Opioid use by cancer status and time since diagnosis among older adults enrolled in the Prostate, Lung, Colorectal, and Ovarian screening trial in the United States

Abstract Background Dosing limits in opioid clinical practice guidelines in the United States are likely misapplied to cancer patients, however, opioid use may be difficult to ascertain as they are largely excluded from opioid use studies. Methods The primary objective was to determine whether cancer patients were more likely to be chronic opioid users after diagnosis. We described prescription opioid use among U.S. older adult cancer patients during two time periods, within 2 years of diagnosis (short‐term) and at least 2 years beyond diagnosis (long‐term), compared to those without cancer (controls). Among participants in the Prostate, Lung, Colorectal, and Ovarian (PLCO) screening trial with linkages to Medicare Part D data during 2011–2015, we used multivariable logistic regression to estimate the association between cancer diagnosis and opioid use outcomes controlling for demographics. The primary outcome of opioid use was measured with the following metrics: Any opioid use, chronic use (90 consecutive days supply of opioid use while allowing for a 7‐day gap between refills), high use (average daily morphine equivalent (MME) ≥120 mg for any 90‐day period), and total MME dose above 2,000 mg (MME2000). Results The short‐term cohort included 1,491 cancer patients and 24,930 controls. Any use in the 2‐year post‐diagnosis period was higher among cancer patients OR 3.3 (95% CI: 3.0–3.7). Chronic use rates were similar by cancer status (4.6% vs. 3.8% for cases and controls, respectively). The long‐term cohort included 4,377 cancer patients and 27,545 controls. Rates of any use were similar among cancer patients and controls (63% vs. 59%). Conclusions Any opioid use was similar among long‐term cancer survivors compared to controls, but differed among short‐term survivors for any opioid use and marginally for chronic opioid use.


| BACKGROUND
Pain is an anticipated consequence of cancer treatment and disease progression. Approximately 38% of cancer patients report moderate to severe pain 1 with variations by pain type and disease characteristics. 2,3 Pain management via opioids is important for cancer patients who have severe pain not adequately managed by alternative medications. Cancer patients may have varying patterns of pain management when compared to patients without cancer. 4 Debate regarding opioid use continues due to concerns for the balance between appropriate use, adverse effects of chronic use, and risk of misue. 5,6 A potential unintended consequence is that patients who benefit from pain management via properly prescribed and used opioids may not receive it as a result of tightened prescribing recommendations generally [7][8][9] and for acute pain. 10,11 Development of persistent use after cancer treatment among opioid-naïve patients is another concern. 12-15 Pain management is integral to cancer care, 16,17 however, opioid dose limits in clinical practice guidelines for non-cancer pain have been misapplied to cancer and palliative care patients. [7][8][9] Efforts to manage opioid misuse may have unintended consequences including complex pain and inadequate pain relief through diagnosis, treatment, and survivorship. [18][19][20] Possibly illustrating this point, a Canadian study found no change in opioid prescribing over time among older cancer patients while opioid prescribing increased among non-cancer patients. 21 A study of outpatient oncology patients found 33% of patients with pain reported receiving inadequate pain management prescribing. 22 Prescription regulations and policies may impact medication prescribed to cancer patients. 23 Inadequate pain management may have negative consequences on health outcomes. 24,25 Cancer patients are often excluded from opioid use studies. 26 Describing pain management may provide insights as to how opioids are used in cancer care and may identify points in the cancer care continuum to ensure appropriate pain management. We examined prescription opioid use among cancer patients at two time points around cancer diagnosis compared to those without cancer to determine whether cancer patients were more likely to be chronic opioid users after diagnosis.

| Data sources and study population
This is a secondary analysis from the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial. 27 Participants were enrolled at 10 U.S. screening centers from 1993 to 2001. Eligible study participants were between 55 and 74 years of age with no previous diagnosis of trial cancers. Participants provided informed consent and each screening center's Institutional Review Board approved the trial. Demographic information was available from a baseline and supplemental questionnaire completed in 2006-2007. Participants were followed up to 13 years for all-cancer incidence and mortality. In 2011, PLCO transitioned to centralized follow-up. Participants reconsented to continue active follow-up, switch to passive follow-up, or refused further follow-up. The active follow-up group was recontacted with questionnaires or consent requests. In 2013, the active group was asked to consent for linkage to Medicare claims. Active and passive participants continued to be followed for mortality and cancer incidence through 2015 based on linkage with the National Death Index and U.S. state cancer registries.

| Part D Medicare claims data
Information on prescription medication use was derived from linked Medicare Part D claims for the years 2011-2015. Part D contains information for each filled prescription including quantity, days' supply, and drug strength. For each linked PLCO study participant, we calculated the morphine milligram equivalents (MME) for each fill and summed over all prescriptions to obtain MME dosage over the relevant study period using CDC conversion tables for analyzing prescription data. 28 Prescription opioid use outcomes of interest were: any opioid use, chronic opioid use, high opioid use, and total MME dose above 2,000 mg (MME 2000 ). Any opioid use was defined as having at least one claim for an opioid prescription fill during the study window. Chronic opioid use was defined as having a record of 90 consecutive days' opioid supply while allowing for a 7-day gap between refills. 29 High opioid use was defined as an average daily MME of at least 120 mg over any consecutive 90-day period and at least 90 days of opioid use during the time period of interest. 30,31

| Statistical analysis
We created two study cohorts to assess opioid use. Participants were excluded if they had <12 months Part D coverage during the assessment periods to ensure sufficient time to assess opioid use. To compare opioid use among short-term cancer survivors, we selected those with any cancer diagnosis between 1 January 2011 and 31 December 2013 (ensuring the potential for at least 2 years of post-diagnosis follow-up to assess Medicare Part D claims), and those without a cancer diagnosis through the end of 2015 as controls (Cohort 1). For cancer patients, opioid use was assessed for the 2-year period following diagnosis, while for controls use was assessed for the 2-year period from their pseudo-diagnosis date (i.e., a randomly selected date between 1 January 2011 and 31 December 2013). The demographic distribution of cancer patients included in the current study were generally representative of all potentially eligible cases (Tables A2 and A3). To determine whether cancer patients were more likely than controls to have had preexisting conditions predisposing them to opioid use, we assessed use in the year before diagnosis or pseudo-diagnosis. We examined the period 6-18 months prior to diagnosis, since cancer symptoms could have been present before diagnosis.
To compare opioid use among long-term cancer survivors (diagnosis more than 2 years before the start of the assessment period) to controls, we selected those with a cancer diagnosis from randomization date until 31 December 2008, and those with no cancer diagnosis through 31 December To estimate the association between cancer diagnosis and the opioid use outcomes we used multivariate logistic regression to calculate odds ratios (ORs). The base model controlled for age and sex; the adjusted model additionally controlled for education, race/ethnicity, income, and marital status.
Because the distribution of cancer types differed by sex, analyses were stratified by sex. The statistical significance of interactions of cancer status by sex were assessed. Since prostate and breast cancer constituted large proportions of the cancers in men and women, respectively, sex-specific analyses were performed stratified by cancer type (breast or prostate vs. all other). Within sex, statistical tests were performed for whether ORs differed by cancer type.
Analyses were conducted using SAS software version 9.4. SAS Institute Inc., Cary, NC.

| RESULTS
Of the 154,887 participants enrolled in PLCO, 49,560 consented to Medicare linkage and 35,855 had at least 12 months of Medicare Part D coverage ( Figure A1). Cohort 1 included 1,491 short-term cancer survivors (first cancer diagnosis from 1 January 2011 through 31 December 2013) and 24,930 controls. Cohort 2 included 4,377 long-term cancer survivors (diagnosis from randomization until 31 December 2008) and 27,545 controls. In Cohort 1, cancer patients were more likely than controls to be men (53.8% vs. 42.0%), and cancer patients were slightly older. In Cohort 2, cancer patients were slightly older than controls (median ages 75 and 73, respectively) and substantially more likely to be men (60.9% vs. 42.5%; Table 1).

| Cohort 1 results
For Cohort 1, mean (SD) months of part D coverage during the assessment period was 23.0 (2.7) and 23.4 (2.2) for cancer patients and controls, respectively. Among cancer patients, for men, 40% were prostate cancer and for women 36% were breast cancer (Table A1).
Rates of any use were higher for cancer patients (68.4%) than for controls (40.2%), as were MME 2000 rates (14.6% vs. 7.9%). In contrast, rates of chronic opioid use were similar by cancer status (4.6% vs. 3.8% for cases and controls, respectively). High opioid use was infrequent in both groups but significantly higher among cancer patients (1.1% vs. 0.3%). For both cancer patients and controls, rates of any use, chronic use, and MME 2000 were higher in women than men, but with similar distribution among cancer patients compared to controls ( Table 2). Median MME dose among users was similar for cases and controls. The most commonly prescribed opioid medications for cancer patients were products containing hydrocodone or oxycodone, and tramadol (Tables A4 and A5). Compared to controls, cancer patients had a higher proportion of oxycodone prescriptions and a lower proportion of tramadol prescriptions (Table A2). Among men with cancer, MME 2000 significantly differed between prostate and non-prostate cancer patients, with rates of 6.6% versus 16.8%, respectively (p < 0.0001). Among women, for breast cancer compared to non-breast cancer patients, the former had significantly lower rates of chronic use (3.2% vs. 7.7%, p = 0.02), high use (0% vs. 1.8%, p = 0.03), and MME 2000 (8.4% vs. 21.6%, p < 0.0001).
Within the 2-year window, any use rates for cases were 59.8% in the first year and 32.0% in the second year; comparable rates for controls were 26.6% and 27.7%. In the period 6-18 months prior to diagnosis or pseudo-diagnosis, opioid use was assessed among the 82% of cases and 81% of controls with complete part D coverage during the period. Opioid use was similar in each group; rates of any use were 26.7% for cases versus 26.6% for controls, and rates of MME 2000 were 4.6% for cases versus 4.7% for controls.
Odds ratios were similar for the base and adjusted models. The OR for any opioid use was elevated, with an OR of 3.3 (95% CI: 3.0-3.7); by sex, the OR was significantly higher for women (4.0) than men (2.8) (p-value for interaction = 0.003). ORs were significantly elevated for MME 2000 , with ORs slightly higher among men when compared to women (adjusted OR 2.3 vs. 2.0, respectively). For chronic use, the OR was modestly elevated (OR = 1.32; 95% CI: 1.02-1.70; Table 3).

| Cohort 2 results
For Cohort 2, the mean (SD) length of part D coverage was 52.7 (13.2) and 53.2 (12.7) months for cancer patients and controls, respectively. The median (25 th /75 th ) time from cancer diagnosis to the start of the assessment period for cases was 7.0 (4.3/10.1) years. In men, 67% of diagnoses were prostate cancer while in women 48% were breast cancer (Table A1). Similar to the short-term cohort, the most common opioid medications prescribed were tramadol and products containing hydrocodone or oxycodone. The distribution of medications was similar for cases and controls. (Tables A4  and A5).
Rates of any opioid use were 63.1% for cancer patients versus 59.0% for controls. Use rates were higher for women than men among both cases and controls, and for each sex, slightly higher for cancer cases than controls (Table 4). Chronic opioid use was similar by cancer status; 5.2% versus 5.4% overall for cases and controls, respectively; chronic use was higher in women for both cases and controls. High opioid use was rare (<1% for both cases and controls).
No opioid use metric differed significantly between breast and non-breast cancer cases for women or between prostate and non-prostate cancer cases for men. Annual rates of any use varied little over the 2011-2015 period, with a range of annual rates T A B L E 3 Odds ratio (OR) estimates for opioid use for short-term cancer survivors versus controls without a cancer diagnosis in Cohort 1

All
Men Women Referent category is individuals without a cancer diagnosis during the study window.
Abbreviation: MME, morphine milligram equivalent. a Model adjusted for sex (in non-sex-specific analyses) and age. b Model additionally adjusted for education, race/ethnicity, income, and marital status. c Chronic opioid use defined 90 consecutive days of opioid use allowing for 7-day gap between refills. d High use defined as greater than or equal to 120 mg or any 90-day period and at least 90 days of opioid use. e Total MME dose >2,000 mg.

| DISCUSSION
We assessed opioid use among patients diagnosed in the United States with multiple cancer types at two points on the cancer care continuum--long-term survivors and those within 2 years of cancer diagnosis (short-term). Any opioid use was similar among long-term cancer survivors compared to controls, but differed among short-term survivors for any opioid use and marginally for chronic opioid use. We observed differences in opioid prescription fills by cancer status and among the two cohorts. Compared to controls, short-term survivors had higher rates of any opioid use and chronic use.
In the long-term cohort, any use was similar. Though rare in both cohorts, high use was significantly more frequent among short-term survivors than controls. These observations may indicate pain prevalence among cancer patients despite increased awareness of the importance of pain management. 1,32 High use among the shortterm cohort is expected and may be explained by proximity of treatment. Specifically, patients in this group may experience pain requiring pain management as a result of surgery, chemotherapy, or radiation. Concerns of inadequate pain relief among cancer patients may not be sufficiently assessed in this study. The window ends in the year before the CDC released opioid management guidelines in 2016. Though the guidelines were explicitly not intended for cancer patients or individuals who receive palliative care, these patients have been affected by guideline adoption. 7 This may explain why any opioid use rates among cancer patients in this study population were higher than non-cancer patients. 8 Future studies should assess change in opioid prescribing rates among older cancer patients after release of the CDC guidelines.
Opioid use patterns at end-of-life likely vary. Though we do not have information on recurrence or progression after diagnosis, we assessed the effect of mortality from cancer on opioid use. In the long-term cohort, 5.1% of patients died from cancer during or within a year of the end of the assessment period. In a sensitivity analysis excluding these patients, chronic use decreased from 5.2% to 4.9% and MME 2000 decreased from 13.6% to 12.2%. To be included in the shortterm cohort, participants had to survive to 2013 to consent to linkage. Therefore, the cancer death rate may be lower than otherwise expected; within 3 years of diagnosis 12.0% had a cancer-specific death.
Little increased rates of chronic use for short-term survivors compared to controls may explain why long-term survivors had little increase of any opioid use and no increase in chronic use compared to controls. This may indicate cancer patients do not become addicted after cancer diagnosis and initial treatment or reflect a shift to use of non-opioids for pain management. Near constant opioid use rates over time was observed in a study of Canadian cancer patients 65 years and older. 21 In a SEER-Medicare population, cancer patients 6 years post-diagnosis were no longer more likely to be chronic users when compared to controls. 30 In contrast, a study of Medicare beneficiaries in Texas diagnosed between 1995 and 2008 reported prolonged high opioid use rates at 5 years after cancer diagnosis and longer. 13 Variability in opioid rates by location, sex, and other factors has been documented. 33 Chronic opioid use rates over time may differ among younger (<64 years) cancer patients. 34 Much of the increase in opioid use in the short-term cohort was observed within the first year after diagnosis which may be due to receipt of treatment. Site-specific treatments such as surgery, chemotherapy, or radiation, symptoms or disease progression, and cancer stage likely contribute to opioid use patterns, thus, the observed higher rates of chronic use and any opioid use among cancer patients when compared to controls is expected. We included multiple cancer types and patients likely received varying treatments specific to their diagnosis. Approximately 10% of patients in the short-term cohort had metastatic (stage IV) disease at diagnosis while 3% of patients in the long-term cohort had metastatic disease at diagnosis. Differences in treatment by cancer site may explain the differences observed in opioid use by sex. The ORs for any opioid use comparing shortterm survivors to controls were 4.1 versus 2.8 for women and men, respectively (p for interaction = 0.002). Among women with breast cancer (about 17% of the short-term cohort and 19% of the long-term cohort), most received surgery, which may result in postoperative pain requiring pain management via opioids, while for prostate cancer (nearly 22% of the short-term cohort and 41% of the long-term cohort), about 50% and 33% received surgery for long-term and short-term cancer survivors, respectively.
This study comprised a population enrolled in Medicare Part D and may not represent individuals enrolled in other prescription drug plans, who may have differing patterns of opioid use. 35 Opioid use represents prescriptions filled under Part D and paid for by Medicare. Information on medication use was not available. We do not know if medication was used as prescribed. Indications were not available and we were unable to determine if medications were prescribed for cancer-related pain, the appropriateness of medication use, or if medication use improved pain and other cancer-related symptoms. We were unable to ascertain PRN use of opioids. While non-cancer disease occurring after baseline was captured using the supplemental questionnaire (2006)(2007), later occurrence of disease that may have required prescriptions for pain management was not captured. Differences in opioid use over time and within subgroups of cancer patients highlight the importance of monitoring opioid use during a patient's cancer trajectory possibly used in combination with other strategies for pain management and reduction of misuse. 20,36 Research is needed within subgroups of cancer patients to differentiate overuse from appropriate use based on treatment and survivorship plans and patient goals. This may include assessing chronic high or low use of specific opioids or describing other medications commonly used with opioids. 37 Though we observed no difference between short-or longterm cancer patients and controls with regard to the number of comorbidities, they may influence opioid use. Future research should examine concurrent use of medications commonly used to manage comorbidities. Opioid use in the 6-18-month period prior to diagnosis or pseudo-diagnosis in the short-term cohort was similar in cases and controls, suggesting the observed higher rate of opioid use in cases post-diagnosis was not due to confounding based on higher rates in cases of non-cancer conditions predisposing to opioid use. Finally, the current study included PLCO trial volunteers who may be healthier than the general U.S. population. The impact of a healthy volunteer effect on outcomes in this trial has been reported elsewhere 38 and could explain why opioid use rates were mainly similar. In a sample of community participants, it may be important to consider how other patientlevel factors impact opioid use among cancer survivors.

| CONCLUSION
Any opioid use was similar among long-term cancer survivors compared to controls, but differed among short-term survivors for any opioid use and marginally for chronic opioid use. Future research should consider concomitant use of opioids and non-opioid medications commonly used to manage pain among cancer patients.

APPENDIX FIGURE A1
Participants included in the study sample by cancer status and time since cancer diagnosis