Comparative effectiveness of trastuzumab emtansine versus lapatinib plus chemotherapy for HER2+ metastatic breast cancer

Flatiron Health (FH) includes information from more than 265 cancer clinics across the US [3]. Data are deidentified thereby protecting patient confidentiality. Research with the database was approved by the Copernicus Group Institutional Review Board. For this study, data from 1 January 2011, through 31 July 2020, were included for patients with a confirmed diagnosis of mBC and documentation of a positive HER2 test result who were aged 18 or over at time of treatment initiation. Patients had to receive T-DM1 or lapatinib plus chemotherapy for second-line treatment of mBC, with the first line of therapy needing to be trastuzumab based. Patients had to have one or more clinic encounters occurring after initiation of second-line treatment.

Data on key baseline demographic and clinical characteristics were extracted from FH for the study population. Eastern Cooperative Oncology Group (ECOG) performance status assessments recorded following the start of first-line treatment and on or before the start of second-line treatment were used. Where there was more than one value, the maximum value was used. In defining a patient’s primary payer, the insurance payer that had a start date prior to and overlapping with the diagnosis of metastatic disease was used. If a patient had more than one payer that met the criteria, the following hierarchy was used to classify a patient: commercial health plan >Medicare >Medicaid >other government plan >patient assistance program >workers compensation >self pay. A patient was considered to have had a diagnosis of metastasis to a given site (bone, lung, liver, brain/central nervous system [CNS]) if the date of diagnosis (to a given site) occurred between the month of metastatic diagnosis and the month of second-line treatment initiation (inclusive). Each independent site of metastasis was used as a separate incident; therefore, a patient could have more than 1 site of metastasis. The number of metastatic sites was the sum of all incidences of metastatic sites identified. Visceral disease was defined as disease with lung, liver or brain metastases. The individual components of the Charlson comorbidity index with the exception of metastatic cancer, were extracted for each patient. A patient was considered to have a diagnosis of a given comorbidity if there was documentation of the diagnosis recorded at any time prior to initiation of second-line treatment, inclusive. First-line and second-line mBC treatments were identified using the derived line of treatment variables that are available in the FH dataset. These derived variables utilize an algorithm to define the timing and constituents of each line of treatment based on the raw treatment data available in the dataset. Time between first-line and second-line treatment initiation was calculated using the start dates of these treatment lines.

Baseline patient demographic and clinical characteristics were summarized using descriptive statistics. OS was defined as the time from initiation of second-line mBC treatment to death from any cause; otherwise, patients were censored at their last visit. The Kaplan–Meier method was used to evaluate median OS. To correct for potential covariate imbalances, all variables listed in Table 1 were considered for inclusion in a logistic regression model to estimate propensity scores. In line with best practice, we sought to construct a propensity score model which included all covariates that were associated with both the treatment and the outcome, or with the outcome only, while avoiding the inclusion of covariates associated with treatment only. As such, any covariate found to be significantly associated with the outcome in preliminary univariate Cox proportional hazard regression models at an alpha <0.1 were included in the propensity score model. Average Treatment effect on Treated (ATT) weights were estimated based on the propensity score and ATT weighted Cox proportional hazards models were used to determine the association of treatment choice with OS. To account for residual confounding not captured by the propensity score all variables were considered for inclusion in the ATT weighted multivariable Cox model, with those having an alpha <0.1 included in the final model.

In total, 278 patients with HER2+ mBC received T-DM1 and 34 lapatinib plus chemotherapy. The majority of patients in the lapatinib plus chemotherapy arm received a capecitabine-based chemotherapy regimen (91.18%) with the remainder receiving vinorelbine or eribulin.

Patient characteristics are described in Table 1. In terms of their first-line treatment of mBC, the use of triple therapy with a taxane and targeted therapy alongside trastuzumab was more common in patients treated with T-DM1 (57.9%) than lapatinib plus chemotherapy (47.1%) and the use of trastuzumab monotherapy was more common in T-DM1 patients (5.0%) than lapatinib plus chemotherapy patients (2.9%). Despite this, the median time between initiation of 1L treatment and initiation of 2L treatment was greater for patients receiving lapatinib plus chemotherapy (1.05 years) than those receiving T-DM1 (0.85 years). Patients treated with lapatinib plus chemotherapy had a greater amount of visceral disease (91.2 vs 73.0% more brain metastases [58.8 vs 22.3%] and a greater number of metastatic sites (median 3 sites vs 2 sites) than T-DM1 patients.

Median follow-up was 28.4 months in the T-DM1 patients and 30.8 months in the lapatinib plus chemotherapy patients. The crude hazard of death was lower in the T-DM1 patients compared with the lapatinib plus chemotherapy patients (HR: 0.55; 95% CI: 0.36–0.85).

The following variables were found to be significantly associated with survival in univariate models and included in the logistic regression model used to estimate the propensity score: age at treatment initiation, metastatic sites (bone, brain/CNS), number of metastatic sites and time between first- and second-line treatment initiation (data not shown). Two patients were dropped from the study population in estimating the propensity score in order to improve population overlap leaving 310 patients for the adjusted analyses. Age and time between first- and second-line treatment had standardized mean differences after weighting of -0.183 and 0.274, respectively; all other variables had standardized mean differences <0.1. The final ATT weighted multivariable Cox model additionally adjusted for ECOG performance status at treatment initiation, metastatic site(s), number of metastatic sites, first-line treatment, insurance status and presence of a Charlson comorbidity. Results from this model were similar to the crude results, with the crude hazard of death lower in the T-DM1 patients compared with the lapatinib plus chemotherapy patients (HR: 0.56; 95% CI: 0.38–0.85; Table 2 & Figure 1).

Overall survival in this real-world cohort of HER2+ mBC patients was significantly longer in patients treated with T-DM1 than in patients treated with lapatinib plus chemotherapy.

Direct comparison of the results presented here with those of the EMILIA trial are complicated by differences in the populations captured in each study. Individuals in the current FH study population were on an average slightly older than those in the EMILIA trial and the current study population included individuals with an ECOG greater than one where the EMILIA population did not. Another difference between the studies is that all patients in the EMILIA trial received a capecitabine-based chemotherapy regimen alongside lapatinib while only 91% of patients in the equivalent arm of our study received a capecitabine-based chemotherapy regimen. Despite these differences, our findings suggest a similar magnitude of relative benefit in those treated with T-DM1 to that reported for the EMILIA trial [1]. Notably, as the population captured in this study is broader than that in the EMILIA trial, the results go beyond those of the trial, supporting the comparative effectiveness of T-DM1 in a broader real-world population receiving a set of treatments reflective of those received in routine clinical practice.

Previous real-world studies of T-DM1 have reported absolute estimates of effectiveness in HER2+ mBC patients treated with T-DM1 and have provided comparisons of T-DM1 effectiveness between T-DM1 patients treated with different prior treatment regimens or with differing severity of disease [4–9], our study provides estimates of absolute effectiveness of T-DM1 in a larger cohort of second-line mBC patients than observed in any of these studies. Further, while these studies focused on absolute effectiveness, to our knowledge ours is the first to report on the real-world effectiveness of T-DM1 in second-line mBC relative to an alternative treatment option in this setting.

Limitations of the FH database and Electronic Health Records in general are well documented and include incomplete information on some variables such as comorbidities, metastases, performance status, date of death and adjuvant/neoadjuvant treatments [3,10]. As expected for a real-world study where treatments are not given at random, differences were observed in patient characteristics and prognostic factors between the two treatment groups. While we did our best to adjust for these in multivariable analyses, as with any nonrandomized study, the possibility of residual confounding must be considered in the interpretation of our results. Only statistical associations rather than causal relationships can therefore be concluded. Our analysis is also limited by the relatively small size of the lapatinib plus chemotherapy arm and the short median follow-up of the cohort. Our analysis should therefore be viewed as exploratory and not definitive, and further work is needed to confirm our findings. Despite these limitations, the results suggest the benefit of T-DM1 demonstrated in clinical trials is being realized in real-world populations, including populations not captured in these trials.

Real-world data supports the effectiveness of T-DM1 in the second-line treatment of HER2+ mBC patients previously treated with trastuzumab.

SV Ramagopalan, J Ray and C Sammon were responsible for study conception and design; SV Ramagopalan, R Pisoni, A Zenin, J Ray, LS Rathore and C Sammon were responsible for data analysis, and drafting and revision of the manuscript.

This study was funded by F Hoffman-La Roche. SV Ramagopalan, R Pisoni, LS Rathore and J Ray receives personal salary from F Hoffman-La Roche, A Zenin receives personal salary from Roche Moscow Ltd and C Sammon receives personal salary from PHMR. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

No writing assistance was utilized in the production of this manuscript.

The authors confirm that all the research meets the ethical guidelines, including adherence to the legal requirements of the study country.

This work is licensed under the Attribution-NonCommercial-NoDerivatives 4.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc-nd/4.0/