Association between progression-free survival and overall survival in women receiving first-line treatment for metastatic breast cancer: evidence from the ESME real-world database

Despite a decrease of the mortality rate, breast cancer (BC) is still the major cause of cancer death among women worldwide [1]. BC is a heterogeneous disease and can be classified according to the tumor immunohistochemical profile, characterized by the presence or absence of hormone-receptor expression (HR-positive [HR +]/HR-negative [HR −] status) and/or human epidermal growth factor receptor 2 (HER2) protein expression and/or HER2gene amplification (HER2 − positive [HER2 +]/HER2 − negative status [HER2 −]) in the tumor cells. Up to 30% of BC patients will experience distant metastases over their lifetime [2]. Metastatic breast cancer (mBC) is a disease with poor prognosis, with median overall survival (OS) ranging from 14.8 months for triple-negative mBC (TN mBC, defined as the lack of HR expression and of HER2 over-expression/amplification) to around 5 years for HER2 + disease respectively [3]. The selection of the most appropriate treatment depends on the characteristics of the patient (age, performance status), the adjuvant treatment (type and duration of therapies for early BC), and the metastatic disease (number of metastatic sites, type of involved organs, time to metastatic occurrence, molecular profile). In the metastatic setting, chemotherapy, endocrine therapy, or targeted therapy are the recommended treatments, as induction or maintenance treatment, according to the tumor molecular profile [4].

In cancer randomized controlled trials (RCT), drug efficacy is assessed using OS, considered as the gold standard efficacy endpoint [5‐7]. However, in the advanced setting, the multiple lines of treatments may affect OS and thus bias the assessment of the true treatment effect. In addition, observing a benefit on OS may require a large number of patients and extensive follow-up, limiting the feasibility of clinical trials based on OS as the primary outcome. In this context, alternative endpoints that could capture treatment benefit accurately and be measurable earlier is central for the evolution of clinical research in oncology. Progression-free survival (PFS), although recognized as presenting some limitations [8], has been used increasingly over the past decades [9] and is now the most common primary efficacy endpoint in mBC clinical trials [10]. The use of PFS relies on the hypothesis that it can adequately replace OS, i.e., be a valid surrogate of OS; otherwise, this might lead to the marketing of drugs that do not ultimately improve OS [11, 12]. PFS however has not been validated as a surrogate endpoint in the context of mBC [13].

Real-world data (RWD) are defined as observational data from other sources than clinical trials, such as electronic medical records, registries, insurance claims, pharmacy records, death certificates, and other patient-generated data [14]. RWD bring information on patient’s profiles not included in RCT and supplement with real-life knowledge on patient management, treatment strategies, and long-term survival. As such, they complement results of RCT by allowing one (i) to assess the generalizability of survival outcomes reported in RCT to the real-life setting, (ii) to expand generalizability of trials’ results to underrepresented or specific populations, and (iii) to generate scientific hypotheses. In mBC, the availability of large datasets for researchers such as the longitudinal Epidemiology Strategy and Medical Economics (ESME) mBC Database are a unique opportunity to investigate real-life survival outcomes for distinct subgroups of mBC patients. These could subsequently be used to validate efficacy data observed in published RCT or generate study hypotheses when estimating sample size for a future RCT. ESME-mBC-derived data have been published, either to describe treatment patterns and patient outcomes for some specific subgroups [15‐20] or to report specifically on OS and associated prognostic factors in mBC [3, 21, 22].

Our primary objective was to describe the individual-level association between rwPFS and OS according to first-line (L1) treatment in women treated for mBC as a potential surrogate endpoint. Secondary objectives included description of treatment patterns, rwPFS and OS, overall and by mBC subtype.

This retrospective analysis provided estimates on individual-level associations between rwPFS and OS in 20,033 women diagnosed in 2008–2017 with mBC who initiated first-line anti-cancer treatment. This large national cohort was a unique opportunity to report on individual-level associations between rwPFS and OS for each L1 treatment pattern, overall and by tumor subtype. Individual-level associations between rwPFS and OS were at least strong for both TN and HR − /HER2 + subtypes (rank correlation coefficient equal or higher than 0.67). For HR + subgroups with considerable variety in therapeutic options, associations were highly variable (weak to strong) depending on the treatment. Overall, within the 4 mBC subgroups, individual-level correlation between rwPFS and OS were at least strong for each dominant treatment class (i.e., chemotherapy for TN mBC, endocrine therapy for HR + /HER2 − , and anti-HER2 in either HER2 + subgroups).

Formal validation of a surrogate endpoint relies on the assessment of both individual- and trial-level associations, the latter being available through meta-analyses of RCTs. As of today however, few surrogate endpoints have been identified [13] and assessment of PFS as a surrogate for OS in mBC is still relevant.

To our knowledge, we report results of the first study assessing individual-patient level association between rwPFS and OS in the largest observational cohort including consecutive patients treated for mBC in the real-world setting. We assessed individual correlation based on the rank correlation approach, a rigorous tool used in the reference method for endpoint surrogacy assessment [28, 29]. Although RWD are not the primary source used for assessing surrogacy, our data source, ESME mBC database, offered a large dataset of individual-patient data [IPD] homogenously centralized to assess multiple endpoints. Secondly, exploring the value of RWD in the search of candidate endpoint for OS surrogacy, our IPD were sourced from a significant quantity of high-quality data. The ESME mBC cohort represented 23,000 + adult patients (without any upper limit regarding the age) consecutively treated in the French 18 participating centers over a 12-year time period. The BC subtypes distribution and de novo mBC disease proportion were consistent with the other observational studies published [30‐33]. We relied on this highly reliable dataset updated on an annual basis (including update on patient status maintained update to date in each center supported by the local use of National registry of death certificates), and this leads to an objective estimation of OS and a strong external validity. Well-designed databases may lead to valid information even if conventional RCTs are still the reference for evidence [34]. Our retrospective analysis considered all 1L therapy and estimated rwPFS are consistent with published data [35, 36]. Finally, over the few past years, the ESME mBC database was extensively used to support post-marketing requirements for heath technology assessment body in France and to supplement clinical data package for marketing authorization file in Europe [37‐42].

The comparison of our findings with published data on surrogate endpoints is complex due to differences in terms of statistical methods (meta-analytical approaches of RCTs using either IPD or aggregated data), recruitment period, distinct target populations (defined by HR expression or HER2 status), or distinct settings (first-line or subsequent lines of therapy). In the study published by Sherrill and al., individual-level association was weak (ρ = 0.38) based on 67 RCTs designed to assess different mBC therapies (mainly chemotherapy or endocrine therapy in monotherapy or combination) [43]. In a meta-analysis of RCTs assessing different types of therapy (including chemotherapy or endocrine therapy in monotherapy or combination) in multiple mBC setting and published between 1990 and 2020, moderate individual-level association was reported [44]. Based on the year of publication, most of RCTs (69.5% published earlier than 2005) were undertaken before our period of interest for mBC diagnosis (2008–2017) and 86.1% of RCTs included in the meta-analysis were not discriminant regarding the HR and HER2 status. Petrelli and al. reported a strong individual-level association (ρ = 0.81) in a meta-analysis of 20 RCTs assessing first-line targeted therapies in patients enrolled again before our period of interest [45]. All aforementioned meta-analyses were based on aggregated data and both PFS and time to progression merged as unique outcome when assessing correlation with OS at individual-patient level. Using IPD for meta-analyses, two studies reported individual-level association data between PFS and OS with a distinct definition for PFS [46, 47]. Burzykowski and al. reported a strong association (ρ = 0.81) at individual level in a meta-analysis of 11 RCTs comparing first-line anthracycline-based therapy with taxanes [46]. Again, no discrimination regarding the HR and HER2 status was retrieved among RCTs in the meta-analysis, which limits the comparison with our estimated associations for patients receiving chemotherapy alone with correlation coefficient ranging from 0.33 to 0.81 for HR + /HER2 + subtype and TN subtype respectively. The second meta-analysis investigated PFS surrogacy with OS in a set of 9 phase II/III RCTs evaluating anti-HER2 targeted therapies (trastuzumab or lapatinib), authors reported a strong individual-level association (ρ = 0.69) but 2 in 9 trials investigated drugs in second-line or more setting [47]. Our estimates of individual-level association between rwPFS and OS in HER2 + mBC patients exposed to L1 targeted therapy (range: 0.67–0.87) were consistent with those results focusing on drugs with similar mechanisms of action although all RCTs pooled in the meta-analysis included patients enrolled before 2008. As surrogacy assessment is specific to a disease and to the mechanisms of action of the drug, comparison of our results to published meta-analyses is complex as few of those are available. Investigations are supported using meta-analysis requiring large amount of individual patient data among terminated RCT assessing drug efficacy with the same mechanism of action.

Results of individual-patient level association using de-identified IPD collected in real-world setting are in line with published results seeking for PFS surrogacy with OS using RCTs data. This work shows the value of RWD to support the search of potential candidate surrogate endpoint for OS.

As RWD, the ESME mBC presents limitations [23, 48]. These include the lack of availability of electronic medical records data required to describe the global mBC management due to the low level of standardization of current electronic medical records as well as the retrospective patient selection-data collection. The interpretation of reported survival estimates may be limited by the presence of confounding factors inherent to the observational nature of RWD, as opposed to randomized controlled studies. As our primary objective was the association between rwPFS and OS, we only reported descriptive data for rwPFS and OS (crude estimates based on Kaplan–Meier) and refer the reader to earlier ESME mBC publications for further data on prognostic factors for rwPFS/OS (3, 21, 22). Concerning overall generalizability and external validity, the cohort centralizes data from patients treated in specialized cancer centers only, which thus may use different clinical practices compared with public hospitals and private institutions. Although we did assess real-world outcomes according to the first-line treatment by tumor subtypes, residual heterogeneity within treatment group remains. Indeed, we did not make any distinctions regarding the mechanism of action of the drugs among each treatment group, which affect individual-level association. As an example, we assigned to “targeted therapy” to both bevacizumab, VEGF-targeted therapy (component of mBC therapy affecting the tumor angiogenesis) and palbociclib, inhibitor of cyclin-dependent kinases 4 and 6 (CDK4/6) involved are the downstream of signaling pathways which lead to cellular proliferation). Antiangiogenics such as bevacizumab are unique in that PFS advantages appear to frequently be completely erased at OS, and this is the only therapy in breast cancer preclinically and clinically implicated to have a reversal effect in second and subsequent lines of therapy. This issue highlights that even within drug classes, residual variability in terms of mechanism of action may persist. Similarly, the impact of subsequent lines of treatment could be further investigated. Finally, in real-world setting, non-systematic RECIST-based progression evaluation may also introduce bias in our analyses, as compared to standardized RECIST-based assessment in RCTs.

This study reports comprehensive information related to individual-level association between rwPFS and OS according to BC subtype and L1 mBC treatment. Overall, within the 4 mBC subgroups, individual-level correlation between rwPFS and OS were at least strong for each dominant treatment class (i.e., chemotherapy for TN mBC, endocrine therapy for HR + /HER2 − , and anti-HER2 in either HER2 + subgroups). Those results support the value of RWD when searching for candidate surrogate endpoints for OS. Data could be used subsequently to investigate or generate research hypotheses for future surrogate endpoint candidates. We also provided researchers with treatment patterns and rwPFS according to the 1L treatment received by tumor subtype. These estimates could be used when designing future research, in particular to provide information not readily available from the published trial literature for underrepresented populations.