Annals of Surgical Oncology

, Volume 19, Issue 7, pp 2088–2094

Tobacco Use Is Associated with Increased Recurrence and Death from Gastric Cancer

  • E. C. Smyth
  • M. Capanu
  • Y. Y. Janjigian
  • D. K. Kelsen
  • D. Coit
  • V. E. Strong
  • M. A. Shah
Healthcare Policy and Outcomes

DOI: 10.1245/s10434-012-2230-9

Cite this article as:
Smyth, E.C., Capanu, M., Janjigian, Y.Y. et al. Ann Surg Oncol (2012) 19: 2088. doi:10.1245/s10434-012-2230-9

Abstract

Background

Tobacco use increases the risk of developing gastric cancer. We examined the hypothesis that gastric cancer developing in patients with a history of tobacco use may be associated with increased risk of cancer-specific death after curative surgical resection.

Methods

From the Memorial Sloan-Kettering Cancer Center Gastric Cancer prospective surgical database, we collected baseline demographic data and tumor characteristics from all patients who had undergone curative resection for gastric cancer between 1995 and 2009 and who had not received pre- or postoperative chemo- or radiotherapy. A smoking history was defined as >100 cigarettes’ lifetime use. The primary end point was gastric cancer disease-specific survival (DSS); secondary end points were 5-year disease-free survival (DFS) and overall survival (OS). Gastric cancer–specific hazard was modeled by Cox regression.

Results

A total of 699 eligible patients were identified with a median age of 70 years (range 25–96 years); 410 (59%) were current or previous smokers. Smoking was associated with gastroesophageal junction/cardia tumors and white non-Hispanic ethnicity. Multivariate analysis included the following variables: tumor stage, age, performance status, diabetes mellitus, gender, and tumor location. In this analysis, the hazard ratio for gastric cancer DSS in smokers was 1.43 (95% confidence interval 1.08–1.91, P = 0.01). Smoking was also an independent significant risk factor for worse 5-year DFS (hazard ratio 1.46, P = 0.007) and OS (hazard ratio 1.48, P = 0.003). Among 516 patients for whom tobacco pack-year usage was available, both heavy (≥ 20 pack-years) and light (<20 pack-years) tobacco use was significantly associated with DSS, DFS, and OS.

Conclusions

Smoking history appears to be an independent risk factor for death from gastric cancer in patients who have undergone curative surgical resection.

Gastric cancer is the second most common cause of cancer death worldwide and is responsible for 700,349 deaths annually.1 Tobacco use is associated with the development of gastric cancer, with a hazard ratio for risk of developing the disease ranging 1.3–1.7 in several series.2, 3, 4 In addition, gastric cancer risk increases with increasing cigarette consumption.2, 3, 4, 5 More than a billion people use tobacco products worldwide, with rising consumption in developing countries where gastric cancer incidence remains the highest.1 Notably, the effect of tobacco use on gastric cancer biology remains unknown.

The prognosis of patients with localized gastric cancer is primarily related to their tumor, node, metastasis system stage grouping, although additional tumor characteristics, such as Lauren’s histology and tumor location, also influence the risk of gastric cancer recurrence and death from gastric cancer after curative-intent resection.6 The significance of prior or current tobacco use on risk of gastric cancer recurrence after resection has not been previously reported. In other tobacco-related malignancies, including squamous cell carcinoma of the head and neck, cervical cancer, anal cancer, and transitional cell carcinoma of the bladder, a history of or current use of cigarettes is associated with increased risk of disease recurrence.7, 8, 9, 10, 11 A molecular rationale for a difference in disease biology has been demonstrated in lung cancer. Specifically, adenocarcinoma of the lung in never smokers, when compared to former smokers, has been demonstrated to be a distinct biologic entity with unique molecular characteristics and behavioral patterns.12,13

We hypothesize that gastric cancers occurring in the setting of prior or current tobacco use may have a worse prognosis, and may be associated with an increased risk of recurrence and gastric cancer related death after curative surgical resection. We specifically examined the influence of tobacco use on gastric cancer disease-specific survival (DSS) after curative resection in a large cohort of consecutive patients undergoing curative resection of gastric cancer in a single academic cancer center.

Methods

This retrospective study was performed after receiving institutional review board approval. The study sponsor had no role in the study design, collection, analysis and interpretation of data, in the writing of the report, or in the decision to submit the paper for publication.

We identified patients with resected gastric cancer from a prospectively maintained surgical database at Memorial Sloan-Kettering Cancer Center (MSKCC). Inclusion criteria were as follows; (1) surgical resection for histologically confirmed primary gastric or gastroesophageal adenocarcinoma between the years 1995 and 2009, (2) no evidence of metastatic disease, and (3) R0 resection (i.e., no positive surgical margins or positive peritoneal cytology). Esophageal carcinomas, defined as the bulk of tumor present in the esophagus (Siewert I tumors) were excluded. Patients who received perioperative or adjuvant therapy were also excluded to avoid the confounding effects of difference in response to or tolerance of treatment between smoking and nonsmoking cohorts.

Review of the electronic medical record of eligible patients was performed systematically by trained physicians by using standardized intake forms. Collected variables were related to patient demographics and pathologic variables and included age at diagnosis, gender, Eastern Cooperative Oncology Group (ECOG) performance score, tumor T and N stage (AJCC 6th edition), location of primary tumor, tumor differentiation, Lauren’s histology, presence of diabetes, and smoking history.14

Smoking status was obtained from the patient history identified in the electronic medical record as documented at the first medical or surgical oncology visit. A positive smoking history was defined as a documented lifetime use of more than 100 cigarettes.15 We defined light tobacco use as lifetime smoking of <20 pack-years, and heavy tobacco use was defined as ≥20 pack-year lifetime use. Patients for whom smoking history was unavailable (n = 26) were excluded.

The primary analysis evaluates DSS defined as the time from surgical resection to death from gastric cancer or to last follow-up. In the death from gastric cancer DSS Cox regression model, deaths due to other causes (n = 62) are censored at the time of death. In addition, in the primary DSS analysis, patients with an unknown cause of death (n = 45) were included as events. We also performed DSS sensitivity analyses in which patients with unknown cause of death were (1) excluded from the analysis, (2) censored at their death date, and (3) included as events if death occurred within 5 years from surgery (n = 26) and censored for deaths beyond 5 years (n = 19). Secondary analyses included overall survival (OS), defined as the time from surgical resection to death or last follow-up, and disease-free survival (DFS) within 5 years after surgical resection, defined as the time from surgery to recurrence or death that occurred in the first 5 years. In the 5-year DFS analysis, patients surviving for longer than 5 years without documented recurrence are censored at 5 years. Subset analysis included evaluation of the effect of pack-years of smoking on DSS, 5-year DFS, and OS. Cox regression was used to assess the association of DSS, OS, disease recurrence, or death within 5 years with known prognostic variables. Associations between categorical variables were assessed by Fisher’s exact test, while the Wilcoxon rank sum test was used for continuous variables.

Results

Patient Characteristics

Between 1/5/1995 and 2/24/2009, 699 eligible patients were identified. As patients who received pre- or postoperative therapy were excluded, the study population was enriched for early stage cancers. Median age at diagnosis was 70 years (range 25–96 years), and the majority of patients were smokers (n = 410, 59%). Smoking history was associated with male gender, white non-Hispanic ethnicity, and proximal/gastroesophageal junction tumors (Tables 1, 2). Lauren’s histology did not vary according to smoking status, although this classification was not reported in 28% of patients (Table 2). Pathologic parameters are provided in Table 3. Tumor depth and nodal status was similar for both smoking and nonsmoking patient groups. Over three-quarters of patients had T1 or T2 tumors, and >60% had no lymph node involvement. There was no difference in vascular or perineural invasion between the smokers and nonsmokers.
Table 1

Patient characteristics

Characteristic

Smoker

Nonsmoker

P

Gender

  

<0.0001

 Male

292 (71%)

124 (43%)

 

 Female

118 (29%)

165 (57%)

 

Age at diagnosis, y

71 (25–96)

70 (27–94)

0.5

Ethnicity

  

<0.0001

 White non-Hispanic

336 (82%)

198 (69%)

 

 White Hispanic

23 (6%)

22 (7%)

 

 Asian

22 (5%)

46 (16%)

 

 African American

25 (6%)

19 (7%)

 

 Unknown/other

4 (1%)

4 (1%)

 

Diabetes

  

0.57

 Yes

59 (14%)

43 (15%)

 

 No

350 (85%)

246 (85%)

 

 Unknown

1 (< 1%)

 

Performance status

  

0.06

 0

338 (82%)

254 (88%)

 

 ≥1

72 (18%)

35 (12%)

 

Data are presented as n (%) or median (range)

Table 2

Tumor characteristics

Characteristic

Smoker

Nonsmoker

P

Site

  

0.006

 Gastroesophageal junction/cardia

168 (41%)

82 (28%)

 

 Body

84 (20%)

66 (23%)

 

 Antrum

143 (35%)

129 (45%)

 

 Other

15 (4%)

12 (4%)

 

Lauren classification

  

0.099

 Intestinal

171 (42%)

131 (45%)

 

 Diffuse

93 (23%)

73 (25%)

 

 Mixed

32 (8%)

25 (9%)

 

 Not reported

114 (28%)

60 (21%)

 
Table 3

Tumor staging

Characteristic

Smoker

Nonsmoker

P

T stage

  

0.97

 T1

173 (42%)

126 (44%)

 

 T2

137 (33%)

92 (32%)

 

 T3

94 (23%)

67 (23%)

 

 T4

6 (2%)

4 (1%)

 

N stage

  

0.10

 N0

251 (61%)

179 (62%)

 

 N1

108 (26%)

71 (25%)

 

 N2

43 (11%)

24 (8%)

 

 N3

8 (2%)

15 (5%)

 

Vascular invasion

  

0.63

 Yes

154 (38%)

115 (40%)

 

 No

242 (59%)

167 (58%)

 

 Not reported

14 (3%)

7 (2%)

 

Perineural invasion

  

0.57

 Yes

144 (35%)

97 (34%)

 

 No

243 (59%)

180 (62%)

 

 Not reported

23 (6%)

13 (5%)

 

Tobacco and Gastric Cancer DSS

At a median follow-up of 76 months, 311 deaths had occurred, 44% of the entire patient cohort. A total of 204 (66%) of these deaths were due to gastric cancer, 62 (20%) were not due to gastric cancer (cardiopulmonary (n = 30), stroke (n = 4), other malignancy (n = 13), postoperative complication (n = 10), other (n = 5), and 45 (14%) were of unknown cause. In univariate analysis, tobacco use was associated with an increased risk of gastric cancer-specific death with hazard ratio (HR) 1.49 [95% confidence interval (CI) 1.15–1.94, P = 0.003], and remained independently significantly associated with worse gastric cancer–specific survival in multivariate analysis (Table 4). Patients with a smoking history had a 43% higher rate of death due to gastric cancer when compared to never smokers. Similarly, when DSS was analyzed with deaths of unknown cause excluded or attributed to gastric cancer if they occurred before 5 years (n = 26), the adverse association of cancer-specific survival and tobacco use remained independently significant; HR 1.39 (95% CI 1.02–1.19, P = 0.036) when deaths of unknown cause were excluded and HR 1.43 (95% CI 1.07–1.93, P = 0.017) when deaths of unknown cause within 5 years of surgery were attributed to disease. When patients with unknown causes of death were not attributed to gastric cancer and were censored at the time of their death, the association of tobacco and DSS demonstrated a HR of 1.28 (95% CI 0.94–1.75, P = 0.11).
Table 4

Multivariate analysis of gastric cancer DSS

Variable

Subcategory

HR

95% CI

P

Diabetes

Yes

1.44

1.02–2.04

0.041

Performance status

≥1

1.27

0.91–1.76

0.157

Smoking

Yes

1.43

1.08–1.90

0.014

T stage

T1

1.00

  
 

T2

2.05

1.40–3.01

0.0002

 

T3

4.64

3.13–6.88

<0.001

 

T4

3.84

1.63–9.05

0.002

N-Stage

N0

1.00

  
 

N1

1.88

1.34–2.58

0.0001

 

N2

4.18

2.83–6.18

<0.001

 

N3

12.58

7.31–21.64

<0.001

Site

Antrum

1.00

  
 

Body

0.96

0.67–1.37

0.809

 

Gastroesophageal junction/cardia

1.38

1.01–1.87

0.042

 

Other

1.50

0.75–3.01

0.248

Age at surgerya

1.01

1.00–1.02

0.043

Sex

F

0.79

0.59–1.06

0.120

Multivariate analysis including tumor stage, location, diabetes, gender, and age. Unknown causes of death were treated as events

aContinuous variable

Tobacco and Disease-free Survival

Disease free survival at 5 years was investigated as a secondary outcome. A total of 453 patients (65%) were alive 5 years after curative-intent gastric cancer surgery and 246 patients (35%) had died, 186 (27%) of recurrent gastric cancer, 34 (5%) of other cause, and 26 (4%) of unknown cause. Eleven patients (1.5%) had experienced disease recurrence but not died within 5 years from surgery. Univariate analysis demonstrated the smoking related HR for 5-year DFS to be 1.51 (95% CI 1.18–1.94, P = 0.001). In multivariate analysis, the HR for smoking in association with 5 year DFS was 1.46 (95% CI 1.13–2.08, P = 0.007). In this analysis, neither ECOG performance status nor diabetes mellitus were significantly independently associated with DFS, whereas T and N stage and age at the time of surgery remained independent predictors of outcome.

Tobacco Use and OS

Tobacco use was significantly associated with worse OS. In univariate analysis, the HR for smoking was 1.57 (95% CI 1.24–1.98, P = 0.0002), and in multivariate analysis, the HR for OS was 1.48 (95% CI 1.15–1.92, P = 0.003). OS was also significantly and independently associated with stage at resection, performance status, site of the primary tumor, age at the time of surgery, and a diagnosis of diabetes mellitus.

Tobacco Pack-year Analysis and Patient Outcome

We also examined the influence of cigarette dose on patient outcome. Of the 410 patients who reported current or previous smoking history, the use of tobacco was quantified in 227 patients (55%). Together with the 289 nonsmokers, 516 patients (74% of the total population) patients were included in this analysis. We defined light smoking (<20 pack-years) or heavy smoking (≥20 pack-years) as variables, along with no smoking history. Light smokers had a significantly worse gastric cancer DSS (HR 1.71, 95% CI 1.22–2.41, P = 0.002), whereas for heavy smokers, the HR for DSS was 1.41 (95% CI 0.94–2.13, P = 0.09). In addition, light and heavy smokers both demonstrated a significantly decreased DSS at 5 years after resection and OS (Table 5).
Table 5

Pack-year tobacco use and outcome

Smoking status

Survival

95% CI

P

Disease Specific Survival

   

 Never

1.0

  

 0–20 pack-years

1.71

1.22–2.41

0.0021

 ≥20 pack-years

1.41

0.94–2.13

0.09

5-year Disease Free Survival

   

 Never

1.0

  

 0–20 pack-years

2.04

1.43–2.91

<0.0001

 ≥20 pack-years

1.69

1.11–2.59

0.0151

Overall Survival

   

 Never

1.0

  

 0–20 pack-years

2.0

1.42–2.81

<0.0001

 ≥20 pack-years

1.76

1.19–2.61

0.005

All unknown causes of deaths were treated as events

Discussion

This study found that for patients with resected gastric cancer, the risk of dying of gastric cancer was significantly worse in those with a smoking history. We examined a large cohort of consecutively resected patients with localized gastric cancer over a 15-year period at a single academic institution. All patients had no evidence of metastatic disease at the time of surgery and underwent resection with curative intent, all with microscopically negative (R0) resection margins. The exclusion of patients who received adjuvant therapy resulted in a relatively uniform study population enriched for early stage gastric cancer and removed any potential source of bias with respect to selection for, response to, or tolerance of perioperative treatment based on physical condition and smoking history. In this population, we found that smokers had 43% higher risk of death due to recurrent gastric cancer after resection than those who had never smoked.

Although the majority of patients were followed closely for recurrent disease or complications from surgery, their follow-up evaluation was not prospectively defined and thus is subject to bias. However, for the primary end point of DSS, the cause of death was attributable for 86% of patients. Furthermore, within 5 years of surgery at which point the vast majority of deaths from gastric cancer would have occurred, the cause of death was attributable in 90% of patients. In addition, both sensitivity analyses in which unknown causes of death were excluded or in which unknown causes of death were attributed to disease if the death occurred within 5 years of resection demonstrated consistent results of a significant independent increased risk of dying of gastric cancer–specific survival after resection amongst smokers (HR 1.39–1.43, P < 0.05 for each). The sensitivity analysis least likely to reflect the course of disease, in which all patients with an unknown cause of death were censored at the date of death, irregardless of when the patient died, was also supportive of the association of tobacco and worse gastric cancer biology with borderline significance (HR 1.28, P = 0.11).

Importantly, the potential clinical effect of smoking history on gastric cancer disease biology has not been previously reported. Cigarette smoke contains more than 4500 biologically active compounds within its gaseous and particulate phases.16 Laboratory studies suggest possible mechanisms to explain the effects of tobacco use on the biological behavior of malignancy. Nicotine, although not carcinogenic by itself, induces proliferation and angiogenesis in several preclinical models.17, 18, 19 Nicotine induced increase in cell motility is associated with decreased E-cadherin expression, and in gastric cancer, this appears to be associated with perturbation of the epithelial-mesenchymal transition (EMT) mediated by the Erk/5-LOX signaling pathway.20,21 Additional factors may play a role including the interaction between Helicobacter pylori infection and tobacco use, which clinically have an additive interaction for the risk of gastric cancer (HR 4.9 for H. pylori and tobacco use together).22 Because many patients in our study received empiric therapy for upper gastrointestinal symptoms before their cancer diagnosis, the true prevalence of H. pylori in our population is unknown. Because H. pylori eradication is negatively affected by tobacco and alcohol use, it is possible that smokers in our cohort could have an increased risk of H. pylori–related cancer.23,24 However, there is no evidence that gastric cancer developing in the setting of H. pylori infection is associated with worse DFS or OS.25

Additionally, the effect of tobacco on the immune system has been well documented. Current and former smokers have demonstrated lower levels of natural killer (NK) cells than never smokers; this depletion may continue for up to 20 years after tobacco cessation.26 Tobacco related defects in NK mediated immune surveillance may lead to an increased risk of metastasis. Lu et al. demonstrated that in a murine lung metastasis model, tobacco exposed mice injected with B16-MO5 melanoma cells developed significantly more lung metastases than those who were not exposed to tobacco smoke.27 Tobacco exposed animals demonstrated both decreased NK cell activation, as measured by CD69 expression, and compromised cytotoxic T-lymphocyte mediated cell lysis, with fewer apoptotic and dead YAC-1 target cells seen after incubation with NK cells taken from smoke exposed mice. Thus, it is possible that both tumor and host factors may contribute to the association between tobacco and gastric cancer–specific survival seen in this study.

The retrospective assessment of tobacco use and lack of data regarding alcohol use provides another potential source of bias in this study. As alcohol intake was not formally quantified, it was not formally assessed in our predictive model. Misclassification of tobacco use in patients with cancer may occur when smoking history is collected in an indirect manner due to recall bias and social desirability issues.28 When an exposure is considered either a personal threat to health, or socially undesirable, as with tobacco consumption in the setting of cancer, underreporting of usage may occur.29 Although both light and heavy smokers appeared to have an adverse association with patient outcome, we did not demonstrate a dose–response effect with tobacco use in this study. This may be due to the unavailability of pack-year data on a significant minority of patients who reported a smoking history, potential biases associated with the mode of data collection, or because a dose–response relationship does not exist. Between 5% and 12% of never smokers are misclassified in epidemiologic studies.30,31 In our study, individuals who may have been misclassified as never smokers would likely only dilute the perceived effects of tobacco usage on DSS.

Tumors of the gastric antrum were significantly more common in the group of never smokers, and cardia/gastroesophageal junction tumors in the group of prior or current smokers (45% vs. 35% and 41% vs. 28%, respectively; P = 0.006). Although tobacco consumption increases the risk of development of both antral and proximal tumors, our data support the previous finding that the relationship appears to be stronger for tumors of the cardia/gastroesophageal junction. In a meta-analysis of 32 epidemiologic studies examining the effect of tobacco on gastric cancer occurrence, Ladeiras-Lopes et al. found that the effect of tobacco consumption on the risk of malignancy was higher for cardia cancers when compared to antral tumors (relative risk of 3.14 vs. 1.60, respectively).3 This was also seen in a large Korean prospective cohort study of 669,570 men in whom the risk for proximal gastric cancer was increased by 90% for former smokers and 120% for current smokers, compared to increases in antral gastric cancer of only 40% for both current and former smokers.32 Notably, neither study examined the association of tobacco use and gastric cancer–specific survival.

There are several examples of tobacco-related molecular differences that may account for unique tobacco-specific tumor biology. Similar to non–small cell lung cancer, in which tobacco-associated disease is molecularly distinct from non-tobacco-related lung cancer, recently published data suggest differences in the molecular profile of colorectal cancer in smokers and nonsmokers. In a study of more than thirty-seven thousand women, Limsui et al. demonstrated that although the HR for colorectal cancer for ever smokers was consistent with the literature (HR 1.19, 95% CI 1.05–1.35), the incidence of cancers with high microsatellite instability, those with a highly methylated CpG island phenotype, and those with BRAF mutations were significantly more common in ever smokers.33 The presence of either a hypermethylated CpG island phenotype in proximal colon tumors or a BRAF mutation in colorectal cancer have each been associated with an increased risk of disease recurrence in curatively resected patients with colon cancer.34,35 Thus, in another gastrointestinal malignancy, tobacco use appears to be associated with a distinct molecular phenotype of malignancy, and that prior smoking may in fact specifically influence disease biology, thereby affecting patient survival. Also in common with the work of Meyerhardt et al. on colorectal cancer, we demonstrated that a diagnosis of diabetes mellitus may also be associated with an increased risk of gastric cancer recurrence.36

In this study, we found that tobacco use was associated with a worse DSS after curative-intent resection for gastric cancer. Tobacco has already been established as an independent risk factor for the development of gastric cancer, particularly for gastroesophageal junction/proximal tumors. Our analysis suggests that gastric cancer that develops in the setting of tobacco use is associated with a higher risk of dying of gastric cancer independent of other known prognostic variables. Additional interesting research questions now arise: is tobacco-related gastric cancer a molecularly distinct phenotype, or do tobacco-related changes in the host tumor environment lead to increased risk of disease recurrence and death? The answers to these questions may lead to the development of future therapeutic targets, which are desperately needed in this common and devastating disease.

Acknowledgment

Supported in part by the DeGregorio Foundation for Gastroesophageal Cancers.

Financial disclosures

There are no financial disclosures from any author.

Copyright information

© Society of Surgical Oncology 2012

Authors and Affiliations

  • E. C. Smyth
    • 1
  • M. Capanu
    • 2
  • Y. Y. Janjigian
    • 1
  • D. K. Kelsen
    • 1
  • D. Coit
    • 3
  • V. E. Strong
    • 3
  • M. A. Shah
    • 1
    • 4
  1. 1.Department of MedicineMemorial Sloan-Kettering Cancer CenterNew YorkUSA
  2. 2.Department of Epidemiology and BiostatisticsMemorial Sloan-Kettering Cancer CenterNew YorkUSA
  3. 3.Department of SurgeryMemorial Sloan-Kettering Cancer CenterNew YorkUSA
  4. 4.Weill Cornell Medical CollegeNew YorkUSA

Personalised recommendations