PD-L1 score among non-small cell lung cancer patients: are we holding out for a hero? Authors: Gilson Gabriel Viana Veloso, Fernanda Telles Lins Taveira, Carlos Gil Ferreira Published: 2026 Background Lung cancer is the most common and deadliest malignancy worldwide (1), and tobacco remains the leading cause of this type of cancer (2). With the advent of precision medicine, the arsenal available for treating non-small cell lung cancer (NSCLC) is more widely than ever (3). Nearly 75% of patients lack actionable driver mutations, and treatment is based on the most common biomarker, the programmed death-ligand 1 (PD-L1), used to predict response to immune checkpoint inhibitors (ICIs) (4). PD-L1 is commonly accessed by immunohistochemistry (IHC) expression, determined by Tumor Proportion Score (TPS), which is the percentage of viable tumor cells showing partial or complete membrane staining at any intensity. The final TPS score usually is divided into three categories: negative (TPS <1%), moderate (TSP =1–49%), and high (TPS ≥50%) (5,6). Recently, the use of liquid biopsies to determine PD-L1 score has gained attention in the oncology field (7). Approximately 20–30% of NSCLC patients have a TPS score <1% (8). Usually, PD-L1 negative patients tend to respond less to ICI than patients with a positive score, yet an important number of subjects have improved, objective rate response (ORR), progression-free survival (PFS) and overall survival (OS) when offered a treatment based on chemotherapy combined with one or more ICIs (9-23) (Table 1). Therefore, the PD-L1 becomes a paradox in Thoracic Oncology, being considered an imperfect biomarker. Table 1Pivotal studies with advanced non-small cell lung cancer and a PD-L1 negative scorePivotal trialYear of first publicationComparative armsAll comersPD-L1 negative5-year follow-up?Median OS (months)HR (95% CI)Median OS (months)HR (95% CI)KEYNOTE-189 (9,17)2018Pembrolizumab + CT vs. CT alone22.0 vs. 10.60.60 (0.50–0.72)17.2 vs. 10.20.59 (0.38–0.92)YesKEYNOTE-407† (10,18)2018Pembrolizumab + CT vs. CT alone17.2 vs. 11.60.71 (0.59–0.85)15.0 vs. 11.00.83 (0.61–1.13)YesIMpower150 (11,19)2018Atezolizumab + bevacizumab + CT vs. atezolizumab + CT vs. bevacizumab + CT19.5 vs. 14.70.80 (0.67–0.95)17.1 vs. 14.10.82 (0.62–1.08)NoIMpower130 (12)2019Atezolizumab + CT vs. CT alone18.6 vs. 13.90.79 (0.64–0.98)15.2 vs. 12.00.81 (0.61–1.08)NoCheckMate 227 (13,20)2019Nivolumab + ipilimumab vs. nivolumab + CT vs. CT aloneNot describedNot described17.4 vs. 12.20.65 (0.52–0.81)YesCheckmate 9LA (14,21)2021Nivolumab + ipilimumab + CT vs. CT alone15.8 vs. 11.00.74 (0.63–0.87)17.7 vs. 9.80.64 (0.49–0.84)YesEMPOWER-Lung 3 (15,22)2022Cemiplimab + CT vs. CT alone21.1 vs. 12.90.66 (0.53–0.83)16.2 vs. 12.41.01 (0.63–1.60)YesPOSEIDON (16,23)2023Durvalumab + tremelimumab + CT vs. durvalumab + CT vs. CT alone14.0 vs. 11.60.76 (0.64–0.89)12.7 vs. 10.90.81 (0.62–1.05)Yes†, squamous histology only. CI, confidence interval; CT, chemotherapy; HR, hazard ratio; OS, overall survival; PD-L1, programmed death-ligand 1. Pivotal studies with advanced non-small cell lung cancer and a PD-L1 negative score †, squamous histology only. CI, confidence interval; CT, chemotherapy; HR, hazard ratio; OS, overall survival; PD-L1, programmed death-ligand 1. The CheckMate 227 overview Unlike CheckMate 9LA, CheckMate 227 is a phase III trial that demonstrates how nivolumab plus ipilimumab without a chemotherapy regimen improved OS among advanced NSCLC patients (13). The pivotal study was first published in 2019 and showed a median OS of 17.1 months in the experimental group vs. 13.9 months with chemotherapy and a rate of OS of 40.1% and 29.7%, respectively, at 2 years (13). Further updates in OS data were made after a 4-year follow-up (24), as well as 5 years (25) and 6 years (20). The latter, with a follow-up of 73.5 months, showed an estimated OS rate was 20% with nivolumab plus ipilimumab vs. 11% with chemotherapy (20). The CheckMate 9LA overview CheckMate 9LA is a Phase III trial that demonstrates how nivolumab (an anti-PD-1 inhibitor) plus ipilimumab (and anti-CTLA-4 inhibitor) combined with two cycles of doublet platinum-based chemotherapy improved OS among advanced NSCLC patients (14). The pivotal study was first published in 2021 with a median follow-up of 13.2 months and yielded a reduced the risk of death by 34% [hazard ratio (HR) 0.66, 95% confidence interval (CI): 0.55–0.80]. The median OS was 15.6 months in the experimental group vs. 10.9 months in the control group (14). Further updates in OS data were made after a 4-year follow-up (26), as well as 5 years (27) and 6 years (21). The latter, with a 68.6-month follow-up, sustaining the OS benefit for double-ICIs against chemotherapy alone (HR 0.74, 95% CI: 0.63–0.87, 6-year OS rates: 16% vs. 10%), irrespective of PD-L1 scores (22). Analysis from the PD-L1 negative subgroup The pooled analysis by Peters et al. (28) presented the long-term efficacy and safety outcomes of nivolumab plus ipilimumab with or without chemotherapy vs. chemotherapy among patients with metastatic NSCLC and tumor PD-L1 lower than 1% from both CheckMate 227 and CheckMate 9LA, including subgroups by baseline brain metastases status and tumor type (squamous or non-squamous NSCLC). Overall, 322 patients with a PD-L1 negative score were analyzed (n=187 from CheckMate 227 and n=135 from CheckMate 9LA). Among these subjects, median age was 64 years, most were males (73%), Eastern Cooperative Oncology Group (ECOG) 1 (64%), current or former smokers (86%), with non-squamous histology (74%). Metastatic sites at baseline included central nervous system (CNS) (11%), liver (20%), and bone (20%) (28). Median OS was 17.4 vs. 11.3 months, respectively (HR 0.64, 95% CI: 0.54–0.76) and 5-year OS rates of 20% vs. 7%, favoring double ICIs against chemotherapy alone (28). Regarding histology, for patients with non-squamous histology, median OS was 17.8 vs. 12.8 months (HR 0.69, 95% CI: 0.57–0.84) favoring double ICIs against chemotherapy alone, while for squamous histology, median OS was 16.8 vs. 8.2 months (HR 0.51, 95% CI: 0.36–0.72) favoring double ICIs (28). The gain in PFS was modest, yet statistically significant: 5.4 vs. 4.9 months for double ICIs and chemotherapy, respectively (HR 0.72, 95% CI: 0.60–0.87) (28). Regarding metastatic CNS at baseline the results are as follows: median OS was 17.4 vs. 9.2 months with nivolumab plus ipilimumab with or without chemotherapy vs. chemotherapy, respectively (HR 0.44, 95% CI: 0.26–0.75), and 5-year OS rates were 6% vs. 0% (28). Finally, the most common treatment related adverse events (TRAEs) were diarrhea (19%), nausea (19%), and fatigue (17%), whereas the most common grade 3 to 4 TRAEs were increased lipase (7%), neutropenia (4%), and diarrhea (3%) (28). Discussion Based on the results of both studies, nivolumab plus ipilimumab is recommended as one potential option for patients with advanced NSCLC in a first-line setting, irrespective of a combined chemotherapy regimen or not. Such regimen among patients with a TPS score below 1% is supported by several guidelines including the National Comprehensive Cancer Network, the American Society of Clinical Oncology, and the European Society for Medical Oncology (28). Although the data showed exciting results for “hard-to-treat” patients, the report can become too simplistic and generic, as no limitations beyond different study design and nivolumab dosage were addressed. For trials with such important data, further questions beyond PD-L1 are needed. Why do these patients have a sustained clinical response and OS after a 6-year follow-up if the main biomarker was negative? Why do these patients have a sustained clinical response and OS after a 6-year follow-up if the main biomarker was negative? What additional molecular features did these patients have, including tumor mutation burden (TMB), and co-mutation statuses such as STK11, KEAP1, KRAS and TP53? The most pressing question remains unanswered: who are the 20% of long-term survivors? Among 322 patients with PD-L1 <1%, approximately 64 were alive at 5 years (28). These tumors were biopsied and clinical data recorded—yet we lack systematic molecular characterization distinguishing survivors from non-responders. Was TMB elevated in responders? Did co-mutations in STK11/LKB1 or KEAP1 differentially impact outcomes? Were baseline immune profiles or tertiary lymphoid structures present in responders but absent in others? Without comprehensive genomic profiling and tumor microenvironment (TME) analysis, we remain unable to prospectively identify which patients will benefit. We treat five patients to save one, exposing four to toxicities without proportional benefit. What additional molecular features did these patients have, including tumor mutation burden (TMB), and co-mutation statuses such as STK11, KEAP1, KRAS and TP53? The most pressing question remains unanswered: who are the 20% of long-term survivors? Among 322 patients with PD-L1 <1%, approximately 64 were alive at 5 years (28). These tumors were biopsied and clinical data recorded—yet we lack systematic molecular characterization distinguishing survivors from non-responders. Was TMB elevated in responders? Did co-mutations in STK11/LKB1 or KEAP1 differentially impact outcomes? Were baseline immune profiles or tertiary lymphoid structures present in responders but absent in others? Without comprehensive genomic profiling and tumor microenvironment (TME) analysis, we remain unable to prospectively identify which patients will benefit. We treat five patients to save one, exposing four to toxicities without proportional benefit. Can these patients be analyzed into oligometastatic vs. polymetastatic at baseline to discuss the OS final data? Can these patients be analyzed into oligometastatic vs. polymetastatic at baseline to discuss the OS final data? Critical is the absence of stratification by metastatic burden. Oligometastatic NSCLC (≤3–5 sites) represents a biologically distinct entity with improved prognosis (29). Could these 20% of survivors be disproportionately oligometastatic patients? Conversely, does polymetastatic disease yield single-digit 5-year survival? The analysis does not report outcomes stratified by number of metastatic sites, organ-specific patterns, or tumor volume—all known to impact immunotherapy efficacy. Without this data, we risk a “one-size-fits-all” approach to a heterogeneous population. TME is becoming more critical than PD-L1 expression in NSCLC management because it provides a more comprehensive, dynamic, and accurate assessment of why patients do or do not respond to immunotherapy. While PD-L1 is a valuable and a pioneer biomarker, it is increasingly seen as insufficient, as patients with low or negative PD-L1 expression can still respond to ICIs, and some high PD-L1 expressors do not (30,31). TMB acts as a distinct, independent biomarker from PD-L1 expression for predicting immunotherapy efficacy in NSCLC, with high TMB identifying responders even among PD-L1-negative cases. Combining TMB with PD-L1 offers superior predictive power, as they reflect different, non-overlapping mechanisms of ICI response (32). Another pitfall for PD-L1 assessment relies on tumor heterogeneity based on biopsy sampling—a tumor can be “negative” in one area but “positive” in another and this could mislead a subjects’ categorization in both clinical trials or clinical practice (33). Overall, PD-L1 “negative” status on a tumor biopsy is not a definitive “no” for immunotherapy; rather, it is a call to action for exploring combination strategies that can overcome immune resistance. While PD-L1 expression is a key biomarker for predicting success, many PD-L1 negative tumors can still respond to immunotherapy when it is combined with other treatment modalities to improve immune cell infiltration and activity (8). Additionally, alternative pathways such as LAG-3 or TIGIT may be active when PD-L1 is absent, and novel combinations are needed to decide the optimal treatment combo is more suitable for each patient (34). To overcome these potential harms, artificial intelligence (AI) has become an important tool to improve assessment of patient’s outcomes. Deep-IO models analyze pre-treatment Hematoxylin and Eosin slides to predict ICIs response and survival. It has been shown to independently predict outcomes better than, or in combination with, PD-L1 scores (35). Finally, unlike Bonnie Tyler’s song, members of the Thoracic Oncology community are no longer waiting for the “PD-L1 hero” by the end of the night and are now focusing how additional tumor features may assist clinical decision making beyond a negative TPS score. Supplementary The article’s supplementary files as