The FDA’s Oncologic Drug Advisory Committee voted against the use of checkpoint inhibitors in first-line advanced gastric cancer with PD-L1 expression less than 1.
In a 2 to 10 vote, the FDA’s Oncologic Drug Advisory Committee (ODAC) voted that the risk-benefit assessment is not favorable for the use of checkpoint inhibitors (CPIs) in first-line advanced HER2-negative gastric and gastroesophageal junction (GEJ) adenocarcinoma in patients with PD-L1 expression less than 1.1
The ODAC discussed this question through data supporting the current approvals of nivolumab (Opdivo) and pembrolizumab (Keytruda) plus chemotherapy, as well as the application of tislelizumab-jsgr (Tevimbra) plus chemotherapy, for this indication.
“I agree based on the totality of the data that there was not a favorable risk-benefit [profile] for this PD–L1-low population,” said ODAC member Neil Vasan, MD, PhD, Columbia University Medical Center. “For me, it was clear that there was benefit in the [PD-L1 expression] greater than 10 and no benefit [with PD-L1 expression less than 1.] It is that intermediate range where we need clinical trials to help answer questions where we have levels of equipoise as with any continuous variable.”
Introductory remarks were delivered by Steven Lemery, MD, MHS, director of the Division of Oncology 3, Office of Oncologic Diseases. He presented the goals of the meeting as considering the class labeling of CPIs, maximizing risk-benefit balance, fostering consistent treatment, and creating a consistent drug developmental landscape. Lemery noted that while PD-L1 is a predictive biomarker, it is an imperfect one.
Based on the studies evaluating CPIs presented to the ODAC during the meeting, CPIs demonstrated an overall survival (OS) benefit in the first line, with a smaller treatment effect observed in PD–L1-low groups. Additionally, subgroup analyses of CPI studies were not always powered to demonstrate a treatment effect.
Lemery summarized that in patients with PD-L1 expression greater than 10, there is a clear benefit with CPI treatment. Among patients with an intermediate PD-L1 expression of 1 to 10, there is a more modest effect and inconsistent benefit. For patients with a negative PD-L1 expression of less than 1, there is a consistent lack of benefit observed.
He also noted that 80% to 90% of patients have a positive PD-L1 expression, meaning that a potential PD-L1 cutoff of 1 would allow most patients to receive treatment with CPIs.
The notable risks of CPIs in all patients, regardless of PD-L1 expression, are immune-mediated adverse effects (IMAEs). Lemery noted that grade 3 or higher IMAEs were typically observed in 3% to 11% of patients. IMAEs could also become chronic and often manifested as endocrine, lung, neurologic, or arthritic conditions. Steroids, the most common line of treatment for these AEs, could compromise patient quality of life.
Lemery proposed 2 potential approaches for how to move forward with the use of CPIs. The first was product-specific PD-L1 expression cutoffs. The pros of this approach would be that there would be statistical rationale based on clinical trial results, and product-specific cutoffs are typically used in recommendations from professional organizations, including the National Comprehensive Cancer Network (NCCN) and the American Society of Clinical Oncology (ASCO). However, variable PD-L1 tests in practice, inconsistent treatment in practice, and inconsistent future drug development presented challenges to this approach. Additionally, as PD-L1 is an imperfect biomarker, the prespecified cut points from clinical trials may not be the optimal choice.
The second approach presented was a pragmatic or uniform PD-L1 cutoff. The pros of this approach were that it considered the totality of data to foster consistent patient management and a consistent approach to future clinical trials. However, this option may also be less statistically sound and presented challenges with consistency of PD-L1 testing devices, which could restrict access of CPIs to patients.
Vaibhav Kumar, MD, MS, clinical reviewer, Division of Oncology 3, Office of Oncologic Diseases, expanded on the FDA perspective. He began with an explanation of 3 ways of performing PD-L1 scoring: tumor proportion score (TPS), combined positive score (CPS), and tumor area positivity (TAP).
Kumar also presented professional organization guidelines from the ASCO and the NCCN. In its recommendations, which were published before the results of KEYNOTE-859 (NCT03675737), ASCO calls for the use nivolumab plus chemotherapy to be assessed on a case-by-case basis in patients with a CPS between 1 and 5. The NCCN considers the nivolumab regimen preferable for patients with a CPS greater than 5 and useful in certain circumstances with a CPS under 5. The pembrolizumab regimen is preferred with a CPS greater than 10 and useful under certain circumstances with a CPS between 1 and 10.
Kumar emphasized that while safety does not differ across PD-L1 score strata, it would not be in patients’ best interest to expose them to a risk without a clear benefit.
Robert Anders, MD, PhD, Division of Gastrointestinal and Liver Pathology at The Johns Hopkins University, and Yelena Janjigian, MD, chief attending of gastrointestinal oncology at Memorial Sloan Kettering Cancer Center, discussed the challenges of PD-L1 testing. Anders highlighted that PD-L1 testing, due to interobserver variability, is a subjective test, and PD-L1 expression varies as a function of time. These issues could possibly present false negatives.
Janjigian explained that immunotherapy is only approved in the first line for gastric cancer in the US, and the response of first-line treatment is crucial for long-term patient outcomes. Based on electronic health record data from Flatiron Health, only 77% of patients had PD-L1 testing prior to the start of first-line therapy, and 50% of patients were treated with immunotherapy in the first line, signaling that CPIs are not being overused in this setting.
Echoing Anders, Janjigian also noted some additional practical limitations of PD-L1 testing. These included difficulty of obtaining sufficient samples, variability in PD-L1 testing in clinical practice, and the challenge of distinguishing intermediate CPS values.
Nivolumab in addition with chemotherapy was approved by the FDA on April 16, 2021, for the first-line treatment of gastric/GEJ cancer without restriction by PD-L1 expression. This was based on
Nivolumab plus chemotherapy led to statistically significant and clinically meaningful improvements in progression-free survival (PFS) and OS compared with chemotherapy and placebo. In patients with a CPS of greater than 5, the PFS HR was 0.68 (95% CI, 0.56-0.81) and the OS HR was 0.71 (98.4% CI, 0.59-0.86). In patients with a CPS of 1 or greater, the OS HR was 0.77 (99.3% CI, 0.64-0.92). The OS HR in all randomized patients was 0.80 (99.3% CI, 0.68-0.94).
In all treated patients, 59% in the nivolumab arm and 44% in the placebo arm experienced grade 3 to 4 treatment-related AEs (TRAEs). According to the 3-year follow-up data, most TRAEs with potential immunologic etiology were grades 1 or 2.2 At follow-ups of 2, 3, and 4 years, the safety profile of the combination in all treated patients was consistent with what was observed in the primary analysis.1
Pembrolizumab plus chemotherapy was
In patients with a CPS greater than 1, the OS was 13.0 months (range, 11.6-14.2) with pembrolizumab vs 11.4 months (range, 10.5-12.0) with placebo (HR, 0.74; 0.65-0.84; P <.0001). The OS was 15.7 months (range, 13.8-19.3) with pembrolizumab vs 11.8 months (range, 10.3-12.7) with placebo in patients with a CPS greater than 10 (HR, 0.65; 0.53-0.79; P <.0001).
Regarding safety, the most common grade 3 to 5 AEs of any cause were anemia, with an incidence of 12% in the pembrolizumab arm vs 10% in the placebo arm, and decreased neutrophil count, occurring in 10% of the pembrolizumab arm vs 8% in the placebo arm. Further, the safety profile was consistent across PD-L1 subgroups.
Cathy Pietanza, MD, vice president of clinical research and global clinical development, late-stage oncology at Merck, stated in her presentation that the KEYNOTE-859 data supported pembrolizumab’s indication in this population and acknowledged that PD-L1 testing outside of clinical trials is variable.
Pietanza emphasized that labeling given by the FDA reflects what is discovered in phase 3 trials. Post hoc subgroup analyses are considered exploratory to assess directional consistency of treatment effects, and using subgroup analyses at various cut points that are not rigorously tested could lead to cofounding findings. Further, Pietanza added that pooled analyses to inform product labeling assumes identical efficacies for all CPIs and identical patient populations.
“PD-L1 expression is a continuum. It can be modulated by other therapies like chemotherapy and is not always predictive of immunotherapy response,” said Pietanza.
Pooja Bhagia, MD, executive director of global clinical development, late-stage oncology at Merck, further discussed efficacy and safety from KEYNOTE-859. She highlighted that PFS and OS were directionally consistent, regardless of PD-L1 cut points. She also noted that risks were consistent across all patients, regardless of PD-L1 expression.
“There is no biological rationale to suggest that the safety profile of pembrolizumab would change based on PD-L1 expression,” added Bhagia.
The
At an interim analysis, the median OS in patients with a PD-L1 TAP greater than 5 was 17.2 months with tislelizumab and chemotherapy vs 12.6 months with placebo (HR, 0.74; 95% CI, 0.59-0.94; P =.006). At a final analysis, the median OS among all randomized patients was 15.0 months with tislelizumab vs 12.9 months (HR, 0.80; 95% CI, 0.70-0.92; P =.001).
Mark Lanasa, MD, PhD, senior vice president and chief medical officer of solid tumors at BeiGene, tislelizumab’s sponsor, highlighted that there was no connection between PD-L1 score and IMAEs. The incidence of IMAEs in the overall population was 31% with tislelizumab vs 12% with placebo. In patients with PD-L1 expression greater than 5, the rate of IMAEs was 29% with tislelizumab vs 13% with placebo.
The discussion question presented was if cumulative data supported the use of PD-L1 expression as a predictive biomarker when selecting patients for treatment with CPIs.
“Patients do not want to see incremental benefits and overall survival. They want to see that tail of the curve and see those durable responses,” said Christopher Lieu, MD, University of Colorado, and acting ODAC chairperson.
“I think there are also some significant challenges here. I see a biomarker that is not binary. This is measured on a gradient, and I think that there is no standardization,” Lieu continued.
“A predictive biomarker at its core is simply [that] there is a different relative effect size by biomarker status,” said ODAC member Daniel Spratt, MD, University Hospitals Seidman Cancer Center, Case Western Reserve University.“There are ways to test this. There are significant interaction effects by subgroup [with] different relative effect sizes.”
“Just because the patient has an amazing response to chemo[therapy and immunotherapy] does not mean they would not have an amazing response to chemo[therapy]. With a hazard ratio of almost 1, they probably had with similar probability of having that benefit,” Spratt added.
“[For scores between 1 and 10], I think that gets into the realm of letting doctors be doctors where they have an opportunity to talk to their patients and make a decision in conjunction with their patient about whether this is worth doing, taking into account the [adverse] effect profile,” said ODAC member William J. Gradishar, MD, Feinberg School of Medicine at Northwestern University.
“I am not particularly impressed with any of the data for the groups [with PD-L1 expression] less than 1,” Gradishar added.
The voting question presented was if a risk-benefit assessment was favorable for the use of CPIs for this indication.
Two members voted that the risk-benefit profile was favorable, while 10 voted that it was not. One member, Ravi Madan, MD, National Cancer Institute, voted to abstain.
“I think our quest for biomarkers has been going on since our quest to develop better therapeutics. I think, unfortunately, most biomarkers fall short. I think PD-L1 has also fallen short in many diseases, including this one because of issues with acquisition, characterization, variability, and sampling error. When it comes to that context, it is hard for me to say that this is the data set to make this decision,” said Madan.
