Nivolumab plus chemotherapy or ipilimumab in gastro-oesophageal cancer

Standard first-line chemotherapy results in disease progression and death within one year in most patients with human epidermal growth factor receptor 2 (HER2)-negative gastro-oesophageal adenocarcinoma1–4. Nivolumab plus chemotherapy demonstrated superior overall survival versus chemotherapy at 12-month follow-up in gastric, gastro-oesophageal junction or oesophageal adenocarcinoma in the randomized, global CheckMate 649 phase 3 trial5 (programmed death ligand-1 (PD-L1) combined positive score ≥5 and all randomized patients). On the basis of these results, nivolumab plus chemotherapy is now approved as a first-line treatment for these patients in many countries6. Nivolumab and the cytotoxic T-lymphocyte antigen-4 (CTLA-4) inhibitor ipilimumab have distinct but complementary mechanisms of action that contribute to the restoration of anti-tumour T-cell function and induction of de novo anti-tumour T-cell responses, respectively7–11. Treatment combining 1 mg kg−1 nivolumab with 3 mg kg−1 ipilimumab demonstrated clinically meaningful anti-tumour activity with a manageable safety profile in heavily pre-treated patients with advanced gastro-oesophageal cancer12. Here we report both long-term follow-up results comparing nivolumab plus chemotherapy versus chemotherapy alone and the first results comparing nivolumab plus ipilimumab versus chemotherapy alone from CheckMate 649. After the 24.0-month minimum follow-up, nivolumab plus chemotherapy continued to demonstrate improvement in overall survival versus chemotherapy alone in patients with PD-L1 combined positive score ≥5 (hazard ratio 0.70; 95% confidence interval 0.61, 0.81) and all randomized patients (hazard ratio 0.79; 95% confidence interval 0.71, 0.88). Overall survival in patients with PD-L1 combined positive score ≥ 5 for nivolumab plus ipilimumab versus chemotherapy alone did not meet the prespecified boundary for significance. No new safety signals were identified. Our results support the continued use of nivolumab plus chemotherapy as standard first-line treatment for advanced gastro-oesophageal adenocarcinoma.

Standard first-line chemotherapy results in disease progression and death within one year in most patients with human epidermal growth factor receptor 2 (HER2)negative gastro-oesophageal adenocarcinoma [1][2][3][4] . Nivolumab plus chemotherapy demonstrated superior overall survival versus chemotherapy at 12-month follow-up in gastric, gastro-oesophageal junction or oesophageal adenocarcinoma in the randomized, global CheckMate 649 phase 3 trial 5 (programmed death ligand-1 (PD-L1) combined positive score ≥5 and all randomized patients). On the basis of these results, nivolumab plus chemotherapy is now approved as a first-line treatment for these patients in many countries 6 . Nivolumab and the cytotoxic T-lymphocyte antigen-4 (CTLA-4) inhibitor ipilimumab have distinct but complementary mechanisms of action that contribute to the restoration of anti-tumour T-cell function and induction of de novo anti-tumour T-cell responses, respectively [7][8][9][10][11] . Treatment combining 1 mg kg −1 nivolumab with 3 mg kg −1 ipilimumab demonstrated clinically meaningful anti-tumour activity with a manageable safety profile in heavily pre-treated patients with advanced gastro-oesophageal cancer 12 . Here we report both long-term follow-up results comparing nivolumab plus chemotherapy versus chemotherapy alone and the first results comparing nivolumab plus ipilimumab versus chemotherapy alone from CheckMate 649. After the 24.0-month minimum follow-up, nivolumab plus chemotherapy continued to demonstrate improvement in overall survival versus chemotherapy alone in patients with PD-L1 combined positive score ≥5 (hazard ratio 0.70; 95% confidence interval 0.61, 0.81) and all randomized patients (hazard ratio 0.79; 95% confidence interval 0.71, 0.88). Overall survival in patients with PD-L1 combined positive score ≥ 5 for nivolumab plus ipilimumab versus chemotherapy alone did not meet the prespecified boundary for significance. No new safety signals were identified. Our results support the continued use of nivolumab plus chemotherapy as standard first-line treatment for advanced gastrooesophageal adenocarcinoma.
We enrolled 3,185 patients, 2,031 of whom were randomized across the 3 treatment groups; of these, 1,581 patients were concurrently randomized to nivolumab plus chemotherapy or chemotherapy (April 2017 to May 2019) and 813 to nivolumab plus ipilimumab or chemotherapy (October 2016 to June 2018). Enrolment to the nivolumab-plus-ipilimumab group was closed early owing to increased rate of adverse events and early deaths relative to the other two study groups, per recommendation from the data monitoring committee. Among randomized patients, the number of patients who received one or more dose of study treatment and those that were discontinued at the data cut-off date for the current analysis (27 May 2021) are shown in Extended Data Fig. 1. The primary reason for treatment discontinuation was disease progression (Extended Data Fig. 1).
Objective responses with nivolumab plus chemotherapy were observed in 226 (60%; 95% confidence interval 55, 65) of 378 patients with PD-L1 CPS ≥ 5 compared with 176 (45%; 95% confidence interval 40, 50) of 390 patients with chemotherapy. In all randomized patients, objective responses were observed in 350 (58%; 95% confidence interval 54, 62) of 603 patients with nivolumab plus chemotherapy versus 279 (46%; 95% confidence interval 42, 50) of 607 patients with chemotherapy (Extended Data Table 1). Additional complete responses were observed with nivolumab plus chemotherapy compared with the prespecified interim analysis at 12-month follow-up (PD-L1 CPS ≥ 5, n = 5 and all randomized, n = 6); there were no additional complete responses with chemotherapy alone. The total number of complete      Table 1). The percentage of patients with PD-L1 CPS ≥ 5 who had more than 50% tumour burden reduction was 53% with nivolumab plus chemotherapy and 44% with chemotherapy; the percentage of patients with more than 80% reduction was 27% and 18%, respectively, with consistent results in all randomized patients (Extended Data Fig. 4).

Subgroup analyses
The hazard ratios for overall survival continued to favour nivolumab plus chemotherapy versus chemotherapy across multiple prespecified subgroups in patients with PD-L1 CPS ≥ 5 and all randomized patients with longer follow-up (Extended Data Figs Fig. 7c, d).

Subsequent therapy
Subsequent therapy was received by 41% and 44% of randomized patients in the nivolumab-plus-chemotherapy and chemotherapy groups and by 48% and 46% of patients in the nivolumab-plus-ipilimumab and chemotherapy groups. The most common subsequent treatment was chemotherapy (36% and 39% in the nivolumab-plus-chemotherapy and chemotherapy groups and 44% and 41% in the nivolumab-plus-ipilimumab and chemotherapy groups). Subsequent immunotherapy was received by 2% and 9% of patients in the nivolumab-plus-chemotherapy and chemotherapy groups and by 3% and 12% of patients in the nivolumab-plus-ipilimumab and chemotherapy groups (Supplementary Information).

Patient-reported outcomes
Since the hierarchically tested secondary endpoint of overall survival with nivolumab plus ipilimumab versus chemotherapy in patients with PD-L1 CPS ≥ 5 was not met, the secondary endpoint of time to symptom deterioration (TTSD) in patients with PD-L1 CPS ≥ 5 and all randomized patients was not statistically tested. An improvement from baseline in the Functional Assessment of Cancer Therapy-Gastric (FACT-Ga) questionnaire total score was observed at all on-treatment assessments ( Supplementary Information). The least-squares mean difference between treatment groups favoured nivolumab plus chemotherapy versus chemotherapy alone (at timepoints with ≥10 patients in each group); however, these differences did not reach the threshold for meaningful change (prespecified as 15.1 points; Extended Data Fig. 9a, b). The proportion of patients who reported not being bothered by treatment side effects over time on the basis of the GP5 item from FACT-Ga was higher with nivolumab plus chemotherapy than with chemotherapy alone, except at baseline when patients had not received treatment (Extended Data Fig. 9c, d).

Discussion
Several targeted and immuno-oncology agents have been evaluated as first-line treatment for HER2-negative gastric or GEJ cancer; however, until recently, none have significantly prolonged survival relative to chemotherapy 1 There is considerable variation in the previously reported prevalence of PD-L1 CPS expression in gastro-oesophageal adenocarcinoma, with PD-L1 CPS ≥ 5 detected in 17-60% of patients [14][15][16][17] . CheckMate 649 is the most robust dataset to date to report PD-L1 CPS ≥ 5 prevalence using an analytically validated assay (28-8 pharmDx) in gastric, GEJ or oesophageal adenocarcinoma. The phase 3 ORIENT-16 trial in China reported a similar PD-L1 CPS ≥ 5 prevalence of approximately 60% using the 22C3 PharmDx assay 17,18 . In CheckMate 649, the magnitude of survival benefit continued to be enriched with nivolumab plus chemotherapy versus chemotherapy in patients with higher PD-L1 CPS, consistent with results at 12-month follow-up 5 . However, in patients with PD-L1 CPS ≥ 5 and ≥10, hazard ratios for overall survival were rather close to each other, with overlapping confidence intervals, and ORR benefit was similar, suggesting no meaningful further enrichment of clinical benefit at or above PD-L1 CPS 10. Furthermore, the ORRs observed with nivolumab plus chemotherapy were higher versus chemotherapy across all evaluated PD-L1 CPS cut-offs, suggesting that clinical benefit with nivolumab plus chemotherapy is not restricted to patients with PD-L1 CPS ≥ 5. Further analyses may help identify factors that are associated with higher magnitude of clinical benefit in patients with lower PD-L1 CPS.
Overall survival continued to favour nivolumab plus chemotherapy versus chemotherapy across multiple prespecified baseline characteristics with longer follow-up. Notably, the magnitude of survival benefit was markedly greater in the MSI-H subgroup for both patients with PD-L1 CPS ≥ 5 and all randomized patients, suggesting that benefit is independent of PD-L1 CPS. Similar results were reported with first-line pembrolizumab plus chemotherapy versus chemotherapy in patients with gastric or GEJ cancer (PD-L1 CPS ≥ 1) who had MSI-H tumours 13 . In CheckMate 649, the overall survival benefit in patients with MSS tumours was similar to that observed in all randomized patients. The secondary endpoint of overall survival with nivolumab plus ipilimumab versus chemotherapy in CheckMate 649 did not meet the prespecified boundary for significance in patients with PD-L1 CPS ≥ 5. The observed response rates with nivolumab plus ipilimumab were lower versus chemotherapy, and there was no enrichment with increasing PD-L1 CPS cut-offs. However, the median duration of response almost doubled with nivolumab plus ipilimumab versus chemotherapy, which is consistent with results in other solid tumours with this combination [19][20][21][22] .
The lack of significant overall survival improvement with nivolumab plus ipilimumab is probably a result of multiple factors. There was an increase in early death rate with nivolumab plus ipilimumab versus chemotherapy; crossing of the Kaplan-Meier curves, which is a known phenomenon with immuno-oncology therapies 13,23,24 , was observed at 12 months, and the overall survival curves remained separated thereafter in favour of nivolumab plus ipilimumab. A higher number of patients receiving subsequent immuno-oncology therapy in the chemotherapy versus nivolumab-plus-ipilimumab group (12% versus 3%, respectively) may have also contributed to these results.
Tumours in gastric, GEJ or oesophageal adenocarcinoma are composed of distinct molecular subtypes 25,26 . Although dual checkpoint inhibition has been proven to be effective in multiple solid tumours [19][20][21][22]27 , further research is needed to evaluate how tumour biology, molecular heterogeneity, dynamics in tumour microenvironment and other patient factors may affect the efficacy of combined PD-L1 and CTLA-4 blockade. Notably, in the small but relevant subgroup of patients with microsatellite instability, which is characterized by high tumour mutational burden and CD8-positive T-cell infiltrates and is susceptible to immune-checkpoint inhibition 25,28-30 , longer overall survival and higher ORR were observed with nivolumab plus ipilimumab versus chemotherapy in CheckMate 649. These data suggest that combined immune checkpoint blockade in this patient population might be of interest to explore in future studies.
No new safety signals were identified with nivolumab plus chemotherapy in CheckMate 649 with similar frequencies of TRAEs relative to the 12-month follow-up. The safety profile of nivolumab plus ipilimumab observed in this trial was consistent with the known safety profile of this combination 12,27 . Limitations of this study have been previously discussed 5 .
In conclusion, the long-term clinically meaningful overall survival and PFS benefit, improved and durable responses, maintained health-related quality of life, and acceptable safety profile indicate a favourable benefit-risk profile of nivolumab plus chemotherapy. These results further support the use of this regimen as a standard first-line treatment in previously untreated patients with advanced gastric, GEJ or oesophageal adenocarcinoma.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-04508-4.

Patients
Adults with unresectable advanced or metastatic gastric, GEJ or oesophageal adenocarcinoma were enrolled, regardless of PD-L1 expression. Patients with known HER2-positive status were excluded, and prior systemic therapy for metastatic disease was not allowed. Other key inclusion criteria were an Eastern Cooperative Oncology Group performance status score of 0 or 1 and the ability to provide a fresh or archival tumour sample to determine PD-L1 status. Additional details on study criteria have been previously described 5 .

Trial design and treatments
CheckMate 649 (NCT02872116) is a randomized, open-label, multicentre, global phase 3 trial of nivolumab plus chemotherapy or ipilimumab versus chemotherapy alone, conducted at 175 hospitals and cancer centres in 29 countries across Asia, Australia, Europe, North America, and South America. Detailed study design and methods for the nivolumab-plus-chemotherapy versus chemotherapy groups have been previously described 5 . In brief, patients were initially randomized 1:1 to nivolumab plus ipilimumab or to chemotherapy from October 2016 to March 2017. The nivolumab-plus-chemotherapy group was added later, and the randomization was switched to 1:1:1 in March 2017. Enrolment to the nivolumab-plus-ipilimumab group was closed early in June 2018, and after this time, the randomization was switched to a 1:1 ratio of nivolumab plus chemotherapy versus chemotherapy to May 2019. Patients already randomized to nivolumab plus ipilimumab could continue treatment per protocol, but the data remained blinded until the pre-planned final analysis. During enrolment, the population for primary endpoints was amended to patients whose tumours expressed PD-L1 CPS ≥ 5 for the nivolumab-plus-chemotherapy versus chemotherapy groups, although patients continued to be enrolled regardless of PD-L1 expression. Additional randomization procedures and stratification by tumour cell PD-L1 status (≥1% versus <1% including indeterminate), region (Asia versus United States and Canada versus rest of world), Eastern Cooperative Oncology Group performance status score (0 versus 1) and type of chemotherapy (CapeOX versus FOLFOX) have been described 5 .
Patients were administered nivolumab (360 mg every 3 weeks or 240 mg every 2 weeks) with investigator's choice of chemotherapy (CapeOX (oxaliplatin 130 mg m −2 on day 1 and capecitabine 1,000 mg m −2 orally twice daily on days 1-14) every 3 weeks or FOL-FOX (leucovorin 400 mg m −2 on day 1, fluorouracil 400 mg m −2 on day 1 and 1,200 mg m −2 on days 1-2, and oxaliplatin 85 mg m −2 on day 1) every 2 weeks); nivolumab (1 mg kg −1 ) with ipilimumab (3 mg kg −1 ) every 3 weeks for 4 cycles, followed by nivolumab (240 mg every 2 weeks); or chemotherapy alone. The dosing for nivolumab 1 mg kg −1 plus ipilimumab 3 mg kg −1 was selected based on results of the Check-Mate 032 study, where this regimen provided numerically higher ORR and longer median overall survival compared with nivolumab monotherapy or nivolumab 3 mg kg −1 plus ipilimumab 1 mg kg −1 , along with a manageable safety profile in heavily pre-treated patients with advanced gastro-oesophageal adenocarcinoma 12 . Treatment was permitted until documented disease progression, unacceptable toxicity, withdrawal of consent or trial end. Nivolumab or ipilimumab were given for a maximum of two years. Patients receiving nivolumab in combination with chemotherapy or ipilimumab were permitted to continue treatment beyond initial disease progression (per Response Evaluation Criteria in Solid Tumors (RECIST), version 1.1), based on the investigator's judgement, until subsequent progression. Dose reductions were not permitted for nivolumab and ipilimumab; dose reductions for chemotherapy were permitted per local standards. Dose delays were allowed for both groups to manage treatment-related toxicity. Nivolumab, ipilimumab, CapeOX and FOLFOX were provided by the sponsor except in certain countries where CapeOX and FOLFOX were procured commercially if allowed by local regulations. Additional details on discontinuation criteria have been previously described 5 .
The trial was conducted according to Good Clinical Practice guidelines developed by the International Council for Harmonisation and in compliance with the trial protocol (Supplementary Appendix). The trial protocol was approved by the institutional review boards or independent ethics committees at each site (NCT02872116). All patients provided written informed consent prior to trial participation per Declaration of Helsinki principles.

Endpoints and assessments
The dual primary endpoints were overall survival (time from randomization to death) and PFS (time from randomization to the date of the first documented tumour progression (by blinded independent central review (BICR) per RECIST, version 1.1) or death) in the nivolumab-plus-chemotherapy versus chemotherapy groups 5 in patients with PD-L1 CPS ≥ 5. Secondary endpoints that were hierarchically tested if the primary endpoints were met were overall survival in patients with PD-L1 CPS ≥ 1 and in all randomized patients in the nivolumab-plus-chemotherapy versus chemotherapy group and overall survival and TTSD in patients with PD-L1 CPS ≥ 5 and in all randomized patients in the nivolumab-plus-ipilimumab versus chemotherapy group. Other key secondary endpoints that were not formally tested included BICR-assessed PFS and ORR evaluated at different PD-L1 CPS cut-offs and in all randomized patients. Key exploratory endpoints included BICR-assessed duration of response; landmark survival rates; PFS2 (time from randomization to progression after subsequent systemic therapy, initiation of second subsequent systemic therapy, or death, whichever is earlier); biomarkers potentially predictive of efficacy; health-related quality of life; and safety and tolerability.
Tumours were assessed using computed tomography or magnetic resonance imaging per RECIST, version 1.1, at baseline, every 6 weeks from the start of cycle 1 for 48 weeks and every 12 weeks thereafter, until disease progression per BICR assessment. Adverse events were assessed throughout the treatment period and during follow-up according to the National Cancer Institute Common Terminology Criteria for Adverse Events, version 4.0.

PRO analyses
FACT-Ga analysis was done for patients with PD-L1 CPS ≥ 5 and all randomly assigned patients who had an assessment at baseline (day 1, assessment before administration of treatment on day of first dose) and at least one subsequent assessment while on treatment. The questionnaire completion rate, defined as the proportion of questionnaires actually received out of the expected number, was calculated and summarized at each assessment point using descriptive statistics. Mean score and mean change from baseline for the FACT-Ga scale were estimated using mixed model for repeated measures. The change from baseline was modelled as a linear function of treatment groups; trial assessment; baseline score; trial stratification factors; interaction terms between treatment group and trial assessment; interaction terms between baseline score and trial assessment; and any potential confounders. A clinically meaningful difference was defined as a 15.1 or greater change from baseline in FACT-Ga total score 31 . The P-value for the difference in least squares means was computed as the two-tailed probability using the t distribution. No adjustments were made for multiple comparisons. In addition, treatment burden was assessed by the individual GP5 item of the FACT-Ga. The GP5 item reads, "I am bothered by side effects of treatment." Frequencies and percentages of the GP5 item question responses ('not at all', 'a little bit', 'somewhat', 'quite a bit' and 'very much') were tabulated at each assessment point with ten or more study subjects in each group.

Statistical analyses
Patients concurrently randomized to the nivolumab-plus-chemotherapy versus chemotherapy groups and the nivolumab-plus-ipilimumab versus chemotherapy groups were included in the respective final overall survival analyses. For the comparison of nivolumab plus chemotherapy and chemotherapy, patients randomized to chemotherapy before the nivolumab-plus-chemotherapy arm was introduced were not included in the analysis. For the comparison of nivolumab plus ipilimumab and chemotherapy, patients randomized to chemotherapy after the closure of nivolumab-plus-ipilimumab arm were not included in the analysis.
For nivolumab plus ipilimumab versus chemotherapy, the analysis of overall survival was pre-planned at a minimum follow-up of approximately 36 months, which corresponded with the pre-planned final analysis of overall survival for nivolumab plus chemotherapy versus chemotherapy at a 24-month minimum follow-up. Since the dual primary endpoints for nivolumab-plus-chemotherapy versus chemotherapy groups were met 5 , the secondary endpoint of overall survival in the nivolumab-plus-ipilimumab versus chemotherapy groups was hierarchically tested in patients with PD-L1 CPS ≥ 5 followed by all randomized patients. If overall survival in the nivolumab-plus-ipilimumab versus chemotherapy groups met the criteria for statistical significance, the secondary endpoint of TTSD in the nivolumab-plus-ipilimumab versus chemotherapy groups was planned to be hierarchically tested in patients with PD-L1 CPS ≥ 5 followed by all randomized patients. In the interim analysis, two-sided alpha levels of 0.02 and 0.03 (type I error) were allocated to the dual primary endpoints of PFS and overall survival, respectively. The comparison of secondary endpoints of overall survival and TTSD for nivolumab plus ipilimumab versus chemotherapy inherited alpha independently from the two primary endpoints (fraction of α transmitted = 0.035) and was tested once after 36 months in patients with PD-L1 CPS ≥ 5 followed by all randomized patients.
The statistical power estimation for the comparison of primary endpoints for the nivolumab-plus-chemotherapy versus the chemotherapy groups has been described previously 5 . Sample size calculations of the primary endpoints were based on simulations in East software, version 6.4.1. (Cytel). The prevalence of patients with PD-L1 CPS ≥ 5 was assumed to be 35% of all randomized patients, based on limited available data 14,24,32 , with 285 patients estimated in the nivolumab-plus-ipilimumab versus chemotherapy analysis. Based on new information from the CheckMate 649 trial, this PD-L1 CPS ≥ 5 prevalence was revised to 60% of all randomized patients, with 489 patients estimated in the nivolumab-plus-ipilimumab versus chemotherapy analysis. For overall survival, the hazard ratio was modelled as a four-piece hazard ratio with an average of 0.7. With 36-month minimum follow-up, it was expected that the 411 events would provide 93% power.
Median PFS, overall survival, and duration of response were estimated using Kaplan-Meier methods, and the corresponding two-sided 95% confidence intervals were calculated using the log-log transformation method. The stratified Cox proportional hazards regression model, with the randomization factors as the stratification factors and treatment group as a single covariate, was used to assess differences between treatment groups in overall survival and PFS. An O'Brien and Fleming α-spending function was employed to determine the hazard ratio for overall survival, using a stratified Cox proportional hazards model. Stratification factors recorded in an interactive web response system were used in the analysis.
The proportion of patients with an objective response and corresponding two-sided 95% confidence intervals were calculated using the Clopper-Pearson method.
Statistical analyses were performed using SAS software, version 9.4 (SAS Institute, Cary, NC).

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability
The Bristol Myers Squibb data sharing policy (https://www.bms. com/researchers-and-partners/independent-research/data-sharingrequest-process.html) is compliant with ICMJE guidelines. Bristol Myers Squibb will honour legitimate requests for clinical trial data from qualified researchers. Data will be shared with external researchers whose proposed use of the data has been approved. Complete de-identified patient data sets will be eligible for sharing 2 years after completion of the CheckMate 649 study. Before data are released, the researcher(s) must sign a Data Sharing Agreement, after which the de-identified and anonymized datasets can be accessed within a secured portal.