Association Between Head and Neck Squamous Cell Carcinoma Survival, Smoking at Diagnosis, and Marital Status

Key Points

Questions  Is there an association between smoking status at diagnosis and head and neck squamous cell carcinoma (HNSCC) survival? Are married patients with HNSCC more likely to be nonsmokers than smokers?

Findings  In this retrospective cohort study of 463 confirmed cases of HNSCC, patients who were smokers at the time of diagnosis had lower survival than nonsmokers. In addition, smokers were more likely to be unmarried.

Meaning  Smoking at the time of HNSCC diagnosis is associated with lower survival than nonsmoking; married individuals are more likely to be nonsmokers.

Importance  While the adverse association between smoking and head and neck squamous cell carcinoma (HNSCC) survival has been well described, there are also inconclusive studies and those that report no significant changes in HNSCC survival and overall mortality due to smoking. There is also a lack of studies investigating the association of marital status on smoking status at diagnosis for patients with HNSCC.

Objective  To examine the association between patient smoking status at HNSCC diagnosis and survival and the association between marital status and smoking in these patients.

Design, Setting, and Participants  This retrospective cohort study was conducted by querying the Saint Louis University Hospital Tumor Registry for adults with a diagnosis of HNSCC and treated at the university academic medical center between 1997 and 2012; 463 confirmed cases were analyzed.

Main Outcomes and Measures  Cox proportional hazards regression analysis was used to evaluate association of survival with smoking status at diagnosis and covariates. A multivariate logistic regression model was used to assess whether marital status was associated with smoking at diagnosis adjusting for covariates.

Results  Of the 463 total patients (338 men, 125 women), 92 (19.9%) were aged 18 to 49 years; 233 (50.3%) were aged 50 to 65 years; and 138 (29.8%) were older than 65 years. Overall, 56.2% of patients were smokers at diagnosis (n = 260); 49.6% were married (n = 228); and the mortality rate was 54.9% (254 died). A majority of patients were white (81.0%; n = 375). Smokers at diagnosis were more likely to be younger (ie, <65 years), unmarried, and to drink alcohol. We found a statistically significant difference in median survival time between smokers (89 months; 95% CI, 65-123 months) and nonsmokers at diagnosis (208 months; 95% CI, 129-235 months). In the adjusted Cox proportional hazards model, patients who were smokers at diagnosis were almost twice as likely to die during the study period as nonsmokers (hazard ratio, 1.98; 95% CI, 1.42-2.77). In the multivariate logistic regression analysis, unmarried patients were 76% more likely to use tobacco than married patients (adjusted odds ratio, 1.76; 95% CI, 1.08-2.84).

Conclusions and Relevance  Smokers were almost twice as likely as nonsmokers to die during the study period. We also found that those who were married were less likely to be smokers at diagnosis. Our study suggests that individualized cancer care should incorporate social support and management of cancer risk behaviors.

There are more than 430 000 survivors of head and neck squamous cell carcinoma (HNSCC) currently in the United States, and the number is expected to rise.¹ Besides improved treatment modalities and cancer care, major drivers of this survival increase are decreasing smoking rates and increasing human papillomavirus (HPV)-associated oropharyngeal cancer, which typically portend better prognosis.^1-3 However, even though smoking rate is decreasing nationally, current smokers, recent quitters (within the past 12 months), and former smokers still account for up to 80% of patients with HNSCC in developed countries.^4-6 Between 72% and 80% of all HNSCC cases are linked to either tobacco alone, alcohol alone, or a synergism of tobacco and alcohol.^6,7 This makes tobacco’s role in HNSCC as relevant today as ever before.

Tobacco use before and after cancer diagnosis is an established prognostic factor in HNSCC survival and mortality.^6,8-12 Besides stage at presentation and tumor site, there is probably no stronger predictor of HNSCC survival than tobacco use.¹³ Even among patients with HPV-positive HNSCC, smokers typically fare worse.^14-16 The same is true for HPV-negative HNSCC.^8-10,12,13 Tobacco use is associated with increased risk of secondary tumors,¹⁷ decreased efficacy of surgical and chemoradiation treatments,¹⁰ and decreased quality of life.¹⁸

Although the association between smoking and HNSCC outcomes has been well described, these findings have been contradicted by other studies that report no significant or inconclusive changes in local disease, disease-specific survival, and overall mortality in HNSCC.^19-23 The present study examines whether tobacco use is associated with reduced HNSCC survival. This finding could have important clinical and policy implications, such as including tobacco cessation in the short- and long-term management of patients with HNSCC.²⁴ While the need to incorporate continuous tobacco screening and cessation programs as part of standard clinical cancer care has been stressed, this is not yet universally available for cancer patients who smoke.^25,26 Since individuals who smoke often cite stress and social environment as smoking triggers,^27,28 understanding the adverse effect of smoking on HNSCC survival also warrants investigating the role of social support in mitigating smoking habits. Previous studies have shown that being married may confer survival benefits for patients with HNSCC.^29,30 However, to our knowledge, no study has investigated the impact of spousal status on smoking status at the time of HNSCC diagnosis. Smoking cessation programs in noncancer patients have achieved success by incorporating couples-oriented interventional strategies,³¹ but this has yet to be replicated in the cancer population, let alone HNSCC. Thus, the objectives of this study were 2-fold. First, we aimed to describe the effect of smoking status at diagnosis on survival in HNSCC, and second, we aimed to investigate the association between marital status and smoking status of patients with HNSCC at the time of diagnosis.

This was a retrospective cohort study based on data obtained from patients who received both outpatient and inpatient care at Saint Louis University Hospital from January 1, 1997, to December 31, 2012. The tumor registry was queried for 463 patients aged 18 to 87 who had a primary diagnosis of HNSCC per International Classification of Diseases for Oncology, Third Edition (ICD-O-3) codes. This study was approved by the Saint Louis University institutional review board, waiving patient written informed consent for this retrospective analysis of deidentified data.

The outcome for the first aim was mortality (alive vs dead), and for the second aim, it was smoking status at diagnosis (yes vs no). Sociodemographic and behavioral characteristics examined included age (18-49, 50-65, or >65 years), sex, race (white, black, or other), marital status (married or not married), insurance (private, government, or no insurance), and alcohol use (yes or no). Cancer stage at diagnosis was classified using the American Joint Commission on Cancer classification. To compensate for the low number of medical records in certain subgroups, we combined stages I and II as “early stage” and stages III and IV as “late stage.”

We compared sociodemographic characteristics and covariates by smoking status at diagnosis using χ² tests. For aim 1, we used time-to-event modeling to describe the association between smoking status at diagnosis and risk of death after adjusting for covariates. A patient was censored from the analysis at the end of follow-up, at the time of death, or at their last hospital visit date, whichever came first. Kaplan-Meier survival curves were used to illustrate unadjusted survival times for patients who were smokers and nonsmokers at the time of diagnosis. An adjusted Cox proportional hazards regression model was used to evaluate the effects of smoking status at the time of diagnosis and other covariates on survival. We also tested for the interaction effect of smoking at diagnosis and marital status on mortality in the Cox proportional hazards regression model. Hazard ratios (HRs) and 95% confidence intervals (CIs) for mortality risk were calculated to compare smokers and nonsmokers at time of diagnosis. For the second aim, we used a multivariable logistic regression model to assess whether marital status was associated with smoking status at diagnosis. All covariates were included in both the adjusted Cox proportional hazards regression and the adjusted logistic regression models regardless of their association with the outcome variables at the bivariate analysis, since the covariates have all been shown in previous studies to be associated with head and neck cancer outcomes.^29,32-34 All tests were 2 sided, and an α ≤ .05 was considered statistically significant. Analyses were performed using SAS software, version 9.4 (SAS Institute Inc).

Table 1 details the distribution of sociodemographic characteristics of our study population. The median follow-up time in the study was approximately 81 months (95% CI, 70-89 months). Overall, 56.2% of patients were smokers at diagnosis (n = 260); 49.6% were married (n = 228); and the mortality rate was 54.9% (254 died). A majority of patients were between 50 and 65 years old (50.3%; n = 233), male (73.0%; n = 338), and white (81.0%; n = 375). Smokers at diagnosis were more likely to be younger (ie, <65 years), unmarried, and to drink alcohol. Figure 1 shows the distribution of smokers at diagnosis stratified by anatomical site. The highest number of smokers had cancers of laryngeal cavity (36.5%; n = 95) followed by the oropharyngeal cavity (32.3%; n = 84), and then the oral cavity (23.1%; n = 60).

The Kaplan-Meier survival curve (Figure 2) revealed a significant difference in median survival time between smokers and nonsmokers at diagnosis. The median survival time for smokers at diagnosis was 89 months (95% CI, 65-123 months), and for nonsmokers at diagnosis it was 208 months (95% CI, 129-235 months). Table 2 lists the adjusted HRs (aHRs) for death from our Cox proportional hazards regression model. After adjusting for prognostic factors, we found that smokers at diagnosis were twice as likely to die during the study period (aHR, 1.98; 95% CI, 1.42-2.77) compared with nonsmokers at diagnosis. Similarly, unmarried patients had almost twice the hazard of death (aHR, 1.87; 95% CI, 1.38-2.53) compared with married patients. Patients aged 65 years or older had more than double the hazard of death (aHR, 2.38; 95% CI, 1.50-3.80) compared with those aged 18 to 49 years. Patients with late-stage disease were more likely to die during the study period (aHR, 1.65; 95% CI, 1.14-2.38) than those with early-stage disease. Finally, patients with oral cavity cancer (aHR, 1.75; 95% CI, 1.11-2.75) and patients with cancer of other or unknown anatomical site (aHR, 2.98; 95% CI, 1.77-5.01) were more likely to die during the study period than those with laryngeal cancer. The interaction between smoking status at diagnosis, marital status, and mortality was not significant.

Since most patients with oropharyngeal cancer may be HPV positive, which we were unable to assess, a sensitivity analysis was performed excluding those with oropharyngeal cancer, and the results are presented in the eTable in the Supplement. The Kaplan-Meier survival curve (eFigure in the Supplement) from sensitivity analysis revealed a significant difference in median survival time between smokers and nonsmokers at diagnosis. The median survival time was the same as in the original analysis; 208 months for smokers and 89 months for nonsmokers at diagnosis. The multivariable Cox proportional model from the sensitivity analysis showed that smokers at diagnosis had twice the hazard of death (aHR, 2.32; 95% CI, 1.50-3.59) compared with nonsmokers at diagnosis. Smoking status hazards of death changed slightly from 1.98 in the original analysis to 2.32 in the sensitivity analysis. Similarly, patients who were unmarried had twice the hazard of death (aHR, 2.04; 95% CI, 1.39-2.99) compared with those who were married. Marital status hazards of death changed slightly from 1.87 in the original analysis to 2.04 in the sensitivity analysis.

Table 3 lists the adjusted multivariate odds ratios (aORs) for the association of marital status and other covariates with smoking status at diagnosis. After adjusting for prognostic factors, we found that marital status, age, insurance, alcohol use, and anatomical site were significantly associated with smoking status at diagnosis. Compared with married patients, unmarried patients were almost twice as likely to be smokers at diagnosis (aOR, 1.77; 95% CI, 1.09-2.87). Patients aged 65 years or older were 78% less likely to be smokers at diagnosis (aOR = 0.22; 95% CI, 0.11-0.46) than those aged 18 to 49 years. Patients who reported alcohol use were almost 3 times as likely (aOR, 2.96; 95% CI, 1.85-4.72), while those who reported using government insurance were approximately twice as likely (aOR, 2.14; 95% CI, 1.24-3.67), to be smokers at diagnosis. Finally, patients with cancer of the oral cavity (aOR, 0.34; 95% CI, 0.17-0.70), oropharyngeal cavity (aOR, 0.30; 95% CI, 0.16-0.56), and cancer of other or unknown sites (aOR, 0.30; 95% CI, 0.12-0.78) were less likely to be smokers at diagnosis than those with laryngeal cancer.

Our study aimed to examine the association between tobacco smoking and marital status in patients with HNSCC and to evaluate the association of smoking status with survival. We established that the smoking rate was very high among this patient population: unmarried patients were more likely to be smokers at the time of diagnosis, and smokers were more likely to die during the study period than nonsmokers. The latest Centers for Disease Control and Prevention report estimates that 15.1% of US adults are current smokers, and that smoking rates continue to drop across the United States.³⁵ However, it is evident that this decreasing smoking rate is not uniform across all groups. Among cancer patients, smoking rates are much higher: nearly 62% of recently diagnosed cancer patients across the United States identified as current smokers, recent quitters, or former smokers, with the highest numbers in lung cancer and HNSCC.¹³ In the present study, approximately 56% of patients were current smokers at diagnosis, which is more than 3 times the current national smoking rate. Previous HNSCC studies have reported varying rates of smoking at diagnosis, ranging from 26.4%⁸ to 56%.⁹ Smoking rates tend to be heterogeneous across head and neck subsites, with the larynx having the higher rates and a greater likelihood of being associated with smoking.^9,36,37 Our study confirms this. We found that patients with laryngeal cancer had a significantly higher smoking rate than patients with cancer of the other subsites. It is important note these differences when tracking smoking status of patients with HNSCC across the cancer continuum.

Not only is tobacco use a risk factor for HNSCC development, but smoking at diagnosis is associated with decreased survival, increased risk for second primary cancers of the lung, esophagus, and even prostate,¹⁷ hypoxic radioresistance during radiation treatment, as well as increased risk of comorbidities and competing causes of death such as cardiovascular and chronic obstructive pulmonary diseases.^38-41 Despite the recent focus on HPV-positive HNSCC, smoking remains as important as oral HPV as a risk factor for patients with HNSCC. Most patients in our study developed primary cancer in subsites that are mostly unrelated to HPV, as described by Chaturvedi and colleagues.⁴² There is evidence that smoking significantly increases the odds of developing persistent oral HPV^43,44 and HPV-associated HNSCC^5,15,45 independent of sexual behaviors. Therefore, there is need for tobacco prevention and control to remain an important aspect of head and neck cancer intervention strategies regardless of HPV status.

We found that patients with HNSCC who were smokers at diagnosis were almost twice as likely to die during the study period as nonsmokers, which is consistent with previous studies.^8,11 There are several proposed mechanisms for smoking-induced effects on survival in HNSCC. Some suggest that tobacco use increases inflammatory reactions,⁴⁶ induces nicotine dose-dependent DNA damage,⁴⁷ and increases production of matrix metalloprotein-2 and -9.⁴⁸ Whatever the mechanism of action, our study adds more evidence that tobacco use at the time of diagnosis is associated with poorer overall survival of patients with HNSCC.

Our findings on the negative association of smoking with mortality before HNSCC treatment have clinically significant implications. Studies have shown that smokers’ knowledge of a causal relationship between smoking and their cancer is a successful motivator for tobacco cessation.^49,50 Our findings add evidence to calls for more standardized tobacco-related assessments, increased smoking cessation efforts, and lifelong surveillance of cancer patients.^11,13,51-53 Although smoking cessation is offered to patients at the time of diagnosis, many physicians do not consistently offer it as part of standard of care for those with HNSCC.^11,24 In fact, a recent survey suggests that fewer than half of oncologists actually offer assistance with smoking cessation.²⁶ The American Association for Cancer Research (AACR) and the American Society of Clinical Oncology (ASCO) both recommend integrating smoking cessation into the care of cancer patients.^54,55 Since 14% to 59% of patients with HNSCC continue to smoke after their cancer diagnosis,^56,57 continuous monitoring of smoking status and tobacco cessation efforts may play a critical role in the care of HNSCC patients with any history of smoking.

We found that HNSCC patients who were married were less likely to be smokers at diagnosis. Although previous studies have noted the association between marital status and survival in HNSCC,^8,30,58,59 ours is the first study to our knowledge to report this association between marital status and smoking at diagnosis after adjusting for other covariates. Previous studies have described the association between marital status and smoking among other populations, such as pregnant women⁶⁰ and successful quitters.⁶¹ Although the present study did not distinguish between recent quitters, quitters, and never smokers, our findings nevertheless highlight the importance of spousal status in tobacco abstinence. We suggest that personalized cancer care, especially for HNSCC patients, should incorporate social support particularly in managing cancer-associated behaviors such as smoking. Kashigar et al⁶² identified the critical role of an individual’s social environment, such as family and peer support, on smoking cessation in HNSCC patients. Involving spouses and other key social support systems may be worthwhile in tobacco smoking reduction effort among HNSCC patients, described as a vulnerable population often with limited social support living in socially deprived environments.⁵⁰ Patients with HNSCC who fit this profile may need extra support to quit smoking,⁵⁷ and marital support may be effective in this regard.

Our study had limitations. First, its retrospective design restricted which covariates we could analyze. Follow-up data on survivorship in our cohort of patients with HNSCC would have provided a more complete and accurate interpretation of the association of smoking and marital status on survival. In the current study, we could not access how changes in marital status affected smoking status after cancer diagnosis. Moreover, we were not able to stratify nonsmokers at diagnosis into recent quitters, long-term quitters, and never smokers. This would have provided a clearer picture of the effect of any tobacco use on HNSCC survivorship. Previous studies have suggested that recent quitters and exsmokers have improved clinical outcomes compared with current smokers.^4,9,11 We recommend that future studies describe both smoking history and intensity. Additionally, future prospective studies should investigate whether changes in marital status affect smoking status. For example, are nonsmokers at diagnosis more likely to smoke after a divorce? Or are former smokers at diagnosis more likely to start smoking again owing to widowhood or divorce? We hypothesize that changes in marital status after diagnosis will affect smoking status and overall outcomes. Furthermore, we did not have information on HPV status that could have affected our findings, since HPV-positive patients tend to have better prognosis. Finally, the status of tobacco use in our study was determined by self-report, without confirmation by biochemical assessments such as carbon monoxide breathalyzers and urine or salivary cotinine tests.

Despite its limitations, the present study demonstrates the interplay between marital status and smoking at diagnosis for patients with HNSCC, and it has added to the existing evidence of an association between tobacco smoking and HNSCC survival. The main strength of the study is the significant finding of the association between smoking status at diagnosis and marital status, an important component of a patient’s social environment and support. This has not been previously reported and represents an important avenue for consideration in smoking prevention and smoking cessation programs in patients with HNSCC cancer. Although overall smoking rates continue to decline, smoking remains a significant causative factor in up to 80% of patients diagnosed with HNSCC.⁴ This warrants increased intervention in this cancer population. Future research should examine the effect of tobacco cessation and timing of cessation on survival. Moreover, given the significant association between patients’ marital status and improved survival, it would be interesting to monitor marital status of cancer patients over the course of their disease and correlate changes with smoking status and survival. In addition, future research should examine other social factors such as the number of friends who smoke, socioeconomic status, and level of education to better identify and target high-risk cancer patients for intervention.

In summary, smoking status at diagnosis was associated with survival of patients with head and neck cancer: patients who were smokers at diagnosis were more likely to die during the study period than those who were not smokers at diagnosis. Married patients were less likely to be smokers at diagnosis. These findings highlight the need for clinicians to discuss and encourage spousal involvement in tobacco cessation efforts to mitigate risky behaviors that negatively affect head and neck cancer survival. Individualized cancer care should incorporate social support and management of cancer risk behaviors. However, since we have shown that marital status may confer survival advantage, there is the unanswered question of how to provide meaningful support for patients with HNSCC who are unmarried. Future studies are needed to identify and evaluate the effectiveness of other feasible, nonmarital support systems for nonmarried patients with HNSCC. Finally, it is important to understand the underlying mechanism in the association between smoking status and marital status at the diagnosis of HNSCC.

Corresponding Author: Nosayaba Osazuwa-Peters, BDS, MPH, CHES, Saint Louis University Cancer Center, 3655 Vista Ave, Third Floor, West Pavilion, St Louis, MO 63110-2539 (nosazuwa@slu.edu).

Accepted for Publication: July 31, 2017.

Published Online: November 9, 2017. doi:10.1001/jamaoto.2017.1880

Author Contributions: Drs Osazuwa-Peters and Varvares had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Osazuwa-Peters, Adjei Boakye, Varvares.

Acquisition, analysis, or interpretation of data: Osazuwa-Peters, Adjei Boakye, Chen, Tobo.

Drafting of the manuscript: Osazuwa-Peters, Adjei Boakye, Chen.

Critical revision of the manuscript for important intellectual content: Osazuwa-Peters, Adjei Boakye, Chen, Tobo, Varvares.

Statistical analysis: Osazuwa-Peters, Adjei Boakye, Tobo.

Administrative, technical, or material support: Osazuwa-Peters.

Study supervision: Osazuwa-Peters, Adjei Boakye, Varvares.

Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.