Infused Therapy and Survival in Older Patients Diagnosed with Metastatic Breast Cancer who Received Trastuzumab

We used Surveillance, Epidemiology, and End Results-Medicare data (2000-2006) to describe treatment and survival in women diagnosed with metastatic breast cancer (MBC) who received trastuzumab. There were 610 patients with a mean age of 74 years. Overall, 32% received trastuzumab alone and 47% received trastuzumab plus a taxane. In multivariate analysis, trastuzumab plus chemotherapy was associated with a lower adjusted cancer mortality rate (Hazard Ratio [HR] 0.54; 95% Confidence Interval [CI] 0.39-0.74; p < .001) than trastuzumab alone among patients who received trastuzumab as part of first-line therapy. Adding chemotherapy to first-line trastuzumab for metastatic breast cancer is associated with improved cancer survival.


INTRODUCTION
In metastatic breast cancer (MBC) among patients with human epidermal growth factor receptor 2 (HER2) overexpressing disease, trastuzumab is indicated for use in combination with paclitaxel for first-line treatment, and as a single agent for those who have previously received one or more chemotherapy regimens (1). The safety and efficacy of trastuzumab in MBC have been evaluated in numerous clinical trials (2)(3)(4)(5)(6)(7)(8)(9)(10). Two were particularly relevant for establishing the current indication in MBC (1). One was a multicenter, randomized, open-label trial conducted in 469 women who had not been previously treated with chemotherapy for metastatic disease (1,8). In this trial, patients were randomly assigned to receive standard chemotherapy alone (n = 234) or standard chemotherapy plus trastuzumab (n = 235). Standard chemotherapy consisted of doxorubicin (or epirubicin) and cyclophosphamide for women who had not previously received adjuvant therapy with an anthracycline, or paclitaxel for women previously treated with an anthracycline ( (8). The other study was a multicenter, open-label, single-arm clinical trial in 222 women who had relapsed following one or two prior chemotherapy regimens for metastatic disease (1,4). In this trial, trastuzumab was studied as a single agent in a population with an average age of 50 years, ranging from 28 to 81 (4). The objective tumor response, as determined by an independent response evaluation committee, was 15% in the intent-to-treat group (4).
Although trastuzumab for MBC has been studied extensively in clinical trials, little has been published on its use in routine clinical practice, especially in populations underrepresented in the trials. Despite the fact that HER2positive disease has a younger age distribution, older patients still comprise a significant proportion of those who might be eligible to receive trastuzumab for MBC. According to the Surveillance, Epidemiology, and End Results (SEER) program, 57% (n = 4,179) of the 7,331 women diagnosed with Stage IV breast cancer in 2004-2006 were aged 65 years or older (11). The objectives of this study were to describe patterns of infused therapy in a cohort of older women who first received trastuzumab following diagnosis of MBC, and to identify factors associated with longer survival.

Data source
The source of data for this study was the National Cancer Institute's (NCI) SEER cancer registry linked to Medicare enrollment and claims data (SEER-Medicare data). This database has been described in detail elsewhere (12). Briefly, as of 2010, SEER collects and publishes cancer incidence and survival data from 17 population-based cancer registries throughout the United States covering approximately 26% of the US population (13). The (198.81, 198.82) or in the lymph nodes (196.XX) based on algorithms for identifying cancer relapse previously reported in the literature (15,16). These algorithms were originally developed for detecting relapse of acute myelogenous leukemia (AML), and the best among them showed a sensitivity of 86% and a specificity of 99% in this disease (15). More recently, they have been applied to, but not validated in, a study on the costs of breast cancer recurrence (16) (17) and Healthcare Common Procedure Coding System (HCPCS) codes (18) within Medicare claims were used to identify intravenous chemotherapy agents and trastuzumab administered during the observation period (19,20).
In addition to de novo Stage, patients were further classified according to whether trastuzumab was part of their first infused therapy regimen during the observation period (first-line trastuzumab), or whether it was started after an initial course of chemotherapy during the observation period (delayed trastuzumab) (Figure 1). If  Infused Therapy in Metastatic Breast Cancer  within two months of the first claim for chemotherapy, or if there was no chemotherapy claim present, the patient was classified as having received first-line trastuzumab. This regimen was then classified hierarchically using ICD-9-CM and HCPCS codes as follows: trastuzumab alone; trastuzumab plus a taxane (paclitaxel or docetaxel), with or without any other agent; and trastuzumab with any other agent except a taxane. Assignment was based on having had at least one claim for each agent during two of the first 6 months following the first chemotherapy claim.
If trastuzumab began at least 2 months after the first claim for chemotherapy, the patient was classified as having received delayed trastuzumab (Figure 1). In this group, the initial chemotherapy regimen (absent trastuzumab) was classified, using the same approach above, as: anthracycline and/or cyclophosphamide with or without any other agent; taxane and/or vinorelbine with any other agent except anthracycline or cyclophosphamide; and other. The subsequent regimen that included trastuzumab was classified the same way as the first-line trastuzumab group described above. Visual inspection of the data prior to the development of this classification scheme showed that in the group receiving delayed trastuzumab, there was very little overlap between the agents used in the initial chemotherapy regimen and those used in the subsequent trastuzumab regimen.
Medicare Part D (prescription drug coverage) data were not available at the time this study was performed. Consequently, it was not possible to identify anti-estrogen or other oral therapy use in this population.

Mortality and censoring
The date of death was assigned by using the Medicare date, if available, even in cases where the SEER date also was available. The Medicare date was preferred because it is more current than the SEER date (21). In cases where the Medicare date was missing but the SEER date of death was available, the SEER date was used. All other patients were assumed to be alive at the end of the observation period (December 31, 2006), although they may have been censored earlier for other reasons, such as switching from fee-for-service to HMO coverage.
The cause of death was classified as cancer or noncancer, using the CODKM variable in the SEER Patient Entitlement and Diagnosis Summary File (PEDSF) through 2006. Cancer mortality included all deaths due to cancer (CODKM = 001-130), and not just due to breast cancer (CODKM = 046). Noncancer mortality included all other identified causes of death, e.g., CODKM = 154 "Diseases of Heart" and CODKM = 148 "Diabetes Mellitus". However, it excluded missing or unspecified cause of death. These patients were censored at the time of death in both the cancer and noncancer survival analysis since exploratory analysis showed that almost 90% of those with a known cause of death died of cancer. Consequently, including them as noncancer deaths could have resulted in significant misclassification. Cancer and noncancer mortality were examined separately since the benefit of cancer therapy should be manifested primarily through differences in cancer mortality, and differences in noncancer mortality could indicate selection bias, particularly confounding by indication (22 (24).
Medicare inpatient (Part A), outpatient, and physician (Part B) claims were used to calculate an NCI Comorbidity Index for each patient (23). This approach (25,26) entailed first removing claims that were considered to have unreliable diagnosis coding, such as those for testing procedures used to rule out conditions. Then, remaining diagnosis and procedure codes were used to identify the 15 noncancer comorbidities in the Charlson Comorbidity Index (CCI) (27). The algorithms used to identify these conditions reflect the Deyo (28) adaptation of the CCI, and include several procedure codes from the Romano (29) adaptation. A weight was assigned to each condition, and the weights were summed to obtain the index for each patient.
In the absence of performance status, we used Medicare claims to construct several medical resource utilization variables that have been shown and validated to predict performance status, (30) consisting of any inpatient admission (yes/no), any admission to the emergency room (yes/no), any use of durable medical equipment (yes/no), and any outpatient visit (yes/no) from 12 months before to 1 month after the diagnosis of MBC. In addition, we used SEER data and Medicare claims to identify prior surgery for breast cancer and radiation treatment (20,31).

Statistical analysis
Cox proportional hazards regression analysis was used to identify treatment, demographic, and clinical factors associated with overall survival. Multivariate analyses were performed on all-cause, cancer and noncancer mortality, including the entire cohort and stratified according to whether the patient received first-line or delayed trastuzumab (total of 9 analyses). We elected to conduct stratified analyses out of concern that these two groups might have significantly different prognostic features that could confound associations between treatment and survival in a single model.
In addition to the standard approach to multivariate analyses, which included individual predictors, multivariate analyses were performed using propensity techniques, (32) in which quintile of propensity score was substituted for all independent variables except initial trastuzumab regimen (total of 9 analyses), and in which the inverse of the propensity  R. I. Griffiths et al.
score was used as a weight in the regression analysis (total of 9 analyses).
Since treatment with trastuzumab following MBC diagnosis was a criterion for inclusion in this study, one concern in specifying this analysis was to avoid immortal time bias, (33) which, according to Suissa, is "a span of cohort follow-up during which, because of exposure definition, the outcome under study could not occur. " In this case, death could not occur between the time the patient was diagnosed with MBC and the time the patient first received trastuzumab. Therefore, the index date for this analysis was advanced from the date of diagnosis to 30 days after the date of the first claim for trastuzumab, to account also for the fact that additional time was required to determine whether or not patients received chemotherapy as part of their initial trastuzumab regimen.

Patterns of Trastuzumab Use
The mean time to initial trastuzumab following diagnosis of MBC was almost 6 months, with a median time of 2 months (  Overall, 70% of the cohort received first-line trastuzumab ( Table 3

). The proportion of patients diagnosed with de novo Stage IV disease was similar between patients who received first-line and those who received delayed trastuzumab. Most first-line trastuzumab patients received either monotherapy (31%) or trastuzumab plus a taxane (48%) as initial therapy following MBC diagnosis. In the delayed trastuzumab group, a slightly higher proportion of patients received trastuzumab alone (35%), and a slightly lower proportion received trastuzumab plus a taxane (45%) relative to firstline trastuzumab patients. During the observation period (2000-2006), the percent of patients receiving trastuzumab alone varied from 23 (2000) to 38% (in both 2003 and 2006).
Longitudinal "lasagna" plots (34) were constructed to illustrate the number of administrations of trastuzumab per month, the duration of trastuzumab, and gaps in administration ( Figure 2) for up to 24 months following the start of treatment. These were stratified by de novo stage and firstline versus delayed use. In general, there was a broad spectrum of use in all four groups, ranging from uninterrupted treatment in excess of one year with four or more administrations in each month, to treatment lasting only one month with only one administration. In stratified analysis, (Table 5) first-line trastuzumab plus chemotherapy was associated with statistically significantly lower cancer mortality (HR 0.54; 95% CI 0.39-0.74; p < .001). However, adding chemotherapy to delayed trastuzumab had no impact on cancer mortality. Findings from the propensity score analysis were consistent with those from the standard regression analysis (Figure 3).  The finding that adding chemotherapy to first-line trastuzumab was associated with improved survival relative to trastuzumab alone raises important questions with regard to the optimal treatment regimen in these patients. According to one recent review, (2) at the time the patients in our study were treated there had not been a randomized study of trastuzumab with or without chemotherapy. However, a recently published randomized trial of trastuzumab monotherapy versus trastuzumab plus docetaxel as first-line therapy in MBC, (10) was stopped early after an interim analysis showed a significantly lower mortality rate in the patients receiving chemotherapy with trastuzumab (HR for survival 2.72 p = .04). Upon inverting the HR for survival, we find that the HR for overall mortality (0.37) in this trial was similar to the HR ratio for cancer mortality among patients receiving first-line trastuzumab plus chemotherapy (0.54) in our study. Therefore, our findings based on data from routine clinical practice appear to support the findings from this recent trial, and suggest they apply to elderly patients.

Survival Analysis
The main limitation of this study, as with any observational study that compares the effects of alternative interventions, is the possibility that patients were selected into the treatment groups for reasons related to survival, (22,35,36) and that we failed to account for these reasons in our analyses. Although no approach completely eliminates confounding by indication in observational studies, (36) we took several steps to minimize its impact here. One possibility is that patients with poorer performance status or more comorbidity were selected to receive trastuzumab monotherapy due to concerns about the toxicity of chemotherapy. (22) SEER-Medicare does not include performance status. Consequently, using Medicare claims we constructed several medical resource use variables that have been validated to predict ECOG and Karnofsky performance status (30). Also, we examined cancer and noncancer mortality based on the rationale that any real benefit of cancer treatment should be observed only through cancer-specific mortality, and that differences in noncancer mortality suggest confounding by indication (22). In this study, we found adding chemotherapy to trastuzumab was associated with a significantly lower rate of cancer but not noncancer mortality.
Although the coefficients for trastuzumab plus chemotherapy were similar for cancer and noncancer mortality, it is important to note that only 24 patients (8% of those with a known cause of death) had a noncancer cause of death. Consequently, the coefficients in these models are unstable and their direction and significance can be determined by a few individuals. Finally, we used two different approaches to multivariate survival analysis: the standard approach in which all patient characteristics were included as independent variables, and propensity techniques (32) in which propensity score quintile was substituted for the vector of independent variables except treatment, and alternatively where the raw propensity score was used as a weight. It should be noted, however, that propensity techniques do not account for unobserved confounding covariates (32). Instrumental variable approaches have been applied to SEER-Medicare analysis, but to our knowledge not in small cohorts such as ours.
A second limitation of this study is that we did not include a comparator group of patients who did not receive trastuzumab. The main reason is that, presently, SEER-Medicare does not include information on HER2 status. It can be inferred that any patient who received trastuzumab had HER2 overexpressing disease, based on the fact that Medicare does not pay for trastuzumab without a positive HER2 test result. However, it is not possible to identify patients who are HER2 positive but who do not receive trastuzumab. Since HER2 overexpression is a significant predictor of both overall survival and time to relapse in patients with breast cancer, (37) the absence of HER2 test results is a significant barrier to conducting comparative effectiveness research (e.g. chemotherapy with versus without trastuzumab) on trastuzumab using this database, because one cannot identify a HER2 positive control group. Second, while trastuzumab was first approved by FDA in September 1998, (38) a Medicare reimbursement code specifically for trastuzumab did not become effective until January 1, 2000. Therefore, we were unable to document the use of trastuzumab during the first year following approval.
Third, as illustrated, patterns of trastuzumab use were diverse, ranging from long, uninterrupted periods with multiple administrations per month to short periods with only