Evaluation of Treatment Patterns and Survival Outcomes in Elderly Pancreatic Cancer Patients: A Surveillance, Epidemiology, and End Results‐Medicare Analysis

In this study, SEER‐Medicare data were used to describe the patterns of care of elderly patients with pancreatic cancer based on stage at presentation. The demographic and clinical determinants of treatment receipt were evaluated, and survival was compared across patient‐, disease‐, and treatment‐specific groups of Medicare patients.

Abstract

Background.

Management of pancreatic cancer (PC) in elderly patients is unknown; clinical trials exclude patients with comorbidities and those of extreme age. This study evaluated treatment patterns and survival outcomes in elderly PC patients using linked Surveillance, Epidemiology, and End Results (SEER) and Medicare data.

Materials and Methods.

Histology codes 8140, 8500, 8010, 8560, 8490, 8000, 8260, 8255, 8261, 8263, 8020, 8050, 8141, 8144, 8210, 8211, or 8262 in Medicare Parts A and B were identified. Data regarding demographic, characteristics, treatments, and vital status between 1998 and 2009 were collected from the SEER. Determinants of treatment receipt and overall survival were examined using logistic regression and Cox proportional hazards models, respectively.

Results.

A total of 5,975 patients met inclusion. The majority of patients were non‐Hispanic whites (85%) and female (55%). Most cases presented with locoregional stage disease (74%); 41% received only chemotherapy, 30% chemotherapy and surgery, 10% surgery alone, 3% radiation, and 16% no cancer‐directed therapy. Patients with more advanced cancer, older age, and those residing in areas of poverty were more likely to receive no treatment. Among patients 66–74 years of age with locoregional disease, surgery alone (hazard ratio [HR] = 0.54; 95% confidence interval [CI]: 0.39–0.74) and surgery in combination with chemotherapy (HR = 0.69; 95% CI: 0.53–0.91) showed survival benefit as compared with the no treatment group. Among patients ≥75 years of age with locoregional disease, surgery alone (HR = 2.04; 95% CI: 0.87–4.8) or in combination with chemotherapy (HR = 1.59; 95% CI: 0.87–2.91) was not associated with better survival.

Conclusion.

Treatment modality and survival differs by age and stage. Low socioeconomic status appears to be a major barrier to the receipt of PC therapy among Medicare patients.

Implications for Practice.

Elderly patients with cancer are under‐represented on clinical trials and usually have comorbid illnesses. The management of elderly patients with pancreatic cancer is unknown, with many retrospective experiences but low sample sizes. Using Surveillance, Epidemiology, and End Results‐Medicare linked data to analyze treatment patterns and survival of elderly patients with pancreatic cancer on a larger population scale, this study highlights treatment patterns and their effect on survival and proposes possible obstacles to access of care in elderly patients with pancreatic cancer other than Medicare coverage.

Introduction

Pancreatic cancer (PC) is the third leading cause of cancer‐related mortality in the U.S. The outcome of PC has not significantly improved over the past decade, with 5‐year survival rates in the range of 8%–10%. Approximately 50% of newly diagnosed PC patients present with local‐ and regional‐stage disease. There is currently no consensus on the role of neoadjuvant therapy or radiation therapy in resected or locally advanced unresectable disease, or combination chemotherapy in nonmetastatic disease. Several randomized clinical trials have demonstrated benefit for systemic chemotherapy in the adjuvant setting [1], [2], [3], [4], [5].

Palliative chemotherapy remains the standard of care for patients with advanced‐stage disease. In recent years, two randomized trials have demonstrated a benefit of combination chemotherapy over single‐agent gemcitabine in patients with advanced‐stage PC [6], [7], [8], [9].

Although clinical standards of care for PC are largely based on the results of randomized clinical trials [6], [7], these trials enrolled a heavily selected group of patients. Patients with advanced age, high comorbidities, and low socioeconomic status (SES) are usually under‐represented on clinical trials. For this reason, information regarding treatment outcomes among older, community‐dwelling PC patients who receive care outside the setting of specialized clinical centers is largely lacking. The Surveillance, Epidemiology, and End Results (SEER)‐Medicare linked database allows this knowledge gap to be filled by providing valuable information on clinical demographic and treatment characteristics of an unselected sample of cancer patients over the age of 65 years.

In this study, SEER‐Medicare data were used to describe the patterns of care of elderly PC patients based on stage at presentation. We evaluated the demographic and clinical determinants of treatment receipt and compared PC survival across patient‐, disease‐, and treatment‐specific groups of Medicare patients.

Materials and Methods

Data Source

The SEER‐Medicare database is maintained by the National Cancer Institute (NCI) and is composed of linked information from registries included in the SEER program and Medicare claims data. The SEER program directs cancer surveillance activities in 18 geographical areas across the U.S., covering 28% of the U.S. population [10]. The available data include demographics, tumor characteristics, first course of treatment, and vital status [11]. The Medicare program is a federally funded health care program administered by the Centers for Medicare and Medicaid Services (CMS), which collect information on medical claims. Medicare covers people aged 65 years and older, persons under the age of 65 with certain disabilities, and end‐stage renal disease patients of any age [12]. CMS and NCI collaborate to link the Medicare claims files with data in the SEER registry [13]. For the purposes of the present study, we obtained the following Medicare files for all PC cases reported to the SEER between 1998 and 2009: home health agency, hospice, Medicare Provider Analysis Review, carrier claims, outpatient claims, and the Patient Entitlement and Diagnosis Summary File (PEDSF).

Study Population

This is a retrospective study of PC patients diagnosed between 1998 and 2009 in the SEER‐Medicare database. The inclusion criteria were as follows: (a) age 66 or older at time of PC diagnosis; (b) histology codes 8140, 8500, 8010, 8560, 8490, 8000, 8260, 8255, 8261, 8263, 8020, 8050, 8141, 8144, 8210, 8211, or 8262 (these codes exclude neuroendocrine tumors, enterochromaffin tumors, and lymphomas); (c) enrolled in Medicare Part A and Part B and no enrollment in a Health Maintenance Organization (HMO) for 12 months before and after PC diagnosis; (d) known month of PC diagnosis; and (e) PC diagnosis not made at autopsy or death certificate. Enrollment in Medicare Part A and Part B without concurrent enrollment in an HMO ensured complete documentation of treatments and comorbidities. Requiring cancer diagnosis at age 66 or older ensured accurate identification of comorbidities in the 12 months prior to cancer diagnosis (Medicare requires 12 months to come into effect, and the median survival is less than 12 months; that is why age 66 was chosen, instead of 65). Patient survival was followed from cancer diagnosis to death or until study completion (end of 2009). Data on sociodemographic characteristics of eligible subjects were obtained from the PEDSF file. The variables in this category included marital status, race, sex, and census tract poverty level. Census tract poverty was dichotomized as <20% and ≥20%, consistent with a definition of a “poverty area” [13].

Treatment‐ and Disease‐Specific Variables

The data for each cohort member were inspected to identify the types of treatment received using predefined Current Procedural Terminology, Healthcare Common Procedure Coding System, and International Classification of Diseases codes (supplemental online Table 1). Treatment was categorized according to five groups: (a) any radiation (with or without other types of therapy); (b) surgery alone; (c) surgery and chemotherapy; (d) chemotherapy alone; and (e) no tumor‐directed treatment. Tumor grade was dichotomized as low (grade 1 or 2) versus high (grade 3 and 4). Patients for whom the grade was unavailable or undetermined were excluded from the analyses. Tumor stage was dichotomized as local/regional versus distant. The Charlson rule‐out and comorbidity macros from the NCI SEER‐Medicare program [14] were run in SAS 9.3 (SAS Institute, Cary, NC) to determine the Charlson Comorbidity Score for each patient. Charlson scores were then categorized as 0, 1, or 2+ [15].

Statistical Analyses

Age‐ and stage‐stratified Kaplan‐Meier survival curves were constructed for the five treatment categories. A binary logistic regression model was used to assess demographic and disease‐specific factors associated with treatment receipt. For this model, the outcome variable was receiving any versus no tumor‐directed treatment. The independent variables of interest were age, sex, race, marital status, census‐tract poverty, Charlson Comorbidity Score, tumor grade, and stage of the disease at presentation. Additionally, a polytomous logistic regression was conducted assessing the same demographic and disease‐specific determinants and comparing each treatment category individually to the outcome reference of no treatment category. The results of the logistic regression analyses were presented as adjusted odds ratios (aORs) and the corresponding 95% confidence intervals (CIs). All logistic regression models were examined for collinearity and interaction. All analyses were conducted in SAS 9.3.

Cox proportional hazards survival models were used to calculate the unadjusted hazard ratios (HRs) and the corresponding 95% CIs for each treatment category using the “no treatment” group as reference. Sociodemographic and disease‐specific variables did not satisfy proportional hazards assumptions; therefore, stratum‐specific, rather than adjusted, hazard ratios were calculated. In the stratified survival analyses, age was dichotomized at the median (75 years), and stage was categorized as local/regional and distant.

Results

Patient Characteristics

A total of 5,975 patients met the criteria for inclusion. Patients’ demographic characteristics across treatment categories are presented in Table 1. The majority of patients were non‐Hispanic whites (n = 5,087, 85%), females (n = 3,288, 55%), and married (n = 3,611, 60%). Most patients presented with local or regional disease (n = 4,422, 74%) rather than distant‐stage PC (n = 1,553, 26%). The most frequent type of treatment was chemotherapy alone (n = 2,452, 41%), followed by chemotherapy and surgery (n = 1,821, 30.4%), and surgery alone (n = 587, 10%); 982 (16%) patients did not receive any cancer treatment. Only 3% of patients received radiation therapy with or without other treatment modalities. Almost half of the patients had tumor of undetermined grade (n = 2,781, 47%). Of the patients with known cancer grade, 65% (n = 2,089) had low‐grade disease (the distribution of chemotherapy treatments by age is shown in supplemental online Table 2). The most common chemotherapy agent was gemcitabine used as single agent (49%), followed by the combination of gemcitabine and fluorouracil (23%).

Table 1. Patients’ demographic characteristics according to treatment received.

‐ Column numbers suppressed due to multiple cells with <11 observations

* Cells with number < 11 suppressed.

Logistic Regression—Any Treatment

The results of the binary logistic regression model comparing any versus no cancer‐directed treatment as the outcome of interest are presented in Table 2. Sex (aOR = 0.84; 95% CI: 0.60–1.2), race (aOR = 0.92; 95% CI: 0.60–1.4), and marital status (aOR = 1.19; 95% CI: 0.85–1.66) were not associated with the type of treatment received. Patients over 80 years of age were significantly less likely to receive treatment compared with younger counterparts. Patients from census tracts categorized as “poverty areas” were significantly less likely to receive treatment (aOR = 0.33; 95% CI: 0.23–0.48) than those residing in census tracts with <20% residents living below the poverty level.

Table 2. Logistic regression model for receiving any type of treatment.

OR >1 indicates higher odds of receiving treatment.

^aAdjusted OR p < .05.

Abbreviations: CI, confidence interval; OR, odds ratio; ref, reference.

Compared with patients with no comorbidities, the aORs (95% CI) for those with Charlson Comorbidity Scores of 1 or 2+ were 1.33 (0.90–1.96) and 0.70 (0.48–1.03), respectively. Tumor grade was not associated with receiving treatment (aOR = 1.02; 95% CI: 0.73–1.42); however, patients with distant‐stage disease were about half as likely to receive therapy (aOR = 0.50; 95% CI: 0.35–0.72) as those diagnosed with local or regional cancer.

Polytomous Logistic Regression

The results of the polytomous logistic regression are presented in Table 3. No differences were observed for any of the treatment categories based on race, marital status, or sex. Patients with a Charlson Comorbidity Score of 2 or more were less likely to receive surgery plus chemotherapy compared with patients with no comorbidities (aOR = 0.67; 95% CI: 0.45–0.99). For all treatment categories, patients living in census tracts with ≥20% of residents living in poverty were less likely to receive therapy. Younger age was associated with greater odds of receiving treatment for all treatment categories except surgery alone. Using age of 80+ years as a reference, patients 70–79 years of age were more likely to receive surgery alone, whereas no difference was observed for the 65–69 or 70–74 age groups.

Table 3. Polytomous logistic regression for each treatment group (vs. no treatment) as the dependent variable of interest.

OR > 1 indicates higher odds of receiving treatment.

^aIndicates statistical significance (p < .05) for at least one treatment category.

Abbreviations: CI, confidence interval; OR, odds ratio; ref, reference.

Survival Outcomes

The results of age‐ and stage‐stratified survival analyses across treatment categories are presented in Figure 1 as survival curves and in Table 4 as stratified hazard ratios. Among patients 66–74 years of age with local‐/regional‐stage tumors, patients treated with surgery alone (HR = 0.54; 95% CI: 0.39–0.74) and surgery in combination with chemotherapy (HR = 0.69; 95% CI: 0.53–0.91) showed statistically significantly better survival compared with the no treatment group. Survival among patients receiving chemotherapy alone (HR = 1.40; 95% CI: 1.05–1.87) in this age and stage‐specific group was worse than among those receiving no treatment. No difference in survival was observed among patients receiving radiation. For patients 66–74 years of age with distant‐stage disease, there was no difference in survival among patients receiving chemotherapy alone versus those receiving no treatment (HR = 1.18, 95% CI: 0.91–1.53).

Figure 1.

Survival curves for the four treatment categories. Left curves, for local/regional stages, right for curves distant stage. Upper curves for ages 66–74, lower curves for ages >75 years.

Table 4. Hazard ratios assessing the association between each treatment modality (vs. no treatment) and overall mortality stratified by age and tumor stage.

Abbreviations: CI, confidence interval; HR, hazard ratio.

Among patients 75 years of age and older with local‐ or regional‐stage disease, survival of patients undergoing surgery (HR = 2.04; 95% CI: 0.87–4.8) or surgery in combination with chemotherapy (HR = 1.59; 95% CI: 0.87–2.91) was not significantly better than in the no treatment group. Patients receiving radiation (HR = 5.13; 95% CI: 2.09–12.56) or chemotherapy alone (HR = 2.62; 95% CI: 1.48–4.65) had worse outcomes compared with the no treatment group. Survival outcomes were not statistically different for any treatment categories among patients 75 years of age and older with distant‐stage tumors.

Discussion

This study evaluated treatment patterns and survival outcomes among a population‐based cohort of elderly PC patients. A substantial proportion of patients, even with early‐stage disease, did not receive any tumor‐directed therapy. The factors associated with no treatment included age over 80 years, high Charlson Comorbidity Score, distant‐stage disease, and low socioeconomic status. This study has certain limitations, which include the retrospective study design, missing data, and miscoding of the disease, its stage, and treatments. Patients with local‐ and regional‐stage diseases included patients who have resectable, borderline resectable, and locally advanced unresectable‐stage PC. It is difficult to account for these stages even in the newest staging criteria because there is no consensus in the staging. This is reflected in the low number of elderly patients (41%) receiving surgery with the local and regional stages. There was an imbalance between the treatment groups analyzed, specifically the radiation treatment group. The SEER‐Medicare data excludes care provided in other settings, such as the Veterans Administration, care for persons with Medicare provided by a secondary payer, and Medicare patients who hold Medigap policies. Data analyzed did not include the details of therapy such as the type of radiation, total dose, timing and location of the radiation delivered, dose of chemotherapy used, number of cycles, and the toxicities and morbidities of the treatments used. The study excluded patients diagnosed after 2009, thereby excluding patients treated with newer combination chemotherapies that have shown survival benefit. Disparities in the receipt of treatment among the SEER‐Medicare population and the limitations of this study are important to understand in future trials utilizing this database. In SEER‐Medicare, treatment allocation is not random, so survival analysis can be confounded by selection bias or by factors that modify the probability of receiving a certain treatment.

The lack of effective therapies and rapid decline in performance status may have contributed to patients with metastatic disease not receiving treatment. Similarly, patients with significant comorbidities may be poor candidates for surgery or systemic therapy. Although it is reassuring that race and gender did not affect treatment, it is concerning that patients from census tracts with ≥20% of residents living below the poverty level were significantly less likely to receive treatment of any kind: chemotherapy, surgery, or radiation. This is in line with previous observations for all cancer patients showing significantly shorter survival times among patients with lower annual incomes or lower educational level than among those with higher income or education, respectively [16], [17], [18], [19].

The fact that the patient population in this study had Medicare coverage for the treatments given (chemotherapy, surgery, or radiation therapy) highlights the importance of factors beyond insurance in the access to health care. There are other indirect challenges to health care access that individuals of low SES face, beyond simply their ability to afford the actual treatment. Issues such as access to transportation play a significant role in health care. A shortage of physicians and nurses exists at higher rates in poor areas, and not all medical providers are willing to treat people with Medicare coverage. Many low SES individuals are unaware of symptoms that need medical attention; this can also be a result of a lower education level, which is more likely among people living in poor areas. Others face language or cultural barriers in seeking the medical care required [20]. More importantly, people who have pancreatic cancer are debilitated enough to be at risk of loss of employment and the associated income losses. Treatment itself requires time away from work, which low SES individuals are less likely to be able to afford. People with low SES tend to have poorer health outcomes compared with those of higher SES [21], which is associated with higher comorbidity scores. Last, although Medicare provides basic health insurance, it is not an all‐inclusive, comprehensive, or free medical plan for the elderly poor [22]. The trend that higher income status is associated with higher receipt of treatment is consistent with previous reports [23], [24]. Our study suggests that high treatment cost is not the only contributor to the disparity in treatment in the elderly population, and that other indirect costs could significantly affect access to treatment.

Chemotherapy was the predominant treatment used in this group of patients. This is in agreement with current standards of care in which chemotherapy has shown benefit in early‐stage and metastatic disease. Similar to previous reports [24], the chemotherapy regimen of choice during the time period of this analysis was gemcitabine. This is a reflection of the pattern of care and may also be related to a reluctance to use combination therapy in elderly patients. Radiation therapy was only used in a small subset of patients. The negative results of recent trials evaluating radiation in the adjuvant setting (EORTC and ESPAC 1) [2], [25] and in locally advanced disease may have contributed to the low utilization of radiation therapy [26].

The results of this study indicate that radiation therapy did not add significantly to survival outcomes in patients 66–74 years of age and may have been harmful in patients over age 75 (HR = 5.13). This is likely related to the fact that patients might have received radiation for palliation of pain or because patients were frail with co‐morbidities making them not candidates to receive surgery of chemotherapy. Radiation therapy adds to the cost of treatment [27] and may have a potential negative impact on quality of life. Based on these facts, prospective trials are still needed to identify the role of radiotherapy in this population prior to its utilization as part of standard therapy.

Surgery provides a clear survival advantage for patients 66–74 years of age. The addition of chemotherapy to surgery in this group clearly improves outcomes compared with no treatment but does not clearly improve outcomes compared with surgery alone. Single‐agent gemcitabine is the approved treatment in the adjuvant setting for resected PC [28]. The benefit from adjuvant chemotherapy has been confirmed in several randomized trials. The difference in outcome between the previous trials and our study may be due to several factors. First, patient selection: randomized trials enroll patients according to strict criteria with timely recovery from surgery, whereas a population‐based study does not exclude patients who have poor recovery after surgery. Second, the median age of patients on adjuvant trials was 62 (Conko‐001) and 65 (ESPAC 4) years, in comparison with our population median age of 75.2 years. Last, the impact of adjuvant therapy may differ between elderly and younger patients. Addressing this question will require conducting a prospective adjuvant trial in the elderly population.

In the age group of >75 years, surgery, radiation, and/or chemotherapy did not provide significant survival advantage over no therapy. This lack of benefit was seen in both early‐ and advanced‐stage disease. Our results highlight the potential risk of applying the results of randomized trials to establish standards of care for the elderly population. The outcomes of patients with advanced‐stage pancreatic and lung cancer on clinical trials was compared with the outcomes of patients treated with the same regimens in the SEER‐Medicare database [29]. This study demonstrated that clinical trials tended to overestimate survival for older Medicare patients. The economic burden of pancreatic cancer in elderly patients is substantial [30]. The mean age of advanced‐stage PC patients enrolled on combination clinical trials was 61–63 years. In the ACCORD‐PRODIGE trial (5‐fluorouracil, leucovorin, irinotecan, oxaliplatin), only 98 of 342 patients (29%) were older than 65 years. In the MPACT trial (gemcitabine/nab‐paclitaxel), the median age of the patients was higher than that in PRODIGE; in fact, 365 patients (42%) were at least 65 years of age, and 10% of the patients were 75 years or older. This under‐representation of the elderly population with PC generates challenges because the results from clinical trials in younger patients cannot be extrapolated to the treatment of the elderly. This analysis of the SEER‐Medicare data was up to the year 2009, when combination chemotherapies were not standard of care, reflecting the low numbers of patients receiving combination in both age cohorts (<75 and >75).

Conclusion

The potential lack of benefit and high cost of current treatment strategies in pancreatic cancer patients over the age of 75 highlights the importance of designing trials to evaluate the best therapeutic approach in this specific population. With the development of new and more intensive treatment strategies (such as combination chemotherapies, stereotactic body radiation, proton therapy, etc.), greater efforts are required that focus on understanding the barriers and identifying the causes of disparities in care for elderly patients with pancreatic cancer.

See http://www.TheOncologist.com for supplemental material available online.

Author Contributions

Conception/design: Walid L. Shaib

Collection and/or assembly of data: Walid L. Shaib

Data analysis and interpretation: Jeb S. Jones, Michael Goodman

Manuscript writing: Walid L. Shaib, Juan M. Sarmiento, Shishir K. Maithel, Kenneth Cardona, Sujata Kane, Christina Wu, Olatunji B. Alese, Bassel F. El‐Rayes

Final approval of manuscript: Walid L. Shaib, Jeb S. Jones, Michael Goodman, Juan M. Sarmiento, Shishir K. Maithel, Kenneth Cardona, Sujata Kane, Christina Wu, Olatunji B. Alese, Bassel F. El‐Rayes

Disclosures

Christina Wu: Vaccinex, Bristol‐Myers Squibb, Boston Biomedical (RF). The other authors indicated no financial relationships.

(C/A) Consulting/advisory relationship; (RF) Research funding; (E) Employment; (ET) Expert testimony; (H) Honoraria received; (OI) Ownership interests; (IP) Intellectual property rights/inventor/patent holder; (SAB) Scientific advisory board