colangelo et al. reported that the risk for CRC mortality was higher for women than for men at the same level of smoking exposure, a phenomenon similar to that observed in patients with lung cancer [19]. However, other studies have reported discordant results. The association between smoking and the risk of CRC mortality was higher in men in a study conducted in Canada [13]; the association was even greater in patients older than 60. Walter et al. and Phipps et al. also reported a greater risk of recurrence or mortality in male smokers [4,24]. In our study, the increased risk in male smokers remained significant and relatively constant at most levels of smoking amount and duration. By contrast, the increases in risk in female smokers were mostly nonsignificant. Although our results may suggest a higher CRC mortality risk in male smokers, the disproportionately low number of female smokers when compared with that of male smokers may attenuate the association in females [4,9,11,18,19,21].
Tumor-related factors associated with an increased risks of recurrence or mortality in smokers include a T3 tumor, one to three affected lymph nodes, nonmetastatic diseases, a mutated KRAS status, and a wild-type BRAF status [4,24]. The effect of MSI remains under debate; some studies have suggested that the associations of smoking with all-cause CRC mortality were higher among patients with microsatellite-stable or MSI-low tumors, whereas others have reported a similar association with MSI-high tumors [4,13,24]. Data regarding genetic analyses of CRC specimens were not available in our database, which precludes the further exploration of the mechanisms underlying the association between cigarette smoking and an increased risk of CRC mortality.
Clinically, cigarette smoking is associated with later stages of CRC at diagnosis, which leads to a poorer prognosis and survival [34]. However, the increased risk persisted in our study despite matching for cancer stage. In addition to a proneoplastic effect, tobacco smoking constitutes a primary risk factor for cardiovascular and pulmonary diseases [32]. Therefore, smoking patients with CRC may incur additional risk or mortality from these causes. A prolonged induction period of more than 35 years is required for smoking to increase the risk of incident CRC [10,11,14,20]. The shorter duration of smoking associated with an increased CRC mortality risk in our study may support the involvement of smokinginduced comorbidities.
Studies have indicated that colon and rectal cancer may have partly different etiologic pathways and should be considered to be two separate entities that differ in susceptibility to carcinogens [11,17]. However, no consensus has been reached regarding whether the risk of incident colon or rectal cancer is more strongly associated with smoking [1,[10][11][12][16][17][18]20,21,27,[35][36][37][38][39]. Similarly, the association between smoking and CRC subsite mortality has been the topic of debate; several studies have reported that smoking was more significantly associated with a worse survival in patients with colon cancer than in those with rectal cancer [4,13,24]. However, others have reported a similar association between smoking and colon and rectal cancer mortality [9]. In our study, smoking was associated with an increased risk of mortality in patients with colon but not rectal cancers, which implies a higher susceptibility in patients with colon cancer. A significant doseresponse relationship also supports the stronger association between smoking and colon cancer mortality.
The most pronounced advantage of our study was the sample size of the nationwide population-based cohort study. The number of events regarding the association between smoking and CRC mortality in our study was greater than those in previous studies [4,9,11,13,19]. We chose ever smoking as the main exposure risk because the effects of smoking may persist after changes in smoking behavior [4,39]. Studies have also reported a similar CRC risk for former and current smokers [15,21]. To obviate the influence of potential confounders, we performed propensity score matching to generate the study cohort. We further adjusted for factors, such as the alcohol consumption and body mass index, because they have been closely associated with both smoking and cancer risk [5,9,10]. The results from this and other studies demonstrate that the effects of smoking in CRC were much smaller than those in cancers of the respiratory and upper gastrointestinal tracts [16]. Nevertheless, quitting smoking for at least 20 years may still significantly reduce the risk of CRC incidence and adverse outcomes, suggesting smoking as a potentially modifiable risk factor of CRC prognosis [4,9,13,18].
It might seem peculiar that patients in the highest smoking amount category did not always have an increased risk of mortality in this study, including various subgroup analyses, as did patients in the highest smoking duration category. This might be attributable to the lower thresholds of smoking duration and amount required to increase the risk of CRC mortality in our study population. Furthermore, we speculate that the differential relationship between smoking intensity or smoking duration and CRC survival may also contribute to this phenomenon, that is, a threshold relationship for smoking intensity and a dose-response relationship for smoking duration. The relatively small number of patients in the highest smoking amount category in the whole cohort and in subgroups might also be accountable. Lastly, it might be postulated that the impact of the smoking duration on the risk of CRC mortality outweighs that of smoking intensity in our study population.
The limitation of this study stemmed largely from the use of administrative databases. The self-reported and retrospective collection of information on smoking and other variables was prone to recall and reporting biases. Data regarding family history of CRC, dietary information, physical activity, CRC screening, and use of cyclooxygenase inhibitors were not comprehensively recorded. Most importantly, other measures of smoking behavior, such as age at initiation, cumulative cigarette pack years, and passive smoking, were not collected. These factors may compromise the accuracy of the analysis. Furthermore, although we performed propensity score matching and adjusted for multiple covariates associated with smoking and CRC prognosis, the possibility of residual confounding cannot be excluded. The misclassification of anatomical subsites of CRC may have occurred, especially for tumors located in the junction of the sigmoid colon and the rectum. Finally, data regarding the molecular derangements of cancer were not available. These factors preclude a further detailed analysis of the differential effects of smoking on subsite CRC mortality.

Conclusions
Our study demonstrated that cigarette smoking was associated with a significantly, though moderately, increased risk of mortality in Asian patients with CRC. The smoking status can plausibly be considered in the risk stratification of CRC, and smoking cessation can be incorporated into comprehensive treatment planning for patients with CRC.