Main

Triple-negative breast cancer (TNBC) is characterized by the lack of expression of estrogen and progesterone receptors and no overexpression or amplification of human epidermal growth factor receptor 2 (HER2)1,2. TNBC represents approximately 10 to 15% of all breast cancer (BC) diagnoses1,3 and is typically very aggressive, with a shorter time to recurrence and poorer prognosis than the other subtypes in both the early and advanced settings4,5.

Programmed cell death-ligand 1 (PD-L1) and its receptor, programmed cell death 1 (PD-1), are targets for immune checkpoint inhibitors (ICIs). Moreover, PD-L1 has been validated as both a prognostic and predictive biomarker for response to immunotherapy across different cancers6. TNBC is considered the most immunogenic among all BC subtypes, characterized by a higher tumor mutational burden and increased infiltration of T cells. Approximately 40% of patients enrolled in immunotherapy trials for advanced TNBC (aTNBC) have PD-L1-positive tumors7,8,9. However, PD-L1 positivity is influenced by the assay used, the algorithm and defined cutoff point used to determine expression, and the metastatic site where PD-L1 status is evaluated10,11,12,13,14.

Current first-line treatment for aTNBC is determined by the expression of PD-L1 (refs. 15,16). The IMpassion130 (ref. 8) and KEYNOTE-355 (refs. 7,17) trials demonstrated that patients with PD-L1-positive aTNBC have a substantial improvement in progression-free survival (PFS) with the addition of ICIs, respectively atezolizumab or pembrolizumab, to standard chemotherapy. In KEYNOTE-355, a statistically significant improvement in overall survival (OS) was also observed7. Conversely, these two pivotal trials did not show a benefit in PFS or OS in patients with PD-L1-negative aTNBC. Consequently, while several clinical trials testing new approaches are ongoing, chemotherapy remains the standard first-line treatment for this population.

Despite treatment advancements, the median OS for aTNBC remains poor, ranging from 16 months in patients with PD-L1-negative tumors to 24 months in those with PD-L1-positive tumors7,8. This underscores the urgent need of developing additional therapeutic strategies for patients with aTNBC. Although PD-L1 expression is used to predict responses to ICI-based regimens in aTNBC, it is not a definitive biomarker, and some patients with PD-L1 negative tumors have exhibited responses to these therapies18,19,20. Therefore, identifying novel predictive biomarkers beyond PD-L1 expression is essential to enhance the efficacy of PD-1/PD-L1-targeted immunotherapies and to broaden their use beyond PD-L1-positive tumors. Additionally, there is also a critical need to explore the potential synergism of these immunotherapies in combination with other targeted or cytotoxic agents, not only to extend the benefits of immunotherapy to patients who have not demonstrated favorable outcomes in previously reported trials but also to enhance therapeutic efficacy in those who do respond, with the goal of delaying disease progression and improving overall outcomes.

Bevacizumab is an antibody directed against vascular endothelial growth factor (VEGF) that reduces the angiogenic and oncogenic activity of VEGF21. When combined with chemotherapy, bevacizumab has demonstrated a synergistic effect in HER2-negative advanced BC, notably improving both PFS and response rates22,23,24,25,26. Beyond its impact on angiogenesis, VEGF can also exert immunosuppressive effects within the tumor microenvironment by impairing antigen presentation by tumor cells. Antiangiogenic agents can counteract this, potentially increasing the efficacy of concurrent treatments27,28. Preclinical studies in TNBC models have shown that combining anti-VEGF therapies with ICIs can improve antitumor activity of these therapies by normalizing tumor vasculature and promoting immune cells infiltration29,30. Indeed, the immunomodulatory properties of bevacizumab make it a promising candidate to use in combination with other immunotherapeutic approaches31,32,33,34,35. A study assessing sequential treatment with bevacizumab and durvalumab showed efficacy in HER2-negative advanced BC, with bevacizumab enhancing the overall response to durvalumab through immune-priming effects36. Moreover, combining bevacizumab with immunotherapy and chemotherapy has also demonstrated clinical benefit in multiple tumors, including HER2-negative advanced BC37.

In this study, we hypothesized that combining atezolizumab plus paclitaxel and bevacizumab could improve outcomes in patients with previously untreated aTNBC, regardless of PD-L1 status. Here, we present the efficacy and safety results of this combination from the ATRACTIB phase 2 study.

Results

From 2 October 2020 to 4 May 2022, a total of 100 patients were recruited at 22 sites in four countries. All enrolled patients were female, with a median age of 55 years (range, 32 to 84), and 75% of them had an Eastern Cooperative Oncology Group (ECOG) performance status of 0 (Table 1). Overall, 29% patients had de novo metastatic BC and 56% presented visceral involvement, including lung, liver, brain, adrenal gland, kidney, spleen and other thoracic and abdominal regions. Additionally, 46% of patients had three or more metastatic sites. Among the 70 patients who had undergone prior (neo)adjuvant therapy, 61 were treated with a taxane-based regimen and none received prior PD-1 or PD-L1 inhibitors or bevacizumab. A tumor sample was available for PD-L1 evaluation in 85 patients (31.8% from a metastatic site), of whom 83 had PD-L1-negative tumors (97.6%, 83 out of 85) (Table 1). For the remaining 15 patients, PD-L1 expression could not be evaluated or samples were not available.

Table 1 Baseline patient characteristics
Full size table

At the time of data cutoff (17 April 2024), all patients had discontinued study treatment, with a median follow-up duration of 19.4 months (range, 0.9 to 37.4). The main reason for treatment discontinuation was disease progression, which occurred in 66% of patients. Nineteen (19%) patients had completed study treatment and transitioned to post-trial access. Other reasons for treatment discontinuation were unacceptable toxicity (7%), the need for surgical intervention related to the disease under study (4%) and withdrawal of consent (4%) (Fig. 1). One patient who discontinued treatment due to unacceptable toxicity simultaneously reported progressive disease.

Fig. 1: Patient disposition of the ATRACTIB trial at the data cutoff date of 17 April 2024.
figure 1

a, Toxicity prompted the decision to discontinue treatment in four patients, but treatment discontinuations were documented in the context of disease progression. b, Patients discontinued treatment in the framework of the study, but were offered to continue receiving treatment outside the trial, as per treating physician decision and in accordance with the marketing authorization holders’ policy. c, Three discontinuations due to AEs (vomiting and cough, digestive toxicity, pulmonary thromboembolism and autoimmune-mediated hepatitis) and four discontinuations due to serious AEs (cerebral accident, sepsis, fatigue and thrombocytopenia, and ejection fraction decreased). n, number of patients.

Full size image

The primary endpoint of this study was met, with a median PFS of 11.0 months (95% CI: 9.0–13.4; P < 0.001) (Fig. 2). Objective response rate (ORR) was 63% (95% CI: 52.8–72.4), 14% of patients had complete responses (3% of them unconfirmed) and 49% had partial responses (6% of them unconfirmed) (Table 2). The median best percentage change from baseline was −59.8% (range, −100.0 to 94.9) (Fig. 3a). Clinical benefit rate (CBR) was 80% (95% confidence interval (CI): 70.8–87.3). Median duration of response (DoR) was 10.1 months (95% CI: 7.3–13.9) and median time to treatment response (TTR) was 1.9 months (95% CI: 1.9–3.7). Median OS was 27.4 months (95% CI: 23.4–37.4); the 12-month OS rate was 81.3% (95% CI: 71.9–87.8), and the 24-month OS was 59.8% (95% CI: 46.3–71.0) (Fig. 3b). A post hoc analysis of the treatment effect on PFS, OS and ORR across key subgroups is presented in Extended Data Figs. 1–3. These analyses suggested a worse PFS in patients with bone (hazard ratio (HR) 1.90; 95% CI 1.18–3.07; P = 0.008) and liver metastasis (HR 1.69; 95% CI 1.02–2.79; P = 0.041), as well as in those with HER2-0 (immunohistochemistry (IHC) 0) status (HR 1.67; 95% CI 1.00–2.78; P = 0.048). In patients with confirmed PD-L1-negative tumors (83 out of 100), median PFS was 9.3 months (95% CI: 7.8–13.2) and median OS was 24.5 months (95% CI: 22.9–not reached) (Extended Data Figs. 1 and 2 and Extended Data Table 1). Additional detailed data on clinical outcomes in patients with PD-L1-negative tumors are provided in Extended Data Table 1.

Fig. 2: PFS in the full analysis set.
figure 2

Kaplan–Meier curve of investigator-assessed PFS in the full analysis set. KM, Kaplan–Meier.

Full size image
Table 2 Objective response per Response Evaluation Criteria in Solid Tumors (RECIST v.1.1) assessed by investigators
Full size table
Fig. 3: Tumor response and survival analysis in the full analysis set.
figure 3

a, Waterfall plot of BOR per RECIST v.1.1. Dotted lines mark the thresholds for partial response (≥30% decrease) and disease progression (≥20% increase). b, OS. Reasons for censoring: reaching the end of the study as per protocol (48%), loss to follow up (7%), patient decision (7%) and investigator decision (1%). aPatients with only non-target lesions.bThree patients discontinued before post baseline assessment, one due to progressive disease and two due to withdrawal of consent.

Full size image

All patients received at least one dose of each study drug. The median relative dose intensities (RDIs) were 93.8% for atezolizumab, 60.7% for paclitaxel and 91.7% for bevacizumab. The median treatment durations were 9.3 months for atezolizumab, 5.6 months for paclitaxel and 9.5 months for bevacizumab.

All patients experienced treatment emergent adverse events (TEAEs), with 61% presenting grade (G) 3 or 4 TEAEs (Extended Data Table 2). Discontinuations of any of the study treatments due to treatment-related TEAEs occurred in 48% of patients, with paclitaxel discontinuations being the most frequently reported (41%) (Extended Data Table 2). There were no treatment-related deaths in this study. The most common non-hematological TEAEs of any grade included fatigue (63%; 7% G3), neurotoxicity (46%; 10% G3 or G4), diarrhea (44%; 4% G3 or G4) and alopecia (41%). Neutropenia (27%; 12% G3 or G4) and anemia (24%) were the most frequent hematologic TEAEs (Table 3). Individual evaluation of safety by study drug is detailed in Supplementary Tables 1–3. Of note, 26% of patients experienced hypertension (6% G3) related to the treatment with bevacizumab, with one patient developing a G4 hypertensive crisis (Supplementary Table 3). Serious TEAEs occurred in 34% of patients (28% G3 or G4) and events of clinical interest (ECIs) were reported in 51% of patients (15% G3 or G4) (Supplementary Tables 4 and 5).

Table 3 TEAEs reported in ≥15% of patients or any TEAEs of grade 4
Full size table

Immune-related adverse events (irAEs) of any grade occurred in 32% of patients, with 8% of them experiencing G3 events (Supplementary Table 6). No G4 irAEs were reported. The most frequently reported irAEs of any grade were thyroid disorders (21%; 1% G3), then immune-mediated hepatitis (4%; 3% G3), gastrointestinal disorders (3%; 3% G3), nephritis (3%; 3% G3) and Addison’s disease (3%; 0% G3).

Discussion

Here, we describe the primary results of the ATRACTIB phase 2 study, which demonstrated encouraging antitumor activity and a manageable safety profile of the atezolizumab, paclitaxel and bevacizumab triplet combination as first-line therapy for patients with aTNBC, enrolled irrespectively of PD-L1 status. The primary endpoint was met, with a median PFS of 11.0 months. Moreover, ORR was 63%, DoR was 10.1 months and the median OS was 27.4 months.

The randomized phase 3 IMpassion130, IMpassion131 and KEYNOTE-355 studies evaluated the role of chemo-immunotherapy as frontline treatment for aTNBC, both in the intention-to-treat and in the PD-L1-positive populations (~40%). The median PFS of 11.0 months observed in ATRACTIB was superior to the median PFS of 7.2, 5.7 and 7.5 months reported among the intention-to-treat populations of the IMpassion130, IMpassion131 and KEYNOTE-355, respectively7,8,9. ATRACTIB was an open-label study with no predefined selection by PD-L1 status, but highly enriched in patients with PD-L1-negative disease, as tumors of 83 of the 85 patients with evaluable basal samples classified as PD-L1-negative (97.6%) by central review. The median PFS and OS in this group were 9.3 and 24.5 months, respectively. Interestingly, in the IMpassion130 study, patients with PD-L1-negative aTNBC treated with atezolizumab plus nab-paclitaxel—a population that closely resembles that of ATRACTIB—had a median PFS of 5.6 months8 and a median OS of 19.7 months38, results considerably lower than those in ATRACTIB. Irrespective of PD-L1 status, our results also surpass those obtained with bevacizumab in combination with paclitaxel in aTNBC, with a median PFS of 8.1 months and a median OS of 18.9 months26. While these comparisons are noteworthy, differences in trial design and patient characteristics should be considered when interpreting efficacy results, particularly given the known variability of immunotherapy outcomes between phase 2 and phase 3 trials39.

In our study, OS was not a primary endpoint; therefore, these data are purely descriptive. The median OS of 27.4 months may have been influenced by the high censoring rate, as 48% of patients were censored due to study completion, limiting its ability to accurately reflect the true survival benefit. However, early survival estimates, before most censures, indicates a 12-month OS rate of 81.3%, which compares favorably with the observed 12-month OS rates of ~72% in the IMpassion130 study38 and ~65.0% in the KEYNOTE-355 study7. Moreover, the OS benefit of bevacizumab treatment in aTNBC has been challenging to demonstrate22,23,24,26, but our findings underscore its potential in combination strategies.

The toxicity profile observed in ATRACTIB was manageable, with G3 or G4 TEAEs occurring in 61% of patients and 48% discontinuing at least one of the treatments due to treatment-related TEAEs, primarily attributed to paclitaxel toxicity (41%). In comparison, IMpassion130 reported a lower incidence of G3 and G4 TEAEs (48.7%) and a discontinuation rate due to TEAEs of 15.9% (ref. 8). However, in the E2100 trial evaluating paclitaxel plus bevacizumab, there was an increased incidence of G3 and G4 adverse events (AEs), including hypertension (14.8%), fatigue (9.1%) and proteinuria (3.6%). Notably, the treatment discontinuation rate due to TEAEs was 51.3%, largely attributed to cumulative toxic effects23. In a phase 2 study evaluating the triple combination of nivolumab, bevacizumab and paclitaxel, the safety profile was comparable to that observed in ATRACTIB, with 58% incidence of G3 and G4 AEs37. Overall, no new safety signals9,23 or treatment-related deaths were reported.

As previously mentioned, ATRACTIB predominantly included patients with PD-L1-negative tumors, for which there is an unmet clinical need. This selection may have resulted from concurrent trials selectively enrolling PD-L1-positive patients, as well as from investigators opting against selecting patients with PD-L1-positive disease because of the lack of clinical benefit observed for atezolizumab plus paclitaxel in IMpassion131 (ref. 9). Moreover, atezolizumab and pembrolizumab were approved as first-line treatments for aTNBC during ATRACTIB recruitment, becoming the standard of care for patients with PD-L1-positive tumors and reducing the need for their clinical trial participation. Finally, differences in PD-L1 assessment methods should also be acknowledged, as the VENTANA SP142 assay—used in ATRACTIB—may classify some tumors as PD-L1-negative that could potentially meet combined positive score positivity thresholds. However, the use of SP142 maintains consistency with prior pivotal trials of atezolizumab in this population8,9, providing support in the methodology and potential comparability of the results.

Other new combinations in aTNBC further reinforce chemo-immunotherapy plus antiangiogenic strategies. The FUTURE-C-PLUS study, evaluating famitinib (VEGF receptor inhibitor), camrelizumab (anti-PD-1) and nab-paclitaxel as first-line treatment for immunomodulatory aTNBC, reported a median PFS of 13.6 months and an ORR of 81.3%, with efficacy in both PD-L1-positive (35.4%) and PD-L1-negative (27.1%) patients40. In comparison, the ATRACTIB trial achieved a median PFS of 11.0 months and an ORR of 63.0% in a predominantly PD-L1-negative population, with 14% of complete responses. Similarly, combining nivolumab, bevacizumab and paclitaxel in a phase 2 study of TNBC has shown an ORR of 59% and a median PFS of 7.8 months, with no correlation between PD-L1 expression and efficacy37. Moreover, other studies evaluating bispecific antibodies, like ivonescimab (targeting VEGF and PD-1)41 and PM8002/BNT327 (targeting VEGF and PD-L1)42, as first-line treatments for aTNBC have also demonstrated promising activity in PD-L1-negative aTNBC. These findings suggest that antiangiogenic agents may be key to paving the way for the use of immunotherapy in PD-L1-negative aTNBC.

Further studies are still needed to identify biomarkers for optimal patient selection. In our study, subgroup analyses were conducted to explore potential predictors of response, and ongoing evaluations—including proteomic plasma profiling and tumor biopsy assessments—may help identify novel biomarkers. The FUTURE-C-PLUS study described PD-L1/CD31 double positivity as a potential biomarker for improved outcomes and revealed other genetic alterations associated with treatment response40,43, warranting further research into new biomarker-driven approaches beyond PD-L1 status.

The ATRACTIB study has two main strengths: a large sample size of 100 patients, which is substantial for a phase 2 trial, and the inclusion of a predominantly PD-L1-negative population, a challenging-to-treat subgroup. However, the predominance of PD-L1-negative tumors also represents a limitation, as it restricts the extrapolation of results to PD-L1-positive patients. Another limitation is the single-arm design, which precludes a formal assessment of the contribution of each agent in the regimen, as well as the potential synergy between bevacizumab and the atezolizumab–paclitaxel combination. Additionally, the high censoring rate for OS data limits its interpretability, rendering the OS results primarily descriptive.

Two important clinical questions arise from our findings. First, whether this regimen may be effective in patients previously treated with immunotherapy, a population not included in the ATRACTIB. Although data on this scenario is limited, the synergy between antiangiogenic agents and immunotherapy (± chemotherapy) has been demonstrated18,44,45,46, suggesting that combining bevacizumab with immunotherapy and chemotherapy may offer additional benefits for this population. Second, the feasibility of including patients with estrogen receptor (ER)-low tumors in future trials evaluating treatment strategies for TNBC, as retrospective analyses suggest biological and clinical similarities between tumors with ER expression levels between 1 to 9% and those with ER expression <1% (refs. 47,48,49,50), including comparable expression of immune biomarkers associated with immunotherapy response. Based on this evidence, it may be reasonable to extrapolate the potential use of our triplet to the ER-low population.

Beyond antiangiogenic combinations, emerging PD-1/PD-L1-targeted immunotherapy plus antibody-drug conjugates (ADCs) strategies are reshaping the first-line treatment landscape in aTNBC. The MORPHEUS-panBC trial (atezolizumab plus sacituzumab govitecan) reported an ORR of 76.7% and a PFS of 12.2 months in patients with PD-L1-positive disease51, while the BEGONIA trial (datopotamab deruxtecan plus durvalumab) showed an ORR of 79% and a PFS of 13.8 months in a predominantly PD-L1-low population (87.1%)52,53. These findings, together with those from ATRACTIB, suggest that immunotherapy benefits may extend beyond PD-L1-positive tumors, particularly in combination with antiangiogenics or ADCs.

In conclusion, the ATRACTIB study demonstrated that the combination of atezolizumab plus paclitaxel and bevacizumab exhibits encouraging antitumor activity as a first-line treatment for patients with aTNBC, irrespective of PD-L1 expression. Importantly, bevacizumab appears to improve the efficacy of immunotherapy in this patient population. In the context of the evolving treatment landscape for aTNBC and the potential introduction of Trop-2-directed ADCs as first-line therapy, our findings warrant further investigation by evaluating the role of bevacizumab in combination with emerging ADCs and immunotherapies, regardless of PD-L1 status.

Methods

Study design and participants

The ATRACTIB study was a single-arm, open-label, multicenter phase 2 trial conducted across 24 sites in Spain, Germany, Italy and France. The trial protocol is available in the Supplementary Information.

Eligible patients were aged 18 years or older with unresectable locally advanced or metastatic TNBC, confirmed by local assessment and defined as <1% expression for ER and progesterone receptor by IHC, and negative for HER2 (0 to 1+ by IHC or 2+ and negative by in situ hybridization test) according to American Society of Clinical Oncology/College of American Pathologists criteria54,55. Patients were included regardless of their PD-L1 status. No prior therapy for advanced disease was allowed and patients who received prior (neo)adjuvant therapies (taxane-based chemotherapy, immunotherapy and/or antiangiogenic agents) were required to have a disease-free interval of at least 12 months, or at least 6 months in the case of non-taxane-based chemotherapy. Patients with measurable or non-measurable disease as per RECIST v.1.1 were included (patients with bone-only lesions were eligible), and availability of recently or newly obtained tumor sample from either primary breast tumor or the metastatic site was required, unless contraindicated due to site inaccessibility and/or participant safety concerns. In those cases, patient eligibility was evaluated by a Sponsor’s qualified designee. Patients with known active uncontrolled or symptomatic central nervous system metastases, or a history of leptomeningeal disease were excluded, as well as patients with active or a history of autoimmune disease, or other clinical situations that could entail higher risk for patients. Patients were enrolled regardless of sex, which was collected according to the identity information provided by the patients. Complete inclusion and exclusion criteria are provided in the Supplementary Methods.

This study was conducted in compliance with the Declaration of Helsinki and was approved by the institutional review boards or independent ethics committees at each site: Comité de Ética de la Investigación con medicamentos del Hospital Universitari Arnau de Vilanova de la Gerencia Territorial de Lleida (Spain), Comité de Protection des Personnes TOURS – Région Center – Ouest 1 (France), Ethik-Kommission II der Universität Heidelberg (Medizinische Fakultät Mannheim) and Ethik-Kommission an der Universität Duisburg-Essen (Germany), and Comitatio Etico IRCCS Pascale and Comitato Etico Territoriale CAMPANIA 1 (Italy). All patients provided written informed consent. This trial is registered at ClinicalTrials.gov, NCT04408118.

Procedures

In treatment cycles of 28 days, patients received an intravenous infusion of atezolizumab (840 mg) on days 1 and 15; if the first infusion was tolerated over 60 min, subsequent infusions were delivered over 30 min. Paclitaxel (90 mg per m2) was administered on days 1, 8 and 15 as an intravenous infusion over 60 min. Bevacizumab (10 mg per kg) was administered as an intravenous infusion over 30 to 90 min on days 1 and 15. Dose reductions were allowed for paclitaxel as defined by prespecified protocol guidelines but not allowed to atezolizumab and bevacizumab per their respective labels. If one of the study drugs was permanently discontinued, the remaining treatments, either as monotherapy or in combination, could continue to be administered if there was no contraindication. Treatment was given until progressive disease, unacceptable toxicity, death, withdrawal or discontinuation from the study for any other reason, whichever occurred first. End of study was defined as 12 months following the enrollment of the last patient or upon the occurrence of at least 67 PFS events with 100 enrolled patients, whichever occurred later. For patients with ongoing treatment at the time of end of study, who were required to discontinue study treatment as part of the study protocol, transition to post-trial access was offered. This decision was made in consultation with the treating physician and in accordance with the marketing authorization holders’ policy.

Tumor assessments were conducted by computed tomography or magnetic resonance imaging according to RECIST v.1.1 at baseline and every 8 weeks up to 12 months. Thereafter, disease assessment was performed every 12 weeks until the end of treatment visit. Bone scans were conducted at baseline and repeated every 24 weeks for patients with bone lesions identified at baseline, unless clinical or biochemical bone progression was suspected.

PD-L1 expression was assessed with the VENTANA SP142 immunohistochemistry assay, and PD-L1 positivity was defined as ≥1% PD-L1 expression on tumor-infiltrating immune cells.

Outcomes

The primary endpoint of this study was investigator-assessed PFS as per RECIST v.1.1. Secondary endpoints were ORR and CBR as per RECIST v.1.1; best percentage change from baseline; DoR; TTR; OS; and safety, including AEs, TEAEs, serious AEs, ECI (the definition of ECI is detailed in the Supplementary Methods), irAEs and rates of discontinuations and dose reductions because of AEs as per Common Terminology Criteria for Adverse Events v.5.0. The definition of all endpoints is provided in the Supplementary Methods.

Exploratory endpoints were immune-related PFS; immune-related ORR; association of tumor and/or immune-related biomarkers (for example, PD-L1, tumor-infiltrating lymphocytes, tumor mutational burden or cytokine profile) with treatment efficacy; and changes in mutation and copy number in oncogenes, tumor suppressor genes and/or genes associated with disease progression, as evaluated in liquid biopsy. Exploratory endpoints are under evaluation and not reported here.

Statistical analysis

We conducted the efficacy and safety analyses in the full analysis set, comprising all enrolled patients who received at least one dose of study drugs. The median PFS was calculated using the Kaplan–Meier method. The maximum likelihood estimation for the exponential distribution test was used to compare differences between observed and historical rates56,57. The PFS analysis was designed to test the null hypothesis that the true median PFS was 7 months or less11. The alternative hypothesis was that the true median PFS was greater than or equal to 9.5 months (HR 0.74). We estimated that enrolling 100 patients would provide 80% power at a nominal level of one-sided α of 0.05. This primary objective would be met with a median PFS of 8.6 months and 67 PFS events observed. The R code used for sample size calculation and primary analysis is provided in the Supplementary Methods. The 95% CI for median PFS, OS, TTR and DoR were estimated using the Greenwood method. ORR and CBR were estimated with the 95% Clopper–Pearson CI based on an exact binomial test. The best percentage change from baseline was described for each patient with a waterfall plot. For forest plot analysis, HR and associated 95% CI for PFS and OS were estimated using Cox proportional hazards models with Breslow’s method of tie handling. The odd ratios and associated 95% CI for ORR were calculated using logistic regression models. RDI for each study drug was calculated from treatment initiation to discontinuation of all study drugs; RDI was defined as the actual dose of the study drug administered to each patient divided by the scheduled dose per study protocol58,59. We used descriptive statistics to summarize baseline characteristics and safety data. Patients without any post-baseline assessment for primary PFS or secondary endpoints OS were censored after one day of treatment. For best overall response, patients lacking post-baseline assessments were considered as non-responders or without clinical benefit. Statistical analyses were performed with SAS software (v.9.4) and with R software (v.4.2.2). Detailed statistical analyses are described in the statistical analysis plan in the Supplementary Information.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.