Study Design and Patients
This was a multinational, double-blind, randomized, parallel-group study registered at ClinicalTrials.gov (NCT02364999) and EudraCT (2014-003878-16). Patients were randomized at 159 centers in 27 countries (see Electronic Supplementary Material [ESM], Table S1). Adult patients were eligible for inclusion if they had histologically or cytologically confirmed, predominantly non-squamous, newly diagnosed Stage IIIB or IV NSCLC (according to lung cancer staging criteria of 2010 [22, 23]) or recurrent NSCLC. For patients with recurrent disease, at least 6 months must have elapsed since completing adjuvant or neoadjuvant treatment. Additionally, patients had at least one measurable lesion per Response Evaluation Criteria in Solid Tumors version 1.1 (RECIST 1.1) , an Eastern Cooperative Oncology Group (ECOG) performance status of 0 or 1, and were eligible to receive study treatment of bevacizumab, paclitaxel, and carboplatin based on the local standard of care for the treatment of advanced or metastatic non-squamous NSCLC. Key exclusion criteria included known central nervous system metastases (treated and stable brain metastases were allowed); known sensitizing EGFR mutations or EML4–ALK translocations (patients with unknown status were permitted to enroll); prior systemic therapy for NSCLC (prior neoadjuvant or adjuvant therapy was allowed if surgical resection for primary disease was performed); and prior treatment with immunotherapy or bevacizumab. Full eligibility criteria can be found in the ESM.
The original protocol (dated 4 November 2014) was amended three times; details of the changes can be found in the final version of the protocol (dated 10 June 2016), which is available at ClinicalTrials.gov.
Randomization and Blinding
Patients were enrolled by study investigators and randomized in a 1:1 ratio to receive either PF-06439535 plus paclitaxel and carboplatin or bevacizumab-EU plus paclitaxel and carboplatin for 4–6 cycles, followed by blinded monotherapy with PF-06439535 or bevacizumab-EU as previously assigned (ESM, Fig. S1). The randomization schedule was computer-generated by the sponsor and included the stratification variables of region (location of the drug depot supplying the site), sex (male/female), and smoking history (never/ever). The schedule was concealed from the sponsor’s personnel directly involved in study conduct and was implemented by the study sites using an interactive web response system.
Treatment assignments were blinded to patients, investigators, and the sponsor’s study team. Limited members of the sponsor’s study team were unblinded at the time of the primary efficacy analysis. Site personnel and patients remained blinded until the completion of the study. PF-06439535 and bevacizumab-EU were provided by the sponsor as blinded supplies in which the external packaging for each vial appeared identical and was identified with a unique container number. Paclitaxel and carboplatin were branded products or generic equivalents available in the local region. Use of nab-paclitaxel in place of paclitaxel was not permitted.
On treatment days when PF-06439535 or bevacizumab-EU was administered in combination with chemotherapy, the order of administration was paclitaxel, carboplatin, and PF-06439535 or bevacizumab-EU. Treatments were administered by intravenous infusion on day 1 of each 21-day cycle. Paclitaxel was administered at an initial dose of 200 mg/m2, carboplatin at an initial dose targeting an area under the concentration versus time curve of 6.0 mg/mL·min, and PF-06439535 or bevacizumab-EU at an initial dose of 15 mg/kg. Paclitaxel and carboplatin dose reductions were allowed for toxicity. No dose reductions were planned for PF-06439535 or bevacizumab-EU, but if deemed necessary the investigator could decrease the dose to 7.5 mg/kg after discussion with the sponsor. Patients were pre-medicated before paclitaxel administration in order to prevent severe hypersensitivity reaction.
After chemotherapy had been discontinued, PF-06439535 or bevacizumab-EU monotherapy could be administered until disease progression (defined per RECIST 1.1), unacceptable toxicity, discretion of the investigator, death, withdrawal of consent, or the end of the study, whichever came first. The dose and regimen for PF-06439535 and bevacizumab-EU were chosen to be consistent with the product labeling of bevacizumab-EU. The chemotherapy given and the regimens used were considered standard of care.
Endpoints and Assessments
The primary endpoint was ORR, defined as the percentage of patients within each treatment group who achieved a best overall response (BOR) of complete response (CR) or partial response (PR) by week 19 in accordance with RECIST 1.1, which was subsequently confirmed on a follow-up tumor assessment by week 25. BOR was derived by the sponsor based on tumor measurements reported by the investigator. This endpoint was considered sufficiently sensitive to detect differences in efficacy between PF-06439535 and bevacizumab-EU. The choice of primary endpoint and the use of investigator-reported tumor measurements were agreed with regulatory authorities.
Secondary efficacy endpoints were duration of response (DOR), 1-year PFS rate, and 1-year OS rate from randomization. DOR was calculated only for the subgroup of patients with an objective response achieved by week 19 and confirmed by week 25. Additional secondary endpoints included safety, peak and trough PF-06439535 and bevacizumab-EU concentrations at selected cycles up to 1 year from randomization, and incidence of anti-drug antibodies (ADAs; including neutralizing antibodies [NAbs]) up to 1 year from randomization.
Tumor assessments included computed tomography (CT) or magnetic resonance imaging (MRI) scans of the head, chest, abdomen (including adrenals), and other disease sites such as the pelvis if clinically indicated. CT scans were performed with contrast media unless contraindicated for medical reasons. MRI was only used when considered more appropriate than CT or when there was a contraindication for use of CT with contrast. For a given patient, the same method of tumor assessment was used throughout the trial. CT or MRI assessments were performed every 6 weeks (± 7 days) until week 25 (based on date of randomization). After week 25, assessments were performed every 9 weeks (± 7 days) until 1 year from randomization. Responses were required to be confirmed by a second set of scans obtained 6 weeks (± 7 days) later in accordance with RECIST 1.1. Additionally, brain scans were performed as clinically indicated and at the time of a confirmatory scan for CR/PR. Patients who continued to receive study treatment after 1 year had tumor assessments performed according to local standard of care.
Blood samples for assessment of immunogenicity were collected pre-dose at specified study cycles and analyzed for the presence or absence of ADAs (anti-bevacizumab or anti-PF-06439535 antibodies). A single, sensitive, specific, and semi-quantitative electrochemiluminescent immunoassay was used. The ADA assay had been validated and used biotinylated and ruthenium-labeled PF-06439535 as reagents. Analysis of ADA samples followed a tiered approach of screening, confirmation, and titer determination. Only those samples confirmed positive for ADAs were further tested for NAbs. The NAb analysis was conducted using a single, validated, quasi-quantitative enzyme-linked immunosorbent assay (ELISA) that utilized PF-06439535 as a reagent. Analysis of NAb samples followed a tiered approach of screening and titer determination.
Drug concentrations of PF-06439535 and bevacizumab-EU were determined using serum samples collected at pre-specified time points. Samples were collected pre-dose; in addition, post-dose samples were collected 1 h (± 0.5 h) after the end of infusion on cycle 1, day 1 and cycle 5, day 1. Concentrations were determined using a validated, sensitive, and specific ELISA. Immunogenicity and PK analyses were carried out at QPS, LLC (Newark, DE, USA).
Safety was characterized by the type, incidence, severity, timing, seriousness, and relationship to study therapy of adverse events (AEs), including cardiotoxicity and infusion-related reactions, and laboratory abnormalities. Other safety evaluations included physical examinations, vital signs, and 12-lead electrocardiogram monitoring. The investigator obtained and recorded all observed or volunteered AEs, the severity of the events (based on Common Terminology Criteria for Adverse Events version 4.03), and his or her opinion of the relationship to study treatment. Treatment-emergent AEs (TEAEs) were those that occurred after the beginning of study treatment, or any pre-existing AE that worsened after the beginning of study treatment. AEs of special interest were arterial thromboembolic events, bleeding/hemorrhage (including pulmonary hemorrhage), cardiac disorders, congestive heart failure, hypertension (only grade 3 or higher), proteinuria or nephrotic syndrome, venous thromboembolic events, gastrointestinal perforations, and wound-healing complications.
After discontinuation from treatment, survival status was collected by telephone contact every 2 months (± 14 days) until death or 1 year from patient randomization. The study was considered complete when the last available patient completed up to 1 year from randomization plus a 28-day follow-up period.
The primary efficacy analysis for the primary endpoint was based on the Miettinen and Nurminen method  without stratification variables and was carried out in the intent-to-treat (ITT) population, defined as all patients who were randomized to study treatment. The estimated ORR risk ratio and ORR risk difference between the PF-06439535 and bevacizumab-EU groups were computed, along with asymptotic two-sided 95% and 90% confidence intervals (CIs). Equivalence was determined based on the following criteria agreed with regulatory authorities. For the FDA, equivalence was considered established if the 90% CI of the ORR risk ratio fell within the margin of 0.73–1.37. For Japan’s Pharmaceuticals and Medical Devices Agency (PMDA), equivalence was considered established if the 95% CI of the ORR risk ratio fell within the margin of 0.729–1.371. Finally, for the European Medicines Agency (EMA), equivalence was considered established if the 95% CI of the ORR risk difference fell within the margin of − 13% to 13%.
The equivalence margins above were derived based on meta-analysis of three randomized studies of reference bevacizumab plus paclitaxel and carboplatin versus paclitaxel and carboplatin alone in patients with NSCLC [8, 26, 27]. The ORR for reference bevacizumab plus chemotherapy was estimated to be approximately 40%, and the ORR for chemotherapy alone was estimated to be 21%. The ORR risk ratio for bevacizumab plus chemotherapy versus chemotherapy alone was 2.17 (95% CI 1.74–2.70). The margin of 0.73–1.37 in the current study maintained 43% of the effect size estimated from the historical ORR data using a log scale, and approximately 50% using a linear scale. Assuming an ORR of 38% in both treatment arms, a sample of 656 patients (328 per treatment arm) would provide approximately 85% power for achieving equivalence according to the FDA criteria above. Considering a possible ~ 7.5% attrition rate for patients reaching evaluation for ORR, a total sample of approximately 710 patients (355 per treatment arm) was to be randomized. Using the EMA and PMDA equivalence criteria, and again assuming an ORR of 38%, the planned sample size would provide power of approximately 86% and 74%, respectively.
As a sensitivity analysis, the Miettinen and Nurminen method  was also conducted without stratification variables using the per-protocol (PP) population, defined as all patients who were randomized and received PF-06439535 or bevacizumab-EU as planned and had no major protocol deviations. The list of patients not included in the PP population and the reasons for exclusion were determined prior to unblinding for the primary efficacy analysis. As secondary analyses in both the ITT and PP populations, the Miettinen and Nurminen method  was repeated with additional stratification variables (region, sex, and smoking history) to assess whether these would affect the ORR risk ratio or risk difference.
Time-to-event endpoints were assessed using the Kaplan–Meier method. A Cox proportional hazard model was used to estimate hazard ratios (HRs); the model included treatment and the covariates of region, sex, and smoking history. The two treatment groups were compared using a 2-sided log-rank test stratified by region, sex, and smoking history.
For the ORR endpoint, if a patient had a missing tumor outcome across all visits or had non-evaluable BOR per RECIST 1.1, he or she was considered a non-responder and was included in the denominator, but not the numerator. For the time-to-event endpoints, missing data were censored.
The safety population, defined as all patients who were randomized and received at least one dose of study treatment, was used for safety and immunogenicity analyses. AEs were summarized by body system and preferred term according to the Medical Dictionary for Regulatory Activities (version 20.1) classification system. The percentage of patients with positive ADA and NAb results was summarized for each treatment and each visit. Patients in the PP population who had at least one drug concentration measurement post-administration of treatment were included in the PK analysis. The drug concentration–time data were summarized by descriptive statistics according to treatment.
The primary efficacy analysis for statistical equivalence was performed when all patients had completed the week 25 visit to support the primary endpoint analysis or had discontinued from the disease evaluation period earlier. All other analyses presented herein are based on final data after study completion. All statistical analyses were conducted as specified in the statistical analysis plan, which can be accessed at ClinicalTrials.gov.