Department of Epidemiology and Biostatistics, SUNY Downstate School of Public Health, Graduate Program in Public Health at SUNY Downstate Medical Center, 450 Clarkson Avenue, Box 43, Brooklyn, NY 11203
Scott Langevin and Joseph Kwok extracted the data from the main database, performed quality control checks, statistical analysis, and contributed to the manuscript writing. Athanassios Argiris completed the clinical writing of the manuscript, Jennifer R. Grandis and Emanuela Taioli conceptualized the study, and William E. Gooding developed the statistical analysis framework. The authors had direct access to anonymous data.
In 2006, it was estimated that 47 million people in the United States are without insurance. Studies have shown that patients who are uninsured or are insured by Medicaid are more likely to present with more advanced cancer. The objective of this study was to examine whether cancer recurrence and mortality of patients diagnosed with squamous cell carcinoma of the head and neck are associated with insurance status, after adjusting for known cancer risk factors. The main outcome measures were overall survival and relapse-free survival.
Retrospective cohort of patients with a biopsy-proven primary squamous cell carcinoma of the oral cavity, pharynx, or larynx diagnosed or treated at the University of Pittsburgh Medical Center between 1998 and 2007. Patients were stratified by their insurance status, including private insurance, uninsured/Medicaid, Medicare disability (Medicare under age 65), and Medicare 65 years + . Covariates included age, gender, race, smoking status, alcohol consumption, anatomic tumor site, treatment, stage at diagnosis, and occupational prestige score. Cox proportional hazards regression was used to estimate the effect of insurance status on overall survival, relapse-free survival, tumor stage, and lymph node involvement.
A total of 1231 patients were included in the analysis. Patients with Medicaid/uninsured (hazard ratio [HR], 1.50; 95% confidence interval [CI], 1.07-2.11) and Medicare disability (HR, 1.69; 95% CI, 1.16-2.48) had significantly lower overall survival compared with patients with private insurance; the result was independent of variables known to affect outcome, such as alcohol and tobacco use. For all squamous cell carcinoma of the head and neck (SCCHN) cancer sites, Medicaid and uninsured patients were significantly more likely to present with an advanced stage tumor at diagnosis (odds ratio [OR] = 2.94; 95% CI, 1.72-5.01) and to present with at least 1 positive lymph node (OR = 1.84; 95% CI, 1.16-2.90) compared with patients with private insurance.
Head and neck cancer includes a variety of cancers that can involve the oral cavity, salivary glands, paranasal sinuses, nasal cavity, pharynx, larynx, and lymph nodes of the neck. Although head and neck cancer constitutes a minority of cancers in the population (approximately 3%-5% of cancers in the United States), it is a cancer with high mortality and morbidity, including disfigurement and difficulties with speech, swallowing, and breathing. The American Cancer Society estimates that in the United States alone, there will be over 35,000 new cases of oral and pharyngeal cancer in 2008, with about 7500 of those resulting in death and 12,250 new cases of laryngeal cancer leading to approximately 3600 deaths.1 Furthermore, the costs of treatment are high, especially when there are treatment-induced complications. One study estimated the cost for radiochemotherapy for head and neck cancer to be approximately $18,000, with an additional $17,000 in the case that the patient develops radiochemotherapy-induced mucositis or pharyngitis.2
In 2006, it was reported that there were approximately 47 million people (15.8% of the population) without health insurance in the United States.3 For a vast majority of these people, lack of insurance often poses a significant barrier to the proper screenings and treatments made available by the increasing technology and understanding of medicine. Access to primary preventive settings, such as smoking cessation programs, is also hampered by lack of insurance coverage. These barriers may then result in increased morbidity and mortality from many diseases, including cancer.
Recent studies suggest that patients without insurance are less likely to receive cancer screenings and are more likely to present with advanced disease and die from their cancer compared with those with insurance, after adjusting for race, sex, age, and zip code estimated income.4-9 For example, the American Cancer Society suggests that the uninsured or those covered by Medicaid are less likely to receive screening procedures for breast and colon cancer and more likely to die from the cancer than their insured counterparts.6 Other studies suggest that those who are uninsured or are covered by Medicaid are more likely than those with other insurance to present with late stage cancer.4, 5, 7, 8 Two studies assessed this relationship in head and neck cancer: 1 included oropharyngeal cancer4 and the other was on cancer of the larynx.5
A common limitation of these studies of survival among head and neck cancer patients by insurance type is the lack of information on specific, highly correlated cancer risk factors such as smoking and drinking habits. These factors are more frequently observed in the low-income strata of the population, which is also less likely to be insured, and are associated with cancer outcomes. Therefore, the study of the effect of insurance status on cancer outcomes cannot dismiss the assessment of the main cancer risk factors.
In this study, we examined the relation between health insurance status and survival for patients with head and neck cancer. By using approximately 10 years of head and neck cancer patient data from the University of Pittsburgh Medical Center, we examined the relationship between insurance status and tumor stage at diagnosis, and the implications insurance status may hold on survival outcomes for head and neck cancer, independent of various behavioral factors, including alcohol and tobacco use.
MATERIALS AND METHODS
Eligible subjects included all patients with a biopsy-proven primary squamous cell carcinoma of the head and neck (SCCHN) arising in the oral cavity, pharynx, or larynx and who were diagnosed or treated at the University of Pittsburgh Medical Center (Pittsburgh, Pa) between January 1998 and October 12, 2007 (N = 1321). Patients with missing health insurance data for the time of diagnosis (n = 63) or who refused treatment (n = 15) were excluded from the study. All patients with private insurance, TRICARE/CHAMPUS, Medicaid, Medicare, or no insurance at the time of diagnosis were included in the study; patients with other types of government-sponsored insurance plans were excluded (n = 12). Thus, there were 1231 patients eligible for the analysis. The derivation of the cohort is illustrated in Figure 1. Approval was obtained from the University of Pittsburgh Institutional Review Board for use of patient data for this study.
The 63 patients with missing insurance data were similar to those with known insurance status with respect to all covariates except anatomic site of tumor origin (P = .03) and treatment (P = .02; data not shown). Compared with those with known insurance status, patients with missing insurance data were more likely to have tumors arise in the oral cavity and less likely to have tumors arising in the pharynx. They were also more likely to be treated with surgery only and less likely to be treated with radiation or chemoradiation only.
All clinical and demographic information on head and neck cancer patients diagnosed or treated at the University of Pittsburgh Medical Center are collected at the time of diagnosis and entered into the University of Pittsburgh Head and Neck Oncology Registry. Patients are prospectively followed to gather clinical data pertaining to subsequent tumors, medical and surgical treatment, and vital status. Sources for registry data abstraction include medical records, the UPMC Network Cancer Registry, and the Social Security Death Index (SSDI). The database is regularly updated to reflect the status of the patient at the most recent follow-up visit. The SSDI is routinely checked on a monthly basis for all subjects known to be alive as of the last date of follow-up, and vital status is updated accordingly.
Health insurance data was obtained through University of Pittsburgh Medical Center electronic billing records. On the basis of insurance status at the time of diagnosis, patients were classified as having private insurance (n = 547), including patients with TRICARE/CHAMPUS and insurance not otherwise specified (NOS), Medicaid or uninsured (n = 128), and Medicare (n = 556), including patients with and without the Part B supplement. All subjects 65 years or older listed in the billing system as having private insurance were categorized as Medicare (n = 126) with the understanding that virtually all Americans 65 years or older are covered by Medicare and, therefore, that the term “private insurance” referred to the Part B supplement. Although TRICARE/CHAMPUS is provided by the military, it was grouped with private insurance because of its similarity to private insurance plans. Of the 128 Medicaid/uninsured patients, 112 (87.5%) were reported as Medicaid at the time of entry in the UPMC system and 17 were reported as uninsured, although patients enrolled in Medicaid before diagnosis and patients who were uninsured at diagnosis and subsequently enrolled in Medicaid for treatment purposes could not be distinguished in this data set. To differentiate patients receiving Medicare disability benefits from Medicare recipients over the age of 65, Medicare patients were further divided into 2 groups: those younger than 65 years of age (n = 81) and those 65 years or older (n = 475).
The major endpoints of interest for this study were overall and relapse-free survival. Overall survival was defined as time from diagnosis of the initial primary tumor to death. Relapse-free survival was defined as time from diagnosis of the initial primary tumor to tumor recurrence, development of a second primary, or metastasis. For the purposes of relapse-free survival, patients dying without tumor recurrence were considered to be censored at the time of death. Prognostic factors at diagnosis, including stage and lymph node positivity, were evaluated as secondary outcomes of interest.
Descriptive statistics of the study population were generated by insurance status. The skewness-kurtosis test10 was used to evaluate if continuous variables were normally distributed. The nonparametric Kruskal-Wallis equality-of-populations rank test11 was used to test for differences in the case of non-normal distribution of a continuous variable. Differences between groups for categorical variables were evaluated with chi-square (χ2) tests. Statistical significance was considered at an alpha level of 0.05.
Kaplan-Meier curves were generated for 5-year overall and relapse-free survival by insurance type, adjusted for age centered at the median. The Wilcoxon log-rank test12 was used to test for differences across insurance groups, with significance considered at P = .05. Multivariate Cox proportional hazards regression13 was used to estimate the effect of insurance status while adjusting for multiple potential confounders. Proportional hazards assumptions were tested statistically using an approach based on the slope of scaled Schoenfeld residuals as a function of time.14
The primary predictor in the Cox models was insurance status, categorized as private, Medicaid/uninsured, Medicare <65 years old, and Medicare ≥65 years old. Other covariates included in the models were age, gender, race, smoking status, alcohol consumption, anatomic tumor site, treatment, stage at diagnosis, and occupational prestige. Age was considered as a continuous variable. Race was categorized as White or non-White. Smoking was categorized as never, light/moderate (defined as ≤10 pack-years, or ≤1 pack per day, or ≤10 years duration), heavy (>10 pack-years, or >1 pack per day, or >10 years duration), and smoker, dose unknown. Alcohol consumption was categorized as never, light (≤6 drinks/week), moderate (7-14 drinks/week); heavy (>14 drinks/week), and drinker, intensity unknown. Anatomic tumor site signified the major primary site in which the tumor arose according to American Joint Committee on Cancer (AJCC) staging manual (sixth edition) specifications,15 and was categorized as oral cavity, pharynx, and larynx. Treatment was categorized as surgery only, radiation and/or chemotherapy only, or combined surgery and radiation and/or chemotherapy. Stage at diagnosis was categorized as stage I, II, III, or IV, based on AJCC stage groupings for head and neck cancer.15 Nakao and Treas occupational prestige scoring was used to adjust for socioeconomic (SES) status16 based on self-reported occupation during working years, categorized as 0-25, >25 and ≤50, >50 and ≤75, and >75. Patients reporting as unemployed were considered to have a prestige score of 0. An additional category, which was not accounted for by the Nakao and Treas scoring system, was created to include homemakers. The occupational prestige score is considered the best way for determining SES from job codes alone and correlates well with other commonly used SES scales.17 Missing values for covariates in the model other than insurance status were coded with indicator variables.
A second model restricted to patients carrying private insurance or Medicaid/uninsured was used to validate the results and further ensure that the results were free of residual confounding; the model was also run on patients <65 years only. In this model, adjustment was performed using propensity score-derived weighting (estimated with a logit model). The same set of covariates used in the original model were used to derive propensity scores.
In addition, separate multivariable logistic regression models18 were used to assess the association of insurance status with prognostic indicators at diagnosis. The primary outcome for the models was stage and lymph node positivity at diagnosis, respectively. Stage was categorized as advanced stage (AJCC stage Group III or IV) versus early stage (AJCC stage Group I or II). The primary predictor in the models was insurance status; covariates in the model were age, gender, race, smoking status, alcohol consumption, anatomic tumor site, income, and education. Model fit was assessed using the Hosmer and Lemeshow goodness-of-fit test, considered a poor fit at P = .05.18
All statistical analyses were performed using Stata software, version 10 (StataCorp LP, College Station, Tex).
A description of the 1231 patients, included in the final analysis, according to insurance status is provided in Table 1. As expected from the stratification performed on the data, significant differences in age were observed by insurance status (P < .001), where Medicare patients 65 years or older had a median age 21 years older than Medicaid/uninsured patients and 16 and 17 years older than those with private insurance and Medicare disability patients, respectively. Medicaid/uninsured and Medicare disability patients were significantly more likely to be African American than other insurance groups (P = .011); patients with Medicaid or no insurance or Medicare disability were more likely to have lower occupational prestige scores (indicating a lower SES status) than those with other insurance types. Patients also differed across insurance groups with respect to smoking status (P < .001) and alcohol consumption (P < .001), with Medicaid/uninsured and Medicare disability patients more likely to report as heavy smokers and heavy drinkers than patients with other insurances. In addition, Medicaid/uninsured patients were more likely to present with advanced stage tumors (stage III and IV) than patients covered by other insurance plans. Treatment modality was also significantly different across insurance plans (P < .001): Medicare subjects were more likely to receive surgery as part of their treatment than other insurance groups.
Table 1. Description of the Study Population According to Insurance Status
The associations between stage and lymph node positivity at diagnosis and insurance status are presented in Table 2. For all sites (oral cavity, pharynx, and larynx), Medicaid and uninsured patients were nearly 3 times as likely to present with an advanced stage tumor at diagnosis (odds ratio [OR], 2.94; 95% confidence interval [CI], 1.72-5.01) and more likely to present with at least 1 positive lymph node (OR,1.84; 95% CI, 1.16-2.90) compared with patients with private insurance. When stratified by tumor site, Medicaid and uninsured patients with laryngeal SCCHN were significantly more likely to present with advanced stage disease (OR, 6.97; 95% CI, 2.57-18.90) and with lymph node involvement (OR, 4.18; 95% CI, 1.85-9.45) than privately insured patients. Medicare disability patients with laryngeal cancer were more likely than those with private insurance to present with lymph node involvement (OR, 2.92; 95% CI, 1.10-7.74) but not advanced stage disease. Having Medicaid or being uninsured was not significantly associated with advanced stage or lymph node positivity at diagnosis for oral or pharyngeal SCCHN. Neither group of Medicare patients was associated with advanced stage at diagnosis or lymph node positivity in comparison to patients with private insurance.
Table 2. Adjusted Odds Ratios and Corresponding 95% Confidence Intervals for Stage and Lymph Node Status at Diagnosis According to Insurance Status
The median follow-up time from initial diagnosis was 22.39 months (10th percentile: 4.44 months; 90th percentile: 65.25 months); 459 deaths occurred among 1231 at risk patients. The median follow-up time for relapse-free survival was 14.55 months (10th percentile: 2.69 months; 90th percentile: 60.98 months), with 406 recurrences among 1231 at risk patients.
The age-adjusted 5-year overall and relapse-free Kaplan-Meier survival curves are displayed in Figure 2 and Figure 3, respectively. Differences by insurance status were observed for overall 5-year survival (P < .001) but not relapse-free 5-year survival (P = .126). Medicaid/uninsured and Medicare disability patients showed lower survival rates compared with those with private insurance or Medicare patients 65 years or older.
In the multivariate Cox model, patients with Medicaid/uninsured (hazard ratio [HR], 1.50; 95% CI, 1.07-2.11) and Medicare disability (HR, 1.69; 95% CI, 1.16-2.48) had significantly lower overall survival compared with patients with private insurance independent of variables known to affect outcome, such as age, gender, race, SES status, alcohol and tobacco use, stage at diagnosis, and treatment. There was no significant difference in survival between Medicare patients 65 years or older and those with private insurance (Table 3). These findings were further substantiated by the propensity score adjusted model restricted to the Medicaid/uninsured and private insurance patient groups, which yielded similar results (HR, 1.51; 95% CI, 1.06-2.15). There were no significant differences in relapse-free survival across insurance groups in the multivariate Cox analysis (Table 3) or the propensity score adjusted Cox analysis restricted to the Medicaid/uninsured and private insurance patient groups (HR, 1.01; 95% CI, 0.69-1.47).
Table 3. Adjusted Hazard Ratios and Corresponding 95% Confidence Intervals for Overall and Relapse-Free Survival According to Insurance Status at Time of Diagnosis
Insurance Status at Time of Diagnosis
Medicare, ≥65 y
Medicare, <65 y
CI indicates confidence interval; HR, hazard ratio.
HRs and corresponding 95% CIs are adjusted for age, gender, race, smoking, alcohol consumption, tumor site, treatment, stage at diagnosis, and socioeconomic status (based on occupational prestige scores).
When the analyses were restricted to Caucasian patients, the results on overall (HR, 1.50; 95% CI, 1.05-2.14) or relapse-free (OR, 1.01; 95% CI, 0.70-1.45) survival are similar to the overall analysis. A subgroup exploratory analysis restricted to patients <65 years, adjusting the data based on propensity score weighting, showed similar results (data not shown). There was no statistically significant interaction between alcohol and smoking on overall survival or relapse-free survival (data not shown).
Previous studies have shown that head and neck cancer patients who are uninsured or insured by Medicaid are at greater risk for presenting with advanced stage disease when compared with head and neck cancer patients with private insurance. Our analysis builds on this premise by examining the differences in survival between head and neck cancer patients with different types of insurance. Furthermore, our analysis addresses one of the limitations of previous studies by assessing the impact of insurance status on survival outcomes for head and neck cancer independent of various factors, including alcohol and tobacco use.
The major finding of this study was that overall survival differed significantly between patients according to types of health insurance after adjustment for covariates that represent cancer outcome risk factors. In particular, our analysis found that patients with Medicaid/uninsured and Medicare disability had a 50% increased risk of death when compared with patients with private insurance, after adjustment for age, gender, race, smoking, alcohol use, site, SES status, treatment, and cancer stage.
Chen et al found that patients diagnosed with advanced stage oropharyngeal4 or laryngeal5 cancer were more likely to have Medicaid or no insurance, present with the largest tumors, and have the most extensive lymph node involvement as compared with those who were privately insured. Our study had similar findings: for all sites, the uninsured and Medicaid patients had an increased risk for presenting with advanced stage tumors and lymph node involvement compared with patients with private insurance. When stratified by tumor site, Medicaid and uninsured patients with laryngeal tumors were nearly 7 times as likely to present with advanced stage disease and were more than 4 times as likely to present with at least 1 positive lymph node than privately insured patients. Having Medicaid or being uninsured was not significantly associated with advanced stage or lymph node positivity at diagnosis for oral or pharyngeal SCCHN. The differences between the risks of advanced stage disease and lymph node involvement among different insurance can be explained by the differences in symptom presentation. Tumors of the oral cavity are often noted in their earlier stages by visual inspection during dental screenings, while tumors of the larynx often present earlier with hoarseness or changes in the voice. In contrast, nasopharyngeal and oropharyngeal tumors may present with nonspecific symptoms such as headache or ear pain or present late with symptoms of dysphagia or a neck mass.
This increased risk for advanced stage cancer with lymph node involvement may be from lack of access to screening procedures. For example, oral cancer screening is typically performed during routine dental cleanings, and Medicaid covers only limited dental care for patients under the age of 21.19 Ward et al9 found that uninsured and Medicaid-insured patients were less likely to be screened for breast and colon cancer and more likely to present with later stage cancer and experience lower survival rates; it is likely that this is true for head and neck cancers as well. Furthermore, the American Cancer Society statistics6 clearly correlate prognosis with the stage of cancer at diagnosis. For example, localized oropharyngeal cancer has an estimated 5-year survival of 82%, while regional and distant stages have 5-year survival of 52% and 27%, respectively, across all races. Thus, as uninsured or Medicaid patients present with more advanced cancers because of less access to healthcare, they are less likely to survive. A case-control study of patients in Cuba suggested that screening may be effective in reducing risk for advanced-stage presentation of oral cancers.20 Furthermore, a cluster-randomized control of patients in India trial suggested that screening for oral cancers in high-risk individuals may be effective for reducing mortality.21
Although insurance status is highly correlated to late stage cancer at presentation, the present multivariate analysis shows that uninsured/Medicaid and Medicare disability patients had decreased survival independent of cancer stage, smoking, alcohol use, age, gender, race, SES status, and treatment modality. Overall decreased survival for Medicaid and uninsured patients may be associated with delay in seeking or receiving treatment. Patients in this population are likely to have multiple barriers to receiving healthcare, including the high cost of healthcare, lack of transportation to medical or dental appointments, or the inability to leave work to attend appointments because of financial stresses. Increased mortality could also be a result of other comorbidities, as patients with limited access to healthcare may likely have other unmanaged medical problems. Increased mortality for Medicare disability patients compared with Medicare patients over age 65 may be attributed to the finding that patients who receive Medicare before the age of 65 are usually severely ill. To receive Medicare before the age of 65, there are several criterion, most of which include that the patient must meet the requirements of the Social Security disability program,22 which requires total, long-term disability.23 The finding that patients with Medicaid or no health insurance experience similar overall survival as a group known to carry severe comorbidities highlights the severity of the health disparities experienced by this patient population.
Health insurance status was not significantly associated with recurrence of head and neck cancer. This suggests that once an individual is diagnosed with SCCHN, they receive comparable care to those patients with insurance and, hence, their likelihood of disease recurrence is not significantly different. Risk of recurrence may be more associated with the extent of the disease rather than health insurance status, given the finding that in the present analysis relapse-free survival was adjusted for stage.
In the studies published so far, the nature of database limited the ability to assess the effect of insurance status on cancer outcomes independently from cancer risk factors. In contrast, the available data for our study allowed the assessment of the effect of insurance status on cancer outcomes (survival and relapse) while adjusting for cancer risk factors, such as smoking and alcohol use, as well as SES and treatments received. The large, clinically well-characterized patient population is also a unique strength of this study.
However, there are several limitations. As it has been reported by another study on the effects of different types of health insurance,4 Medicaid enrollment often happens for an uninsured patient at time of diagnosis. The group of Medicaid-insured patients may comprise a heterogeneous group of patients, some of which belonged to the uninsured group up to the time of diagnosis, thus introducing some misclassification in the statistical analysis. The effects of this misclassification may be illustrated by a study that suggested that patients enrolled in Medicaid postdiagnosis were more likely to have late stage disease compared with patients enrolled in Medicaid before diagnosis.24 In addition, the database used for the present analysis does not provide information on patient income, although SES was assessed here indirectly using Nakao and Treas occupation prestige scores, which correlates well with other commonly used measures of SES.16 Moreover, the database does not provide information on comorbidities. This is a limitation of the present study, as patients with certain insurance types, such as Medicaid and particularly Medicare disability, may have more comorbidities, which can affect overall patient survival.
Because smoking and alcohol use are significant risk factors for head and neck cancer, specifically for squamous cell carcinomas, it would have been useful to have more complete quantitative information on smoking and alcohol use; however, a limitation of the data set is the lack of information on the smoking dose in 25.5% of patients reporting a smoking history and on the alcohol intensity in 23.7% of patients reporting a history of alcohol use. Furthermore, reliable information on duration of tobacco and alcohol use is unavailable, thus preventing the assessment of lifetime alcohol and smoking exposure. Last, the time lag between diagnosis and surgical resection or other additional treatment (such as radiation or chemotherapy) was not available. This measurement may have allowed us to analyze any differences in time to treatment among the different types of insurance.
In conclusion, our analysis shows that patients with squamous cell carcinoma of the head and neck who are uninsured, receive Medicaid, or receive Medicare but are under the age of 65 have increased mortality when compared with those patients with private insurance, even after adjusting for covariates that are independently associated with outcome. This may reflect a lack of preventative care in these populations as well as barriers to healthcare that may delay treatment after diagnosis. Further investigation into our current Medicaid and Medicare disability system is warranted to determine exactly where improvements can be made to improve head and neck cancer outcomes for these populations.
Health insurance status is an independent predictor of head and neck cancer survival.
CONFLICT OF INTEREST DISCLOSURES
This article was partially supported by NIH grant numbers P20CA132385-01 (E.T.), P50CA097190 (J.G.), and 5TL1RR024155-03 (S.M.L.).