Pancreatic cancer represents the fourth leading cause of cancer-related death, and the 5-year survival for all stages together ranges from 5% to 10%. Although “distant disease” spread represents the hallmark of pancreatic cancer, local tumor progression can lead to substantial patient morbidity. Recent data suggest that nearly 1 in 3 patients with pancreatic cancer will die of local disease progression.1 In addition, cancer progression within the pancreas can cause intractable pain, gastrointestinal ulceration and bleeding, biliary obstruction, or pancreatitis.
Existing research documenting patterns of disease spread typically uses radiographic progression to define local or nonlocal/distant tumor progression.2 However, radiographic progression—often defined as tumor diameter increasing by 30%—may not correlate with meaningful clinical outcomes, such as patient quality of life. Understanding the clinically meaningful patterns of disease spread will help inform providers and patients about patient-oriented consequences associated with pancreatic cancer. The purpose of this study was to characterize disease progression by evaluating patterns of hospitalization in a large cohort of patients with pancreatic cancer.
A population-based retrospective cohort study was conducted using the SEER-Medicare linked database. The SEER database contains demographic and clinical data pooled from individual population-based cancer registries covering approximately 30% of the US population. Medicare provides federally funded health insurance, including inpatient and outpatient services, for individuals aged ≥65 years in the United States. The SEER-Medicare linkage includes Medicare claims data for patients within SEER, and claims extend from prior to diagnosis through death, which allows for assessment of patterns of care and hospitalizations over the entire course of a patient’s disease.
A total of 32,475 patients aged ≥66 years diagnosed with histologically confirmed malignant adenocarcinoma of the pancreas in 2000 through 2011 were identified (supplemental eTable 1, available with this article at JNCCN.org). Our analysis was restricted to those aged ≥66 years at diagnosis to allow a full year of claims data from which to calculate a comorbidity score. We excluded 338 patients with prior malignancies or with pancreatic cancer diagnosed at autopsy or on the death certificate alone. Another 10,637 patients with incomplete data were excluded, including those without continuous Medicare coverage (Parts A and B) or with any Part C coverage (HMO enrollment) from 12 months prior to diagnosis through time of death or last follow-up. The final study cohort comprised 21,500 patients.
Demographic and tumor-related information, such as patient age and cancer stage at diagnosis, year of diagnosis, sex, race, marital status, histologic grade, tumor location, and tumor size were extracted from SEER. Patients were grouped into geographic region based on location of their SEER registry: West (San Francisco, San Jose, Los Angeles, and Greater California; Seattle, Washington; New Mexico; Utah; Hawaii), Midwest (Detroit, Michigan; Iowa), South (Atlanta, rural, and greater Georgia; Kentucky; Louisiana), and East (New Jersey; Connecticut). Socioeconomic status was assessed with median regional household income as measured from 2000 census tract data. Inpatient and outpatient Medicare claims from the year before diagnosis were used to assess preexisting comorbidity using the Deyo adaptation of the Charlson comorbidity index.3 Care at a teaching hospital was defined as any indirect medical education payment noted during a hospitalization after the patient’s cancer diagnosis.4
Treatment variables, including chemotherapy, surgery, and radiotherapy (RT), were identified from Medicare claims data. Administration of chemotherapy was determined through Medicare billing claims, including ICD-9 diagnosis and procedure codes and Healthcare Common Procedure Coding System (HCPCS) codes, using previously described methods (supplemental eTable 1).5 RT was identified from claims related to its planning and delivery,6 and a course of RT was defined as a cluster of radiation-related Medicare claims. Radiation is delivered in daily fractions, and for this study we assumed that each daily claim of a radiation treatment represented a single fraction. We further assumed that any break in RT of >30 days represented a separate course of treatment. In patients who received multiple courses of RT, we included only the first course in the analysis. Pancreatic surgery included partial, distal, or total pancreatectomy and a pancreaticoduodenectomy (Whipple procedure) (supplemental eTable 1).
Compared with categorizing patients by classic AJCC staging, we categorized them into functional groups based on stage and operability: resectable, unresectable, or metastatic. This approach is consistent with both NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines) for Pancreatic Adenocarcinoma7 and other studies.8 Patients who presented with stage IV disease at diagnosis composed the metastatic group; those with stage I–III disease at presentation who underwent cancer-directed surgery were included in the resectable group; and the remaining patients with stage I–III disease at presentation were placed into the unresectable group. The unresectable group likely included a mix of patients with technically unresectable disease and those with potentially resectable disease who did not undergo surgery due to medical comorbidity or patient choice. SEER does not include data to assess resectability and lacks information about patient preferences, so we were unable to further categorize this unresectable group.
Study End Points
The primary end point was hospitalization after pancreatic cancer diagnosis. All hospitalizations were identified from the inpatient Medicare Provider Analysis and Review (MedPAR) file and further categorized by the ICD-9 diagnosis codes during that hospitalization. We first categorized hospitalizations as either cancer-related or non–cancer-related. Among the cancer-related hospitalizations, these were further classified as due to either local disease progression or nonlocal/distant disease. Hospitalizations for local disease progression included those due to duodenal or gastric ulceration, upper gastrointestinal bleed, disorders of the biliary tree (eg, biliary obstruction, cholangitis), local pain (eg, postoperative, abdominal), problems of the pancreas itself (eg, pancreatitis), and radiation toxicity. Hospitalizations for nonlocal/distant disease included those due to thromboembolic events, cytopenia, dehydration, nausea or vomiting, malnutrition and cachexia, ascites, pathologic fracture, and chemotherapy toxicity. Complications were selected from the NCCN Guidelines for Pancreatic Adenocarcinoma.7 Readmissions were included, but the initial hospitalization for pancreas-directed surgery was not.
To describe the long-term rates of hospitalization, we used competing risk models to account for the competing risk of death,9 which included unadjusted cumulative incidence analyses with differences between groups assessed with Gray’s test for inequality. Multivariable Fine-Gray regression models were used to identify independent predictors of hospitalization while controlling for potential confounding factors. Covariables included in the multivariable models were defined a priori and included age at diagnosis, marital status, race/ethnicity, residence in a metropolitan area, geographic region, treatment at a teaching hospital, median income, Charlson comorbidity index, disease group (resectable, unresectable, metastatic), tumor location and size, histologic grade, year of diagnosis, and sex. Because chemotherapy and RT are often delivered after diagnosis, to account for the immortal time bias we modeled receipt of these treatments as time-dependent covariables.10,11 All statistical tests were 2-sided, and P<.05 was considered significant. Analyses were conducted using SAS 9.4 (SAS Institute Inc).
Of the 21,500 patients in the study cohort, 11,298 (52.5%) presented with metastatic disease, 3,092 (14.4%) were in the resectable group, and 7,110 (33.1%) were in the unresectable group. Complete demographic and clinical characteristics are presented in Table 1.
Patient Demographic and Clinical Characteristics
Incidence of Hospitalization
Overall, 43.5% of patients (n=9,347) required hospitalization for at least one local complication and 60.9% (n=13,101) required hospitalization for at least one nonlocal complication. Among patients who were hospitalized for a complication of local disease progression, 50.6% were hospitalized more than once and 13.6% were hospitalized >3 times. The median length of hospitalization was 6 days (interquartile range, 3–10 days).
Among all patients, the 2-month incidence of hospitalization was 32.8% for local and 35.7% for nonlocal complications (Figure 1), and the 12-month incidence of hospitalization was 40.9% for local and 54.9% for nonlocal complications. The most common causes of hospitalization from local progression were biliary disorders and pancreas-related (Table 2). The most common causes of hospitalization from nonlocal progression were cytopenia and malnutrition/cachexia (Table 2). Supplemental eTable 2 presents causes of hospitalization stratified by receipt of RT and chemotherapy.
Twelve-Month Cumulative Incidence of Local and Nonlocal Complications Requiring Hospitalizationa
Causes of Hospitalization by Patient Group
Risk of hospitalization varied by disease stage at presentation (Figure 2). Patients from the resectable and unresectable groups tended to be hospitalized because of local complications early in their clinical course, although over time the risk of hospitalization from nonlocal complications surpassed that from local complications. The 12-month cumulative incidence of local complications was highest for patients from the unresectable group (53.1%), followed by the resectable group (39.5%), and lowest for the metastatic group (33.7%). The 12-month cumulative incidence of nonlocal complications was highest for patients from the metastatic group (57.0%), followed by those from the unresectable (56.8%) and resectable groups (42.8%) (Table 2).
Predictors of Complications Requiring Hospitalization
On multivariable analysis identifying predictors of hospitalization, compared with patients with metastatic disease, those with unresectable disease were at a 49% increased risk of hospitalization for local complications (subdistribution hazard ratio [SDHR], 1.49; 95% CI, 1.43–1.57), whereas those with resectable disease were at a 13% decreased risk (SDHR, 0.87; 95% CI, 0.81–0.94) (Figure 3). Other factors that increased risk of hospitalization for local complications included younger age, Hispanic ethnicity, living in the Midwest, care at a teaching hospital, higher comorbidity, smaller tumor size, and tumor in the head of the pancreas. Neither chemotherapy nor RT were associated with a significant decrease in the risk of local complications.
When considering predictors of hospitalization for nonlocal disease, we found that patients with resectable disease had a 32% decreased risk of hospitalization compared with those with metastatic disease (SDHR, 0.68; 95% CI, 0.65–0.72). There was no difference in the risk of hospitalization for a nonlocal complication between the unresectable and metastatic groups (SDHR, 0.96; 95% CI, 0.92–1.00) (Figure 3). Other factors that increased the risk of hospitalization for complications of nonlocal disease included younger age, black race, living in a metropolitan area, living in the Midwest, care at a teaching hospital, lower income (bottom quartile), higher comorbidity, larger tumor size, and tumor in the head of the pancreas (Figure 3). Neither chemotherapy nor RT were associated with a significant decrease in the risk of nonlocal complications (Figure 3).
This population-based study is likely the first to define the specific patterns and causes of hospitalization for a large real-life cohort of elderly patients with pancreatic cancer. The key finding of this study is the high observed rates of hospitalization for both local and nonlocal disease progression. Pancreatic cancer represents an aggressive lethal cancer characterized by early and aggressive tumor spread. Most patients have overt metastatic or micrometastatic disease on presentation, which leads researchers and clinicians to often consider pancreatic cancer a problem of “distant” metastatic disease. Improving outcomes among this challenging disease will certainly require novel advances in systemic therapy; however, the findings of our study highlight the clinical importance of considering both the local pancreatic tumor and nonlocal/distant metastatic disease in optimizing patient care.
Our findings complement other research that demonstrates the clinical importance of local pancreatic tumor progression. A group from Johns Hopkins performed autopsies on 76 patients with pancreatic cancer and found that 30% died of local tumor destruction.1 Furthermore, a single-center study from our institution found that the risk of hospitalization from complications of the local tumor was 45%.12 Local tumor progression can lead to pain, biliary obstruction, and gastric ulcers and bleeding, all of which reduce patient quality of life. Although RT has the capacity to improve local tumor control and reduce complications from local tumor progression, randomized clinical trials have produced mixed survival results for patients with pancreatic cancer treated with RT.13,14 Furthermore, the absolute local control benefit of using conventionally fractionated radiation seems to be small. The randomized LAP07 trial13 found that chemoradiotherapy among patients with locally advanced pancreatic cancer decreased the risk of local tumor progression by 14%. Similarly, our study found that RT did not significantly reduce the risk of hospitalization from local tumor progression. Although alternative radiation strategies such as stereotactic body RT15 and dose-escalated RT16 hold promise, researchers lack randomized evidence supporting widespread use. Optimizing patient quality of life requires improved strategies to mitigate the complications from local tumor progression. These strategies could include ablative local therapies, such as cryotherapy17; irreversible electroporation18; or palliative measures, such as biliary stenting/bypass19 and celiac plexus block.20 Another possible intervention includes monitoring patient-reported outcomes, which in select populations of patients with cancer may reduce the risk of hospitalization21 and improve survival.22 Finally, an option that should be explored is early integration of palliative care, which has been shown to relieve symptoms, reduce hospitalizations, and improve survival in metastatic non–small cell lung cancer.23 Given the high cost of hospitalizations,24 improvements in locally directed palliative interventions may reduce the overall economic impact of pancreatic cancer. In general, research efforts in pancreatic cancer most often strive to improve survival, although the impact of treatment on long-term quality of life and cost of care must also be considered.
This study builds on a prior study by Reddy et al25 of 1,730 Medicare beneficiaries with resectable pancreatic cancer who underwent pancreatectomy. In this study, which resulted in a readmission rate of 53% at 1 year, 25% of readmissions likely arose from operative complications and 48% were potentially associated with recurrence. Our study sought to define risk factors for hospitalization to help better risk-stratify patients (Figure 3). With hospitalization due to local tumor progression, we found the highest risk among patients with unresectable disease, and for hospitalization from nonlocal disease spread, the highest risk was among those with metastatic disease. We also found that younger patients had a higher risk of hospitalization. Recent studies have shown that younger patients have longer overall survival,26,27 raising the question of whether disease biology may differ by patient age.
In addition, we found that the risk of hospitalization was higher in nonwhite patients, which could reflect known health disparities that influence patterns of care,28,29 and that the risk of hospitalization varied by geography. Other research has documented variability in hospitalization rates by geographic region, which may reflect regional differences in patient behavior, referral patterns, or treatment options offered to patients.24,30
In this study, receiving care at a teaching hospital was associated with a significantly increased risk of both local and nonlocal hospitalization. These data lack the granularity to determine the etiology of the link between academic teaching hospitals and hospitalizations. However, this observation could arise due to underlying differences in patient complexity or difference in institutional treatment patterns.31,32
One unexpected finding was the observed association between larger primary tumor size on presentation and paradoxically lower rates of hospitalization caused by local complications. The underlying explanation of this association is unknown; however, it could reflect differences in presentation. Perhaps patients who present with larger primaries have tumors anatomically located in regions that elicit fewer symptoms and have a decreased likelihood of leading to hospitalization. Overall, understanding these associations could help clinicians better identify patients at a higher risk of hospitalization from pancreatic cancer.
Limitations of this study largely arise from those surrounding the data within the SEER-Medicare linked database, which lack specific details of patient characteristics, performance status, metastatic disease burden, and treatment, all of which very likely influence a patient’s risk of hospitalization. Another limitation is an inability to account for complications managed on an outpatient basis or in an emergency department, including pain, lethargy, nausea, and dysphagia, all of which contribute substantially to pancreatic cancer morbidity but may not be severe enough to warrant an inpatient admission.33 Along these lines, we suspect that the number of emergency department visits is higher than the number of hospitalizations. In addition, we determined the cause of hospitalization through billing codes, which raises the possibility of misclassification of this study’s primary outcome. Unmeasured confounding factors, such as patient anatomy and performance status, could also impact these study results. Furthermore, distinguishing whether complications arise from local or nonlocal disease can pose a challenge in retrospectively collected data, particularly when using billing claim data as a proxy for patient complications. For example, biliary obstruction could arise from the primary tumor obstructing bile flow (local) or from a metastatic tumor in the hilum of the liver causing bile obstruction. Although one would expect most biliary obstruction cases in this study to arise from local tumor progression, misclassification remains a possibility. Despite these limitations, this study offers useful insight into the morbidity that patients experience with pancreatic cancer.
Combating metastatic disease represents a key therapeutic goal in the overall management of pancreatic cancer, although this study demonstrates the clinical importance of considering the local tumor and nonlocal disease in the overall management of patients with pancreatic cancer.
Previous publication: This study was published as an abstract in J Clin Oncol 2018;36(Suppl 4):Abstract 236.
Author contributions: Study concept: Sarkar, Fero, Murphy. Data curation: All authors. Formal analysis: Sarkar, Fero, Seible, Panjwani, Murphy. Software: Sarkar, Fero, Matsuno, Murphy. Investigation: Sarkar, Fero, Seible, Murphy. Methodology: All authors. Validation: Sarkar, Fero, Panjwani, Murphy. Visualization: Sarkar, Fero, Matsuno. Supervision: Fero, Seible, Murphy. Resources: Murphy. Writing—original draft: Sarkar, Fero, Murphy. Writing—review and editing: Sarkar, Seible, Panjwani, Murphy.
Disclosures: The authors have not received any financial consideration from any person or organization to support the preparation, analysis, results, or discussion of this article.
Iacobuzio-Donahue CAFu BYachida S. DPC4 gene status of the primary carcinoma correlates with patterns of failure in patients with pancreatic cancer. J Clin Oncol 2009;27:1806–1813.
Eisenhauer EATherasse PBogaerts J. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer 2009;45:228–247.
Deyo RACherkin DCCiol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 1992;45:613–619.
Simpson DRMartínez MEGupta S. Racial disparity in consultation, treatment, and the impact on survival in metastatic colorectal cancer. J Natl Cancer Inst 2013;105:1814–1820.
Warren JLHarlan LCFahey A. Utility of the SEER-Medicare data to identify chemotherapy use. Med Care 2002;40(Suppl 8):IV-55–61.
Virnig BAWarren JLCooper GS. Studying radiation therapy using SEER-Medicare-linked data. Med Care 2002;40(Suppl 8):IV-49–54.
Tempero MAMalafa MPAl-Hawary M. NCCN Clinical Practice Guidelines in Oncology: Pancreatic Adenocarcinoma. Version 2.2018. Accessed August 14 2018. To view the most recent version of these guidelines visit NCCN.org.
O’Neill CBAtoria CLO’Reilly EM. Costs and trends in pancreatic cancer treatment. Cancer 2012;118:5132–5139.
Fine JGray RJ. A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc 1999;94:496–509.
Shintani AKGirard TDEden SK. Immortal time bias in critical care research: application of time-varying Cox regression for observational cohort studies. Crit Care Med 2009;37:2939–2945.
Fero KESeible DMCardillo N. Natural progression of symptoms after diagnosis of pancreatic cancer [abstract]. J Clin Oncol 2016;34(Suppl):Abstract e15748.
Hammel PHuguet Fvan Laethem JL. Effect of chemoradiotherapy vs chemotherapy on survival in patients with locally advanced pancreatic cancer controlled after 4 months of gemcitabine with or without erlotinib: the LAP07 randomized clinical trial. JAMA 2016;315:1844–1853.
- Search Google Scholar
- Export Citation
)| false . , Hammel P , Huguet F , van Laethem JL Effect of chemoradiotherapy vs chemotherapy on survival in patients with locally advanced pancreatic cancer controlled after 4 months of gemcitabine with or without erlotinib: the LAP07 randomized clinical trial. 2016; 315: 1844– 1853. 10.1001/jama.2016.4324 27139057
Loehrer PJ SrFeng YCardenes H. Gemcitabine alone versus gemcitabine plus radiotherapy in patients with locally advanced pancreatic cancer: an Eastern Cooperative Oncology Group trial. J Clin Oncol 2011;29:4105–4112.
Herman JMChang DTGoodman KA. Phase 2 multi-institutional trial evaluating gemcitabine and stereotactic body radiotherapy for patients with locally advanced unresectable pancreatic adenocarcinoma. Cancer 2015;121:1128–1137.
- Search Google Scholar
- Export Citation
)| false . , Herman JM , Chang DT , Goodman KA Phase 2 multi-institutional trial evaluating gemcitabine and stereotactic body radiotherapy for patients with locally advanced unresectable pancreatic adenocarcinoma. 2015; 121: 1128– 1137. 10.1002/cncr.29161 25538019
Krishnan SChadha ASSuh Y. Focal radiation therapy dose escalation improves overall survival in locally advanced pancreatic cancer patients receiving induction chemotherapy and consolidative chemoradiation. Int J Radiat Oncol Biol Phys 2016;94:755–765.
- Search Google Scholar
- Export Citation
)| false . , Krishnan S , Chadha AS , Suh Y Focal radiation therapy dose escalation improves overall survival in locally advanced pancreatic cancer patients receiving induction chemotherapy and consolidative chemoradiation. 2016; 94: 755– 765. 26972648 10.1016/j.ijrobp.2015.12.003
Martin RC II. Irreversible electroporation of locally advanced pancreatic neck/body adenocarcinoma. J Gastrointest Oncol 2015;6:329–335.
Boulay BRParepally M. Managing malignant biliary obstruction in pancreas cancer: choosing the appropriate strategy. World J Gastroenterol 2014;20:9345–9353.
Wong GYSchroeder DRCarns PE. Effect of neurolytic celiac plexus block on pain relief, quality of life, and survival in patients with unresectable pancreatic cancer: a randomized controlled trial. JAMA 2004;291:1092–1099.
- Search Google Scholar
- Export Citation
)| false . , Wong GY , Schroeder DR , Carns PE Effect of neurolytic celiac plexus block on pain relief, quality of life, and survival in patients with unresectable pancreatic cancer: a randomized controlled trial. 2004; 291: 1092– 1099. 10.1001/jama.291.9.1092 14996778
Basch EDeal AMKris MG. Symptom monitoring with patient-reported outcomes during routine cancer treatment: a randomized controlled trial. J Clin Oncol 2016;34:557–565.
Basch EDeal AMDueck AC. Overall survival results of a trial assessing patient-reported outcomes for symptom monitoring during routine cancer treatment. JAMA 2017;318:197–198.
Temel JSGreer JAMuzikansky A. Early palliative care for patients with metastatic non-small-cell lung cancer. N Engl J Med 2010;363:733–742.
Wang YSchrag DBrooks GA. National trends in pancreatic cancer outcomes and pattern of care among Medicare beneficiaries, 2000 through 2010. Cancer 2014;120:1050–1058.
Reddy DMTownsend CMKuo YF. Readmission after pancreatectomy for pancreatic cancer in Medicare patients. J Gastrointest Surg 2009; 13:1963–1974discussion 1974–1975.
He JEdil BHCameron JL. Young patients undergoing resection of pancreatic cancer fare better than their older counterparts. J Gastrointest Surg 2013;17:339–344.
Sehgal RAlsharedi MLarck C. Pancreatic cancer survival in elderly patients treated with chemotherapy. Pancreas 2014;43:306–310.
Wray CJCastro-Echeverry ESilberfein EJ. A multi-institutional study of pancreatic cancer in Harris County, Texas: race predicts treatment and survival. Ann Surg Oncol 2012;19:2776–2781.
Singal VSingal AKKuo YF. Racial disparities in treatment for pancreatic cancer and impact on survival: a population-based analysis. J Cancer Res Clin Oncol 2012;138:715–722.
Guadagnolo BALiu CCCormier JN. Evaluation of trends in the use of intensity-modulated radiotherapy for head and neck cancer from 2000 through 2005: socioeconomic disparity and geographic variation in a large population-based cohort. Cancer 2010;116:3505–3512.
- Search Google Scholar
- Export Citation
)| false . , Guadagnolo BA , Liu CC , Cormier JN Evaluation of trends in the use of intensity-modulated radiotherapy for head and neck cancer from 2000 through 2005: socioeconomic disparity and geographic variation in a large population-based cohort. 2010; 116: 3505– 3512. 10.1002/cncr.25205
Bakens MJvan Gestel YRBongers M. Hospital of diagnosis and likelihood of surgical treatment for pancreatic cancer. Br J Surg 2015;102:1670–1675.
Youngwirth LMNussbaum DPThomas S. Nationwide trends and outcomes associated with neoadjuvant therapy in pancreatic cancer: an analysis of 18,243 patients. J Surg Oncol 2017;116:127–132.
Keane MGHorsfall LRait G. A case-control study comparing the incidence of early symptoms in pancreatic and biliary tract cancer. BMJ Open 2014;4:e005720.