pharmacoepidemiology and drug safety 2015; 24: 301–309 Published online 15 December 2014 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/pds.3733
ORIGINAL REPORT
A competing risk analysis of hormone therapy interruption in Asian women with breast cancer† Kun-Pin Hsieh1,2, Li-Chia Chen3, Kwok-Leung Cheung4 and Yi-Hsin Yang1* 1
School of Pharmacy, College of Pharmacy, Kaohsiung Medical University, Kaohsiung, Taiwan Department of Pharmacy, Kaohsiung Medical University Hospital, Kaohsiung, Taiwan 3 Division for Social Research in Medicines and Health, School of Pharmacy, University of Nottingham, Nottingham, UK 4 School of Medicine, University of Nottingham, Royal Derby Hospital Centre, Derby, DE22 3DT, UK 2
ABSTRACT Purpose This study aimed to use a competing risk approach to evaluate the probability of the occurrence of hormone therapy (HT) interruption and to assess its associated predictors in Asian women with breast cancer. Methods This retrospective cohort study used the Taiwan Health Insurance Research Database from 2003 to 2011. Reimbursement data for women with newly diagnosed primary breast cancer were extracted. Interruption (gap ≥ 180 days) and time to first interruption of HT were identified. The probability of interruption was analysed by Kaplan–Meier (KM) method and cumulative incidence competing risk (CICR) method. Competing risk regressions were used to assess the predictors of interruption. Results The 5-year cumulative incidence of first HT interruption was 14% versus 13% estimated by the KM and the CICR methods, respectively. The estimated incidences from CICR method tended to be around 11% lower than KM method in various HT utilization patterns. Younger (≤50 years) age at diagnosis, switching HT and the presence of HT-related adverse events were identified as predictors of interruption in competing risk regressions. Conclusions The competing risk approach provided lower probabilities and estimates when investigating the incidence of first interruption than the standard survival analysis. The competing risk method, which takes into account the competing risks from cancer recurrence and death, should be considered in future analysis. In terms of improving persistence of HT, it is important to focus on patients of younger age at diagnosis, HT switching and experiencing adverse events. Copyright © 2014 John Wiley & Sons, Ltd. key words—drug utilization; breast cancer; hormone therapy; persistence; interruption; competing risks; pharmacoepidemiology Received 8 June 2014; Revised 2 November 2014; Accepted 3 November 2014
INTRODUCTION
*Correspondence to: Y.-H. Yang, School of Pharmacy, College of Pharmacy, Kaohsiung Medical University, 100, Shih-Chuan 1st Road, Kaohsiung, 807, Taiwan. Email: [email protected] † This study contains original unpublished work and is not being submitted for publication elsewhere at the same time. Parts of the results were reported in podium presentations at the following two international conferences: Kun-Pin Hsieh, Li-Chia Chen, Chao-Sung Chang, Yi-Hsin Yang. How EarlyStage Non-Persistence and Non-Adherence to Adjuvant Hormonal Therapy Associated With Postmenopausal Breast Cancer Mortality in Taiwan. Podium presentation (CN2), International Society of Pharmacoeconomics and Outcome Research (ISPOR) 5th Asia-Pacific Conference, 3 September 2012 at the Taipei International convention Center, Taipei, Taiwan (Best Presenter Award). Kun-Pin Hsieh, Li-Chia Chen, Chao-Sung Chang, Yi-Hsin Yang. The Associations between Discontinuation/Nonadherence to Tamoxifen and Mortality of Breast Cancer Patients: A Population Claim-based Data Analysis in Taiwan. Podium presentation. 28th International Conference on Pharmacoepidemiology & Therapeutic Risk Management (ISPE), Barcelona, Spain, 22–26 August 2012. Pharmacoepidemiology and Drug Safety, 2012; 21: (Suppl. 3): 9.
Copyright © 2014 John Wiley & Sons, Ltd.
Discontinuation or treatment interruption of hormone therapy (HT) may lead to an increase in breast cancer (BC) mortality.1–4 However, concerns have been raised by several studies that reported a suboptimal adherence and persistence to HT in BC patients.5–7 Although a variety of methods were used, HT interruption rates are consistently high ranging from 19 to 33% at 3 years after treatment initiation8–10 and reaching 50% at 5 years.1,8 Current evidence of detrimental BC outcomes associated with HT interruption is predominately based on research in western countries, where the peak BC diagnosis age (60 to 70 years old) is approximately 10 years older than it is for those in Asian countries (40 to 50 years old).11 Moreover, the incidence of HT interruption and its
302
k.-p. hsieh et al.
associated predictors have not been fully studied in an Asian population. Previous studies8,10,12 mostly used standard survival analysis methods, that is, the Kaplan–Meier (KM) estimates and Cox regression, which do not incorporate competing risks to assess the risk of HT interruption and its associated predictors, although some studies excluded either BC recurrence13 or death14 before analysis. A competing risk is an event that may hinder the observation of a clinical event or that may modify the chance for the occurrence of event.15,16 Clinical progression of BC, such as recurrence and mortality, may be considered as competing risks to the chance of HT interruption. For a local or systemic BC recurrence, HT is likely to be discontinued, and patients then may undergo surgery or receive other therapies before their HT is resumed. Similarly, mortality is also considered as a competing risk preventing the occurrence of HT interruption for a patient who has died during the recommended 5-years of HT. Therefore, it is of interest to evaluate how competing risk analysis differs from the standard survival analysis in analysing incidence and time to HT interruption. In Taiwan, BC is the most common cancer in women and the fourth leading cause of cancer death for women. The incidence of BC per 100 000 women increased from 40 in 1999 to 56.1 in 2008. BC treatment is covered by the Taiwan National Health Insurance (NHI), which was established in 1995 and provides comprehensive medical care for 99.6% of the entire population (23.74 million in 2010).17 Sixty-two percent of women with BC received adjuvant HT,18 including tamoxifen and aromatase inhibitors (AIs), that is, anastrozole, letrozole or exemestane under NHI reimbursement. Tamoxifen is normally the agent of choice, and AIs are also reimbursed under limited conditions. This study aimed to use a competing risk approach to evaluate the risk of HT interruption and to assess the associated predictors to HT interruption in women with BC in Taiwan. METHODS Study design and data source This retrospective cohort study used a populationbased reimbursement dataset, the Taiwan National Health Insurance Research Database (NHIRD) from 2003 to 2011. Data of all BC patients were extracted, including outpatient and inpatient medical claims, dispensing claims from community pharmacies, and the Registry for Catastrophic Illness (i.e. a patient file in the NHIRD) as the research data sources. This study Copyright © 2014 John Wiley & Sons, Ltd.
was approved by the Institutional Review Board of Kaohsiung Medical University Hospital (KMUHIRB-20120047). Study cohort Adult (age ≥ 18 years) women with newly diagnosed BC from 2003 to 2010 who received more than 28 days HT were included in this study. Women with BC were identified by screening BC-related International Classification of Diseases Revision 9 (ICD-9) codes (174 for BC) on NHIRD and the Registry for Catastrophic Illness. Patients whose BC diagnosis dates (disease index date) recorded from 2003 to 2010 were selected, and those who had other cancers (ICD-9 codes 140–208, except 174) or benign lesions (ICD-9 codes 210–239) recorded before the disease index date were excluded. Furthermore, patients whose HT-dispensed prescription duration summed to less than 28 days during the study period were also excluded as their utilization information was not sufficient for quantifying HT interruption. All study cohorts were followed from the disease index date until death or the end of the study (December 2011). Interruption of hormone therapy Each individual patient’s HT prescription duration was followed from the first HT prescription date to the last date that a HT prescription covered. An interruption was defined as any non-HT-covered period between two consecutive HT prescriptions for more than 180 days. As Taiwan NHI reimburses the maximum of three monthly refills of any prescription for chronic and stable conditions, a gap of 180 days without HT would indicate that at least two clinical visits had been missed. The first episode of HT interruption and the time to first interruption from the first HT prescription date to the date of the first interruption were identified as the primary outcome. Competing risks for interruption Breast cancer recurrence and death were considered as competing risks because their occurrence cannot be treated as independent censoring when analysing the time to HT interruption. Individual patients who received surgery (OP; including breast-conserving surgery or mastectomy), chemotherapy (CT), radiotherapy (RT) or targeted therapy (TT; i.e. trastuzumab) within the windows of the first HT interruption and occurred 1 year after the BC diagnosis index date were analysed as an event indicative of BC recurrence. Pharmacoepidemiology and Drug Safety, 2015; 24: 301–309 DOI: 10.1002/pds
hormone therapy interruption for breast cancer
Death and date of death were identified from the Registry for Catastrophic Illness data.
Covariates Each patient’s age at diagnosis, baseline comorbidities, BC treatment strategies, HT utilization patterns and prescriptions for managing the HT-related adverse events (AEs) were also recorded as covariates to adjust the outcome measures. The age at diagnosis was further categorized into four groups: younger than 50 years, 50 to 64 years, 65 to 69 years and 70 years or older. Each patient’s comorbidities recorded 1 year before BC diagnosis were identified and converted to a Charlson Comorbidity Index score,19,20 which was further categorized into three groups (0, 1 and ≥2). Each patient’s BC treatment strategies, including OP, CT, RT, HT or TT, were identified by corresponding codes for medical orders and drugs and were dichotomized individually as covariates. OP, CT or RT that occurred within 12 months after the disease index date was considered as the primary BC treatment strategies. HT administered after OP was considered as adjuvant HT; in contrast, HT administered without any prior primary OP was recognized as primary HT. Because NHIRD did not include information regarding disease status, it was assumed that women in the primary HT group (i.e. no OP) had late-stage BC according to the 2010 Taiwan Cancer Registry Annual Report, showing that 59% of women with stage IV BC did not have primary OP compared with 5% of women with stage III BC without OP.18 Each patient’s HT utilization patterns during the follow-up period were categorized into four groups: (1) received tamoxifen only; (2) started with tamoxifen and switched to AIs; (3) received AIs only; and (4) started with AIs and switched to tamoxifen. Switching between tamoxifen and AIs was defined when an alternative HT was administered more than 84 days (estimated duration for three maximum refills; i.e. 28 days multiplied by three refills) after discontinuation of the previous HT course. Medicines attempting to manage HT-related AEs were used as the surrogate indicators to identify relevant AEs associated with tamoxifen or AI treatment. Prescriptions of medicines for managing HT-related AEs issued with HT at the same outpatient visit during the windows of the 4th to 12th month in the first year of HT were identified, including anti-inflammatory agents and analgesia for musculoskeletal and joint symptoms (MSAEs), antidepressants and gabapentin for vasomotor symptoms, lipid-modifying agents for Copyright © 2014 John Wiley & Sons, Ltd.
303
lipid metabolism disorders, hypnotics and sedatives for insomnia, and anxiolytics for anxiety. Data analysis Descriptive statistics were used to report patients’ characteristics. The probabilities of the first interruption episode were estimated by using the KM method as well as the cumulative incidence competing risk (CICR) method, which incorporates competing risks of cancer recurrence and death in the cumulative incidence function.15 Analyses using both KM and CICR methods21 were conducted on the three different age groups and the four HT utilization groups as mentioned earlier. Sub-hazard ratios (SHRs) were estimated by competing risk regressions for modelling the time to interruption as a main event of interest after controlling for competing events of recurrence and death.22 The adjusted covariates included age at diagnosis, Charlson Comorbidity Index score, BC treatment strategies, HT utilization patterns, HTrelated AE management, income, residential areas of NHI divisions, HT prescription duration and year of HT initiation. The competing risk regression models were also implemented in two sub-group analyses: (1) patients with 5 or more years of follow-up and (2) patients with adjuvant HT. All analyses were conducted using SAS version 9.3 (SAS Institute Inc., Cary, NC, USA), and SAS macros %CIF21 and %PSHREG23 were used for competing risk analyses. RESULTS Patient characteristics A total of 37 391 women with newly diagnosed BC (mean age at BC diagnosis ± standard deviation: 52.1 ± 11.6 years) were included in this study, accounting for 166 693.89 follow-up patient-years (median 4.2 years). There were 5080 (13.6%) patients who had at least one interruption during the study period (Table 1). Patients with a younger age at diagnosis had a significantly higher proportion of interruption, that is, 14.7% for those younger than 50 years, 13.3 and 12.4% for those diagnosed at 50–64 and 65–69 years, respectively, and 10.0% for those 70 years or older (p < 0.0001). The proportion of patients experiencing HT interruption was significantly higher in patients who received primary HT (21.6%) than those who received adjuvant HT (13.2%), earlier HT-initiated year, and it was also higher in those whose HT was switched (p < 0.0001) (Table 1). Pharmacoepidemiology and Drug Safety, 2015; 24: 301–309 DOI: 10.1002/pds
304
k.-p. hsieh et al.
Table 1. Baseline characteristics of persistence and interruption Characteristics †
Number of patients (%) Age at diagnosis (%)† Younger than 50 years 50–64 years 65–69 years ≥70 years or older CCI score 0 1 2 Primary therapy (%)† Mastectomy BCS Adjuvant HT Primary HT HT utilization patterns (%)† Started with tamoxifen without HT switched Started with AIs without HT switched Started with tamoxifen with HT switched Started with AIs with HT switched HT initiation year (%)† 2003–2005 2006–2008 2009–2011 Patient had other adjuvant therapy (%)† Chemotherapy (yes) Radiotherapy (yes) Targeted therapy (yes) HT-related side effects management before the first interruption‡ (%)† MSAEs (yes) VMSs (yes) Lipid metabolism disorders (yes) Insomnia (yes) Anxiety (yes)
Persistence
Interruption
p-value
32 311 (86.4)
5080 (13.6)
14 950 (85.4) 12 177 (86.7) 2202 (87.6) 2982 (90.0)
2566 (14.7) 1872 (13.3) 311 (12.4) 331 (10.0)
ORIGINAL REPORT
A competing risk analysis of hormone therapy interruption in Asian women with breast cancer† Kun-Pin Hsieh1,2, Li-Chia Chen3, Kwok-Leung Cheung4 and Yi-Hsin Yang1* 1
School of Pharmacy, College of Pharmacy, Kaohsiung Medical University, Kaohsiung, Taiwan Department of Pharmacy, Kaohsiung Medical University Hospital, Kaohsiung, Taiwan 3 Division for Social Research in Medicines and Health, School of Pharmacy, University of Nottingham, Nottingham, UK 4 School of Medicine, University of Nottingham, Royal Derby Hospital Centre, Derby, DE22 3DT, UK 2
ABSTRACT Purpose This study aimed to use a competing risk approach to evaluate the probability of the occurrence of hormone therapy (HT) interruption and to assess its associated predictors in Asian women with breast cancer. Methods This retrospective cohort study used the Taiwan Health Insurance Research Database from 2003 to 2011. Reimbursement data for women with newly diagnosed primary breast cancer were extracted. Interruption (gap ≥ 180 days) and time to first interruption of HT were identified. The probability of interruption was analysed by Kaplan–Meier (KM) method and cumulative incidence competing risk (CICR) method. Competing risk regressions were used to assess the predictors of interruption. Results The 5-year cumulative incidence of first HT interruption was 14% versus 13% estimated by the KM and the CICR methods, respectively. The estimated incidences from CICR method tended to be around 11% lower than KM method in various HT utilization patterns. Younger (≤50 years) age at diagnosis, switching HT and the presence of HT-related adverse events were identified as predictors of interruption in competing risk regressions. Conclusions The competing risk approach provided lower probabilities and estimates when investigating the incidence of first interruption than the standard survival analysis. The competing risk method, which takes into account the competing risks from cancer recurrence and death, should be considered in future analysis. In terms of improving persistence of HT, it is important to focus on patients of younger age at diagnosis, HT switching and experiencing adverse events. Copyright © 2014 John Wiley & Sons, Ltd. key words—drug utilization; breast cancer; hormone therapy; persistence; interruption; competing risks; pharmacoepidemiology Received 8 June 2014; Revised 2 November 2014; Accepted 3 November 2014
INTRODUCTION
*Correspondence to: Y.-H. Yang, School of Pharmacy, College of Pharmacy, Kaohsiung Medical University, 100, Shih-Chuan 1st Road, Kaohsiung, 807, Taiwan. Email: [email protected] † This study contains original unpublished work and is not being submitted for publication elsewhere at the same time. Parts of the results were reported in podium presentations at the following two international conferences: Kun-Pin Hsieh, Li-Chia Chen, Chao-Sung Chang, Yi-Hsin Yang. How EarlyStage Non-Persistence and Non-Adherence to Adjuvant Hormonal Therapy Associated With Postmenopausal Breast Cancer Mortality in Taiwan. Podium presentation (CN2), International Society of Pharmacoeconomics and Outcome Research (ISPOR) 5th Asia-Pacific Conference, 3 September 2012 at the Taipei International convention Center, Taipei, Taiwan (Best Presenter Award). Kun-Pin Hsieh, Li-Chia Chen, Chao-Sung Chang, Yi-Hsin Yang. The Associations between Discontinuation/Nonadherence to Tamoxifen and Mortality of Breast Cancer Patients: A Population Claim-based Data Analysis in Taiwan. Podium presentation. 28th International Conference on Pharmacoepidemiology & Therapeutic Risk Management (ISPE), Barcelona, Spain, 22–26 August 2012. Pharmacoepidemiology and Drug Safety, 2012; 21: (Suppl. 3): 9.
Copyright © 2014 John Wiley & Sons, Ltd.
Discontinuation or treatment interruption of hormone therapy (HT) may lead to an increase in breast cancer (BC) mortality.1–4 However, concerns have been raised by several studies that reported a suboptimal adherence and persistence to HT in BC patients.5–7 Although a variety of methods were used, HT interruption rates are consistently high ranging from 19 to 33% at 3 years after treatment initiation8–10 and reaching 50% at 5 years.1,8 Current evidence of detrimental BC outcomes associated with HT interruption is predominately based on research in western countries, where the peak BC diagnosis age (60 to 70 years old) is approximately 10 years older than it is for those in Asian countries (40 to 50 years old).11 Moreover, the incidence of HT interruption and its
302
k.-p. hsieh et al.
associated predictors have not been fully studied in an Asian population. Previous studies8,10,12 mostly used standard survival analysis methods, that is, the Kaplan–Meier (KM) estimates and Cox regression, which do not incorporate competing risks to assess the risk of HT interruption and its associated predictors, although some studies excluded either BC recurrence13 or death14 before analysis. A competing risk is an event that may hinder the observation of a clinical event or that may modify the chance for the occurrence of event.15,16 Clinical progression of BC, such as recurrence and mortality, may be considered as competing risks to the chance of HT interruption. For a local or systemic BC recurrence, HT is likely to be discontinued, and patients then may undergo surgery or receive other therapies before their HT is resumed. Similarly, mortality is also considered as a competing risk preventing the occurrence of HT interruption for a patient who has died during the recommended 5-years of HT. Therefore, it is of interest to evaluate how competing risk analysis differs from the standard survival analysis in analysing incidence and time to HT interruption. In Taiwan, BC is the most common cancer in women and the fourth leading cause of cancer death for women. The incidence of BC per 100 000 women increased from 40 in 1999 to 56.1 in 2008. BC treatment is covered by the Taiwan National Health Insurance (NHI), which was established in 1995 and provides comprehensive medical care for 99.6% of the entire population (23.74 million in 2010).17 Sixty-two percent of women with BC received adjuvant HT,18 including tamoxifen and aromatase inhibitors (AIs), that is, anastrozole, letrozole or exemestane under NHI reimbursement. Tamoxifen is normally the agent of choice, and AIs are also reimbursed under limited conditions. This study aimed to use a competing risk approach to evaluate the risk of HT interruption and to assess the associated predictors to HT interruption in women with BC in Taiwan. METHODS Study design and data source This retrospective cohort study used a populationbased reimbursement dataset, the Taiwan National Health Insurance Research Database (NHIRD) from 2003 to 2011. Data of all BC patients were extracted, including outpatient and inpatient medical claims, dispensing claims from community pharmacies, and the Registry for Catastrophic Illness (i.e. a patient file in the NHIRD) as the research data sources. This study Copyright © 2014 John Wiley & Sons, Ltd.
was approved by the Institutional Review Board of Kaohsiung Medical University Hospital (KMUHIRB-20120047). Study cohort Adult (age ≥ 18 years) women with newly diagnosed BC from 2003 to 2010 who received more than 28 days HT were included in this study. Women with BC were identified by screening BC-related International Classification of Diseases Revision 9 (ICD-9) codes (174 for BC) on NHIRD and the Registry for Catastrophic Illness. Patients whose BC diagnosis dates (disease index date) recorded from 2003 to 2010 were selected, and those who had other cancers (ICD-9 codes 140–208, except 174) or benign lesions (ICD-9 codes 210–239) recorded before the disease index date were excluded. Furthermore, patients whose HT-dispensed prescription duration summed to less than 28 days during the study period were also excluded as their utilization information was not sufficient for quantifying HT interruption. All study cohorts were followed from the disease index date until death or the end of the study (December 2011). Interruption of hormone therapy Each individual patient’s HT prescription duration was followed from the first HT prescription date to the last date that a HT prescription covered. An interruption was defined as any non-HT-covered period between two consecutive HT prescriptions for more than 180 days. As Taiwan NHI reimburses the maximum of three monthly refills of any prescription for chronic and stable conditions, a gap of 180 days without HT would indicate that at least two clinical visits had been missed. The first episode of HT interruption and the time to first interruption from the first HT prescription date to the date of the first interruption were identified as the primary outcome. Competing risks for interruption Breast cancer recurrence and death were considered as competing risks because their occurrence cannot be treated as independent censoring when analysing the time to HT interruption. Individual patients who received surgery (OP; including breast-conserving surgery or mastectomy), chemotherapy (CT), radiotherapy (RT) or targeted therapy (TT; i.e. trastuzumab) within the windows of the first HT interruption and occurred 1 year after the BC diagnosis index date were analysed as an event indicative of BC recurrence. Pharmacoepidemiology and Drug Safety, 2015; 24: 301–309 DOI: 10.1002/pds
hormone therapy interruption for breast cancer
Death and date of death were identified from the Registry for Catastrophic Illness data.
Covariates Each patient’s age at diagnosis, baseline comorbidities, BC treatment strategies, HT utilization patterns and prescriptions for managing the HT-related adverse events (AEs) were also recorded as covariates to adjust the outcome measures. The age at diagnosis was further categorized into four groups: younger than 50 years, 50 to 64 years, 65 to 69 years and 70 years or older. Each patient’s comorbidities recorded 1 year before BC diagnosis were identified and converted to a Charlson Comorbidity Index score,19,20 which was further categorized into three groups (0, 1 and ≥2). Each patient’s BC treatment strategies, including OP, CT, RT, HT or TT, were identified by corresponding codes for medical orders and drugs and were dichotomized individually as covariates. OP, CT or RT that occurred within 12 months after the disease index date was considered as the primary BC treatment strategies. HT administered after OP was considered as adjuvant HT; in contrast, HT administered without any prior primary OP was recognized as primary HT. Because NHIRD did not include information regarding disease status, it was assumed that women in the primary HT group (i.e. no OP) had late-stage BC according to the 2010 Taiwan Cancer Registry Annual Report, showing that 59% of women with stage IV BC did not have primary OP compared with 5% of women with stage III BC without OP.18 Each patient’s HT utilization patterns during the follow-up period were categorized into four groups: (1) received tamoxifen only; (2) started with tamoxifen and switched to AIs; (3) received AIs only; and (4) started with AIs and switched to tamoxifen. Switching between tamoxifen and AIs was defined when an alternative HT was administered more than 84 days (estimated duration for three maximum refills; i.e. 28 days multiplied by three refills) after discontinuation of the previous HT course. Medicines attempting to manage HT-related AEs were used as the surrogate indicators to identify relevant AEs associated with tamoxifen or AI treatment. Prescriptions of medicines for managing HT-related AEs issued with HT at the same outpatient visit during the windows of the 4th to 12th month in the first year of HT were identified, including anti-inflammatory agents and analgesia for musculoskeletal and joint symptoms (MSAEs), antidepressants and gabapentin for vasomotor symptoms, lipid-modifying agents for Copyright © 2014 John Wiley & Sons, Ltd.
303
lipid metabolism disorders, hypnotics and sedatives for insomnia, and anxiolytics for anxiety. Data analysis Descriptive statistics were used to report patients’ characteristics. The probabilities of the first interruption episode were estimated by using the KM method as well as the cumulative incidence competing risk (CICR) method, which incorporates competing risks of cancer recurrence and death in the cumulative incidence function.15 Analyses using both KM and CICR methods21 were conducted on the three different age groups and the four HT utilization groups as mentioned earlier. Sub-hazard ratios (SHRs) were estimated by competing risk regressions for modelling the time to interruption as a main event of interest after controlling for competing events of recurrence and death.22 The adjusted covariates included age at diagnosis, Charlson Comorbidity Index score, BC treatment strategies, HT utilization patterns, HTrelated AE management, income, residential areas of NHI divisions, HT prescription duration and year of HT initiation. The competing risk regression models were also implemented in two sub-group analyses: (1) patients with 5 or more years of follow-up and (2) patients with adjuvant HT. All analyses were conducted using SAS version 9.3 (SAS Institute Inc., Cary, NC, USA), and SAS macros %CIF21 and %PSHREG23 were used for competing risk analyses. RESULTS Patient characteristics A total of 37 391 women with newly diagnosed BC (mean age at BC diagnosis ± standard deviation: 52.1 ± 11.6 years) were included in this study, accounting for 166 693.89 follow-up patient-years (median 4.2 years). There were 5080 (13.6%) patients who had at least one interruption during the study period (Table 1). Patients with a younger age at diagnosis had a significantly higher proportion of interruption, that is, 14.7% for those younger than 50 years, 13.3 and 12.4% for those diagnosed at 50–64 and 65–69 years, respectively, and 10.0% for those 70 years or older (p < 0.0001). The proportion of patients experiencing HT interruption was significantly higher in patients who received primary HT (21.6%) than those who received adjuvant HT (13.2%), earlier HT-initiated year, and it was also higher in those whose HT was switched (p < 0.0001) (Table 1). Pharmacoepidemiology and Drug Safety, 2015; 24: 301–309 DOI: 10.1002/pds
304
k.-p. hsieh et al.
Table 1. Baseline characteristics of persistence and interruption Characteristics †
Number of patients (%) Age at diagnosis (%)† Younger than 50 years 50–64 years 65–69 years ≥70 years or older CCI score 0 1 2 Primary therapy (%)† Mastectomy BCS Adjuvant HT Primary HT HT utilization patterns (%)† Started with tamoxifen without HT switched Started with AIs without HT switched Started with tamoxifen with HT switched Started with AIs with HT switched HT initiation year (%)† 2003–2005 2006–2008 2009–2011 Patient had other adjuvant therapy (%)† Chemotherapy (yes) Radiotherapy (yes) Targeted therapy (yes) HT-related side effects management before the first interruption‡ (%)† MSAEs (yes) VMSs (yes) Lipid metabolism disorders (yes) Insomnia (yes) Anxiety (yes)
Persistence
Interruption
p-value
32 311 (86.4)
5080 (13.6)
14 950 (85.4) 12 177 (86.7) 2202 (87.6) 2982 (90.0)
2566 (14.7) 1872 (13.3) 311 (12.4) 331 (10.0)
