Ipatasertib plus Paclitaxel for Patients with PIK3CA/AKT1/PTEN-Altered Locally Advanced Unresectable or Metastatic Triple-Negative Breast Cancer in the IPATunity130 Phase III Trial Open Access

The phosphoinositide 3-kinase/protein kinase B (PI3K/AKT) pathway is frequently activated by genomic aberrations in cancer (1). In triple-negative breast cancer (TNBC), AKT can be activated by loss of function of negative regulators (PTEN, INPP4B, PHLPP, and PP2A), gain of function of positive regulators [PI3K, AKT, receptor tyrosine kinases (e.g., HER2)], and chemotherapy-induced survival response (2, 3). Approximately 35% of TNBCs harbor alterations in PIK3CA, AKT1, and/or PTEN (4–7). Consequently, there has been keen interest in targeting AKT.

Ipatasertib is a highly selective ATP-competitive AKT inhibitor. Studies in tumor xenograft models suggested that ipatasertib enhanced the activity of taxane chemotherapy (8). In the randomized phase II LOTUS trial, the addition of ipatasertib to paclitaxel improved progression-free survival (PFS) compared with paclitaxel alone in an unselected population of patients with locally advanced/metastatic TNBC (9). The effect on PFS was more pronounced in the subgroup of 42 patients with alterations in PIK3CA, AKT1, and/or PTEN [hazard ratio 0.44, 95% confidence interval (CI), 0.20–0.99], providing the rationale for prospectively evaluating ipatasertib plus paclitaxel in a biomarker-selected population. A similar effect was observed in the subgroup of patients with PIK3CA/AKT1/PTEN alterations in the randomized phase II PAKT trial evaluating another AKT inhibitor, capivasertib, in combination with paclitaxel (10). Final overall survival (OS) results from LOTUS showed a numerical trend favoring ipatasertib-containing therapy in both the intent-to-treat (ITT) and the biomarker-selected populations, with median OS exceeding 2 years (11).

The phase III IPATunity130 trial evaluated ipatasertib plus paclitaxel combination therapy in two independent randomized cohorts (Cohort A in TNBC and Cohort B in hormone receptor-positive HER2-negative unresectable locally advanced or metastatic breast cancer). A third cohort, the single-arm signal-seeking Cohort C, evaluated ipatasertib and paclitaxel in combination with atezolizumab in TNBC (12). The two randomized cohorts were powered independently and designed to be analyzed separately. Results from Cohort B have been published (13) and some elements of the methodology for Cohort A (reported below) are identical to the previously published Cohort B. However, the patient population, and thus the biology, prognosis, and clinical outcomes, differs between the two cohorts. Here, we report the methodology and results from Cohort A.

In Cohort A of the international randomized double-blind placebo-controlled phase III IPATunity130 (NCT03337724) trial, eligible patients had PIK3CA/AKT1/PTEN-altered, measurable [by Response Evaluation Criteria in Solid Tumours (RECIST) version 1.1], locally recurrent inoperable or metastatic TNBC. Receptor status was based on local assessment of the most recent biopsy [i.e., recurrent or metastatic tissue where applicable and if safely accessible, or nonfine-needle aspiration sample], or assessed on-study if not available locally, according to the ASCO/CAP guidelines (HER2 immunohistochemistry 0 or 1+, in situ hybridization negative; estrogen and progesterone receptor <1% of tumor cell nuclei immunoreactive to the respective hormonal receptor). Tumor PIK3CA/AKT1/PTEN alterations (activating alterations in PIK3CA and/or AKT1, and/or inactivating alterations in PTEN, described in detail in Supplementary Table S1) were assessed from the most recently available tumor tissue sample using the Foundation Medicine Inc. (Cambridge, Massachusetts) next-generation sequencing clinical trial assay (or appropriately validated local laboratory test result for PIK3CA/AKT1/PTEN alteration using the commercial tissue-based next-generation sequencing FoundationOne CDx). Patients had to be considered by the investigator to be candidates for taxane monotherapy (unsuitable for endocrine therapy, and without rapid clinical progression, life-threatening visceral metastases, or the need for rapid symptom and/or disease control that may require combination chemotherapy) and have Eastern Cooperative Oncology Group performance status 0 or 1. Patients who had previously received chemotherapy for advanced TNBC or whose diagnosis of advanced TNBC was <1 year since their last (neo)adjuvant chemotherapy were ineligible, as were patients with a history or known presence of brain or spinal cord metastases or leptomeningeal carcinomatosis.

The study protocol was approved by an institutional review board or independent ethics committee at each site. The trial was carried out in accordance with the principles of the International Council for Harmonisation Guidelines for Good Clinical Practice, the Declaration of Helsinki, and all applicable national and local laws. All patients provided written informed consent.

Using an interactive web-response system, investigators randomized patients in a 2:1 ratio to receive either oral ipatasertib (400 mg daily on days 1–21) plus intravenous paclitaxel (80 mg/m² on days 1, 8, and 15) of a 28-day cycle, or placebo plus the same paclitaxel regimen. Randomization was stratified by three criteria: prior (neo)adjuvant chemotherapy (yes vs. no), geographic region (Asia-Pacific vs. Europe vs. North America vs. rest of world), and tumor alteration status (PIK3CA/AKT1-activating mutation vs. PTEN alteration without PIK3CA/AKT1-activating mutation). Treatment was continued until disease progression according to RECIST version 1.1, unacceptable toxicity, or patient withdrawal. Dose modifications (interruption or dose reduction) of paclitaxel were performed as clinically appropriate based on the investigator’s medical judgment, with one dose reduction (to 65 mg/m²) allowed. Two dose reductions of ipatasertib/placebo (first to 300 mg, then to 200 mg) were permitted. Patients discontinuing paclitaxel or ipatasertib/placebo because of toxicity could continue the other agent as a single-agent treatment. Crossover from placebo to ipatasertib was not permitted. To improve the management of diarrhea, the protocol specified that prophylactic loperamide should be administered during the first cycle according to local guidance.

Tumors were assessed every 8 weeks by the investigators according to RECIST version 1.1. After discontinuing treatment, patients were followed up every 3 months for survival and subsequent anticancer therapies. Adverse events (AEs) were assessed and graded according to Common Terminology Criteria for Adverse Events (version 4.0). Details of the methodology for collecting and analyzing patient-reported outcomes are provided in the Supplementary Materials.

The primary objective was to assess the efficacy of the ipatasertib plus paclitaxel combination as determined by investigator-assessed PFS. PFS was defined as the interval between randomization and the first occurrence of disease progression (investigator-assessed per RECIST version 1.1) or death from any cause, whichever occurred first.

OS (defined as the interval between randomization and death from any cause) was the key secondary endpoint. Other secondary endpoints included the confirmed objective response rate (investigator-assessed per RECIST version 1.1), clinical benefit rate (complete or partial response, or stable disease sustained for ≥24 weeks) in patients with measurable disease at baseline, patient-reported outcomes, and safety.

Tumor samples from a formalin-fixed paraffin-embedded tissue block (or freshly cut unstained serial tumor slides from the most recent tumor tissue sample) were evaluated for gene expression by RNA sequencing using TruSeq RNA Access (Illumina, Inc., San Diego, California) and Q2 Lab Solutions (formerly Expression Analysis, Morrisville, North Carolina). Samples were classified into subtypes by gene expression based on the Absolute Intrinsic Molecular Subtyping method (14) and the method developed by Burstein and colleagues (15). PDL1 expression was assessed using the Ventana PDL1 (SP142) immunohistochemistry assay (Ventana Medical Systems, Tucson, Arizona) by pathologists at HistoGeneX (Antwerp, Belgium), trained by Ventana for PDL1 (SP142) specifically for TNBC. PDL1-positive status was defined as tumor-infiltrating immune cell PDL1 expression on ≥1% of the tumor area.

The planned sample size was 250 patients. For the primary analysis, 125 PFS events in the ITT population were required to detect a hazard ratio of 0.50 with 95.5% power at a two-sided 5% significance level. This corresponds to an increase in median PFS from 6 months in the control arm to 12 months in the ipatasertib-containing arm. Secondary endpoints were to be tested only if the primary analysis of PFS was statistically significant.

Efficacy analyses were based on all randomly assigned patients (ITT population) according to the treatment arm to which patients were allocated. Safety analyses were based on all patients who received at least one dose of ipatasertib, placebo, or paclitaxel; patients were analyzed based on the treatment actually received. Exploratory biomarker analyses were performed in all treated patients with evaluable samples.

This trial is registered with ClinicalTrials.gov, NCT03337724.

Qualified researchers may request access to individual patient-level data through the clinical study data request platform (https://vivli.org). Further details on Roche’s criteria for eligible studies are available here: https://vivli.org/members/ourmembers. For further details on Roche’s Global Policy on the Sharing of Clinical Information and how to request access to related clinical study documents, see here: https://www.roche.com/research_and_development/who_we_are_how_we_work/clinical_trials/our_commitment_to_data_sharing.htm.

Authors from F. Hoffmann-La Roche/Genentech were involved in data analysis and interpretation.

Between February 6, 2018, and April 8, 2020, 255 patients were randomized: 168 to ipatasertib plus paclitaxel and 87 to placebo plus paclitaxel (Fig. 1). All but two patients (both in the ipatasertib arm) received study treatment.

Baseline characteristics were generally well balanced between treatment arms and representative of TNBC trial populations (Table 1; Supplementary Table S2), although a slightly lower proportion of patients in the ipatasertib plus paclitaxel arm than the placebo plus paclitaxel arm had metastatic (rather than locally advanced unresectable) disease at baseline (79% vs. 87%, respectively) and had received (neo)adjuvant chemotherapy (48% vs. 55%, respectively). In 14% of patients, previous systemic anticancer therapy included endocrine treatment [22 patients (13%) in the ipatasertib plus paclitaxel arm, 14 patients (16%) in the placebo plus paclitaxel arm], most commonly tamoxifen and letrozole, suggesting hormone-receptor-positive early breast cancer before diagnosis of advanced TNBC. In both arms, approximately half of the patients had PIK3CA/AKT1-activating mutations (predominantly E17K in AKT1 and E545K in PIK3CA; Supplementary Table S3) and half had PTEN alterations without a PIK3CA/AKT1-activating mutation. PDL1 status was unknown in approximately one-third of patients.

At the clinical cutoff date for the primary analysis (May 7, 2020), the median duration of follow-up in the overall population was 8.3 months (range 0–26.8 months). PFS events had occurred in 55% of patients in each treatment arm. There was no difference in investigator-assessed PFS between the two treatment arms (stratified hazard ratio 1.02, 95% CI, 0.71–1.45; log-rank P = 0.92). Median PFS was 7.4 (95% CI, 5.6–8.5) months in the ipatasertib plus paclitaxel arm and 6.1 (95% CI, 5.5–9.0) months in the placebo plus paclitaxel arm (Fig. 2A). Similarly, there was no difference in the confirmed objective response rate [39% (95% CI, 31%–47%) with ipatasertib plus paclitaxel vs. 35% (95% CI, 25%–46%) with placebo plus paclitaxel] or clinical benefit rate [47% (39%–55%) vs. 45% (35%–56%), respectively].

Subgroup analyses of PFS showed consistent results across all populations analyzed and no subgroup deriving a benefit from ipatasertib was identified (Fig. 2B). In both arms, median PFS approached 1 year among patients with a chemotherapy-free interval of >3 years, representing 20% of the trial population. In both treatment arms, median PFS in patients whose tumors harbored a PIK3CA/AKT1-activating mutation was approximately double that in patients whose tumors had PTEN alterations (without a PIK3CA/AKT1-activating mutation), with no effect of ipatasertib detected in either subgroup.

At the time of the primary analysis, OS results were immature (deaths in 20% of patients). The final OS analysis was performed with a data cutoff date of October 30, 2021 (median duration of follow-up: 18.8 months in the ipatasertib plus paclitaxel arm and 16.8 months in the placebo plus paclitaxel arm). By this date, deaths had been recorded in 126 patients (49%). The stratified hazard ratio was 1.08 (95% CI, 0.73–1.58), and the median OS was 24.4 (95% CI, 19.2–29.4) months in the ipatasertib plus paclitaxel arm and 24.9 (95% CI, 16.9–40.4) months in the placebo plus paclitaxel arm (Fig. 2C).

At least one question of the European Organisation for Research and Treatment of Cancer Quality of Life Questionnaire Core 30 was completed by all but two patients at baseline. Mean changes from baseline global health score/quality of life scores during the first five cycles (when ≥50% of patients completed questionnaires) were <3 points and were similar in the two treatment arms. There was no clinically meaningful (≥10-point) deterioration from baseline in either arm (Supplementary Fig. S1).

The proportion of patients with recorded postprogression therapy was similar in the two treatment groups: 104 patients (62%) in the ipatasertib plus paclitaxel arm and 54 (62%) in the placebo plus paclitaxel arm. The most commonly used postprogression therapies were capecitabine, platinum, and gemcitabine (Supplementary Table S4). Only two patients (one in each arm) received a PARP inhibitor (talazoparib) and 16 [8 (5%) in the ipatasertib plus paclitaxel arm and 8 (9%) in the placebo plus paclitaxel arm] received an anti-PD(L)1 agent at progression.

A total of 179 samples were available for RNA sequencing analysis. Most samples were from primary tissue. In a small proportion of patients [20 patients (12%) in the ipatasertib plus paclitaxel arm and 16 (18%) in the placebo plus paclitaxel arm], samples from the primary tumor indicated estrogen receptor-positive early breast cancer but TNBC in the metastatic sample; an additional two patients in the ipatasertib plus paclitaxel arm had progesterone receptor-positive primary tumors.

Half of the samples were classified by PAM50 as basal, with some differences between Asian and non-Asian populations in the distribution of samples according to Burstein and intrinsic classifiers (Supplementary Table S5). In exploratory analyses according to PAM50 or Burstein molecular subtype, there was no difference in treatment effect on PFS in any of the subgroups (Fig. 3; Supplementary Fig. S2). For OS, there was no apparent difference between the two treatment arms in the subgroups with basal and/or nonluminal androgen receptor (LAR) tumors. However, patients in the nonbasal subgroup (Fig. 3) and the LAR subgroup (Supplementary Fig. S2) treated with ipatasertib-containing therapy seemed to have worse OS than those receiving placebo plus paclitaxel. With deaths in 49% of the ipatasertib plus paclitaxel arm and 24% of the placebo plus paclitaxel arm of the nonbasal subgroup, the median OS was 27.3 versus 40.4 months, respectively (Fig. 3). Although the medians should be interpreted with extreme caution given the small sample sizes, low event rates, and extensive censoring before the median, both the medians and the event rates suggest a better prognosis associated with the nonbasal and LAR subtypes.

At the final efficacy analysis, six patients remained on treatment: five (3%) in the ipatasertib plus paclitaxel arm and one (1%) in the placebo plus paclitaxel arm (Table 2; Fig. 1). Dose reductions were more common in the ipatasertib plus paclitaxel arm, including ipatasertib dose reductions for diarrhea in 14% of patients. However, the proportion of patients discontinuing treatment because of AEs was similar in the two arms (Table 2).

Similar percentages of patients in the ipatasertib and placebo arms experienced grade ≥3 AEs (51% vs. 46%, respectively; Table 3). The proportion of patients experiencing AEs meeting the criteria for “serious” AEs was similar in the two treatment arms (20% of patients in the ipatasertib plus paclitaxel arm and 23% in the placebo plus paclitaxel arm). Grade 5 AEs occurred in two patients (1%) in the ipatasertib plus paclitaxel arm, comprising one case each of cardiopulmonary failure and pulmonary embolism, and two patients (2%) in the placebo plus paclitaxel arm (one case each of tumor lysis syndrome and gastric cancer).

The most common AEs were diarrhea (85% in the ipatasertib plus paclitaxel arm vs. 31% with placebo plus paclitaxel), alopecia (47% vs. 44%, respectively), and nausea (39% vs. 25%; Table 3). The most common grade ≥3 AEs with ipatasertib plus paclitaxel were diarrhea (9%) and neutropenia (7%).

Adding ipatasertib to paclitaxel did not significantly improve PFS (primary endpoint) in patients with PIK3CA/AKT1/PTEN-altered TNBC in Cohort A of the randomized phase III IPATunity130 trial. Furthermore, at the final OS analysis, there was no signal of improved OS with the addition of ipatasertib to paclitaxel in this biomarker-selected population. Results from this trial differ from findings in both of the randomized phase II trials of AKT inhibition in advanced TNBC: LOTUS (paclitaxel with/without ipatasertib; ref. 9) and PAKT (paclitaxel with/without capivasertib; ref. 10). The reasons for the contrasting findings in IPATunity130 Cohort A are not clear, although further characterization of the patient population and molecular subtyping hint at enrollment bias and patient selection, as discussed below.

The median OS of 25 months with paclitaxel is remarkable, emphasizing the heterogeneity of TNBC and suggesting potential selection bias toward a population with a particularly good prognosis. Enrollment of patients with highly aggressive tumor types in clinical trials is challenging, particularly if biomarker results are required for stratification, and it is therefore possible that the patients enrolled in this trial had more indolent disease than those in other recent phase III trials in TNBC or in populations presenting in routine practice. For example, median OS was 18.7 months with nab-paclitaxel in the IMpassion130 trial (16), 15.5 months with investigator-selected chemotherapy in the KEYNOTE-355 trial (17), and 22.8 months with paclitaxel in IMpassion131 (18). Compared with the LOTUS trial, the population treated in IPATunity130 Cohort A included a smaller proportion of Asian patients, a smaller proportion of patients previously treated with (neo)adjuvant chemotherapy, and a larger proportion with a chemotherapy-free interval of ≥12 months. Furthermore, patients with a chemotherapy-free interval of <12 months were enrolled into LOTUS but were ineligible for IPATunity130 Cohort A. Of note, the numerical trend favoring ipatasertib in chemotherapy-pretreated patients in LOTUS (9, 11) was not replicated in IPATunity130 Cohort A (nor in the PAKT trial evaluating capivasertib in a similar setting; ref. 10). Although OS suggests enrollment of a population with better-than-expected outcomes, median PFS of 6.1 months in the control arm matches the assumptions made when the trial was designed, and therefore the negative outcome for the primary endpoint cannot be attributed to “overperformance” of the control arm.

RNA sequencing analysis revealed some important differences between the population enrolled in IPATunity130 Cohort A and those treated in the earlier LOTUS trial. Only 50% of patients in IPATunity130 had basal-type tumors, compared with 81% in the ITT population of LOTUS (11) and 69% in the PIK3CA/AKT1/PTEN-altered (biomarker-selected) population of LOTUS [Roche data on file]. This lower prevalence in a biomarker-selected population is as expected, given the known enrichment of PIK3CA and AKT1 mutations in LAR/nonbasal subtypes. Conversely, only 15% of IPATunity130 Cohort A had the basal-like immune-activated subtype, compared with 34% in the LOTUS ITT population and 28% in the LOTUS biomarker-selected population [Roche data on file]. The LAR subtype was over-represented in IPATunity130 Cohort A (51% prevalence, compared with 21% and 34% in the LOTUS ITT and biomarker-selected populations, respectively), which may have influenced the trial outcome in two ways. First, these patients tend to have a better prognosis, suggesting that the population enrolled in IPATunity130 Cohort A was not a “typical” TNBC trial population. In the control arm, the median PFS was 10.9 months in the LAR subgroup versus 5.3 months in the non-LAR population; the median OS was 40.4 versus 12.2 months, respectively. One hypothesis is that patients needing to start treatment urgently for particularly aggressive diseases may not have been enrolled, as the delay while waiting for molecular results for stratification was not considered acceptable. In support of this hypothesis, the median time from diagnosis of metastatic disease to randomization was almost 2 months in IPATunity130 Cohort A, compared with 1 month in the LOTUS trial [Roche data on file], in which patients could be randomized before the central test results were available (9). However, this hypothesis does not explain the lack of difference between treatment arms, as the patients with better prognosis should have been evenly distributed between treatment arms through randomization. Second, the numerically worse OS with ipatasertib in the nonbasal/LAR subgroups may have contributed to the lack of effect in the overall population. However, these observations in such small subgroups (≤8 events in subgroups of ≤30 patients in the control arm) could be chance findings or attributable to postprogression therapy, particularly as this effect was not observed for PFS, and selection bias cannot be excluded. In addition, most samples were collected from primary rather than metastatic tissue, and some known mutations (including AKT1) are typically acquired (19, 20). Therefore, the samples may not reflect the molecular characteristics at the time of first-line treatment. Consequently, this can be considered only as a hypothesis-generating observation requiring validation in other datasets. Interpretation is complicated further by variation in the distribution of molecular subtypes according to race, consistent with previously reported observations suggesting that the LAR subtype tends to be enriched in Asian populations (21). Furthermore, the definitions of PIK3CA/AKT1/PTEN alterations in this trial differ slightly from the definition used in the PAKT trial of capivasertib. Finally, in the LOTUS trial, there seemed to be an enhanced ipatasertib treatment effect on OS in the basal-like subgroup, but this was not observed in the basal subgroup of the present trial and may have been a chance finding in LOTUS given the small sample size. Likewise, the apparently worse OS in the nonbasal subgroup of IPATunity130 Cohort A should be treated with caution, especially given the lack of such an effect for PFS, the small number of OS events in both arms, and the potential of postprogression therapy to have an impact on OS.

Ipatasertib plus paclitaxel was well tolerated, and the safety profile of the regimen was consistent with the known risks of each agent. No new safety signals were identified. Ipatasertib was associated with more all-grade diarrhea, nausea, and vomiting, similar to findings with ipatasertib plus paclitaxel in the LOTUS trial (9). The incidence of grade ≥3 diarrhea with ipatasertib plus paclitaxel was considerably lower in IPATunity130 Cohort A than in the randomized phase II LOTUS trial (9% vs. 23%, respectively), probably reflecting enhanced awareness and more active prophylaxis and management of this side effect.

Although this trial was not positive and the biomarker-selected population did not benefit from the addition of ipatasertib to paclitaxel, the translational work may help improve our understanding of the PI3K/AKT pathway, which is poorly understood in TNBC. These results may also challenge our approach to trial design: while enrolling a biomarker-selected population based on PFS findings from the randomized phase II LOTUS trial was in accordance with the drive toward personalized medicine, it brings new challenges if biomarker selection introduces bias toward certain subtypes of TNBC and nonrepresentative patient populations are enrolled into confirmatory trials. Furthermore, the enhanced PFS benefit in the biomarker-selected population of both LOTUS and PAKT was not evident for OS with longer follow-up. The ongoing CAPItello-290 trial, which is evaluating the AKT inhibitor capivasertib in an “all-comer” population of patients with advanced TNBC, may help us to understand whether targeting the PI3K/AKT pathway can be effective in a broader, unselected population of patients with TNBC.

R.A. Dent reports travel sponsorship and/or honoraria as part of advisory board committees from AstraZeneca, MSD, Pfizer, Eisai, Novartis, Daiichi Sankyo, and Roche. S.-B. Kim reports research funding from Novartis, Sanofi-Aventis, and DongKook Pharm Co; consultancy in advisory boards for Novartis, AstraZeneca, Lilly, Dae Hwa Pharmaceutical Co. Ltd., ISU Abxis, OBI pharma, Beigene, and Daiichi Sankyo; and stocks in Genopeaks and NeogeneTC. M. Oliveira reports grants, personal fees, and nonfinancial support from Roche during the conduct of the study, as well as grants, personal fees, and nonfinancial support from AstraZeneca and Gilead; personal fees from Cureos Science, iTEOS, Lilly, MSD, Relay Therapeutics, and Daiichi Sankyo; grants and personal fees from Roche and Seagen; grants from Zenith Epigenetics, Ayala Pharmaceuticals, Genentech, and Immutep; and nonfinancial support from Eisai and Pierre-Fabre outside the submitted work. C. Barrios reports grants from Nektar, Pfizer, Polyphor, Amgen, Daiichi Sankyo, Sanofi, Exelixis, Regeneron, Novartis, GSK, Janssen, OBI Pharma, Lilly, Seagen, Roche, BMS, MSD, AstraZeneca, Novocure, Aveo Oncology, Takeda, PharmaMar, Gilead Sciences, Servier, Tolmar, Nanobiotix, Dizal Pharma, TRIO, Labcorp, ICON, IQVIA, Parexel, Nuvisan, PSI, Worldwide, Latinaba, Fortrea, PPD, and Syneos Health during the conduct of the study, as well as personal fees from Gilead, Boehringer Ingelheim, GSK, Novartis, Pfizer, Roche/Genentech, Eisai, Bayer, MSD, AstraZeneca, Zodiac, Lilly, Sanofi, Daiichi Sankyo, and Roche outside the submitted work; and stock ownership in Thummi and MedSIR. J. O’Shaughnessy reports personal fees from Agendia, Aptitude Health, Daiichi Sankyo, Eisai, G1 Therapeutics, Genentech, Gilead Sciences, Lilly, Loxo Oncology, Merck, Novartis, Pfizer, Pierre Fabre Pharmaceuticals, Puma Biotechnology, Roche, Samsung Bioepis, Ontada, Seagen, Stemline Therapeutics, AstraZeneca, Fishawack Health, and Veru outside the submitted work. S. Saji reports grants from AstraZeneca during the conduct of the study, as well as personal fees from Kyowa Kirin, Pfizer, Ono, Exact Sciences, and Nippon Kayaku; grants and personal fees from Taiho, MSD, Novartis, Eisai, Takeda, Daiichi Sankyo, Eli Lilly, Gilead, AstraZeneca, and Chugai; and grants from Sanofi outside the submitted work. M. Philco reports other support from Roche and Bayer outside the submitted work. D. Bradley reports other support from Roche Products Limited during the conduct of the study, as well as other support from Roche Products Limited outside the submitted work; in addition, D. Bradley has a patent for HER2 treatment pending. H. Hinton reports being an employee and shareholder of F. Hoffmann La-Roche. M.J. Wongchenko reports other support from Genentech/Roche during the conduct of the study, as well as other support from Genentech/Roche outside the submitted work. N. Turner reports advisory board honoraria from AstraZeneca, Lilly, Pfizer, Roche/Genentech, Novartis, GlaxoSmithKline, Repare Therapeutics, Relay Therapeutics, Gilead, Inivata, Guardant, and Exact Sciences and research funding from AstraZeneca, Pfizer, Roche/Genentech, Merck Sharpe and Dohme, Guardant Health, Invitae, Inivata, Personalis, and Natera. No disclosures were reported by the other authors.

R.A. Dent: Conceptualization, resources, supervision, methodology, writing-original draft, writing–review and editing. S.-B. Kim: Resources, writing–review and editing. M. Oliveira: Resources, writing–review and editing. C. Barrios: Resources, writing–review and editing. J. O’Shaughnessy: Resources, writing–review and editing. S.J. Isakoff: Resources, writing–review and editing. S. Saji: Resources, writing–review and editing. R. Freitas-Junior: Resources, writing–review and editing. M. Philco: Resources, writing–review and editing. I. Bondarenko: Resources, writing–review and editing. Q. Lian: Software, formal analysis, validation, visualization, writing–original draft, writing–review and editing. D. Bradley: Data curation, writing–review and editing. H. Hinton: writing–review and editing. M.J. Wongchenko: Conceptualization, data curation, formal analysis, supervision, investigation, methodology, writing–review and editing. S.-J. Reilly: Data curation, supervision, writing–original draft, project administration, writing–review and editing. N. Turner: Conceptualization, resources, supervision, methodology, writing–original draft, writing–review and editing.

We thank the patients, their families, the investigators and study staff at participating sites, members of the independent data monitoring committee, and Bruno Kovic (Genentech, Inc., South San Francisco, California) for his work on the analysis of patient-reported outcomes. This work was supported by Genentech/Roche. Medical writing assistance was provided by Jennifer Kelly, MA (Medi-Kelsey Ltd.), funded by F. Hoffmann-La Roche Ltd., Basel, Switzerland. No grant number is applicable.