We drew upon the Surveillance, Epidemiology and End Results program (SEER)-Medicare database,(14) a linkage of cancer registry and Medicare enrollment and claims data. This linked database includes cancer cases through 2011 and Medicare claims through 2013. Medicare Part A and B claims provide information on diagnoses and procedures in the hospital and outpatient setting and Part D claims provide information on prescription drug dispensing (available from 2007–2012).

For this study, we identified adults aged 66 years and older who were diagnosed with a first, primary cancer of the colon (American Joint Commission on Cancer 6^th Edition (AJCC) stage II or III), breast (AJCC stage I–III), or lung (AJCC stage I–II) from 2007–2011. These cancer sites and stages were selected to identify populations that might receive chemotherapy, excluding older adults diagnosed with advanced stage disease, where the risk-benefit assessment of PIM use is less clear. To be included in the study cohort, individuals had to have: (1) at least 12 months of continuous Medicare enrollment in Parts A and B prior to their diagnosis date (set to the first day of the month of diagnosis) to assess relevant comorbid conditions, (2) Medicare Part D (prescription drug) coverage during the month of diagnosis, and (3) at least one prescription medication dispensed in the month of diagnosis. Individuals who were diagnosed at autopsy, did not survive throughout the month of diagnosis, or had a missing month of diagnosis were excluded.

Demographic and tumor characteristics were obtained in the month of diagnosis, including age, sex, race (White, Black, Other), marital status (married, single, divorced/widowed/ separated), year of diagnosis, and AJCC stage. Using Medicare claims data, we assessed comorbidity using the Charlson Comorbidity Index(15) (categorized as 0, 1, 2+) during the 12-months prior to the month of diagnosis and medication burden using a count of the number of unique prescriptions (generic name level) in the month prior to the month of diagnosis (categorized as 0–2, 3–5, 6–9, 10+). Using the same 12-month period, we estimated each individual’s predicted probability of being frail based on an internally(16) and externally(17) validated Medicare claims-based model including 20 unique variables (e.g., diagnosis of weakness, wheelchair claim, home oxygen claim). The resulting predicted probabilities were categorized as 0–<10% (low probability of frailty), 10–<20% (low-intermediate), 20–<50% (intermediate-high), and 50%+ (high). Finally, we constructed a variable for whether an individual underwent surgical resection or received chemotherapy or radiation in the 6-months following diagnosis. In addition, we constructed indicators for the use of specific chemotherapeutic agents in each month from diagnosis through month 11. Codes used to define cancer treatments from Medicare claims are listed in the Supplementary Materials.

We identified PIMs according to the 2012 Beers criteria,(4) a medication screening tool developed to help healthcare providers optimize medication use in older adults. The Beers criteria, originally developed in 1991, have been regularly updated by the American Geriatrics Society. The 2012 Beers criteria include 34 drugs to avoid in older adults and 18 drugs that should be avoided as they could exacerbate a coexisting disease (i.e., drug-disease interactions). Prevalence of any PIM dispensing was evaluated monthly, starting 6 months before and going through 23 months following the diagnosis month (month 0). The pre-cancer diagnosis period (months −6 to −1) was used to establish baseline PIM use patterns prior to a cancer diagnosis and to facilitate comparison with published prevalence estimates of PIM in the general older adult population. We selected the 23 months following the month of cancer diagnosis to evaluate patterns of PIM use during the initial treatment (month 0–11) and continuing (month 12–23) phases of cancer care,(18, 19) as the transitions in PIM prevalence related to cancer treatment were of particular interest. All analyses were anchored at the month of cancer diagnosis (month 0).

Of particular interest were PIMs related to the alleviation of cancer symptoms and treatment-related side effects (referred to as cancer-related PIMs). We specifically examined the broad Beers criteria categories of “Pain” and “Central Nervous System” to identify cancer-related PIMs for pain, anxiety/depression, and insomnia and then identified specific PIMs frequently used to manage nausea, diarrhea, and appetite in cancer patients. For presentation purposes, cancer-related PIM analyses were limited to specific PIMs that had a >1% prevalence in at least one month for at least one cancer site.

To be included in the denominator for a given month, individuals had to have: (1) at least 12 months of continuous Medicare enrollment in Parts A and B prior to the given month of interest, (2) Medicare Part D (prescription drug) coverage during the month, and (3) at least one prescription medication dispensed (or days’ supply carried over) in a given month. Dispensing of a prescription medication was a requirement for an individual to contribute to the denominator, consistent with prior studies,(20–22) as an adult who is not receiving any prescription medications cannot be exposed to a PIM. Because eligibility was determined on a month-by-month basis, the number of individuals contributing to monthly prevalence measures changes over time.

PIMs were identified using Medicare Parts A, B, and D claims as described by Jiron et al.(22) We first used the Anatomical Therapeutic Chemical (ATC) classification system to identify all medications and classes of medications listed as part of the 2012 Beers criteria and then developed a crosswalk of these medications to their specific National Drug Codes (NDCs). For PIMs due to drug-disease interactions, we used International Classification of Diseases, 9^th Revision, Clinical Modification (ICD-9) codes to identify specific conditions in the 12-months prior to the given month of interest.

We identified potential drug-drug interactions involving chemotherapeutics through a review of the literature(13, 23–29) with confirmation by two clinical pharmacists with expertise in oncology and geriatrics. Our review was limited to potential interactions with chemotherapeutic agents used as initial treatment for stage I–III breast, II–III colon, or I–II lung cancer. We again used the ATC classification system to identify all medications included in our review and developed a crosswalk of these medications to their specific NDCs. A clinical pharmacist further classified each potential drug-chemotherapeutic interaction as minor (caution advised), moderate (monitor/modify therapy), or major (avoid/use alternative) using Micromedex® Online (Micromedex, Inc., Ann Arbor, MI, USA).

The 12-month period prevalence of potential drug-chemotherapy interactions was evaluated on the specific chemotherapy agent-level during the initial treatment phase of care (month 0–11). To be included in the denominator for the period prevalence analyses, individuals had to have: (1) continuous Medicare enrollment in Parts A, B, and D (and no managed care coverage) during the entire initial treatment phase and (2) at least one claim for the administration of a specific chemotherapeutic agent of interest in at least one month during this period. To be included in the numerator, patients had to have a prescription claim for a potential interacting drug and overlapping days’ supply with the administration of a specific chemotherapeutic agent of interest. For presentation purposes, we restricted our descriptive analyses to specific chemotherapies that had more than 100 patients in the denominator in an attempt to avoid imprecise estimates. All prevalence with a numerator of <11 were suppressed due to SEER-Medicare privacy rules. Specific chemotherapeutics included were 5-fluorouracil (5-FU)/capecitabine (colon), cyclophosphamide (breast), doxorubicin (breast), methotrexate (breast), paclitaxel (breast, lung), carboplatin (lung), cisplatin (lung), etoposide (lung), and gemcitabine (lung). Tamoxifen (breast) was also included in analysis, but is considered endocrine therapy.

We estimated the monthly point prevalence of any PIM among the three cancer cohorts from the 6 months before through the 23 months following the month of cancer diagnosis (month 0). Given the specific interest in the influence of chemotherapy on the use of PIMs as supportive care agents, month-level analyses were stratified by chemotherapy receipt (yes versus no). The monthly prevalence of specific cancer-related PIMs was computed among the three cohorts. Finally, we estimated the 12-month period prevalence of potential drug-chemotherapy interactions during the initial treatment phase, stratifying results by cancer site. All statistical analyses were performed in SAS version 9.4 (Cary, NC). This study was performed after approval by the University of North Carolina at Chapel Hill Institutional Review Board.

After applying all study inclusion criteria, there were 19,318 stage I–III breast, 7,283 stage II–III colon, and 7,237 stage I–II lung cancer patients included in our baseline cohorts (Supplemental Table 1). Demographic and clinical characteristics of these patients are reported in Table 1. Median age at cancer diagnosis was similar across cohorts at 75 years for breast and lung cancer to 78 years for colon cancer. Variation in the burden of comorbidity at the time of cancer diagnosis was observed across the three cohorts, where lung cancer patients had the highest proportion of patients with a Charlson comorbidity score of 2 or more (lung: 36%, colon: 26%, and breast: 19%) and were dispensed the greatest number of prescription medications in the month prior to cancer diagnosis (% receiving 6+ medications, lung: 33%, colon: 27%, and breast: 25%). However, colon cancer patients had the highest probability of being frail (using a claims-based model), while breast cancer patients had the lowest probability. During the six months following cancer diagnosis, 98% of all colon cancer patients and 94% of breast cancer patients underwent surgical resection, compared with only 46% of lung cancer patients. Chemotherapy was most common among colon cancer patients (34%), while radiation was frequent among breast cancer patients (44%).

The monthly prevalence of any PIM prior to cancer diagnosis was similar across all three cancer cohorts, hovering between 37–40% (Figure 1), similar to general population estimates among Medicare beneficiaries.(22) However, following cancer diagnosis, different patterns emerged. The prevalence of PIM among breast cancer patients consistently decreased over the period following diagnosis, whereas PIM prevalence sharply increased in the first few months following a colon or lung cancer diagnosis, and slowly decreased back to pre-diagnosis levels over the following 23 months. Decreases in PIM prevalence in the breast cancer cohort were attributable to decreased dispensing of estrogen following diagnosis (9% at six months prior to diagnosis to 0.5% one year after diagnosis).

We further plotted the monthly prevalence of any PIM dispensing stratified by chemotherapy receipt (Figures 2a–c). Longitudinal patterns across cancer sites were consistent showing a sharp immediate increase in the monthly prevalence of both PIM dispensing among individuals who initiated chemotherapy within the first six months following cancer diagnosis. In contrast, the monthly prevalence of PIM remained relatively constant among those who do not initiate chemotherapy.

The monthly prevalence of cancer-related PIM use is presented in Figure 3a–c. Across all three cancer cohorts, the prevalence of amitriptyline use (a tricyclic antidepressant) was high, but remained relatively constant over the study period at 1.5–2.5%. The monthly prevalence of cyclobenzaprine (a muscle relaxant) was also steady, but lower across cohorts ranging from 0.5–1.5%. The lowest cancer-related PIM prevalence was for dicyclomine (an antispasmodic), which was also stable across the trajectory of care, with the exception of the colon cancer cohort, where a small spike in the month of diagnosis was observed.

For breast cancer, promethazine (an anti-emetic) was the most common cancer-related PIM, with a monthly prevalence that was elevated in the first nine months following cancer diagnosis (1%–3%), but returned to pre-diagnosis levels (<1%) thereafter. In the lung cancer cohort, the prevalence of promethazine and megestrol (a drug indicated to increase appetite) use increased after cancer diagnosis and remained elevated throughout the following year (promethazine: 1–3%; megestrol: 2–4%). The colon cancer cohort had the most cancer-related PIM use, including a high prevalence of metoclopramide, a pro-motility drug used to speed transit after surgery, (3–4%) and promethazine (3%) use during the initial months following diagnosis. Increasing prevalence in belladonna alkaloid use, an antispasmodic, (2–3%) and megestrol (1–3%) were largely sustained during the year following cancer diagnosis.

The 12-month period prevalence for selected potential drug-chemotherapeutic interactions are presented in Table 2 along with a brief description of the potential interaction outcome (e.g., increased chemotherapy effect). Overall, the reported period prevalence of potential drug-chemotherapy interactions ranged from 1–31%. The most prevalent potential interactions in the colon cancer cohort were for 5-FU/capecitabine and hydrochlorothiazide (a diuretic used to manage blood pressure, 22%) and warfarin (a drug used to treat blood clots, 15%), both classified as moderate potential drug interactions. In the breast cancer cohort, the most prevalent potential interactions classified as major included cyclophosphamide and hydrochlorothiazide (31%) and methotrexate and non-steroidal anti-inflammatory drugs (drugs used to inflammation, pain, and fever, 16%), while the most prevalent moderate interaction was for methotrexate in combination with proton-pump inhibitors (drugs used for suppression of gastric acid, 29%). Finally, in the lung cancer cohort, warfarin was considered a major interaction when used together with etoposide (14%) and gemcitabine (15%).