Introduction

Globally, gastric cancer is one of the most common and deadly cancers, with more than one million new cases diagnosed annually [1]. Many patients present with advanced-stage disease, for which second-line treatment options include single-agent chemotherapy with a taxane or irinotecan and the antivascular endothelial growth factor receptor 2 antibody ramucirumab, alone or combined with paclitaxel [2]. Pembrolizumab is a selective, humanized monoclonal antibody against programmed death 1 (PD-1) that prevents interaction between PD-1 and its ligands, PD-L1 and PD-L2 [3], and that has demonstrated antitumor activity and manageable safety in patients with advanced gastric or gastroesophageal (GEJ) cancer across multiple lines of therapy [4,5,6,7,8]. Based on data from cohort 1 of the KEYNOTE-059 study [9], pembrolizumab was approved in the United States for the treatment of patients with recurrent locally advanced or metastatic advanced gastric/GEJ adenocarcinoma expressing PD-L1 (combined positive score [CPS] ≥ 1) that progressed on at least two previous lines of therapy [3].

KEYNOTE-061 was a randomized, open-label, phase 3 trial of pembrolizumab compared with paclitaxel for previously treated advanced gastric/GEJ cancer [7]. In patients with CPS ≥ 1 tumors (data cutoff date: October 26, 2017), pembrolizumab did not significantly improve overall survival (OS) compared with paclitaxel (hazard ratio [HR], 0.82; 95% CI 0.66–1.03; one-sided P = 0.0421) or progression-free survival (PFS; HR, 1.27; 95% CI 1.03–1.57). Duration of response (DOR) was substantially longer with pembrolizumab than with paclitaxel (median, 18.0 vs 5.2 months), and pembrolizumab demonstrated a better safety profile than paclitaxel [7]. Herein we present results from KEYNOTE-061 based on two additional years of follow-up.

Methods

Study design

The study design for KEYNOTE-061 has been reported [7]. In brief, eligible patients had histologically or cytologically confirmed adenocarcinoma of the stomach or GEJ that was metastatic or locally advanced but unresectable, disease progression per Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1 after first-line therapy with a platinum and fluoropyrimidine, and Eastern Cooperative Oncology Group performance status (ECOG PS) 0 or 1. Histology was evaluated by investigator. Patients were randomly allocated 1:1 to intravenous pembrolizumab 200 mg every 3 weeks for up to 2 years or paclitaxel 80 mg/m2 on days 1, 8, and 15 of each 4-week cycle or until disease progression, intolerable toxicity, physician decision, or patient withdrawal of consent. Randomization was stratified according to geographic region (Europe, Israel, North America, and Australia vs Asia vs rest of the world), time to progression on first-line therapy (< 6 months vs ≥ 6 months), and PD-L1 expression status (CPS < 1 vs ≥ 1). After 489 patients were enrolled, the independent data monitoring committee recommended that enrollment be restricted to patients with CPS ≥ 1 tumors on the basis of outcomes in patients with CPS < 1 tumors [7]. Consequently, all final 103 patients had CPS ≥ 1 tumors.

PD-L1 expression was assessed in archival or newly collected tumor samples at a central laboratory using PD-L1 IHC 22C3 pharmDx (Agilent) and measured using the CPS, defined as the number of PD-L1–staining cells (tumor cells, lymphocytes, macrophages) as a proportion of the total number of viable tumor cells, multiplied by 100.

The study protocol and all amendments were approved by the institutional review board or ethics committee at each institution. The study was conducted in accordance with the protocol and its amendments and Good Clinical Practice guidelines. All patients provided written informed consent before enrollment.

Outcomes

The primary objectives of this analysis were OS (defined as the time from randomization to death from any cause) and PFS (defined as the time from randomization to radiologic disease progression assessed per RECIST v1.1 by masked and independent central review or death from any cause) in the population with CPS ≥ 1 tumors.

Additional exploratory objectives included OS and PFS in the populations with CPS ≥ 5 and CPS ≥ 10 tumors; response rate (defined as the proportion of patients with complete response [CR] or partial response [PR]) and DOR (defined as the time from first documented CR or PR to radiologic disease progression or death from any cause), both assessed per RECIST v1.1 by masked and independent central review and by investigator assessment in the populations with CPS ≥ 1, CPS ≥ 5, and CPS ≥ 10 tumors; and safety in all patients, irrespective of CPS.

Statistical analysis

The analyses of the intention-to-treat population and the PD-L1 CPS ≥ 1 population were prespecified, whereas the analyses of the CPS ≥ 5 and CPS ≥ 10 subgroups were post hoc.

OS, PFS, and response rate were analyzed in the intention-to-treat population, defined as all patients who were randomly allocated to treatment, irrespective of whether they received the treatment. DOR was analyzed in all patients whose best response was CR or PR. Safety was assessed in all patients who received at least one dose of study treatment.

SAS version 9.4 (SAS Institute) was used for all statistical analyses. OS, PFS, and DOR were estimated using the Kaplan–Meier method. HRs and their associated 95% CIs were calculated using stratified Cox proportional hazards models with Efron’s method of tie handling. Kaplan–Meier analysis of OS was also analyzed in the protocol-specified subgroup of ECOG PS 0 or 1 based on the pembrolizumab treatment effect previously observed [7].

This trial is registered with ClinicalTrials.gov, NCT02370498.

Results

The time from randomization to the data cutoff date of October 7, 2019, was 4 years and 4 months. At the time of analysis, 18 of 194 patients (9.3%) in the CPS ≥ 1 population completed 2 years of treatment with pembrolizumab, and the remaining 176 of 194 patients (90.7%) discontinued before the 2-year limit; all paclitaxel-treated patients had already discontinued treatment at the time of the protocol-specified analysis (Fig. 1). Baseline demographics and disease characteristics were generally balanced between treatment groups in the total population and in the CPS ≥ 1 population (Table 1). Baseline characteristics for the CPS < 1, CPS ≥ 5, and CPS ≥ 10 populations are reported in Online Resource 1; the prevalences of most characteristics for each population were comparable to those of the total population.

Fig. 1
figure 1

Patient disposition. AE adverse event, CPS combined positive score, CR complete response, PD progressive disease. aDefined as clinical progression or progressive disease. bThere was no maximum number of doses of paclitaxel

Full size image
Table 1 Baseline characteristics in the overall and the PD-L1 CPS ≥ 1 intention-to-treat populations
Full size table

At the time of data cutoff, 366 patients in the CPS ≥ 1 population had died (176/196 [89.8%] in the pembrolizumab group and 190/199 [95.5%] in the paclitaxel group); median OS was 9.1 months (95% CI 6.2–10.7) for pembrolizumab and 8.3 months (95% CI 7.6–9.0) for paclitaxel (HR for death, 0.81; 95% CI 0.66–1.00) (Fig. 2a). The 24-month OS rates were 19.9% for pembrolizumab and 8.5% for paclitaxel. In the CPS ≥ 5 population, 170 of 186 patients (91.4%) had died (84/95 [88.4%] in the pembrolizumab group and 86/91 [94.5%] in the paclitaxel group); median OS was 10.4 months (95% CI 6.7–15.5) for pembrolizumab and 8.3 months (95% CI 6.8–9.4) for paclitaxel (HR for death, 0.72; 95% CI 0.53–0.99) (Fig. 2b). The 24-month OS rates were 24.2% for pembrolizumab and 8.8% for paclitaxel. In the CPS ≥ 10 population, 95 of 108 patients (88.0%) had died (44/53 [83.0%] in the pembrolizumab group and 51/55 [92.7%] in the paclitaxel group); median OS was 10.4 months (95% CI 5.9–18.3) for pembrolizumab and 8.0 months (95% CI 5.1–9.9) for paclitaxel (HR for death, 0.69; 95% CI 0.46–1.05) (Fig. 2c). The 24-month OS rates were 32.1% for pembrolizumab and 10.9% for paclitaxel.

Fig. 2
figure 2

Kaplan–Meier analysis of overall survival in the populations with (a) CPS ≥ 1, (b) CPS ≥ 5, and (c) CPS ≥ 10 tumors. CPS combined positive score, HR hazard ratio, Pts patients

Full size image

In the CPS ≥ 1 population, OS favored pembrolizumab across all subgroups, with the exception of diffuse histology (Fig. 3a). We also performed a Kaplan–Meier analysis of OS based on ECOG PS. Among patients with CPS ≥ 1 tumors and ECOG PS 0, median OS was 12.3 months (95% CI 9.7–15.9) for pembrolizumab and 9.3 months (95% CI 8.3–10.5) for paclitaxel (HR for death, 0.69; 95% CI 0.50–0.95) (Fig. 3b); the 24-month OS rates were 23.9% and 10.9%, respectively. Among patients with CPS ≥ 1 tumors and ECOG PS 1, median OS was 5.4 months (95% CI 3.7–7.7) for pembrolizumab and 7.5 months (95% CI 5.3–8.4) for paclitaxel (HR for death, 0.98; 95% CI 0.74–1.31) (Fig. 3c); the 24-month OS rates were 16.7% and 6.6%, respectively. Factors affecting the treatment benefit in the overall population were explored with a similar subgroup analysis, shown in Online Resource 2.

Fig. 3
figure 3

Overall survival analysis by (a) subgroups and Kaplan–Meier analysis in the population with CPS ≥ 1 tumors and ECOG PS (b) 0 or (c) 1. CPS combined positive score, ECOG PS Eastern Cooperative Oncology Group performance status, GEJ gastroesophageal junction, HR hazard ratio, Pts patients

Full size image

In the population with CPS ≥ 1 tumors, 377 of 395 patients (95.4%) experienced disease progression or died (185/196 [94.4%] in the pembrolizumab group and 192/199 [96.5%] in the paclitaxel group); median PFS was 1.5 months (95% CI 1.4–2.0) for pembrolizumab and 4.1 months (95% CI 3.2–4.3) for paclitaxel (HR for disease progression or death, 1.25; 95% CI 1.02–1.54) (Fig. 4a). In the population with CPS ≥ 5 tumors, 174 of 186 patients (93.5%) experienced disease progression or died (87/95 [91.6%] in the pembrolizumab group and 87/91 [95.6%] in the paclitaxel group); median PFS was 1.6 months (95% CI 1.4–2.8) for pembrolizumab and 4.0 months (95% CI 2.8–4.4) for paclitaxel (HR for disease progression or death, 0.98; 95% CI 0.71–1.34) (Fig. 4b). In the population with CPS ≥ 10 tumors, 97 of 108 patients (89.8%) experienced disease progression or died (45/53 [84.9%] in the pembrolizumab group and 52/55 [94.5%] in the paclitaxel group); median PFS was 2.7 months (95% CI 1.4–4.3) for pembrolizumab and 4.0 months (95% CI 2.7–4.4) for paclitaxel (HR for disease progression or death, 0.79; 95% CI 0.51–1.21) (Fig. 4c).

Fig. 4
figure 4

Kaplan–Meier analysis of progression-free survival in the populations with (a) CPS ≥ 1, (b) CPS ≥ 5, and (c) CPS ≥ 10 tumors. CPS combined positive score, HR hazard ratio, Pts patients

Full size image

In the population with CPS ≥ 1 tumors, confirmed responses were observed in 32 of 196 patients in the pembrolizumab group (response rate, 16.3%) and in 27 of 199 patients in the paclitaxel group (response rate, 13.6%); CRs were observed in nine patients (4.6%) and five patients (2.5%), respectively (Table 2). In the population with CPS ≥ 5 tumors, confirmed responses were observed in 19 of 95 patients in the pembrolizumab group (response rate, 20.0%) and in 13 of 91 patients in the paclitaxel group (response rate, 14.3%); CRs were observed in seven patients (7.4%) and two patients (2.2%), respectively (Table 2). In the population with CPS ≥ 10 tumors, confirmed responses were observed in 13 of 53 patients in the pembrolizumab group (response rate, 24.5%) and in 5 of 55 patients in the paclitaxel group (response rate, 9.1%); CRs were observed in seven patients (13.2%) and one patient (1.8%), respectively (Table 2). Median DOR was longer in the pembrolizumab group than in the paclitaxel group, regardless of CPS status; median DOR in the pembrolizumab group increased with increasing PD-L1 enrichment (CPS ≥ 1, 19.1 months; CPS ≥ 5, 32.7 months; CPS ≥ 10, not reached) (Table 2). More than 60% of responders in the pembrolizumab group had responses lasting ≥ 12 months (Table 2).

Table 2 Response by PD-L1 combined positive score
Full size table

Treatment-related adverse events (AEs) occurred in 157 of 294 patients (53.4%) treated with pembrolizumab and 233 of 276 patients (84.4%) treated with paclitaxel (Table 3); grade 3–5 treatment-related AEs were reported in 44 of 294 patients (15.0%) and 97 of 276 patients (35.1%), respectively. The most common grade 3–5 treatment-related AEs (≥ 2% in either group) were fatigue and anemia in the pembrolizumab group and decreased neutrophil count, anemia, fatigue, neutropenia, decreased white blood cell count, and peripheral neuropathy in the paclitaxel group. Four patients died of treatment-related AEs (pembrolizumab, n = 3; paclitaxel, n = 1).

Table 3 Adverse events in the overall as-treated population
Full size table

Immune-mediated AEs and infusion reactions occurred in 55 of 294 patients (18.7%) treated with pembrolizumab and 21 of 276 patients (7.6%) treated with paclitaxel (Table 3). Grade 3–5 immune-mediated AEs occurring in two or more patients treated with pembrolizumab were hepatitis (n = 4), hypophysitis (n = 2), and pneumonitis (n = 2).

Discussion

After approximately two additional years of follow-up in previously treated patients with gastric/GEJ cancer, the results of this long-term analysis from KEYNOTE-061 were consistent with those of the primary analysis [7]. However, second-line pembrolizumab did not significantly improve OS compared with paclitaxel at the primary analysis or after two additional years of follow-up. The Kaplan–Meier curve confirmed the beneficial treatment effects observed at 24 months in the primary analysis [7] after two additional years, with higher 24-month OS rates with pembrolizumab than with paclitaxel (19.9% vs 8.5%). Of note, the difference in 24-month OS rates between pembrolizumab and paclitaxel groups increased as the PD-L1 CPS cutoff increased (CPS ≥ 5, + 15.4%; CPS ≥ 10, + 21.3%), suggesting the utility of CPS to enrich for patients likely to benefit from pembrolizumab. Additionally, response rates were numerically higher with pembrolizumab, as indicated by two additional patients achieving CR (n = 9) compared with the primary analysis (n = 7). Durable responses were also observed after almost 4 years, some of which were ongoing at the data cutoff date. Of significance, the benefit of pembrolizumab in PFS and ORR also incrementally increased with PD-L1 enrichment. The safety profile for pembrolizumab remained consistent with that of the primary analysis and with that previously observed in other monotherapy trials, and no new safety signals were observed with long-term follow-up [7].

The observation that selection of patients by CPS enriched the long-term efficacy of pembrolizumab is consistent with previous clinical data in gastroesophageal cancer [10,11,12]. In the phase 3 CheckMate-649 study in patients with previously untreated, unresectable, non–HER2-positive gastric, GEJ, or esophageal adenocarcinoma, nivolumab plus chemotherapy resulted in significant improvements in OS and PFS vs chemotherapy alone in patients with PD-L1 CPS ≥ 5 (primary endpoint), patients with PD-L1 CPS ≥ 1, and all randomly assigned patients, and the treatment effect was more pronounced in the CPS ≥ 5 population. However, these studies, including KEYNOTE-061, were not powered to assess treatment effect specifically in the PD-L1 CPS < 1 population. In addition, during KEYNOTE-061, the independent data monitoring committee recommended that enrollment be restricted to patients with CPS ≥ 1 tumors, resulting in a small enrollment of patients with PD-L1 CPS < 1. With the available evidence, PD-L1 CPS remains valuable in deciding treatment strategies.

In the subgroup analysis of OS in patients with CPS ≥ 1 tumors, pembrolizumab-treated patients with ECOG PS 0 had a numerically longer 24-month OS rate than patients with ECOG PS 1 (23.9% vs 16.7%, respectively). These findings are also consistent with data reported at the primary analysis [7] and emphasize the need for further exploration in these patients.

Long-term data from immune checkpoint inhibitors in the second-line setting in patients with gastric/GEJ cancer are limited. In the phase 3 ATTRACTION-2 study, patients received third line or later nivolumab for advanced gastric/GEJ cancer [13]. After a median follow-up of 27 months, a higher 2-year OS rate was observed with nivolumab (10.6%) than with placebo (3.2%) in patients previously treated with at least two chemotherapy regimens; these findings were irrespective of PD-L1 status (assessed retrospectively on tumor cells using the 28–8 pharmDx assay). Although cross-trial comparisons should be interpreted with caution based on different patient populations and treatment lines, both the KEYNOTE-061 (more globally distributed in the second-line treatment setting) and the ATTRACTION-2 (predominantly Asian population [Japanese, South Korean, Taiwanese] in the third line or later treatment setting) studies suggested long-term OS benefits of anti–PD-1 therapy for patients with gastric/GEJ cancer.

As previously discussed [7], limitations of the study include its open-label design. As a result, there was an imbalance in the number of patients who were randomly allocated but who did not receive study treatment in the paclitaxel group compared with the pembrolizumab group. Consequently, patients in the paclitaxel group likely received other therapies, and this could have affected the study results and impacted the relative benefit of pembrolizumab vs paclitaxel. Subsequent therapy between the two treatment groups was also likely to be different, which could have affected the OS outcomes reported here. Furthermore, although the treatment groups were well balanced at baseline, the exclusion of patients whose tumors expressed CPS < 1 after 83% of patients were enrolled and the change in stratification factors after 21% of patients were enrolled might have introduced bias that affected the results [7].

Our findings suggest the potential for an increased treatment benefit with pembrolizumab monotherapy in patients with PD-L1 CPS ≥ 5 and CPS ≥ 10 tumors and in patients with better ECOG PS. Additionally, the safety profile of pembrolizumab remained favorable, showing fewer treatment-related AEs compared with paclitaxel. Taken together, these long-term data add insight to the existing body of evidence and support further exploration of pembrolizumab as monotherapy and as part of combination therapy in other gastric cancer settings.