Error bars indicate 95% confidence intervals as determined by the adjusted Wald method.
Shapiro M, Chen Q, Huang Q, Boosalis VA, Yoon CH, Saund MS, Whang EE, Gold JS. Associations of Socioeconomic Variables With Resection, Stage, and Survival in Patients With Early-Stage Pancreatic Cancer. JAMA Surg. 2016;151(4):338-345. doi:10.1001/jamasurg.2015.4239
Socioeconomic variables including sex, race, ethnicity, marital status, and insurance status are associated with survival in pancreatic cancer. It remains unknown exactly how these variables influence survival, including whether they affect stage at presentation or receipt of treatment or are independently associated with outcomes.
To investigate the relationship between socioeconomic factors and odds of resection in early-stage, resectable pancreatic adenocarcinoma and to determine whether these same factors were independently associated with survival in patients who underwent resection.
Design, Setting, and Participants
This was a retrospective cohort study of patients diagnosed as having T1 through T3 M0 pancreatic adenocarcinoma between January 1, 2004, and December 31, 2011, identified from the Surveillance, Epidemiology, and End Results database.
Main Outcomes and Measures
Socioeconomic and geographic variables associated with utilization of resection and disease-specific survival.
A total of 17 530 patients with localized, nonmetastatic pancreatic cancer were identified. The resection rate among these patients was 45.4% and did not change over time. Utilization of resection was independently associated with white vs African American race (odds ratio [OR] = 0.76; 95% CI, 0.65-0.88; P < .001), non-Hispanic ethnicity (for Hispanic, OR = 0.72; 95% CI, 0.60-0.85; P < .001), married status (OR = 1.42; 95% CI, 1.30-1.57; P < .001), insurance coverage (OR = 1.63; 95% CI, 1.22-2.18; P = .001), and the Northeast region (vs Southeast, OR = 1.67; 95% CI, 1.44-1.94; P < .001). Stage at presentation correlated with sex, race, ethnicity, marital status, and geographic region (ethnicity, P = .003; all others, P < .001); however, the factors associated with increased resection correlated with more advanced stage. Patients who underwent resection had significantly improved disease-specific survival compared with those who did not undergo resection (median, 21 vs 6 months; hazard ratio [HR] for disease-specific death = 0.32; 95% CI, 0.31-0.33; P < .001). Disease-specific survival among the patients who underwent surgical resection was independently associated with geographic region, with patients in the Pacific West (HR for death = 0.706; 95% CI, 0.628-0.793), Northeast (HR for death = 0.766; 95% CI, 0.667-0.879), and Midwest (HR for death = 0.765; 95% CI, 0.640-0.913) having improved survival in comparison with those in the Southeast (all P < .001).
Conclusions and Relevance
Disparities in the utilization of surgical resection for patients with early-stage, resectable pancreatic cancer are associated with socioeconomic variables including race, ethnicity, marital status, insurance status, and geographic location. Of these factors, only geographic location is independently associated with survival in patients undergoing resection.
Pancreatic adenocarcinoma has one of the poorest outcomes of all malignant neoplasms, with a 5-year survival of 7%.1 While pancreatic cancer is currently the fourth leading cause of cancer-related death, it has been projected that by 2030 it will become the second leading cause.2 Unfortunately, most patients present with unresectable disease, which severely limits therapeutic options and the likelihood of cure.
Recently, several authors have shown that various socioeconomic variables such as marital status, race, ethnicity, insurance status, and level of education predict poor outcome in this and other cancers.3- 8 The precise nature of the associations between these variables and survival remains unclear. It is possible that these associations are due to differences in stage at presentation and receipt of treatment or are independently associated with outcome. Survival in pancreatic cancer is largely based on clinical disease characteristics such as tumor size, grade, and lymph node status, all of which help determine treatment modalities. In turn, the receipt of treatment, particularly surgical resection, strongly influences outcome.
Unfortunately, the rate of surgical resection remains low, even among patients with early-stage disease. Using the National Cancer Data Base, Bilimoria et al4 found that of all stage I pancreatic adenocarcinoma diagnosed between 1995 and 2004, 71% of patients did not undergo surgery. While a more recent study using data from the Surveillance, Epidemiology, and End Results (SEER) database from 2004 through 2007 shows an increased rate of resection, it still indicates that only slightly more than half of patients with localized disease are undergoing resection.9 Recent investigations have uncovered associations of rates of resection with patient socioeconomic and demographic characteristics, including race, ethnicity, sex, and insurance status.3,5,8,10- 15 The goal of this study was to determine whether socioeconomic factors are associated with disparities in the utilization of surgical resection in early-stage pancreatic cancer and whether these variables are independently associated with cancer-specific survival in patients selected to undergo resection.
Patients diagnosed between January 1, 2004, and December 31, 2011, as having T1 through T3 pancreatic adenocarcinomas (International Classification of Diseases for Oncology, Third Edition codes 8000-5, 8010-15, 8020-2, 8030-5, 8046, 8050, 8070-8074, 8140-45, 8260-63, 8310, 8323, 8440, 8450-3, 8470-71, 8480-1, 8490, 8500-4, 8510, 8521-3, 8550, 8560, 8570, 8574) without distant metastasis at diagnosis were extracted from the SEER database. Cases in which the surgical resection status was unknown or in which a local excision was performed were removed. Similarly, cases in which pancreatic cancer was not the first cancer diagnosed in the patient’s lifetime were excluded as treatment decisions and survival could be affected by the other malignant neoplasms. Cases diagnosed at autopsy were excluded. This research was approved by the VA Boston Healthcare System Research and Development Committee and was deemed exempt from institutional review board oversight. Informed consent was not required because the study used deidentified data from the SEER database.
All variables used in analysis were available in the database. Only patients who received external beam radiation were included as having received radiation. Information on chemotherapy administration and patient comorbidities was not available in the SEER database.
Geographic region was divided into 4 areas based on the original registry location. All active SEER database registries, encompassing approximately 28% of the US population, were included. The area defined as Pacific West included San Francisco, California; Oakland, California; Hawaii; New Mexico; Seattle, Washington; Utah; Alaska; San Jose/Monterey, California; Los Angeles, California; and Greater California. The Midwest included metropolitan Detroit, Michigan, and Iowa. The Southeast included metropolitan Atlanta, Georgia; rural Georgia; Kentucky; Louisiana; and Greater Georgia. The Northeast included Connecticut and New Jersey.
Disease-specific survival was calculated using the SEER cause-specific death classification data.
A 1-way analysis of variance was performed to examine the linear trend for resection over time. The t test and Pearson χ2 test were performed for continuous and categorical variables, respectively, to test associations of socioeconomic variables with stage of disease. Univariate and multivariate logistic regressions were used to determine associations of socioeconomic variables with resection. Possible interaction terms were examined during model development; however, none were statistically significant. The log-rank test and Cox regression models were used to determine associations with cancer-specific survival. P < .05 was considered statistically significant. Statistical analysis was performed with SAS version 9.2 (SAS Institute, Inc) and IBM SPSS Statistics version 22 (IBM Corp) software.
A total of 17 530 patients with localized, nonmetastatic pancreatic cancer (defined as T1-T3, M0) diagnosed between January 1, 2004, and December 31, 2011, were identified from the SEER database (Table 1). Among these patients, 47.8% were male and the median age was 70 years (interquartile range [IQR], 61-79 years). The majority of the patients were white (14 147 of 17 463 patients [81.0%]). More than half the patients (9653 of 16 962 [56.9%]) were married, and only 295 of 11 416 patients (2.6%) were uninsured. The median tumor size was 3.4 cm (IQR, 2.5-4.4 cm), with 12 795 of 16 401 patients (78.0%) presenting with stage II disease. The majority of the tumors were located in the head of the pancreas (12 923 of 15 285 tumors [84.5%]). Among 16 892 patients, 5063 (30.0%) underwent external beam radiation. The rate of surgical resection was 45.4% without significant variation over time during this study (Figure 1).
We first aimed to determine the factors associated with the utilization of surgical resection (Table 2). We found that all of the socioeconomic and demographic variables included except year of diagnosis and insurance status were associated with resection on univariate analysis. Patients had an increased likelihood of resection if they were male (odds ratio [OR] = 1.20; 95% CI, 1.13-1.27); younger (OR = 0.95; 95% CI, 0.94-0.95); white (OR = 1.16; 95% CI, 1.08-1.26) or non–African American (for African American, OR = 0.80; 95% CI, 0.73-0.88); non-Hispanic (for Hispanic, OR = 0.83; 95% CI, 0.74-0.93); married (OR = 1.78; 95% CI, 1.67-1.89); or located in the Northeast (OR = 1.62; 95% CI, 1.50-1.76) (all P < .001). Tumor size and tumor location were also associated with resection, with smaller tumors and tumors located in the body and tail more likely to be resected (tumor size, OR = 0.995; 95% CI, 0.994-0.997; tumors in head, OR = 0.74; 95% CI, 0.68-0.80; P < .001 for both). Tumor grade was similarly found to be associated with resection on univariate analysis (for well differentiated vs moderately differentiated, poorly differentiated, and undifferentiated, OR = 1.29; 95% CI, 1.17-1.42; P < .001); however, the vast majority of patients with information regarding grade had undergone resection (7194 of 9454 patients [76.1%]), possibly skewing the results, and thus tumor grade was not included in subsequent analysis.
We then performed a multivariate analysis including all of the socioeconomic variables as well as the tumor characteristics that were significant on univariate analysis. We found that the likelihood of resection was independently associated with race, ethnicity, marital status, geographic location, and insurance status. African American patients were less likely to undergo resection compared with the white population (OR = 0.76; 95% CI, 0.65-0.88; P < .001). While the odds of resection were lower for American Indian patients, the difference did not reach statistical significance (OR = 0.65; 95% CI, 0.36-1.17; P = .15), possibly owing to the small population size represented. Hispanic patients had a decreased rate of resection compared with non-Hispanic patients (OR = 0.72; 95% CI, 0.60-0.85; P < .001). Married patients were more likely to undergo resection than unmarried patients (OR = 1.42; 95% CI, 1.30-1.57; P < .001). The rate of resection in the Northeast was higher compared with the Southeast (OR = 1.67; 95% CI, 1.44-1.94; P < .001). While an association between insurance status and resection was not significant on univariate analysis, insurance status was independently associated with surgical resection when included in the multivariate model, with patients having insurance also having higher odds of resection (OR = 1.63; 95% CI, 1.22-2.18; P = .001). Tumor characteristics, such as tumor size (OR = 0.990; 95% CI, 0.986-0.993; P < .001) and location in the pancreas (head, OR = 0.74; 95% CI, 0.65-0.83; P < .001), were also associated with resection, as was patient age (OR = 0.95; 95% CI, 0.94-0.95; P < .001).
We next examined for associations between the socioeconomic factors and stage of disease, as tumor stage can affect both the odds of resection and overall survival (Table 3). All the factors examined except insurance status were related to stage at presentation (insurance status, P = .52; ethnicity, P = .003; all others, P < .001). Male, white, non-Hispanic, and married patients all had higher rates of stage IIB disease compared with those who were female; African American, American Indian, and Asian/Pacific Islander; Hispanic; and unmarried, respectively. Patients in the Northeast had a higher rate of stage IIB tumors and a slightly lower rate of stage I tumors compared with patients in other regions.
We then tested associations with survival. All patients had survival data available. The median follow-up of all patients was 8 months (IQR, 3-17 months), and the median follow-up of patients without disease-specific death was 12 months (IQR, 4-30 months). A total of 12 179 patients (69.5%) died of disease (1049 patients [6.0%] died of other causes). Not surprisingly, patients who underwent surgical resection had increased survival compared with those who did not (hazard ratio [HR] for disease-specific death = 0.32; 95% CI, 0.31-0.33; P < .001) (Figure 2). The median disease-specific survival of patients who underwent resection was 21 months (95% CI, 20.3-21.7), while the median disease-specific survival of patients who did not was 6 months (95% CI, 5.8-6.2).
As several socioeconomic and demographic variables purported to be associated with survival were associated with utilization of a treatment (surgical resection), we next sought to explore whether these variables were independently associated with survival in the patients who received this treatment. We constructed a multivariate survival model including relevant patient, tumor, and treatment variables using the resected cohort (Table 4). The median follow-up of patients who underwent resection was 15 months (IQR, 7-27 months), with a follow-up of 18 months (IQR, 7-38 months) for patients without disease-specific death. Of 7951 patients who underwent resection, 4482 (56.4%) died of disease (420 patients [5.3%] died of other causes). Our analysis revealed that of the socioeconomic and demographic factors assessed, only geographic location was independently associated with disease-specific survival. Patients located in the Southeast had the greatest likelihood of disease-specific death with comparatively longer survival in the Pacific West (HR = 0.706; 95% CI, 0.628-0.793; P < .001), the Northeast (HR = 0.766; 95% CI, 0.667-0.879; P < .001), and the Midwest (HR = 0.765; 95% CI, 0.640-0.913; P < .001). Sex, race, ethnicity, marital status, and insurance status were not independently associated with disease-specific survival after resection.
Pancreatic cancer remains a major cause of cancer-related death. As progress is made in the prevention and treatment of other malignant neoplasms without comparable progress in pancreatic cancer, its relative burden continues to increase. While breakthroughs in prevention and treatment are desperately needed for this disease, in the absence of such advances, progress can still be made by addressing disparities in the delivery of available treatment.
Recently there has been considerable attention directed at the effects of socioeconomic factors on survival for pancreatic cancer and other malignant neoplasms. Nevertheless, it is not completely understood how socioeconomic variables influence outcome. It is possible that associations between socioeconomic variables are mediated through disparities in the receipt of treatment or that these socioeconomic variables are independently associated with outcome. In this study, we investigated associations between socioeconomic variables and the utilization of surgical resection in resectable, early-stage pancreatic cancer as well as associations between these socioeconomic variables and survival in patients selected for surgical resection. Using a population-based registry, our research revealed that race, ethnicity, marital status, insurance status, and geographic location were all independently associated with resection. The associations between socioeconomic factors and stage of disease revealed that in most cases, factors associated with increased rates of resection were also associated with tumor stage; however, opposite of what would be expected, these factors were associated with more advanced stage. Thus, associations between socioeconomic factors and stage do not explain the disparities in utilization of resection for resectable pancreatic cancer. Finally, we determined that of all of the socioeconomic variables examined, only geographic location was associated with increased disease-free survival among patients who underwent resection.
Various studies have examined racial disparities in relation to pancreatic cancer treatment.3,5,10,13,14 Historically, black patients have been less likely to undergo resection than their white counterparts.5,11,16 Riall et al5 used SEER-Medicare linked data to look specifically at racial disparities in the treatment of pancreatic cancer and found that black patients diagnosed as having locoregional pancreatic cancer were less likely than white patients to be seen by a surgeon and were less likely to undergo resection if they did visit a surgeon. Our study similarly found that African American patients had decreased rates of resection, indicating a persistent racial treatment bias. Thus, the racial disparity in the utilization of resection represents an area for improvement in the care of patients with pancreatic cancer.
Disparities in the use of resection for pancreatic cancer have also been attributed to economic status, education level, and insurance coverage. Patients with lower annual incomes and/or lower levels of education have been found to be less likely to undergo resection.4 Smith et al8 investigated an association between insurance coverage and resection and found that pancreatic cancer surgery was more likely to be recommended and performed in areas of middle to high insurance coverage compared with areas of low coverage. We similarly found that patients who were insured were more likely to undergo resection than those who were uninsured. It is possible that patients who are uninsured are less likely to find a surgeon willing to perform a pancreatic cancer resection, which is a costly procedure that requires a significant time commitment from the surgeon for perioperative care.
Given that pancreatic cancer treatment ranges from radiation and chemotherapy to surgical procedures that have high morbidity, having strong social support is likely an important factor in making the decision to undergo treatment. Being married is typically indicative of a base level of social support from which most single Americans do not benefit. Aizer et al7 recently examined the connection of marital status with disease presentation and disease treatment. The trends were similar in all cancers. For pancreatic cancer in particular, they found that married patients were more likely to undergo definitive treatment, including surgical intervention. We also found that married patients were more likely to undergo surgery, independent of tumor characteristics and patient demographic characteristics. This indicates that assessing the social needs of unmarried patients is likely warranted to ensure that they are willing and able to undergo resection of pancreatic cancer.
Among the most novel findings in this study are the regional differences in the likelihood of surgical resection for pancreatic cancer that were present even after controlling for available socioeconomic factors and tumor characteristics. We found that patients in the Northeast were more likely to undergo surgery compared with those in the Pacific West, Midwest, and Southeast. A potential explanation for our results is variation in practice trends and availability of specialized surgeons and centers in the geographic regions sampled. Pancreatic surgery is technically difficult, and surgeons without specialty training in pancreatic cancer surgery may be less likely to offer resection. Furthermore, surgeons in academic hospitals are more likely than those in community hospitals to offer surgery to a potentially resectable patient. Further testing the hypothesis that regional variations in concentrations of specialized surgeons and centers account for the regional differences in utilization of resection for pancreatic cancer would require gathering specific information regarding the category of hospital where pancreatic surgery is performed (eg, community vs teaching hospital) as well as the number of subspecialty-trained pancreatic surgeons in each hospital. These data are not contained in the SEER database.
As tumor stage could likely influence resection rates, we examined the relationship between stage and socioeconomic variables. We found that patients who were male, married, non-Hispanic, or white presented with later-stage cancer compared with those who were female; single; Hispanic; or African American, American Indian, and Asian/Pacific Islander, respectively. Although some studies show that African American patients tend to present at a later stage of disease,1,17 we found that both African American and Hispanic patients had slightly lower rates of advanced-stage disease compared with white and non-Hispanic patients. While previous studies have found that being married is associated with earlier disease presentation,7,13 we found that married patients in our study had increased rates of advanced-stage disease compared with unmarried patients. We did not find a difference in cancer stage at diagnosis between patients who were insured vs those who were uninsured. Finally, we looked at stage in relation to geographic region. We found that patients located in the Northeast had the highest concentration of advanced tumors but the highest odds of undergoing resection. These discrepancies are possibly related to the fact that we limited our analysis to stage IA through IIB disease. It should be noted that the relevance of the associations with stage may be limited given that the differences between groups were small. Nevertheless, these results indicate that the disparities in the utilization of resection for early pancreatic cancer cannot be explained by different stage at presentation of disease and lend increased support to inherent racial, socioeconomic, and regional biases in the treatment decisions.
Our research further confirmed that, as expected, resection is associated with improved survival. Many studies have concluded that socioeconomic factors that tend to lead to more aggressive treatment are similarly associated with increased cancer survival. What has not been determined is the nature of this association. Thus, in this study we specifically examined whether socioeconomic variables were independently associated with survival in patients selected for resection of pancreatic cancer. We found that after accounting for tumor characteristics and receipt of radiation, the only factor independently associated with disease-specific survival was geographic location. Survival was improved in the Pacific West, the Northeast, and the Midwest compared with the Southeast. There was no significant association between survival and race, ethnicity, level of insurance coverage, or marital status. We are not aware of other studies demonstrating such an association between geographic location and disease-specific survival for pancreatic cancer. Surgical and oncologic care of patients with pancreatic cancer is complex, and the best outcomes are generated from high-volume centers.18 While in this study we could not examine location and concentration of high-volume hospitals in relation to outcomes, it is possible that such a difference could explain these results. Further validation of regional differences in survival after resection of pancreatic cancer and study of the reasons accounting for these differences are needed.
There are several limitations inherent to this study. The SEER database does not contain information on chemotherapy administration, which can also contribute to outcomes. It also does not include information on patient comorbidities, nor does it offer in-depth explanations of why patients do or do not undergo treatment. The lack of data on comorbidities, for instance, may influence the association between socioeconomic variables and the receipt of surgical resection, as the variables that are associated with increased likelihood of resection may identify a healthier cohort who are more likely to be offered surgery. Furthermore, data regarding margin status following surgery are not available. An additional limitation is that we did not include tumor stage in our analysis of factors associated with the utilization of resection. Staging based on imaging may not always be comparable to that determined pathologically; thus, comparison between resected and unresected patients may not be appropriate. However, we did include tumor size in this analysis, as size can be reasonably estimated clinically. It should be noted that while the SEER registry directs registrars to include pathologic size when available, tumor size based on imaging is also allowed. No indication is given regarding how size is determined for individual patients. Presumably, pathologic size is used for the patients who underwent resection, and those who did not undergo resection likely had tumor size derived from radiologic studies. There may be differences between these 2 measurements of size and even between pathologic measurements of tumor size depending on whether the tumor is measured before or after fixation.
To our knowledge, this is the first study demonstrating that while socioeconomic factors are associated with whether patients with localized pancreatic adenocarcinoma undergo therapeutic resection, only geographic location is independently associated with disease-specific survival in patients selected to undergo resection. Understanding the factors involved in treatment and survival of patients with pancreatic cancer is an essential part of targeting areas for improvement of outcomes with this disease.
Corresponding Author: Jason S. Gold, MD, Surgery Service, VA Boston Healthcare System (112), 1400 VFW Pkwy, West Roxbury, MA 02132 (firstname.lastname@example.org).
Accepted for Publication: July 23, 2015.
Published Online: November 18, 2015. doi:10.1001/jamasurg.2015.4239.
Author Contributions: Drs Shapiro and Chen 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: Shapiro, Yoon, Saund, Gold.
Acquisition, analysis, or interpretation of data: Shapiro, Chen, Huang, Boosalis, Whang, Gold.
Drafting of the manuscript: Shapiro, Yoon, Saund, Gold.
Critical revision of the manuscript for important intellectual content: Shapiro, Chen, Huang, Boosalis, Saund, Whang, Gold.
Statistical analysis: Shapiro, Chen, Gold.
Administrative, technical, or material support: Huang, Yoon, Gold.
Study supervision: Saund, Whang, Gold.
Conflict of Interest Disclosures: None reported.
Previous Presentation: This paper was presented at the 39th Annual Meeting of the Association of VA Surgeons; May 3, 2015; Miami Beach, Florida.