Five years of treatment provides the optimum duration of tamoxifen therapy for the prevention of breast cancer recurrence and mortality. Durations of adjuvant tamoxifen therapy less than 5 years are associated with poorer outcomes for breast cancer patients. The purpose of the study was to assess rates of tamoxifen nonpersistence (early discontinuation) in women aged 35 years or older using prescription refill data from a national prescribing database.
A cohort of 2816 women commencing tamoxifen as initial hormonal therapy was identified between January 2001 and January 2004. The cumulative tamoxifen persistence rate was calculated for these women and the relation between nonpersistence and clinical and demographic variables assessed.
Within 1 year of commencing treatment the cumulative tamoxifen nonpersistence rate was 22.1%. This is twice the rate of treatment discontinuation observed in other studies by this time. By the end of follow-up at 3.5 years, the cumulative nonpersistence rate had increased to 35.2%. Determinants of nonpersistence identified included age and a history of antidepressant use.
Tamoxifen, a selective estrogen receptor modulator, has been the standard adjuvant hormonal treatment for hormone-responsive early breast cancer for over 20 years. Five years of adjuvant tamoxifen is the recommended duration of treatment and results in a reduction in the relative breast cancer recurrence risk of 46% and the relative risk of death of 26%.1 It is estimated that use of tamoxifen in the adjuvant setting can prevent 40,000 breast cancer-related deaths worldwide every year.2 Comparisons of treatment durations indicate that women receiving less than 5 years of tamoxifen therapy have significantly higher breast cancer recurrence rates and mortality.1, 3 For this reason nonadherence to tamoxifen therapy, in particular early discontinuation of treatment, is likely to result in significantly worse outcomes for breast cancer patients.
Adherence is defined as the extent to which patients take medications as prescribed.4, 5 The length of time from initiation to discontinuation of treatment is a specific aspect of adherence and is referred to as persistence.4, 6 The measurement of tamoxifen persistence offers advantages over adherence, as results can be easily interpreted in relation to breast cancer outcome data. Rates of discontinuation or nonpersistence reported in clinical trials of adjuvant tamoxifen range from 16% to 32% at 5 years.7–11 These are low rates for a chronic therapy but not unusual in clinical studies where patients are carefully selected, highly motivated, and closely monitored. Persistence rates for patients not participating in clinical trials are typically lower, even with treatments that are documented to reduce the risk of life-threatening events and mortality—for example, statins.12 Studies of tamoxifen nonpersistence outside of clinical trial settings have reported nonpersistence rates of 17% at 2 years to 31% at 5 years,13–15 rates that are comparable to those achieved in clinical trials. These studies, however, focused on elderly populations and used patient self-report as a measure of treatment discontinuation, a method that has considerable limitations and may significantly underestimate the true rate of tamoxifen nonpersistence.16, 17 In addition, the self-selection of patients for these studies may have introduced significant bias to the results.18 To date there have been no published studies with objective measurements of tamoxifen persistence outside of the clinical trial setting.
This study aims to evaluate persistence with tamoxifen therapy, in women aged 35 years or older, using prescription refill data from a national prescribing database. Prescription refill data provide accurate and objective estimates of medication use in large populations over long periods of time5, 6 and have been used widely to measure persistence with other chronic therapies.19–21 This is the first study to evaluate persistence with tamoxifen therapy using this method and the results may provide an insight into the medication-taking behavior of women prescribed chronic oral therapies for the treatment of breast cancer. How well and for how long oncology patients take prescribed oral therapies is of growing importance as the availability and use of oral antineoplastic agents increases. Oncology patients express a preference for oral treatments,22 hence, identifying those at risk of nonpersistence, quantifying the extent of nonpersistence, and developing strategies to improve persistence are of vital importance.
MATERIALS AND METHODS
Source of Data
The study population was identified from the Irish Health Services Executive (HSE) Primary Care Reimbursement Services (PCRS) pharmacy database. The HSE-PCRS provides free health services including the provision of medicines to 1.15 million people in Ireland, approximately one-third of the total population. Eligibility for the service is primarily by means test and age, with patients over 70 years of age automatically qualifying for entry. Therefore, groups such as the elderly and the socially disadvantaged are overrepresented with respect to the general population. Basic demographic information such as age and sex are available for all patients and full details of every drug dispensed within the scheme are recorded. For this study national prescribing data was available for the years 2000 to 2005. No data on diagnosis are available in the database.
The study cohort was identified from the HSE-PCRS database and included all women over the age of 35 who commenced tamoxifen as initial hormonal therapy between January 2001 and January 2004. The date of the first prescription for tamoxifen during this period was identified as the index date for each patient. Women with no tamoxifen prescribed in the 12 months before the index date were identified as commencing tamoxifen. Initial hormonal therapy with tamoxifen was defined as having no other hormonal therapy (toremifene, anastrozole, letrozole, or exemestane) prescribed in the 12 months before the index date. In addition to these criteria for inclusion, patients were also required to have received at least 1 prescription for any item in the 12 months preceding the index date. This excluded patients with a preexisting prescription for tamoxifen who had become newly eligible for inclusion in the HSE-PCRS scheme.
Measurement of Tamoxifen Persistence
Tamoxifen nonpersistence was defined as 180 consecutive days of no tamoxifen supply without alternative hormonal therapy during that time. A longitudinal dataset of medication supply was constructed for each patient and from this nonpersistence with tamoxifen was identified. The number of days supply of tamoxifen was calculated from quantity and dosage information associated with each prescription record. Patients were initially classified as nonpersistent if they had a period of 180 consecutive days of no tamoxifen supply after their index date. The date of first tamoxifen discontinuation was recorded for each of these patients. From this initial group of patients those with no prescription for any item in the 12 months after their tamoxifen discontinuation date were reclassified as lost to follow-up. Patients commencing an alternative hormonal therapy before the end of the 180 consecutive days of no tamoxifen supply were reclassified as treatment switchers. These patients were not included in the nonpersistent group of patients. Patients restarting tamoxifen or starting another hormonal therapy after 180 days of no tamoxifen supply remained classified as nonpersistent along with patients who discontinued tamoxifen and received no further hormonal therapy. All patients were followed for between 1 and 3.5 years (median, 2.7 years) from their index date to identify tamoxifen nonpersistence and for a minimum of 1 year (median, 1.7 years) after their tamoxifen discontinuation date to identify subsequent hormonal therapy use.
Potential Determinants of Nonpersistence
Variables considered as potential determinants of nonpersistence were chosen with reference to previous studies of tamoxifen persistence13–15 and adherence.22–25 These included age; the number of comorbidities such as diabetes, cardiovascular disease, and respiratory disease; and the number of different pharmacological agents prescribed in the year before tamoxifen initiation. Additional variables also considered were the use of antidepressant, antipsychotic, or anxiolytic/hypnotic agents, and the presence of cognitive or functional impairment such as Parkinson disease or dementia.
Age was identified for patients at the index date according to 10 year age-bands specified in the HSE-PCRS database. Patients were identified as having antidepressant, antipsychotic, benzodiazepine anxiolytic/hypnotic, or benzodiazepine-related (zopiclone, zolpidem, zaleplon) hypnotic use if they received at least 1 prescription for any of these drugs in the year preceding the index date. Cognitive or functional impairment was identified by the use of anti-Parkinson disease or antidementia agents during the same period. Comorbidities were defined as follows from drug use in the year preceding the index date: diabetes by the use of insulin or oral hypoglycemic agents; cardiovascular disease by the use of at least 1 cardiovascular agent per month; and respiratory disease by the use of at least 2 prescriptions for an inhaled beta agonist or any other pharmacological agent used for respiratory disease. The mean number of pharmacological agents prescribed per month was calculated for the year preceding the index date.
The cumulative tamoxifen nonpersistence rate was estimated using a Kaplan-Meier analysis. Study subjects were censored at the time of loss to follow-up, treatment switch, or end of follow-up, whichever occurred first. A Cox proportional hazards regression model was used to estimate the relation between nonpersistence with tamoxifen and the demographic and clinical variables described previously for a maximum follow-up of 3.5 years. A stepwise selection procedure with criteria for entry of P < .1 was used to select the final optimal model. Crude and adjusted hazard ratios (HR) and 95% confidence intervals (CI) are presented for independent variables. To assess the impact of censoring from loss to follow-up, a sensitivity analysis was performed using the method of extremes and assuming that patients either discontinued tamoxifen immediately after censoring or persisted with tamoxifen to the end of follow-up. Significance at P < .05 was assumed unless stated otherwise. SAS v. 9.1 (SAS Institute, Cary, NC) was used for all statistical analyses.
A cohort of 2816 women commencing tamoxifen as initial hormonal therapy between January 2001 and January 2004 was identified from the HSE-PCRS database. The characteristics of these women are presented in Table 1. In Ireland, primary therapy for early breast cancer and surveillance thereafter is delivered from tertiary referral centers and provided by a multidisciplinary team comprising surgical, medical, and radiation oncologists. Data from the Irish National Cancer Registry indicates that there are almost 2100 new breast cancer diagnoses in Ireland every year,26 with AJCC stage III breast cancer accounting for approximately 16% of these and stage IV approximately 13%. Previous research indicates that almost 75% of breast cancers are hormone receptor-positive.27
Table 1. Characteristics of Patients Starting Tamoxifen on the HSE-PCRS Database Between January 2001 and January 2004 (n = 2816)
HSE-PCRS indicates the Irish Health Services Executive Primary Care Reimbursement Services.
One-third of patients in the cohort of women identified were above the age of 75, reflecting the overrepresentation of elderly patients in the HSE-PCRS database but also the higher rates of estrogen receptor-positive breast cancer seen in this population. In the year before tamoxifen initiation antidepressant and antipsychotic prescribing rates are comparable to the mean yearly prescribing rates for these drugs in the total HSE-PCRS population of women over the age of 35. Benzodiazepine and benzodiazepine-related prescribing rates, however, were higher in the study cohort, 41.3% and 13.8% vs 31.2% and 8.5%. It is possible that this higher prescribing rate in study subjects may be accounted for by the use of these agents for women receiving chemotherapy or radiotherapy before tamoxifen initiation. The prevalence of diabetes, respiratory, and cardiovascular diseases was 4.2%, 13.7%, and 21.3%, respectively; 32.3% of women in the study cohort had at least 1 of these comorbidities. Parkinson disease or dementia was identified in 2.9% and 0.9% of patients, respectively. The median number of pharmacological agents received per month in the year preceding initiation of tamoxifen was 1.92 (interquartile range, 0.67, 3.91).
The outcomes of the 2816 women commencing tamoxifen as initial hormonal therapy between January 2001 and January 2004 women are presented in Figure 1. The number of women identified as nonpersistent with tamoxifen treatment was 26.5%. Of these women 1.2% switched to another hormonal therapy after 180 consecutive days of no tamoxifen supply; 3.7% restarted tamoxifen between 180 days and 1 year after discontinuing; and 1.4% restarted tamoxifen more than 1 year after discontinuing. The Kaplan-Meier plot of tamoxifen persistence is shown in Figure 2. The cumulative nonpersistence rate was 11.1% (95% CI: 10.0%, 12.3%) 30 days after commencing tamoxifen; 14.5% (95% CI: 13.2%, 15.9%) at 90 days; 22.1% (95% CI: 20.6%, 23.7%) by 1 year, 28.4% (95% CI: 26.6%, 30.3%) by 2 years, and 35.2% (95% CI: 32.5%, 37.9%) by the end of follow-up at 3.5 years.
The number of women switching to another hormonal therapy within 180 days of discontinuing tamoxifen was 25.4% (anastrozole 20.9%, letrozole 3.7%, and exemestane 0.8%). The median time from the index date to treatment switch for these patients was 1.5 years. Loss to follow-up occurred in 16.7% of women and can be accounted for by death, loss of eligibility for the HSE-PCRS scheme, or tamoxifen nonpersistence; 31.4% of women persisted with tamoxifen to the end of follow-up. The sensitivity analysis for censoring due to loss to follow-up demonstrated that by 3.5 years the minimum nonpersistence rate, assuming all patients persisted with tamoxifen after censoring, was 29.9% (95% CI: 28.0%, 31.8%) and the maximum, assuming all discontinued, was 54.2% (95% CI: 51.6%, 56.7%).
Relation Between Patient Characteristics and Nonpersistence
Predictors of nonpersistence in univariate and adjusted models appear in Table 2. Patients between the age of 35 and 44 (HR 1.36, 95% CI: 1.01, 1.83) and over the age of 75 (HR 1.46, 95% CI: 1.16, 1.83) were most likely to discontinue tamoxifen when compared with the reference group of women aged 45–54. Use of antidepressant agents in the year preceding tamoxifen initiation was associated with a significantly increased risk of early discontinuation of tamoxifen (HR 1.41 95% CI: 1.18, 1.70). The presence of Parkinson disease or dementia was also associated with an increase in the risk of nonpersistence (HR 1.72 95% CI: 1.24, 2.39). The risk of nonpersistence decreased as the mean number of pharmacological agents prescribed per month in the year before commencing tamoxifen increased. The use of anxiolytics, hypnotics, or antipsychotics or the presence of 1 or more of the identified comorbidities was not associated with any increased risk of nonpersistence.
Table 2. Crude and Adjusted Hazard Ratios (HRs) for Potential Determinants of Tamoxifen Nonpersistence in Patients Starting Tamoxifen Between January 2001 and January 2004 (n = 2816)
This is the largest study of tamoxifen persistence to date and the first to use objective measures of treatment discontinuation outside of the clinical trial setting. The results demonstrate that persistence with tamoxifen in clinical practice is lower than previously reported and that nonpersistence must be considered as a barrier to the effective use of oral hormonal therapies in breast cancer patients. The observed nonpersistence rate of 35.2% at 3.5 years is higher than the 5-year nonpersistence rates seen in self-reported studies of persistence (31%)13 and in clinical trials of adjuvant tamoxifen (16%–32%).7–11 In addition, it is likely that persistence with tamoxifen will have declined further by the time the cohort of patients in this study has completed 5 years of treatment. It is of particular concern that 14.5% of women obtained no more than 90 days supply of tamoxifen and in total 22.1% of women discontinued tamoxifen within 1 year of commencing treatment. This is twice the rate of nonpersistence reported in previous studies at 1 year.13, 14 These high rates of nonpersistence, especially in the early stages of treatment, are likely to have a negative impact on the therapeutic success of tamoxifen. Therefore, it is doubtful that the full benefits of tamoxifen, as demonstrated in randomized studies, will be achieved in clinical practice.
The analysis of potential determinants of tamoxifen nonpersistence showed that treatment discontinuation was associated with the extremes of age. Reduced persistence in the elderly may be influenced by psychosocial issues such as decreased social supports and an increasing incidence of cognitive and functional impairment.28 This is evidenced by the higher risk of discontinuation observed in patients receiving anti-Parkinson disease or antidementia treatments. The reasons for nonpersistence in younger women are less clear. It is possible that these women may not adjust to a diagnosis of breast cancer as well as older women.29, 30 They may also be less willing to accept or more likely to experience side effects associated with tamoxifen use.15 As younger women have the potential for a longer life expectancy, the negative consequences of nonpersistence with tamoxifen are significant, particularly as the benefits of 5 years of tamoxifen therapy have now been shown to extend to 15 years.3
The significant association between persistence with treatment and increasing numbers of prescribed medications has been observed in previous studies of tamoxifen persistence13, 14 and more recently in reports from the international breast intervention study, IBIS-I.31 It has been suggested that patients with a large medication burden develop routines to promote adherence.13, 31 It is also possible that the healthy drug user effect32 may play a role in this association, whereby patients with good health-seeking behavior who are in turn more likely to seek and receive medical treatment for conditions are also those patients with good medication-taking behavior.
Oncologists have cited psychological problems as a prime determinant of poor medication-taking behavior in oncology patients33; therefore, the significant association observed here between antidepressant use and reduced persistence with tamoxifen is unsurprising. Depression is a well-recognized risk factor for poor medication-taking behavior34, 35 and the link between depression and early discontinuation of tamoxifen has been noted previously.15 Depression has also been associated with reduced survival in women with breast cancer,36–38 although the evidence from some studies has been inconclusive.39 Whereas a number of mechanisms have been proposed to explain this association, results from this study suggest that nonpersistence with tamoxifen may in part provide an explanation.
Tamoxifen's side effects have been identified in previous studies as a significant reason for nonpersistence with treatment.13, 15 Consequently, effective strategies for reducing tamoxifen's side effect burden may be expected to increase persistence. The selective serotonin reuptake inhibitor (SSRI) and selective serotonin-norepinephrine reuptake inhibitor (SSNRI) antidepressant classes have been used successfully to reduce the frequency and severity of vasomotor symptoms associated with breast cancer and tamoxifen use.40–43 It is therefore possible that concomitant use of tamoxifen and certain SSRIs/SNRIs could improve tamoxifen persistence by minimizing side effects. Whether women continued with or commenced antidepressant use after initiation of tamoxifen was not recorded for this study and as a result it is not possible to draw conclusions about the effect of concurrent antidepressant use on tamoxifen persistence. Additional work is required to explore the relation between tamoxifen persistence, antidepressant use, depression, and breast cancer survival.
Whereas the use of prescription refill data provides an objective measure of tamoxifen persistence, there are a number of limitations to this study. First, without access to diagnostic or staging information it is inevitable that a small proportion of this study cohort will be receiving tamoxifen for an indication other than early breast cancer, such as advanced breast cancer. Second, the assessment of patients' beliefs about tamoxifen or the influence of side effects was beyond the scope of this study. Both of these factors have been shown to predict tamoxifen persistence. Third, it is not possible to differentiate between tamoxifen discontinuation by the patient or at the direction of the patient's prescriber. Finally, the nonpersistence rates reported here may only provide a conservative estimate of treatment discontinuation. There are a number of reasons for this: patients prescribed tamoxifen but who did not fill any prescription will not have been included in the study cohort; persistence data provide no information on whether patients take the tamoxifen they have obtained; treatment discontinuations after switching hormonal therapy were not recorded and loss to follow-up may have in some instances been due to nonpersistence.
This study demonstrates that persistence with tamoxifen cannot be assumed and raises concerns about persistence with other oral hormonal therapies for breast cancer and oral antineoplastics in general. As the availability and use of these agents increases, the challenge is for oncologists to identify those at risk of nonpersistence and to develop strategies to improve persistence. This is of particular importance, as longer durations of adjuvant therapy may be recommended for breast cancer in the future and as cancer survivorship becomes a priority area in clinical practice and research. Unfortunately, many interventions to improve adherence and persistence have not been shown to be effective and those interventions that are beneficial are costly and time-consuming.44 Nevertheless, their use may be worthwhile in oncology patients where the potential benefits of improved adherence and persistence are significant.
We thank the HSE Primary Care Reimbursement Services for supplying the data on which the study was based, and the National Cancer Registry Ireland