Background:

Adherence and persistence studies face several methodologic difficulties, including short-term mortality. We compared approaches to quantify adherence and persistence to first line (1L) oral targeted therapy (TT) in patients diagnosed with metastatic renal cell carcinoma (mRCC).

Methods:

Patients with mRCC ages 66 years or more who initiated TTs within 4 months of diagnosis were identified in the Surveillance, Epidemiology, and End Results Medicare-linked database (2007–2015). Adherence [proportion of days covered (PDC) >80%] was calculated using (i) PDC with a fixed 6-month denominator including then excluding patients who died within the 6 months and (ii) PDC with a denominator measuring time on treatment. Risk of nonpersistence was obtained by censoring death or treating death as a competing risk using cumulative incidence functions.

Results:

Among 485 patients with mRCC initiating a 1L oral TT (sunitinib, 64%; pazopanib, 25%; other, 11%), 40% died within 6 months. Adherence was higher after restricting to patients who survived (60%) compared with including those patients and assigning zero days covered after death (47%). Risk of nonpersistence was higher when censoring patients at death, 0.91 [95% confidence interval (CI), 0.88–0.94], compared with treating death as a competing risk, 0.75 (95% CI, 0.71–0.79).

Conclusions:

Different approaches to handling death resulted in different adherence and persistence estimates in the metastatic setting. Future studies should explicitly report the proportion of patient deaths over time and explore appropriate methods to account for death as competing risk.

Impact:

Use of several approaches can provide a more comprehensive picture of medication-taking behavior in the metastatic setting where death is a major competing risk.

The development of oral oncologic therapies has dramatically changed the clinical landscape in oncology. In the last decade, over 50 oral anticancer agents received FDA approval. Currently, they represent approximately 25% to 30% of all anti-neoplastic agents in development (1, 2). Increasing use and development of oral anticancer agents is predominantly driven by patient preference over intravenous treatment as they are easily administered, noninvasive, and can be received at home, thus requiring fewer clinic visits and more flexibility in timing and location of administration (3–8).

However, the shift in treatment paradigm from intravenous to oral administration places the responsibility of drug acquisition and administration from a trained medical worker to the patient (9). Moreover, studies have shown that the complexity of many oral anticancer regimens and the difficulty managing side effects of these drugs in the home are associated with suboptimal adherence (10). Poor adherence to treatment can reduce drug efficacy, increase treatment failure, and result in early discontinuation of therapy. This is especially concerning for patients with metastatic disease as oral targeted therapy (TT) is more often used, and patients treated in the metastatic setting are more likely to be frail and have higher risks of toxicities (11).

Because of the rapid increase in drug approvals for oral TTs for metastatic disease, nonadherence to oral therapies is an emerging issue. Moreover, treatment pathways and the advent of new combination regimens due to pharmaceutical innovation will only increase regimen complexity over time. However, there is no gold standard for measuring adherence or persistence, nor is there guidance for how these should be assessed in high mortality populations (12–14). Prescription refill data can provide insight to adherence and persistence to treatment. The most common claims-based approaches to estimating adherence are defined by an 80% threshold of proportion of days covered (PDC) or medication possession ratio, with most adherence studies for oral agents being conducted in patients with nonmetastatic diseases (15). Application of these same measures is challenging in the metastatic setting where death is a major competing risk for therapy continuation. To implement many of the traditional adherence measures, such as measuring adherence over a fixed time period, patients either have to survive up to that time period or those who die are considered nonadherent after death. Therefore, key decisions about how to handle mortality must be made when designing and analyzing studies of medication adherence and persistence in metastatic cancer populations. To this end, the objective of this study was to compare different measures of and approaches to quantifying adherence and persistence to oral oncologic therapy using a motivating example of 1L oral TTs in older adults diagnosed with metastatic renal cell carcinoma, a particularly high-mortality population.

Data source

We used the Surveillance Epidemiology and End Results (SEER) program-Medicare linked dataset, representing a linkage of >3.3 million individuals in SEER with their Medicare enrollment and claims data. The NCI's SEER program is a surveillance system collecting demographics, clinical and tumor data, selected treatments, vital status, and cause of death for individuals diagnosed with cancer within one of 18 SEER regions, currently covering 34.6% of the cancer population in United States (16). Medicare claims provide information on service dates, diagnoses, procedures, and prescription dispensing for covered individuals (16).

Study population

Because Medicare Part D data are only available in the linked SEER-Medicare starting on January 1, 2007, we identified patients diagnosed with a first primary mRCC ages 66 years or older from July 2007 to September 2015, who initiated a 1L oral TT within 4 months of diagnosis. Individuals diagnosed with RCC were identified using the International Classification of Disease for Oncology, Third Edition site code: C64.9 and histology codes indicative of RCC (8260, 8310, 8316–8320, 8510, and 8959). Stage IV RCC cases were identified according to the American Joint Committee on Cancer, 6th Edition staging system. Because the day of diagnosis is not available in SEER, we assigned the first day of the diagnosis month as the diagnosis date (17). To be eligible, patients must have continuous enrollment in Medicare Part A, B, and D from 6 months prior to and up to 4 months after diagnosis (including the month of diagnosis) or death, whichever came first. This 10-month window was determined by clinician input. The rationale was to account for symptoms and clinical diagnostic work-up occurring within 6 months prior to a de novo mRCC diagnosis and subsequent initiation of treatment or death within 4 months after the diagnosis. Prescriptions dispensed for oral TTs available during the study period (sunitinib, sorafenib, pazopanib, axitinib, and everolimus) were identified in Medicare Part D data using National Drug Codes. The index date was defined as the date of treatment initiation. Patients missing stage or reported by autopsy or death certificate were excluded. To prevent inclusion of a recurrent mRCC case, patients who received a nephrectomy prior to the date of diagnosis were also excluded.

Measures of adherence and nonpersistence

Adherence to 1L TTs were calculated using the PDC with fixed and variable observation periods. A “fixed” PDC was defined as the total number of days covered by therapy over a fixed interval period (18–20). We also computed a “variable” PDC, which represents a patient's time on treatment. This was defined as the total number of days covered by therapy over the number of days between the first fill and the last refill plus its days of supply or until death, whichever comes first (18–20). For both fixed and variable PDCs, we defined adherence at an threshold of 80% PDC. This cutpoint is widely used and has shown to be a reasonable cutpoint for stratifying patients with low and high adherence (21). To account for the recommended dosing schedule for sunitinib (4 weeks on and 2 weeks off), when the PDC for periods where patients had 4-week days' supply and 4 weeks of pills dispensed, their days' supply was set to 6 weeks (22). Guideline approved dosing schedules for other TTs had a once or twice daily schedule with no “off weeks” in treatment.

Persistence to 1L TTs was defined as the time from treatment initiation to discontinuation. Patients were persistent to therapies if the duration of one claim (including days' supplied) overlapped with the date of a subsequent claim. A 30-day grace period was applied between successive prescriptions to allow for delays in regular refilling. For sunitinib, the 30-day period was applied after the end of the 6-week cycle. Nonpersistence was defined as discontinuing treatment due to a switch to a different therapy or lack of a subsequent claim for 1L TT by the end of the days' supply plus the grace period. The date of discontinuation was assigned to (i) the date of treatment switch, (ii) date of death, or (iii) last day of days' supply for the last refill for patients who did not die or have a treatment switch prior to the last day of days' supply.

For overlapping prescription fills for the same drug, the second overlapping prescription shifted forward to begin at the end of the days' supply for the previous prescription. Hospitalized patients were assumed to have received their medications through the facility during hospitalization. Therefore, if a patient had a drug on hand when hospitalized, their hospital length of stay (days) was counted as a fill. If a patient accumulated an extra supply of their medication because of their hospital stay, we assumed that supply could be used once he/she returned home (23). For gaps between hospitalization discharge and prescription fill, we shifted days' supply that overlapped with the stay to the days following the end of the relevant stay (23).

Patient demographics and clinical characteristics

The following patient demographic and clinical characteristics were summarized at the time of mRCC diagnosis: age, sex, race (white, black, other), marital status (married/domestic partner, divorced/widowed/separated, unmarried, unknown), nonclear cell or clear-cell histologies, tumor node involvement (N0, N1, unknown), and tumor size (T1a, T1b, T2, unknown). Residence in an urban or rural area, Medicare low-income subsidy status (full, partial, none), and a census-tract poverty indicator (percentage of residents living below poverty) were used to describe socioeconomic status. Frailty and comorbidities were evaluated 6 months prior to mRCC diagnosis. The predicted probability of frailty was calculated using a validated Medicare claims-based algorithm developed by Faurot and colleagues (24) These predictors included demographic characteristics, durable medical equipment charges, and geriatric syndromes. Gagne comorbidity score (25), a single numeric comorbidity score for predicting short- and long-term mortality, was categorized as ≤0 (low), 1 (intermediate), and 2 or more (high).

Statistical methods

Adherence and persistence to oral TTs were described at 3, 6, and 12 months from the index date. Adherence was calculated using a variable PDC, and two fixed PDCs where (i) patients who died at any point within the specified time period were excluded and (ii) patients who died at any point within the specified time period were included but assigned zero days covered after death. A sensitivity analysis was conducted restricting the study population to those who had at least two prescription fills.

Baseline hazard functions for the cumulative incidence of nonpersistence were obtained using two approaches. We first fitted a Cox proportional hazard model (26) to obtain the baseline Kaplan–Meier estimates, censoring patients at death. Cumulative incidence functions (CIF) for nonpersistence indicating death as a competing risk were generated using Aalen–Johansen risk functions (27). These functions and the cumulative incidence of death were plotted to illustrate the impact of mortality on the risk of nonpersistence. Sensitivity analyses for nonpersistence assessed 15- and 60-day grace periods, re-categorizing discontinuations as deaths that occurred within the 30-day grace period, and restriction to patients with at least two prescription fills.

All analyses were conducted using SAS version 9.4. This study was approved by the University of North Carolina at Chapel Hill Institutional Review Board (18–0632).

After applying inclusion and exclusion criteria, 485 incident patients with mRCC who initiated a 1L oral TT within 4 months of their diagnosis were identified (Fig. 1). Of these patients, the majority initiated sunitinib (64%) followed by pazopanib (25%), with 60% having at least two prescription fills. The median age at diagnosis was 74.2 years (IQR, 69.8–78.9) with a predominance of men (59%) and white race (80%). Within 3 months of treatment initiation, 27% of patients died, increasing to 40% by 4 months, and 61% by 12 months. There were 84 patients (17%) who went entered hospice care before the date of their last days supply of their last fill. Among these patients, the median days between the start of 1L treatment to the date of hospice start was 40 days (IQR, 21–86). The median days between the date start of hospice and the end of the last days' supply was 18 days (IQR 7–26). More detailed demographic and clinical characteristics of the cohort are presented in Table 1.

Figure 1.

Study population selection criteria.

Figure 1.

Study population selection criteria.

Close modal
Table 1.

Baseline demographic and clinical characteristics of patients who initiated an oral TT in the first-line (n = 485).

Median, nIQR (%)
1L therapy type 
 Sunitinib 310 63.9 
 Pazopanib 119 24.5 
 Othera 56 11.6 
Had a nephrectomyb 
 No 360 74.2 
 Yes 125 25.8 
Year of diagnosis 
 2007–2008 110 22.7 
 2009–2010 76 15.7 
 2011–2012 108 22.3 
 2013–2015c 191 39.3 
Age at diagnosis 74.3 69.8, 78.9 
 65–74 263 54.2 
 74–84 196 40.4 
 85+ 26 5.4 
Sex 
 Male 288 59.4 
 Female 197 40.6 
Race/ethnicity 
 White 387 79.8 
 Black 28 5.8 
 Other 70 14.4 
Census tract % < poverty 
 1st quartile (lowest) 100 20.6 
 2nd quartile 127 26.2 
 3rd quartile 139 28.7 
 4th quartile (highest) 119 24.5 
Low-income subsidy 
 None 308 63.5 
 Full 158 32.6 
 Partial 19 3.9 
Urban/rural 
 Urban 422 87.0 
 Rural 63 13.0 
SEER region 
 Northeast 68 14.0 
 South 129 26.6 
 North central 69 14.2 
 West 219 45.2 
Marital status 
 Married/domestic partner 289 59.6 
 Divorced/widowed/separated 138 28.4 
 Unmarried 43 8.9 
 Unknown 15 3.1 
Gagne Combined Comorbidity Index 
 Low 340 70.1 
 Intermediate 104 21.4 
 High 41 8.5 
Faurot frailty prediction 0.041 0.033, 0.057 
 <10% 383 79.0 
 10%–<20% 58 12.0 
 20% + 44 9.0 
Hospice before end of 1L treatment 
 No 401 82.7 
 Yes 84 17.3 
Histology group 
 Non–clear cell 73 15.0 
 Clear cell 412 85.0 
Nodal involvement 
 N0 269 55.5 
 N1 73 15.0 
 Unknown 143 29.5 
Tumor size 
 T1a 32 6.6 
 T1b 59 12.1 
 T2 89 18.4 
 Unknown 305 62.9 
Median, nIQR (%)
1L therapy type 
 Sunitinib 310 63.9 
 Pazopanib 119 24.5 
 Othera 56 11.6 
Had a nephrectomyb 
 No 360 74.2 
 Yes 125 25.8 
Year of diagnosis 
 2007–2008 110 22.7 
 2009–2010 76 15.7 
 2011–2012 108 22.3 
 2013–2015c 191 39.3 
Age at diagnosis 74.3 69.8, 78.9 
 65–74 263 54.2 
 74–84 196 40.4 
 85+ 26 5.4 
Sex 
 Male 288 59.4 
 Female 197 40.6 
Race/ethnicity 
 White 387 79.8 
 Black 28 5.8 
 Other 70 14.4 
Census tract % < poverty 
 1st quartile (lowest) 100 20.6 
 2nd quartile 127 26.2 
 3rd quartile 139 28.7 
 4th quartile (highest) 119 24.5 
Low-income subsidy 
 None 308 63.5 
 Full 158 32.6 
 Partial 19 3.9 
Urban/rural 
 Urban 422 87.0 
 Rural 63 13.0 
SEER region 
 Northeast 68 14.0 
 South 129 26.6 
 North central 69 14.2 
 West 219 45.2 
Marital status 
 Married/domestic partner 289 59.6 
 Divorced/widowed/separated 138 28.4 
 Unmarried 43 8.9 
 Unknown 15 3.1 
Gagne Combined Comorbidity Index 
 Low 340 70.1 
 Intermediate 104 21.4 
 High 41 8.5 
Faurot frailty prediction 0.041 0.033, 0.057 
 <10% 383 79.0 
 10%–<20% 58 12.0 
 20% + 44 9.0 
Hospice before end of 1L treatment 
 No 401 82.7 
 Yes 84 17.3 
Histology group 
 Non–clear cell 73 15.0 
 Clear cell 412 85.0 
Nodal involvement 
 N0 269 55.5 
 N1 73 15.0 
 Unknown 143 29.5 
Tumor size 
 T1a 32 6.6 
 T1b 59 12.1 
 T2 89 18.4 
 Unknown 305 62.9 

aOther treatments include sorafenib, axitinib, and everolimus.

bNephrectomy occurred within 4 months of diagnosis.

cPatients diagnosed through September 2015.

Adherence

After assigning days covered as zero after death, the median PDC with a 6-month fixed denominator was 0.47 (IQR, 0.23–0.80). Alternatively, restricting to those who survived 6 months after treatment initiation resulted in a higher PDC of 0.70 (IQR, 0.33–0.93). The PDC was highest when calculated with a variable denominator (i.e., time on treatment) at 0.98 (IQR, 0.74–1.00). Sensitivity analyses restricted to patients who had at least two prescription fills showed increased 6-month median PDCs including and excluding patients who died, 0.70 (IQR, 0.47–0.93) and 0.82 (IQR, 0.50–0.97), respectively. No differences were observed in the variable PDC after applying the exclusion.

Figure 2 illustrates how the alternative mortality approaches resulted in different estimates of adherence (PDC ≥ 0.80) over time. Adherence estimates were higher in the analysis restricting to patients who survived up to 6 months (40%; 95% CI, 35%–46%) compared including these patients and assigning days covered as zero after their death (25%; 95% CI, 21%–29%). Adherence was consistently higher after restricting to patients who survived across time periods. Differences ranged between 13 and 17 percentage points. Adherence estimates, calculated using the variable PDC, were highest at 78% (95% CI, 74%–82%). Sensitivity analyses restricting to patients who had at least two prescription fills showed higher estimates of adherence when calculated using a fixed PDC, whereas adherence estimates calculated using a variable PDC showed a marginal increase (Supplementary Fig. S1).

Figure 2.

Proportion of patients with mRCC adherent to oral TTs. Adherence to oral TTs calculated at fixed 3, 6, and 12 months are shown for (i) patients who died and were defined as nonadherent after death and (ii) patients who did not survive through the time period of interest and were excluded. Adherence measured at varying times on treatment is also depicted.

Figure 2.

Proportion of patients with mRCC adherent to oral TTs. Adherence to oral TTs calculated at fixed 3, 6, and 12 months are shown for (i) patients who died and were defined as nonadherent after death and (ii) patients who did not survive through the time period of interest and were excluded. Adherence measured at varying times on treatment is also depicted.

Close modal

Nonpersistence

Most patients discontinued treatment or died within 12 months of treatment initiation (93%) with a median time to discontinuation of 77 days (IQR, 42–162). After restricting to those who survived 12 months, the median time to discontinuation was substantially higher at 127 days (IQR, 55–286). Figure 3 reports the cumulative mortality and risk of nonpersistence to oral TTs in patients with mRCC, showing the two methods of handling death. The 12-month risk of nonpersistence was 0.91 (95% CI, 0.88–0.94) when censoring follow-up at death versus 0.75 (95% CI, 0.71–0.79) when accounting for death as a competing risk. As expected, sensitivity analyses assessing 15- and 60-day gap periods showed nonpersistence generally decreased with increasing gap periods. Sensitivity analyses restricting to patients who had at least two prescription fills showed lowered risk estimates for nonpersistence Analyses reclassifying discontinuations due to death occurring within the grace period as a death also showed lowered estimates (Supplementary Table S1).

Figure 3.

CIF for nonpersistence with death as a competing risk and Kaplan–Meier curve for death and nonpersistence where death is censored. The cumulative incidence is presented at days 90, 180, and 365 below. The 95% confidence bands are represented in gray.

Figure 3.

CIF for nonpersistence with death as a competing risk and Kaplan–Meier curve for death and nonpersistence where death is censored. The cumulative incidence is presented at days 90, 180, and 365 below. The 95% confidence bands are represented in gray.

Close modal

This descriptive study explored methodologic approaches to account for mortality when estimating adherence and nonpersistence to oral TTs in incident patients with mRCC. By restricting to patients who survived, adherence using a fixed PDC (e.g., 6 months) was substantially higher compared with defining patients as nonadherent (i.e., no covered days) after death. Both approaches provide meaningful information about adherence in high short-term mortality populations; however, these results highlight methodologic decisions that must be considered and clearly reported. For instance, researchers interested in informing policies regarding the use and cost of treatment may assign nonadherence after death to obtain the proportion of those who achieve adherence within a specific time period. With this approach, lowered adherence estimates should be expected as nonadherence will be predominantly driven by patient deaths. Alternatively, researchers may only include those who survived if they are interested in evaluating adherence through gaps in prescription fills. However, estimates can only be generalizable to a unique subpopulation with superior prognosis.

Changes in adherence and death over time, must also be considered. Adherence to oral TTs generally decreases over time from treatment initiation (28). Our findings also showed that the magnitude of difference between both approaches increased at each time point as a result of increasing mortality. To this end, calculating adherence using a fixed period may be more useful for medical conditions in which patients are expected to survive and remain on long-term therapy. Researchers may instead become interested in measuring adherence during time on treatment. Compared with a fixed period, adherence using a variable PDC was higher. However, variation in follow-up time due to differences in treatment response, demographics, or clinical characteristics within the study population must be considered. Patients with shorter follow-up will also have fewer prescription fills, providing little time to assess the extent to which a patient's actual refilling behavior corresponds to their prescribed regimen. Moreover, PDCs using a variable denominator may be difficult to compare across studies because of differences between patient populations and follow-up times.

Given these limitations, obtaining the CIF for nonpersistence may be a better measure for understanding medication taking behavior in this population. Plots of the CIF may also provide better characterization of nonpersistence in the metastatic setting over time. Yet, most studies reporting persistence will exclude patients who died, inducing similar concerns related to selection bias, as stated above or account for death as part of the definition for discontinuation. Patients who discontinue treatment due to death may have substantially different profiles including more severe disease or comorbid conditions than those who intentionally discontinue treatment (e.g., treatment failure or switch) and survive. Other studies will use Kaplan–Meier curves to estimate risk of nonpersistence over time and censor patients at their time of death. Although this approach accounts for death by assuming that the competing risk can be prevented (29), we propose estimating the CIF as it is arguably of much more clinical interest. In closed cohorts with no loss to follow-up, it corresponds to an observable quantity: the number of cases divided by the population size at baseline. Further, misinterpretation of the Kaplan–Meier estimator as a 1 − CIF estimator can lead to upward bias in risk estimates because conditional risk will overestimate CIF. This study illustrates the extent of this bias by showing a lower risk of nonpersistence with death as a competing risk versus censoring death. Thus, obtaining the CIF using the Aalen–Johansen estimator is recommended to provide an unbiased estimate in this setting.

There are several limitations to this study. First, if a medication was dispensed, it was assumed to be taken as prescribed, as claims-based measures of adherence and persistence rely on prescription fill dates and days' supply. Studies have shown drug exposures obtained through prescription fills and other procedure codes have been shown to be more reliable than records of physician-ordered prescriptions (30). Second, the reason for treatment discontinuation (e.g., adverse event, patient/provider preference, disease progression) is also not available in claims data. Therefore, a discontinuation was assumed for those who did not refill their prescription within the 30-day grace period. Another limitation of this study is small sample size. Differences in results would be expected within certain subgroups where prognosis is worse, such as those with higher comorbidity, frailty scores, or different site of metastasis. However, results could not be further stratified by these characteristics due to sample size constraints. Larger and more robust studies, are needed to further enhance our understanding of adherence and persistence in these different patient populations. Results may also not be generalizable to younger incident mRCC patients as this study was conducted in older adults who likely have a higher baseline risk of mortality. However, the recommended methodologic considerations remain applicable in younger age groups with metastatic disease. Finally, these analyses are only conducted in those diagnosed with mRCC and should also be replicated in other cancer types to understand the generalizability of these findings.

Our study has several strengths. This is one of few studies to evaluate adherence and persistence in the metastatic cancer setting. The majority of existing studies were conducted in cancer populations that either explicitly excluded patients who died or excluded those with metastatic disease (1, 2, 19, 28, 31–35). Studies will also often require patients to meet extended continuous enrollment criteria throughout the study period of interest (36–38), thereby implicitly excluding those who died. Importantly, several studies describing methods for defining adherence and persistence have been published. Yet, to date, the impact of mortality has not been explored. This study demonstrates the effects of higher death rates on the magnitude of adherence and persistence estimates.

The cost of treatment for patients with mRCC is rising as more effective combination therapies with oral TTs are approved (e.g., axitinib/pembrolizumab and nivolumab/cabozantinib). The cost and benefit of these expensive drugs to patients who have a high risk of death are a growing concern among payers and policy makers. Yet, measures of adherence and persistence are not the same in the trial setting. Therefore, understanding how to best characterize medication adherence to oral therapies in the real-world is needed. Future studies should leverage advanced statistical methods to evaluate the impact of these alternative methods for defining medication adherence and their effects on oncologic outcomes.

Conclusion

Medication adherence and persistence is an emerging issue in the metastatic setting. Yet, death is a major competing risk. Future studies should consider reporting the proportion of deaths, the potential for bias, and implications for generalizability depending on how death was handled. Researchers should also consider obtaining CIFs to account for the competing risk of death and plot functions of nonpersistence to better characterize longitudinal patterns in patients diagnosed with metastatic cancers. Use of several approaches can provide a more comprehensive picture of medication taking behavior in the metastatic setting. Our study aimed to characterize adherence to first line therapy in patients with mRCC. Future studies evaluating the impact of these alternative methods in other cancer types and subsequent lines of therapy are warranted.

D.S. Chun reports other support from Bristol Myers Squibb during the conduct of the study as well as other support from Takeda Pharmaceuticals outside the submitted work. M.J. Funk reports grants from NIH during the conduct of the study as well as salary support from the Center for Pharmacoepidemiology, Department of Epidemiology at UNC Chapel Hill (current members: GlaxoSmithKline, UCB Biosciences, Takeda, AbbVie, Boehringer Ingelheim); in addition, J. Funk is a member of the Scientific Steering Committee (SSC) for a post-approval safety study of an unrelated drug class funded by GSK. All compensation for services provided on the SSC is invoiced by and paid to UNC Chapel Hill. K. Gooden reports personal fees from Bristol Myers Squibb outside the submitted work. T. Stürmer reports other support from BMS during the conduct of the study as well as other support from Novartis, Roche, and Novo Nordisk outside the submitted work; in addition, T. Stürmer receives investigator-initiated research funding and support as a principal investigator (R01 AG056479) from the National Institute on Aging (NIA), and as co-investigator (R01 HL118255, R01MD011680) from the NIH and also receives salary support as Director of Comparative Effectiveness Research (CER), NC TraCS Institute, UNC Clinical and Translational Science Award (UL1TR002489), the Center for Pharmacoepidemiology (current members: GlaxoSmithKline, UCB BioSciences, Takeda, AbbVie, Boehringer Ingelheim), from pharmaceutical companies (Novo Nordisk), and from a generous contribution from Dr. Nancy A. Dreyer to the Department of Epidemiology, University of North Carolina at Chapel Hill. T. Stürmer does not accept personal compensation of any kind from any pharmaceutical company. T. Stürmer owns stock in Novartis, Roche, and Novo Nordisk. J.L. Lund reports grants from AbbVie, Inc. and other support from GlaxoSmithKline, Boehringer Ingelheim, UCB Biosciences, Takeda, and Merck outside the submitted work. No disclosures were reported by the other authors.

D.S. Chun: Conceptualization, data curation, formal analysis, funding acquisition, visualization, methodology, writing–original draft. B. Hicks: Conceptualization, formal analysis, funding acquisition, methodology, writing–review and editing. S.P. Hinton: Data curation, software, funding acquisition, writing–review and editing. M.J. Funk: Methodology, writing–review and editing. K. Gooden: Conceptualization, methodology, writing–review and editing. A.P. Keil: Conceptualization, formal analysis, methodology, writing–review and editing. H.-J. Tan: Conceptualization, formal analysis, funding acquisition, writing–review and editing. T. Stürmer: Conceptualization, methodology, writing–review and editing. J.L. Lund: Conceptualization, resources, formal analysis, supervision, funding acquisition, visualization, methodology, writing–original draft, writing–review and editing.

This study was funded by a Lineberger Comprehensive Cancer Center Developmental Grant. Dr. Jessica Edwards, PhD, Department of Epidemiology, University of North Carolina, Chapel Hill, North Carolina, contributed to the data analysis.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Geynisman
DM
,
Wickersham
KE
.
Adherence to targeted oral anticancer medications
.
Discov Med
2013
;
15
:
231
41
.
2.
Bassan
F
,
Peter
F
,
Houbre
B
,
Brennstuhl
MJ
,
Costantini
M
,
Speyer
E
, et al
.
Adherence to oral antineoplastic agents by cancer patients: definition and literature review
.
Eur J Cancer Care
2014
;
23
:
22
35
.
3.
Borner
M
,
Scheithauer
W
,
Twelves
C
,
Maroun
J
,
Wilke
H
.
Answering patients' needs: oral alternatives to intravenous therapy
.
Oncologist
2001
;
6
:
12
6
.
4.
Liu
G
,
Franssen
E
,
Fitch
MI
,
Warner
E
.
Patient preferences for oral versus intravenous palliative chemotherapy
.
J Clin Oncol
1997
;
15
:
110
5
.
5.
Weingart
SN
,
Brown
E
,
Bach
PB
,
Eng
K
,
Johnson
SA
,
Kuzel
TM
, et al
.
NCCN Task Force Report: oral chemotherapy
.
J Natl Compr Canc Netw
2008
;
6
:
S1
14
.
6.
Teuffel
O
,
Cheng
S
,
Ethier
MC
,
Diorio
C
,
Martino
J
,
Mayo
C
, et al
.
Health-related quality of life anticipated with different management strategies for febrile neutropenia in adult cancer patients
.
Support Care Cancer
2012
;
20
:
2755
64
.
7.
Beusterien
K
,
Grinspan
J
,
Kuchuk
I
,
Mazzarello
S
,
Dent
S
,
Gertler
S
, et al
.
Use of conjoint analysis to assess breast cancer patient preferences for chemotherapy side effects
.
Oncologist
2014
;
19
:
127
34
.
8.
Schott
S
,
Schneeweiss
A
,
Reinhardt
J
,
Bruckner
T
,
Domschke
C
,
Sohn
C
, et al
.
Acceptance of oral chemotherapy in breast cancer patients: a survey study
.
BMC Cancer
2011
;
11
:
129
.
9.
Barton
D
.
Oral agents in cancer treatment: the context for adherence
.
Semin Oncol Nurs
2011
;
27
:
104
15
.
10.
Jacobs
JM
,
Ream
ME
,
Pensak
N
,
Nisotel
LE
,
Fishbein
JN
,
MacDonald
JJ
, et al
.
Patient experiences with oral chemotherapy: adherence, symptoms, and quality of life
.
J Natl Compr Canc Netw
2019
;
17
:
221
8
.
11.
Hurria
A
,
Lichtman
SM
.
Clinical pharmacology of cancer therapies in older adults
.
Br J Cancer
2008
;
98
:
517
22
.
12.
Arnet
I
,
Kooij
MJ
,
Messerli
M
,
Hersberger
KE
,
Heerdink
ER
,
Bouvy
M
.
Proposal of standardization to assess adherence with medication records: methodology matters
.
Ann Pharmacother
2016
;
50
:
360
8
.
13.
Lam
WY
,
Fresco
P
.
Medication adherence measures: an overview
.
Biomed Res Int
2015
;
2015
:
217047
.
14.
Bjarnadottir
MV
,
Czerwinski
D
,
Onukwugha
E
.
Sensitivity of the medication possession ratio to modelling decisions in large claims databases
.
Pharmacoeconomics
2018
;
36
:
369
80
.
15.
Barillet
M
,
Prevost
V
,
Joly
F
,
Clarisse
B
.
Oral antineoplastic agents: how do we care about adherence?
Br J Clin Pharmacol
2015
;
80
:
1289
302
.
16.
SEER-Medicare: Brief Description of the SEER-Medicare Database
.
2018
. https://healthcaredelivery.cancer.gov/seermedicare/overview/.
Accessed October 3, 2018.
17.
SEER-Medicare: Defining the Date of Diagnosis & Treatment
.
2018
. https://healthcaredelivery.cancer.gov/seermedicare/considerations/date.html.
Accessed October 14, 2018.
18.
Wang
S
,
Huang
Z
,
Trubenberg
S
.
Measuring medication adherence with simple drug use and medication switching
.
SAS Global Forum
2013
,
2013
19.
Nau
D
.
Proportion of days covered (PDC) as a preferred method of measuring medication adherence
.
Pharmacy Quality Alliance
2012
.
20.
Sperber
CM
,
Samarasinghe
SR
,
Lomax
GP
.
An upper and lower bound of the medication possession ratio
.
Patient Prefer Adherence
2017
;
11
:
1469
78
.
21.
Karve
S
,
Cleves
MA
,
Helm
M
,
Hudson
TJ
,
West
DS
,
Martin
BC
.
Good and poor adherence: optimal cut-point for adherence measures using administrative claims data
.
Curr Med Res Opin
2009
;
25
:
2303
10
.
22.
Motzer
RJ
,
Jonasch
E
,
Agarwal
N
,
Bhayani
S
,
Bro
WP
,
Chang
SS
, et al
.
Kidney cancer, version 2.2017, NCCN Clinical Practice Guidelines in Oncology
.
J Natl Compr Canc Netw
2017
;
15
:
804
34
.
23.
Centers for Medicare and Medicad Services
.
Medicare 2018 Part C & D Star rating Technical notes
;
2017
.
24.
Faurot
KR
,
Jonsson Funk
M
,
Pate
V
,
Brookhart
MA
,
Patrick
A
,
Hanson
LC
, et al
.
Using claims data to predict dependency in activities of daily living as a proxy for frailty
.
Pharmacoepidemiol Drug Saf
2015
;
24
:
59
66
.
25.
Gagne
JJ
,
Glynn
RJ
,
Avorn
J
,
Levin
R
,
Schneeweiss
S
.
A combined comorbidity score predicted mortality in elderly patients better than existing scores
.
J Clin Epidemiol
2011
;
64
:
749
59
.
26.
Cox
D
.
Regression models and life-tables
.
J Roy Stat Soc: Ser B
1972
;
34
:
187
220
.
27.
Aalen
O
,
Johansen
S
.
Empirical transition matrix for nonhomogeneous markov-chains based on censored observations
.
Scand J Stat
1978
;
5
:
141
50
.
28.
Greer
JA
,
Amoyal
N
,
Nisotel
L
,
Fishbein
JN
,
MacDonald
J
,
Stagl
J
, et al
.
A systematic review of adherence to oral antineoplastic therapies
.
Oncologist
2016
;
21
:
354
76
.
29.
Cole
SR
,
Edwards
JK
,
Naimi
AI
,
Munoz
A
.
Hidden imputations and the Kaplan-Meier estimator
.
Am J Epidemiol
2020
;
189
:
1408
11
.
30.
Funk
MJ
,
Landi
SN
.
Misclassification in administrative claims data: quantifying the impact on treatment effect estimates
.
Curr Epidemiol Rep
2014
;
1
:
175
85
.
31.
Wigertz
A
,
Ahlgren
J
,
Holmqvist
M
,
Fornander
T
,
Adolfsson
J
,
Lindman
H
, et al
.
Adherence and discontinuation of adjuvant hormonal therapy in breast cancer patients: a population-based study
.
Breast Cancer Res Treat
2012
;
133
:
367
73
.
32.
Wulaningsih
W
,
Garmo
H
,
Ahlgren
J
,
Holmberg
L
,
Folkvaljon
Y
,
Wigertz
A
, et al
.
Determinants of non-adherence to adjuvant endocrine treatment in women with breast cancer: the role of comorbidity
.
Breast Cancer Res Treat
2018
;
172
:
167
77
.
33.
Farias
AJ
,
Hansen
RN
,
Zeliadt
SB
,
Ornelas
IJ
,
Li
CI
,
Thompson
B
.
The association between out-of-pocket costs and adherence to adjuvant endocrine therapy among newly diagnosed breast cancer patients
.
Am J Clin Oncol
2018
;
41
:
708
15
.
34.
Oberguggenberger
AS
,
Sztankay
M
,
Beer
B
,
Schubert
B
,
Meraner
V
,
Oberacher
H
, et al
.
Adherence evaluation of endocrine treatment in breast cancer: methodological aspects
.
BMC Cancer
2012
;
12
:
474
.
35.
Farias
AJ
,
Du
XL
.
Association between out-of-pocket costs, race/ethnicity, and adjuvant endocrine therapy adherence among Medicare patients with breast cancer
.
J Clin Oncol
2017
;
35
:
86
95
.
36.
Weaver
KE
,
Camacho
F
,
Hwang
W
,
Anderson
R
,
Kimmick
G
.
Adherence to adjuvant hormonal therapy and its relationship to breast cancer recurrence and survival among low-income women
.
Am J Clin Oncol
2013
;
36
:
181
7
.
37.
Miller
LA
,
Stemkowski
S
,
Saverno
K
,
Lane
DC
,
Tao
Z
,
Hackshaw
MD
, et al
.
Patterns of care in patients with metastatic renal cell carcinoma among a U.S. payer population with commercial or Medicare advantage membership
.
J Manag Care Spec Pharm
2016
;
22
:
219
26
.
38.
Vogelzang
NJ
,
Pal
SK
,
Ghate
SR
,
Swallow
E
,
Li
N
,
Peeples
M
, et al
.
Clinical and economic outcomes in elderly advanced renal cell carcinoma patients starting pazopanib or sunitinib treatment: a retrospective Medicare claims analysis
.
Adv Ther
2017
;
34
:
2452
65
.

Supplementary data