Opioid use and associated factors among pancreatic cancer patients diagnosed between 2007 and 2015

Abstract Background Opioid therapy provides essential pain relief for cancer patients. We used the population‐based Surveillance Epidemiology and End Results (SEER) linked with Medicare database to identify the patterns of opioid use and associated factors in pancreatic adenocarcinoma cancer patients 66 years or older. Patients and Methods We assessed opioid types, dispensed days, opioid uptake rates, and factors associated with opioid use after pancreatic adenocarcinoma cancer diagnosis in Medicare beneficiaries between 2007 and 2015 from the SEER‐Medicare data. Multivariable regression analysis was used to adjust for a variety of patient‐related factors. Results We identified a cohort of 10,745 pancreatic cancer patients with a median age of 76 years old and median survival of 7 months; 75% of patients‐initiated opioids after cancer diagnosis. African Americans had the lowest rate of opioid use of 69.1% compared with all other race/ethnicity groups at around 75%. No significant yearly trend of prescribing opioids was detected. Hydrocodone was the most frequently prescribed opioid type. Regression analysis revealed that age ≤80 years, residing in Southern or Western SEER registries, residing in urban/less urban versus big metro areas, having stage IV cancer at diagnosis, longer survival time, and undertaking cancer‐directed treatment or using palliative care were positively associated with opioid initiation, more prescribed opioid types, and higher opioid doses. Discussion While a range of sociodemographic variables were associated with opioid use in unadjusted analysis, the associations between race/ethnicity, gender, and socioeconomic status with opioid initiation disappeared when sociodemographic factors, tumor characteristics, and cancer treatment were adjusted. Conclusion Health care professionals' opioid prescription pattern for pancreatic cancer patients does not parallel the U.S. opioid epidemic. Racial/ethnic disparities in opioid treatment were not identified.


| BACKGROUND
Pancreatic cancer has a high mortality rate, which is largely due to late presentation of symptoms and advanced disease diagnosis. Pain is one of the most common symptoms of pancreatic cancer, affecting more than 80% of pancreatic cancer patients. 1 As pancreatic cancer progresses, at least 40% of patients experience severe pain, leading to reduced performance status, reduced ability to tolerate cancer treatment and decreased survival. 1 Opioid-based pharmacotherapy is the primary strategy to manage moderate and severe pain for patients 65 years or older with progressive and advanced pancreatic cancer in the hope to improve quality of life for these patients. 2,3 Nearly half of pancreatic cancer patients 65 years or older require use of strong opioids (e.g., morphine) to alleviate severe pain. 4,5 Pancreatic cancer is the third leading cause of cancer death in the U.S. and more than two-thirds of pancreatic cancer patients are older than 65 years. [6][7][8] With the surging aging population in the U.S., incidence of pancreatic cancer is anticipated to rise by 55% by 2030. 6,9 The increasing incidence of pancreatic cancer in the aging population has created a great need for a better understanding of how these patients use opioids to control pain and symptoms. Most previous studies on patterns of opioid use analyzed patients in different age groups and with different cancer types as a cohort and thus overlooked the heterogeneities of opioid consumption in pancreatic cancer patients, especially among patients 66 years or older.
Thus, we aimed to assess the patterns of opioid use in pancreatic patients older than 65 years diagnosed between 2007 and 2015 using the National Cancer Institute's Surveillance, Epidemiology, and End Results linked with Medicare Database (SEER-Medicare). Our study focused on pancreatic adenocarcinoma since it is the most common form of pancreatic cancer accounting for more than 90% of pancreatic cancer diagnosis. Using a population-based approach and a nationwide dataset allowed us to grasp more comprehensively the use of opioids in older pancreatic cancer patients. We also evaluated the association of factors including patients' sociodemographic characteristics and clinical features, including cancer-directed treatment, with opioid use. Our results could help respond to the soaring urgency of providing effective and safe pain management for older pancreatic cancer patients, in addition to elaborating on racial/ethnic disparities in opioid initiation experienced by this vulnerable population.

| Data sources and patients
The study was exempt by the Institutional Review Board of the University of Texas MD Anderson Cancer Center. Based on the National Cancer Institute's SEER-Medicare data from 2007 to 2015, we used the following criteria to select the study cohort: (1) being diagnosed with histologically confirmed primary pancreatic cancer between January 1, 2007 andDecember 31, 2015; (2) the histology type was adenocarcinoma; (3) being at least 66 years old at diagnosis; (4) having full and continuous Medicare Parts A and B coverage from at least 12 months before their pancreatic cancer diagnosis and till the end of the study period, and (5) having full and continuous Medicare Part D from the first month of their pancreatic cancer diagnosis until death or the end of the study. The end of the study period was defined as end of full coverage of Medicare Parts A & B without HMO, end of full coverage of Part D, death, or December 31, 2016 (whichever came first; see Figure 1).

| Opioid use and patient covariates
Opioid use was identified from Medicare Inpatient, Outpatient, Hospice, Durable Medical Equipment and Part D files using revenue codes and generic drug names. We examined patients' opioid use by summarizing number of dispense days, opioid dispensed doses, and types of opioid medications in the units of each opioid prescription, each study patient, and the whole study cohort during the study follow-up duration. Opioid dispensed doses (morphine equivalent doses after pancreatic cancer diagnosis were calculated based on an equation: Strength per Unit multiplied by (Number of Units divided by Days of supply) multiplied by conversion factor. 10,11 Specifically, we computed, during the study follow-up duration, (1) the number of patients with at least one opioid prescription; (2) the sum of opioid medication prescriptions a given study patient received; (3) sum of supply days of a given opioid medication and sum of total supply days of all opioid medications prescribed to a given study patient; and (4) sum of dispensed doses of a given opioid medication and sum of total dispensed doses of all opioid medications prescribed to a given study patient, which were reported using median and interquartile measures. Additionally, number of study patients who had their first opioid prescription occurred in the first month after pancreatic cancer diagnosis, and frequencies of opioid medications by opioid types prescribed to the whole study cohort were assessed. F I G U R E 1 Pancreatic cancer cohort selection consort diagram By classifying patients as opioid users versus nonusers, we assessed factors associated with opioid use by using logistic regression analysis. The covariates controlled for in the regression were diagnosis age, diagnosis year, cancer stage, sex, race/ethnicity, marital status, residence urbanization level, SEER registry area, and duration from pancreatic cancer diagnosis to the end of study. 12,13 Quartile rankings of census-level median income and the proportion of residents with less than a high school education, and state buy-in enrollment were used as economic status indicators. 7,8 The Charlson Comorbidity Index (CCI) was used to classify comorbidity severity based on claims made during the period from 12 months to 1 month before cancer diagnosis. 9,14 Cancer-directed surgery, chemotherapy, radiation therapy, immunotherapy were identified using Medicare claims from 1 month before until 6 months after cancer diagnosis. 8 The International Classification of Diseases, 9th Revision and 10th Revision, Clinical Modification (ICD-9-CM and ICD-10-CM) codes were used to identify the above cancer treatment and palliative care use (Table A1). 15,16

| Statistical analyses
The Kruskal-Wallis, Mann-Whitney U, and chi-square tests were used to assess the differences of continuous variables and categorical variables by patients' opioid use. Multivariable logistic regression modeling was used to assess factors associated with opioid use and risk ratios (RR) and 95% confidence intervals (95% CI) were reported. The data analysis was conducted using SAS Enterprise Guide 7.1 (SAS Institute, Inc.). All tests were two-tailed, and the significance level was set at less than 0.05.

| Patient characteristics
We identified 10,745 pancreatic patients, with median age of 76 years old and median survival time of 7 months, of whom 75.0% used opioids. Based on univariable data analysis, patients who were 66-80 years old, women, or married were significantly more likely to use opioids (Table 1). In this initial analysis, African Americans had the lowest opioid use rate of 69.1% compared with any other race/ethnicity in our cohort. Those diagnosed with stage III pancreatic adenocarcinoma had the highest opioid use rate of 79.5% while those with stage IV had the lowest opioid use rate of 73.5%. Further, opioids were more likely to be used in patients who had less comorbidities and survived for 13 months or longer after pancreatic cancer diagnosis. Opioid users were more likely to live in the Southern SEER registries, and in rural areas. Having any cancer-directed treatment (i.e., radiotherapy, chemotherapy, surgery, or immunotherapy) or using palliative care were also positively associated with opioid use. No significant difference of opioid use was detected by cancer diagnosis year from 2007 to 2015, with the opioid use rate fluctuating between 72.3% and 77.2% over the study period.

| Summary of opioid consumption among opioid users
We assessed opioid consumption by the follow-up time of the study patients during the first 12 months after their pancreatic cancer diagnosis ( Table 2). We summarized opioid use in the following four follow-up intervals: 0-2, 3-5, 6-12, and >12 months. The opioid use rate, number of opioid types prescribed, and total opioid doses consumed significantly increased as the study follow-up time increased. Patients with follow-up time of 6-12 months had the highest number of dispense days per opioid type, total opioid dispense days, and doses consumed per opioid type. Hydrocodone was the most frequently prescribed opioid among the study patients followed for at least 2 months or 3 to 5 months. Oxycodone was the most frequently prescribed opioid among those followed for at least 6 months or longer.

| Factors associated with opioid use
Based on the multivariable logistic regression analysis (Table 3), we found that patients with more than 1 year of follow-up were 20% more likely to use opioids compared with those with <3 months follow-up time with 95% CI (1. 16, 1.22). Other factors such as being female, residing in the Southern and Western SEER registries, urban/less urban versus big metro environment, surviving longer after pancreatic cancer diagnosis, advanced cancer stage, and undertaking cancer-directed treatment were associated with higher likelihood of opioid use. Patients older than 80 years were 6% less likely to use opioids compared with those 66-70 years old. After adjusting for patients' sociodemographic characteristics, tumor characteristics, and cancer treatment, African American patients as well as other racial/ethnic minorities had the same odds of receiving opioid medication compared to Whites.    [19][20][21] Palliative care use was associated with more frequent prescription of opioids in our univariable analysis. Wang et al., 22 Tse et al., 23 and Fisher et al., 24 reported older pancreatic cancer patients using palliative care consumed opioids significantly more frequently than those receiving standard oncology care. We suggest that the higher likelihood of using opioids among those with palliative care use might be attributed to the practice of prescribing strong and high dose opioids to manage severe pain for cancer patients in the palliative care stage. 17,18 Our data also indicated that patients with longer follow-up time, less comorbidities, and continuing with cancer-directed treatment consumed more types and higher doses of opioids, which might be due to opioids being the first line pharmacotherapy for controlling pain or symptoms resulting from undertaking cancer-directed treatment for a longer time. Mercadante et al. 25 demonstrated cancer patients' consumption of opioid doses increased as their illness progressed. Our study found that patients residing in the Southern and Western SEER Registries, semi-rural areas were more likely to use opioids. Such geographic variations in opioid use in the U.S. were reported in Beccaro et al., 26 Stiefel et al., 27 and Roeland et al. 28 We argue that pancreatic adenocarcinoma cancer patients' geographic location was one of the main factors influencing the access to and consumption of opioids. 29,30 In our study, we found no racial/ethnic disparities in opioid use among pancreatic cancer patients after adjusting for patients' sociodemographics, tumor characteristics, and cancer treatment. Although African Americans had the lowest opioid initiation rate in our univariable analysis, this difference disappeared after adjusting for the aforementioned covariates in multivariable logistic regression analysis. In contrast to our finding, a previous study by Pletcher et al. 31 reported Black patients were less likely to receive opioid analgesic medication than White patients to manage their pain, with odds ratio of 0.66 and 95% CI (0.62, 0.70). Several factors may explain the discrepancy between our findings and those of Pletcher and colleagues. Pletcher's study was based on data collected from emergency department visits, and the study population was not cancer patients. Moreover, the data had limited information on patients' chronic disease conditions and disease-related treatment. 31 In our study, patients' opioid use was obtained from the full spectrum of care including inpatient and outpatient services. Moreover, we found that opioid use was significantly associated with a range of variables including patients' age, cancer stage, cancer treatment, and follow-up time. Therefore, racial/ ethnic disparities of opioid use were no longer significant after adjusting for patients' sociodemographics, tumor characteristics, cancer treatment, follow-up time, and comorbidity in the multivariable model.
Our study has many strengths. We used the populationbased and longitudinal SEER-Medicare data to identify the patterns and factors associated with opioid use in pancreatic cancer patients 66 years or older. Previous studies of the patterns of opioid use had limited sample sizes or focused on multiple cancer types, making it a challenge to generalize the study results to pancreatic cancer patients. Our results, generated from a large sample size of older pancreatic cancer patients, present a more accurate assessment of changing patterns of opioid use along the disease trajectory for this population. Further, our analysis of opioid use stratified by the study follow-up time revealed the complex interactions between opioid use and pancreatic cancer survival time.
Despite these strengths, our study had some limitations. Since we used administrative data, no information regarding patients' preference of pain medications, oncologists' training in opioid prescribing, and regulatory policies surrounding opioid prescribing was captured, which might have led to selection bias and uncontrolled confounders. As our study results were based on a cohort of Medicare beneficiaries with median age of 76 years old, the generalizability of the results to younger pancreatic cancer patients remains unknown. Further, due to using the SEER-Medicare data from 2007 to 2015 to assess the utilization patterns of opioids among pancreatic cancer patients, the follow-up time of our study patients was limited by the availability of our study data and the unique characteristics of claims data. We might overlook the information on study patients' opioid use if their opioid use was not covered by their Medicare health insurance coverage periods available in our study data or the study patients remained alive after December 31, 2016. Yet, 95.1% of total 10,745 patients ended their follow-up time due to death. Thus, our study fully captured majority of our study cohort's use of opioids from pancreatic diagnosis to their death. The likelihoods of using opioids in the study patients died during the study follow-up period increased significantly along with the increasing of patients' survival time. Also, since our study time ended in 2015, it will be interesting to determine if our conclusions remain unchanged for more recent years by analyzing newer SEER-Medicare data.
Our study indicated that the study patients with longer survival time were more likely to use opioids and used more different types of opioids. Yet, those survived more than 12 months were less likely to initiate opioid use during the first month after cancer diagnosis than those survived less.
Such findings indicated potential relations between opioid initial prescription timing and survival time in pancreatic cancer patients, echoing the argument of the correlations between opioid initial prescription timing and prolonged opioid use in post-surgery patients. 32 Clinicians need to take into consideration the timing of starting prescribing opioids for opioid naïve patients 65 or older with pancreatic cancer to ensure the optimized pain management and avoidance of possible opioid overuse. Additionally, our study found that fentanyl was more frequently used in those survived 4-12 months. Oxycodone was most used in those survived for more than 12 months. Fentanyl and oxycodone have been associated with overdose deaths from synthetic opioids according to CDC in the United States. 33 Although pancreatic cancer patients 65 years or older are not a high-risk population of overdose deaths from fentanyl and oxycodone, clinicians need to consider the effects of prolonged use of fentanyl and oxycodone in this population given their unpredictable responses to opioids.
Using opioid-based pharmacotherapy to manage cancer pain remains an essential component to improve pancreatic adenocarcinoma cancer patients' quality of life. A consensus understanding of opioid use is that opioids can control cancer pain effectively when prescribed and used correctly and safely. 19 Our results could help health care professionals make clinical decisions on balancing opioidbased pain management and safe opioid prescribing for pancreatic adenocarcinoma cancer patients 66 years or older while facing their unpredictable opioid responses, demand for high quality of life, and the ongoing crisis of opioid misuse and overuse in the U.S.

ETHICAL APPROVAL
This study was considered exempt of institutional review board approval because the data are de-identified.