Adjusted odds ratios and 95% confidence intervals for associations of adherence and nonadherence for 1 or more PMD for 6 chronic conditions to nonadherence to hormonal therapy in women with early-stage breast cancer in the MarketScan database from January 1, 2010, to December 31, 2012. Error bars represent 95% confidence intervals. PMD indicates prior medication(s).
Association between adherence to prior medication for each of 6 chronic conditions (by disease) and nonadherence to hormonal therapy in women with early-stage breast cancer in the MarketScan database from January 1, 2010, to December 31, 2012, as compared with those with no prior medications for the 6 chronic conditions included in the study.
eTable. Univariate and multivariate analysis of demographic and clinical predictors of non-adherence to endocrine therapy among women with early-stage breast cancer (2010-2012) in the MarketScan database.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Neugut AI, Zhong X, Wright JD, Accordino M, Yang J, Hershman DL. Nonadherence to Medications for Chronic Conditions and Nonadherence to Adjuvant Hormonal Therapy in Women With Breast Cancer. JAMA Oncol. 2016;2(10):1326–1332. doi:10.1001/jamaoncol.2016.1291
Is adherence and/or nonadherence to chronic medications associated with adherence and/or nonadherence to oral hormonal therapy (HT) in women with early-stage breast cancer?
In this retrospective cohort study of 21 255 women with breast cancer, those who used 1 or more medication(s) prior to HT and were adherent (n = 9223) had a 9.8% nonadherence rate to HT, while those who were nonadherent to their medications (n = 4214) had a 23.1% nonadherence rate to HT.
Those who are nonadherent to chronic medications are at increased risk for nonadherence to HT and could benefit from vigilance and possible future interventions.
While adjuvant hormonal therapy (HT) reduces mortality for women with nonmetastatic breast cancer, nonadherence to HT is common.
We investigated the association between patterns of prior nonadherence to medications for chronic conditions with HT nonadherence.
Design, Setting, and Participants
For this retrospective cohort study, the MarketScan database was scanned for women 18 years and older who had been diagnosed with nonmetastatic breast cancer between January 1, 2010, and December 31, 2012, and who filled 2 or more prescriptions for tamoxifen and/or an aromatase inhibitor.
Main Exposures and Outcomes
Nonadherence to medications for 6 chronic conditions (hypertension, hyperlipidemia, gastroesophageal reflux disease, thyroid disease, diabetes, osteoporosis) in the 12 months before diagnosis was defined as a medication possession ratio (MPR) less than 80%. Nonadherence to HT was defined as an MPR less than 80% between the first and last prescription for HT up to 2 years.
Multivariable logistic regression was used to determine the association between prior medication nonadherence and HT nonadherence.
Of 21 255 women treated with adjuvant HT, 3314 (15.6%) were nonadherent, and age (<55 or ≥75 years vs 55-64 years), higher 30-day out-of-pocket costs, and increased comorbidities were associated with nonadherence. Women without prior medications for 1 of the chronic conditions (n = 7828 [37%]) had an 18.4% nonadherence rate to HT. Those who used 1 or more medication prior to HT and were adherent (n = 9223 [43%]) had a 9.8% nonadherence rate to HT (relative to those without prior medications: odds ratio [OR] 0.56; 95% CI, 0.50-0.61), while those who were nonadherent to their chronic medications (n = 4214 [20%]) had a 23.1% nonadherence rate to HT (OR 1.43; 95% CI, 1.30-1.58). Adherence and nonadherence for medications for each of the 6 medical conditions was associated with adherence or nonadherence for HT, respectively.
Conclusions and Relevance
We found that nonadherence to medications for chronic conditions prior to HT was associated with greater nonadherence to oral HT in patients with breast cancer. Medication nonadherence history may play an important role in determining patients at risk for nonadherence to a subsequent medication for a different illness, such as HT, and a potential target for future interventions.
Patients often do not take their medications as prescribed or discontinue them early, and these patterns are associated with treatment failure, poor outcomes, and increased health care costs. This has been a particular issue for medications for chronic diseases that require long-term use and are often without apparent direct effect on symptoms, such as drugs for hypertension or hyperlipidemia. As a consequence, there has been a great effort to determine reasons and predictors for nonadherence in an effort to increase adherence.1-3
Given the fact that medications used in oncology are potentially life-saving or life-prolonging, it is surprising that nonadherence to these medications is also common and thus also a topic of research interest. In particular, there is a large pool of growing literature on lack of adherence to oral drugs, which now constitute about 25% of antineoplastic agents.4,5 This literature has focused on oral hormonal therapy (HT) used as adjuvant therapy for patients with early-stage breast cancer. Multiple studies indicate that 30% to 50% of women do not fully complete 5 years of therapy.6-16 This nonadherence can negatively affect breast cancer recurrence and mortality.17,18 Predictors of nonadherence include patients who are younger or older, African-American, being treated by a specialist other than a medical oncologist, have poor belief in the treatment’s efficacy, have higher copayments for the drug, and have lower financial resources.19,20
A few studies have suggested that prior adherence to chronic medications could predict subsequent adherence to newly prescribed cardiovascular or psychiatric drugs.21-23 We hypothesized that prior nonadherence to medications for chronic illnesses could similarly be associated with nonadherence to the use of oral HT in women with breast cancer. We used a large population-based database to address this question.
Data for this study came from the Truven Health Analytics MarketScan Database, a medical and pharmacy insurance claims database that includes information on 49 million active employees, early retirees, and COBRA (Consolidated Omnibus Budget Reconciliation Act) continuers and their dependents insured by approximately 100 employer-sponsored plans. The database contains information on outpatient and inpatient services, prescription drugs, demographic information, eligibility status, type of health plan, and cost of services.
We identified 21 255 women 18 years or older diagnosed with breast cancer (International Classification of Diseases, Ninth Revision [ICD-9] code 174.x) who took oral HT between January 1, 2010, and December 31, 2012, within the MarketScan database. A claims-based algorithm was used to identify women with nonmetastatic breast cancer; details can be found elsewhere.24 This algorithm was adapted from previously used algortihms.25,26 In part, early-stage breast cancer was identified in the algorithm by the use of curative surgery (mastectomy or breast conserving surgery), axillary lymph node dissection, and radiation therapy. To be eligible, patients were required to be continuously enrolled 12 months or more before and after the diagnosis date and to have filled at least 2 prescriptions for either tamoxifen or an aromatase inhibitor (ie, anastrozole, letrozole, and/or exemestane) during the study period.
The index date for the purposes of demographic characteristic data assessment was the time of breast cancer diagnosis. Age at diagnosis was categorized as younger than 45, 45 to 54, 55 to 64, 65 to 74, and 75 years or older. Geographic location was categorized as northeast, north central, south, and west. Health plan type was grouped into preferred provider organization, health maintenance organization, comprehensive, and other. In addition, patients were categorized by year of HT initiation (2010-2012).
Comorbidity was estimated using the Deyo classification of the Charlson comorbidity score (0, 1, or >1).27 Prior comorbidity was based on data from the 1 year prior to the breast cancer diagnosis and used to control for the health status of the study subjects.
The copayment for HT was defined as the amount paid by a subscriber for a 30-day prescription. For payments for more than a 30-day (ie, 60-day, 90-day) prescription, copayment was adjusted to the 30-day copayment amount. Copayment was categorized as less than $5, $5 to $9, $10 to $14, $15 to $19, and $20 or more based on the common copayment amount.
Prior medications included common drugs used for 6 chronic conditions: diabetes, hypertension, hyperlipidemia, thyroid disease, osteoporosis, and gastroesophageal reflux disease (Table 1). Patients were required to have at least 2 prescription claims for any medication included among the 6 chronic conditions. The 6 chronic conditions were not mutually exclusive; those with prescriptions for more than 1 condition were included in more than 1 disease cohort. Medication adherence was defined as a medication possession ratio (MPR)28 between the first and last prescriptions of 80% or greater. The MPR for the prior chronic medication was calculated based on the period between the first and last prescription for that medication during the 12 months prior to the index diagnosis of breast cancer and before the initiation of HT.
Unless stated otherwise, if adherence for an individual was calculated for multiple diseases or drugs, it was estimated as the average adherence for the drugs for all the conditions (ie, the numerator was the sum of the days that all the medications were taken divided by the sum of the number of days each medication was intended to be taken). An average MPR 80% or greater for the combination of the drugs that were taken was considered adherent. It was therefore possible that a patient could be overall adherent on average but nonadherent to 1 or more of the individual medications, or vice versa.
Analyses were also conducted that were stratified by the disease or condition (ie nonadherence was estimated solely for the drugs specific to that condition). In that case, the average MPR was limited to the drugs of that condition and the other drugs ignored.
The primary outcome was defined as adherence (yes/no) to HT, defined as an MPR of 80% or more. The MPR was calculated as the ratio of the sum of all the days supplies filled by the pharmacy divided by the number of days between the first and last prescription of HT over a minimum follow-up of 1 year and a maximum of 2 years.3
All patients were followed for up to 2 years, through December 31, 2013. We censored patients at the date at which they disenrolled from MarketScan or at the end of the study period. A change in HT from tamoxifen to an aromatase inhibitor, or vice versa, was not considered discontinuation.
Both univariate and multivariate logistic regression models were used to assess the association between covariates and the outcome of nonadherence to HT. For multivariate analysis, all the above covariates were included. The main factor of interest—nonadherence to drugs for chronic conditions prior to a breast cancer diagnosis—was categorized based on the number of prior medications and overall average adherence to these medications as previously described. For patients who were found to have had medications for multiple conditions prior to the initiation of HT, nonadherence was estimated by the average MPR to all of these prior medications. It was thus possible to be considered adherent, yet to have been nonadherent to 1 or more individual medications. In addition, an alternative multivariate model was fitted to detect the dose-response relationship using the same outcome and adjusting for the same covariates. The reference group for most of the analyses consisted of those with no prior medications. Analyses were also conducted stratifying by condition, (ie, estimating nonadherence to medications for each condition separately and the association of nonadherence for each condition with subsequent HT nonadherence). We rejected the null hypothesis at the .05 level of significance in all models. Statistical analyses were conducted using SAS software version 9.4 (SAS Institute).
This study used deidentified data and was deemed exempt by the institutional review board at Columbia University Medical Center.
We identified 21 255 women diagnosed with early-stage breast cancer who initiated HT between January 1, 2010, and December 31, 2012. Of these, 3314 (15.6%) were nonadherent (17.9% for tamoxifen and 13.6% for aromatase inhibitors; P = .07). The clinical and demographic characteristics of the cohort are presented in the eTable in the Supplement. The majority were older than 55 (57%), from a preferred provider organization, and had no comorbidities (82%). Of the total, 13 229 patients (62%) initially used aromatase inhibitors, while 8026 (38%) started with tamoxifen. Almost two-thirds (63%) used at least 1 prior medication for a chronic condition.
We found an association between age and nonadherence to HT. Compared with women with breast cancer between 55 and 64 years of age, those younger than 45 years (OR, 2.00; 95% CI, 1.77-2.25) and between age 45 and 54 years (OR, 1.43; 95% CI, 1.30-1.58) were more likely to be nonadherent, as were those 75 years or older (OR, 1.24; 95% CI, 1.06-1.45). As the 30-day out-of-pocket costs rose from $5, there was a gradual increase in nonadherence, with those paying more than $20 having odds of nonadherence of 2.09 (95% CI, 1.77-2.46) compared with those with an out-of-pocket cost of $0 to $4. Those from the south region also had greater nonadherence (eTable in the Supplement).
The mean rates of nonadherence to the chronic medications in the year prior to the HT were quite substantial: 37% for diabetes; 28%, hypertension; 30%, hyperlipidemia; 21%, thyroid disease; 32%, osteoporosis; and 38%, gastroesophageal reflux disease.
Both univariate and multivariate analysis showed an association between adherence to chronic medications prior to breast cancer diagnosis and subsequent HT adherence (eTable in the Supplement). While those with no prior medications for any of the 6 chronic conditions had an HT nonadherence rate of 18%, those who had at least 1 prior medication had a nonadherence rate of 13% (P < .001). Among women who were on average adherent to 1 or more drugs prior to their breast cancer diagnosis, their nonadherence rate to HT was 9.8% (n = 900 of 9223) while women who were on average nonadherent to their medications prior to their breast cancer were nonadherent to HT at a rate of 23.1% (n = 975 of 4214) (Table 2). The adjusted odds ratios were 0.56 (95% CI, 0.50-0.61) and 1.43 (95% CI, 1.30-1.58), respectively.
Figure 1 illustrates that patients taking more medications for chronic diseases are more likely to be nonadherent to subsequent HT if they are consistent in being nonadherent to all of their medications. Conversely, if patients were adherent to all of their medications, they were more likely to be adherent to subsequent HT. Generally speaking, patients taking more than 1 chronic disease medication were consistent in being either adherent or nonadherent to all the medications they were taking. Overall, for the 21 medications included in the analysis for the 6 chronic conditions, there was a 1% average discordance rate for adherence and/or nonadherence. The average discordance rate for diabetes medications (2 medications) was 1%; 1.5%, hypertension (8 medications); less than 1%, hyperlipidemia (5 medications); not applicable to thyroid disease (1 medication); less than 1%, osteoporosis (2 medications); and 1%, gastroesophageal reflux disease (3 medications).
Table 2 shows the results of a multivariate analysis for each individual chronic condition. These results are illustrated in Figure 2. They demonstrate that for each condition, as compared with those with no prior medication, patients who were adherent to their medication were less likely to be nonadherent to their subsequent HT (OR, 0.49-0.74). For each condition, with the exception of osteoporosis, patients who were nonadherent to their prior medication were more likely to be nonadherent (OR, 1.21-1.88).
In this study, we confirmed that nonadherence to HT was common, with 15.6% of women having a MPR less than 80% within the first 2 years of starting their medication. We also found that a history of nonadherence to medications in the year prior to patient’s breast cancer diagnosis was associated with increased nonadherence to oral HT, and similarly, adherence to medications for chronic conditions was associated with decreased HT nonadherence. Furthermore, regardless of the number of medications a woman was taking, nonadherence to HT was the lowest when women were adherent to all of their chronic condition medications. As women increased the number of medications to which they were nonadherent, the odds of nonadherence to HT increased. In addition, we confirmed that younger and older age, higher out-of-pocket costs, being from the South, and a greater number of comorbidities were also associated with greater nonadherence.
The use of MPR greater than 80% has been a common measure of nonadherence3 and widely used in the assessment of nonadherence to breast cancer HT.19,20,29 Prior studies have found rates of nonadherence during the first 1 to 2 years in the 10% to 20% range, consistent with our finding of a 15.6% nonadherence rate during this timeframe.
Only a few studies have evaluated the association between prior adherence patterns and subsequent medication adherence. One study22 of adherence to cardiovascular medications collected self-reported adherence to medications in the week prior to a hospital admission for 646 cardiovascular patients with a mean of 8 medications and found that a lower average adherence predicted worse postdischarge average adherence. A 1% increase in adherence was associated with a 1.6% increase in the postdischarge adherence. Another study used MarketScan to assess in patients with schizophrenia the effect of adherence to their antipsychotic medications on later adherence to their medications for chronic conditions and found an association.23 A third study investigated 1433 members of United Healthcare who were prescribed statins and found their adherence (MPR >85%) over an initial 6-month period to be 82.7% and decreasing to 78.8% in the second 6-month period. Prior adherence predicted subsequent adherence with the area under the receiver operating curve of 0.78 (P < .05).21 Interestingly, studies have suggested that the converse may occur; breast cancer diagnosis and treatment may reduce adherence to medications for chronic conditions.30,31
It has also been found in the context of cancer screening that those who are compliant with one screening modality tend to be more adherent to another modality of cancer screening.32 Thus the finding that adherence to medications for 1 illness and associated adherence to subsequent HT for breast cancer seems logical and appears to reflect a form of health-conscious behavior. It has been shown, for example, that patients who are adherent in the placebo arms of randomized trials tend to have lower mortality rates than those who are not, and thus this behavior tends to reflect underlying traits that are positive for the individual.33
As noted previously, nonadherence to HT for breast cancer can have a significant effect on survival outcomes.17,18 The hope is that by identifying patients at highest risk for nonadherence, interventions can be developed and targeted at higher-risk groups.19,34
At the least, such information can alert the oncologist to be vigilant in patients who have previously been nonadherent and thus make greater efforts to inquire at regular visits and encourage compliance. Such efforts may include behavioral, educational, or other interventions.2 No intervention has yet been proven to be effective in the oncology context. Our own group is currently conducting a randomized trial within SWOG (formerly the Southwest Oncology Group) (S1105) that randomizes postmenopausal women receiving aromatase inhibitors to either receive biweekly text messaging or not. Results from this trial are pending. Furthermore, a variety of interventions have been evaluated to reduce toxic effects of these agents in hopes of improving adherence.35
The finding with regard to out-of-pocket costs confirms previous studies by our group that demonstrated that higher copayments were associated with decreased adherence to HT.36 More directly, Hershman and colleagues20 have also shown that increased financial net worth was associated with increased adherence. Other studies also suggested that costs play a role in adherence to HT.7 This has been demonstrated both in the context of medications for chronic conditions as well as in the context of oncology drugs. It would suggest that efforts should be made to reduce out-of-pocket costs for life-saving drugs such as HT for early breast cancer. Indeed, a recent article from our group37 found that the introduction of generic aromatase inhibitors with a concomitant decline in cost led to a significant increase in adherence to the medication.
We found that both younger and older age were associated with nonadherence to HT. This confirms findings from prior studies.29,38-40 It is unclear exactly how age affects nonadherence. It can be associated with race, physician-patient interaction, and other factors, as well as financial factors, the same issues that play a role at every age.38
Our study had several strengths. It used a large database that is population-based. Follow-up of the study population over the 3 years of the study was excellent with very little dropout.
There were limitations as well. The population was almost entirely insured; therefore, results cannot easily be generalized to those who are not insured. We did not have detailed information on cancer stage, and therefore used a validated claims-based algorithm to determine that patients were not metastatic. Also this database does not provide us with detailed information on other potentially important confounders, such as race. Finally, we did not have detailed information on why patients were nonadherent to their HT or medications for chronic conditions. In particular, we could not comment on whether adverse effects affected the rate of nonadherence. This may have resulted in some misclassification. Finally, this paper was limited to an investigation of nonadherence and did not investigate early discontinuation.
We found that nonadherence to adjuvant HT among women with breast cancer was less likely in those who were adherent to other medications prior to breast cancer diagnosis compared with those receiving no medications and was more likely among women who were nonadherent to medications for chronic conditions. We also found this relationship was dose dependent such that the more medication classes a patient was adherent to, the more likely they were to be adherent to HT; and vice-versa, nonadherence to multiple medication classes was associated with increased nonadherence to HT. Our findings suggest that a detailed history of medication use and nonadherence may help identify patients who are at risk for nonadherence to breast cancer HT and who may benefit from targeted interventions.
Corresponding Author: Alfred I. Neugut, MD, PhD, Division of Medical Oncology, Columbia University Medical Center, Columbia University, 722 W 168th St, Rm 725, New York, NY 10032 (firstname.lastname@example.org).
Published Online: June 9, 2016. doi:10.1001/jamaoncol.2016.1291
Author Contributions: Dr Neugut had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Neugut, Hershman.
Acquisition, analysis, or interpretation of data: Neugut, Zhong, Wright, Accordino, Yang, Hershman.
Drafting of the manuscript: Neugut, Zhong, Accordino, Hershman.
Critical revision of the manuscript for important intellectual content: Neugut, Wright, Yang, Hershman.
Statistical analysis: Zhong, Wright, Yang.
Obtained funding: Neugut, Hershman.
Administrative, technical, or material support: Hershman.
Study supervision: Neugut.
Conflict of Interest Disclosures: Dr Neugut has served as a consultant to Pfizer, Teva Pharmaceuticals, Takeda Pharmaceuticals, United BioSource Corporation, and serves on the Medical Advisory Board of EHE, Intl. No other conflicts are reported.
Funding/Support: Supported by a grant to Dr Neugut from the Department of Defense (grant no.BC043120). Dr. Hershman is supported by grants from the Breast Cancer Research Foundation, the American Society of Clinical Oncology, and the Witten Family Foundation.
Role of the Funder/Sponsor: The funders/sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.