Utility of various programmed death-ligand 1 (PD-L1) assays in predicting the clinical benefit of immune checkpoint inhibitors in PD-L1-expressing non-small-cell lung cancer patients: a systematic review.

Introduction
Immune checkpoint inhibitors (ICIs) targeting the programmed death ligand 1 (PD-L1)/programmed death-1 pathway restore tumor-specific immunity and have improved survival of patients with advanced non-small cell lung cancer (NSCLC). ICIs, either as monotherapy or in combination with chemotherapy, are now standard of care for patients with advanced NSCLC in both first-line and subsequent lines of treatment. Although ICIs can provide clinical benefit across PD-L1 expression levels, higher PD-L1 expression has been more consistently associated with greater benefit in landmark trials. Clinical benefit, in terms of overall survival (OS) and progression-free survival (PFS), has been associated with PD-L1 expression for many ICIs in several landmark trials, including the IMpower110, IMpower150, and OAK trials for atezolizumab (1-3); KEYNOTE-010, KEYNOTE-024, KEYNOTE-042, and KEYNOTE-189 for pembrolizumab (4-8); CheckMate-227 for nivolumab (9); and the PACIFIC study for durvalumab (10). Based on these findings, tumor cell (TC) PD-L1 expression is now a prerequisite for prescribing most ICIs for NSCLC in the first-line setting with pembrolizumab monotherapy or atezolizumab monotherapy for PD-L1 ≥50% (11-14). For second-line treatment, the requirement for pembrolizumab monotherapy is PD-L1 ≥1%, while atezolizumab monotherapy or nivolumab monotherapy may be prescribed for any level of PD-L1 expression (11,12,14).
The determination of PD-L1 expression was validated in clinical trials for each drug using a companion diagnostic (CDx) originally developed alongside each specific ICI. The CDx for pembrolizumab, nivolumab, atezolizumab, durvalumab, and avelumab are PD-L1 immunohistochemical (IHC) assays 22C3 pharmDx (22C3) (15), PD-L1 IHC 28-8 pharmDx (28-8) (9), Ventana PD-L1 SP142 (SP142) (2), Ventana PD-L1 SP263 (SP263) (10), and PD-L1 IHC 73-10 pharmDx (16,17), respectively. Regulatory bodies like the Food and Drug Administration (FDA) and European Medicines Agency (EMA) have approved these CDx assays for their corresponding ICIs; however, assay availability varies globally, with regions relying on single platforms or laboratory-developed tests (18,19). Subsequently, several studies have examined the interchangeability of these PD-L1 assays and concordance of PD-L1 staining. The Blueprint phase 2 study of 81 lung cancer specimens of various histological and sample types used five trial-validated PD-L1 assays and examined the PD-L1 staining of TCs and tumor-infiltrating immune cells (ICs) (20). In TC staining, the 22C3, 28-8, and SP263 assays demonstrated concordant results, whereas the SP142 assay identified fewer positive cases and the 73-10 assay detected a higher number of positives (20). Additionally, the SP263 assay identified slightly more positive cases than the 22C3 and 28-8 assays (20). Another study also examined the concordance of staining between 22C3, SP263, and 28-8 on 493 formalin-fixed paraffin-embedded NSCLC samples, across a range of PD-L1 expression cut-offs (1%, 10%, 25%, and 50% of tumor membrane staining) (21). A concordance of >90% was found between all three assays across all the levels of expression (21), providing strong evidence for assay interchangeability at the level of PD-L1 assessment in lung cancer. Despite these results, there is a lack of direct comparative data on PD-L1 assays in ICI-treated NSCLC patients.
Despite strong analytical concordance among PD-L1 assays, the ability of alternative (non-companion) assays to predict clinical benefit in corresponding ICI-treated NSCLC populations has not been confirmed. In clinical practice, specific assays for PD-L1 staining may not always be available, in turn limiting the choice of ICIs for patients. The use of distinct clones, detection instruments, and scoring algorithms in each assay has raised concerns regarding the interchangeability of test results between different assays. A better understanding of concordance between assays will help inform physicians on a broader range of appropriate assays for selection when prescribing PD-L1/PD-1 inhibitors. This systematic literature review specifically aimed to assess whether PD-L1 expression results obtained from ‘alternative’ (non-CDx) assays, can predict treatment outcomes in patients receiving ICI monotherapy for advanced/metastatic NSCLC. Demonstrating such predictive value could support wider clinical use of these assays and provide clinicians with greater flexibility in selecting ICIs based on locally available testing platforms. We present this article in accordance with the PRISMA reporting checklist (22) (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-361/rc).

Methods

Literature search methods
To identify publications that assessed the clinical benefits of ICI monotherapy in patients with NSCLC who were tested for PD-L1 status using an alternative assay, we conducted a literature search in PubMed (including MEDLINE) and Google Scholar on December 20, 2023, with no limits on year or publication date. We limited our search to publications in the English language. The specific search strategy is shown in Table S1. In brief, the search terms ‘NSCLC’ and ‘PD-L1 inhibitor’ versus ‘NSCLC’ and ‘22C3, SP142, SP263, or 28-8’ were used. Additionally, reference lists of included studies were searched to avoid missing potential studies. All searched literature was imported, and duplicates were removed.

Eligibility criteria
We included clinical trials (phases I to IV) and real-world studies (including observational studies). Further inclusion criteria specified only studies that provided the PD-L1 expression status of patients with NSCLC, involved those treated with ICI monotherapy, and reported associated clinical outcomes. We included only studies that used at least one non-CDx assay (henceforth termed ‘alternative’ assay, defined as an assay not specifically developed together with the ICI) to assess PD-L1 status. The use of the CDx was allowed but not required. Unpublished studies were excluded. The outcomes included objective response rate (ORR), PFS, and OS. The details of the key inclusion and exclusion criteria are provided in Table S2. Two researchers independently reviewed the titles, abstracts, and full texts of potential citations to select included studies. Any disagreement between the two assessors was resolved by a third researcher.

Data abstraction
Two researchers independently extracted the necessary information from included studies using a data extraction form in Excel. The extracted data included citation, location, study design, ICI, alternative assays used, patient characteristics [age, sex, disease stage, histology, performance status, line of treatment, smoking status, presence of liver metastasis, epidermal growth factor receptor (EGFR)/V-Ki-ras2 Kirsten rat sarcoma viral oncogene homologue (KRAS) mutations, and anaplastic lymphoma kinase (ALK) rearrangements], and outcome results (ORR, PFS, and OS).

Results
The initial search identified 2,077 records after removing duplicates. Following title and abstract screening, we assessed the eligibility of 845 potentially relevant full-text reports (involving an ICI as a treatment for lung cancer). After full-text screening, four studies in four reports [study identifiers: Gadgeel et al. 2022 (23); Herbst et al. 2020 (1); Tamiya et al. 2020 (24); Fujimoto et al. 2018 (25)] were included in this review and underwent data extraction (Figure 1).
Of the four studies (comprising a total of 1,364 patients with evaluable PD-L1 data from at least one alternative assay), two were open-label, phase III randomized controlled trials (RCTs) [Gadgeel et al. 2022 (23); Herbst et al. 2020 (1)], and two were retrospective, real-world cohort studies [Tamiya et al. 2020 (24); Fujimoto et al. 2018 (25)] (Table 1). Only one study (Herbst et al. 2020) included patients with NSCLC undergoing first-line therapy; all other studies included patients undergoing later lines of therapy. Patients in all studies had stage IIIB–IV disease, and data were included for stage, histology, patient age, sex, Eastern Cooperative Oncology Group (ECOG) status, smoking history, and oncogene status (ALK, EGFR, and KRAS). Atezolizumab was used in both the phase III RCTs, and nivolumab was used in the two retrospective studies; no studies using pembrolizumab or durvalumab met the eligibility criteria and were therefore excluded. All four studies used the CDx along with one or more alternative assays and reported PFS as an outcome (Table 2). OS data were not available in one study (Fujimoto et al. 2018), while another (Tamiya et al. 2020) did not have data on ORR.
The two RCTs included in this review compared atezolizumab with chemotherapy. Gadgeel et al. 2022 (OAK trial, NCT02008227) analyzed data from the previously treated patients with metastatic NSCLC to evaluate whether SP142 (CDx) and 22C3 (alternative) PD-L1 assays could predict the clinical outcome of atezolizumab treatment versus docetaxel (23). The study prospectively assessed archival or fresh tumor samples (blocks or formalin-fixed paraffin-embedded slides) for PD-L1 expression on TC and tumor-infiltrating ICs using SP142. Retrospective 22C3 staining was conducted on freshly cut tissues or tissue sections <6 months old, stored under suitable conditions, and utilizing a tumor proportion score (TPS). Among 577 patients in the 22C3 biomarker-evaluable population (22C3-BEP), TC1/2/3 or IC1/2/3 on SP142 had a 60% inter-assay concordance with TPS ≥1% on 22C3, while TC3 or IC3 had a 64% inter-assay concordance with TPS ≥50%. In the same population (22C3-BEP), atezolizumab improved OS compared with docetaxel across all SP142 and 22C3 PD-L1 subgroups (median 9.9–20.4 vs. 7–9 months). OS benefits from atezolizumab were highest in the SP142-defined PD-L1-high [TC3 or IC3: hazard ratio (HR), 0.39; 95% confidence interval (CI): 0.25–0.63] and 22C3-defined PD-L1-high (TPS ≥50%: HR, 0.56; 95% CI: 0.38–0.82) and low in 22C3-defined (TPS 1% to <50%: HR, 0.55; 95% CI: 0.37–0.82) subgroups. PFS improved with increasing PD-L1 expression for both assays. Notably, both assays also identified single-positive patients with non-overlapping PD-L1 expression. These results pertain particularly to the PD-L1 high subgroup. Among the broader PD-L1 positive population, an OS benefit was observed in both 22C3-uniquely positive and double-negative groups, while no significant benefit was noted in the SP142-uniquely positive group (HR, 0.90; 95% CI: 0.62–1.29).
Herbst et al. 2020 (1) (IMpower110, NCT02409342), assessed PD-L1 expression using SP142 (CDx), 22C3, and SP263 assays in patients with EGFR and ALK wild-type tumors (N=554). SP142 identified 554 patients with ≥1% PD-L1 expression (TC1/2/3 or IC1/2/3). The selected patients had any level of PD-L1 expression assessed by SP142 assay. A central laboratory performed IHC analyses on archival tumor tissue or biopsy tissue that was obtained during patient screening. Among 534 patients evaluated using 22C3 assay, 78% had TPS ≥1%, and 49% had TPS ≥50%. Of the 546 patients with evaluable SP263 results, 77% had TC ≥1%, and 54% had TC ≥50%. High overlap was observed between the TPS ≥50% subgroup on the 22C3 assay and the same TC ≥50% subgroup on the SP263 assay. Among patients with high PD-L1 expression by SP142 (TC3 or IC3), the OS was significantly longer in the atezolizumab group than in the chemotherapy group (median 20.2 vs. 13.1 months; HR, 0.59; 95% CI: 0.40–0.89; p=0.01). Patients with TPS ≥50% via the 22C3 assay had a median OS of 20.2 months for atezolizumab and 11.0 months for chemotherapy (HR, 0.60; 95% CI: 0.42–0.86). Those with TC ≥50% via the SP263 assay had median OS values of 19.5 months and 16.1 months, respectively (HR, 0.71; 95% CI: 0.50–1.00). In conclusion, both RCTs provide clinical evidence demonstrating the effective selection of responders to atezolizumab using alternative PD-L1 assays (22C3 and SP263) beyond the CDx SP142. Patients with high PD-L1 expression as identified through 22C3, SP263, and SP142 assays were the subgroup deriving the most clinical benefit from atezolizumab.
Tamiya et al. 2020 (24) (retrospective study) included 201 patients treated with nivolumab monotherapy. The percentage of PD-L1-positive cells in the overall tumor sections was estimated in increments of 5%, except for 1%. The pathologists were blinded to the clinical data. PD-L1 expression was evaluated using 22C3 (alternative) and 28-8 (CDx), which showed that 17.4% and 14.4% had TPS 1–49%, 11.9% and 14.9% had TPS ≥50%, and 36.3% and 36.8% were PD-L1-negative, respectively. The median OS of the entire cohort was 12.27 months. Multivariate Cox hazards model of survival found that compared with TPS <50%, TPS ≥50% via the 22C3 assay did not significantly predict OS (HR, 0.90; 95% CI: 0.40–2.02; p=0.30). Survival outcomes stratified by PD-L1 level using 28-8 were not reported.
Fujimoto et al. 2018 (retrospective cohort study), analyzed data from 40 patients with previously treated advanced NSCLC who received nivolumab (25). Samples were anonymized and scored independently by the pathologists who were blinded to the clinical data. The study found good concordance of PD-L1-stained TCs between 28-8, 22C3, and SP263 assays (weighted κ coefficient: 0.64–0.71), while SP142 showed lower concordance with other assays (weighted κ coefficient: 0.39–0.55). PFS was significantly longer in patients with highly positive PD-L1 staining (TPS ≥50%) than in patients with negative staining (TPS <1%) using 28-8 (median not reached vs. 2.1 months; p=0.005), 22C3 (median not reached vs. 2.0 months; p=0.01), and SP263 assays (median not reached vs. 2.0 months; p=0.048). In contrast, PFS was not significantly different between patients with TC ≥50% and patients with TC <1% by SP142. Using receiver operating characteristic analysis, Fujimoto et al. 2018 showed that 28-8, 22C3, and SP263 had equivalent predictive performance for response to nivolumab [area under the curve (AUC) 0.75–0.82] (26) whereas the SP142 assay showed lower predictive performance (AUC 0.68).

Discussion
ICIs, the standard for PD-L1-positive advanced/metastatic NSCLC, are each associated with a specific PD-L1 CDx, with varying scoring methods reflecting on TCs or both TC and tumor-infiltrating IC expression (27).
Antibody epitope mapping study of SP263, SP142, 22C3, 28-8, and E1L3N showed distinct differences in binding sites, but concluded that discordances likely result from assay protocols and tumor heterogeneity, not from antibody specificity (28). A meta-analysis similarly suggested that properly designed assays may perform similarly to the CDx (29). However, further clinical studies are needed to confirm the predictive value for ICIs’ clinical outcomes in NSCLC patients expressing PD-L1 using alternative assays (29). A systematic review of the literature was conducted to address this evidence gap.
Our review found that for atezolizumab monotherapy-treated patients with advanced/metastatic NSCLC, two RCTs showed clinical benefit regardless of the PD-L1 assay used, suggesting similar predictive ability to the CDx. Limited evidence exists for nivolumab-treated patients, with two smaller retrospective studies showing inconsistent predictive value, particularly with 22C3. Clinical data supporting alternative assay use for pembrolizumab or durvalumab in NSCLC patients are lacking despite analytical concordance data for PD-L1 staining (20). In addition to atezolizumab and pembrolizumab, cemiplimab has demonstrated significant clinical benefit in patients with advanced NSCLC and high PD-L1 expression. In the phase III EMPOWER-Lung 1 trial, cemiplimab monotherapy significantly improved median OS compared to platinum-doublet chemotherapy in patients with PD-L1 expression ≥50% (median OS not reached vs. 14.2 months; HR, 0.57; 95% CI: 0.42–0.77) (30). The benefit was even more pronounced in patients with very high PD-L1 expression (≥90%), with a median OS not reached vs. 11.7 months in the chemotherapy arm (HR, 0.37; 95% CI: 0.23–0.58). Median PFS was also longer with cemiplimab (8.2 vs. 5.7 months; HR, 0.54; 95% CI: 0.39–0.73). These findings further support the predictive value of PD-L1 expression levels in identifying patients most likely to benefit from ICI monotherapy and underscore the importance of accurate and reliable PD-L1 testing in clinical decision-making.
Gadgeel et al. 2022 included all patients with NSCLC regardless of TC PD-L1 expression levels and found that despite differences in scoring algorithms/sensitivity levels for PD-L1 detection, patients with high PD-L1 expression, stratified by both SP142 (CDx) and alternative 22C3 assays, derived the most clinical benefit from atezolizumab treatment compared with docetaxel. These results align with the real-world data from Taiwan where 94% of patients were tested using 22C3 (31), consistent with data from the pivotal OAK study in 2017 (31).
The second RCT (Herbst et al. 2020) examining atezolizumab included treatment-naïve patients with metastatic NSCLC and PD-L1 expression on ≥1% of TC or tumor-infiltrating IC covering ≥1% of the tumor area (determined by the SP142 assay). However, some patients were additionally assessed using the 22C3 and SP263 assays, showing high overlap between the TPS ≥50% subgroup by 22C3 and the TC ≥50% subgroup by SP263, both predicting the clinical benefit of atezolizumab. While Gadgeel et al. 2022 included all patients regardless of PD-L1 status, Herbst et al. 2020’s preselected PD-L1-positive population limits direct comparison, though both studies indicated greater benefit in patients with high PD-L1 expression. Nonetheless, Gadgeel et al. 2022 reported the greatest clinical benefit in patients with high PD-L1 expression, consistent with the findings of Herbst et al. 2020. However, the presence of discordant or borderline cases highlights the need for caution around key clinical thresholds such as 1% and 50%, where misclassification may lead to under- or overtreatment.
Taken together, current evidence suggests that 22C3 and SP263 may potentially be used to predict the clinical benefit of atezolizumab in advanced/metastatic NSCLC. This finding may be of clinical value in the real-world setting where 22C3 is widely available. A study in the US found that from 2015 to 2017, almost 60% of all PD-L1 testing used 22C3 (with an increasing trend from 39% at the start of the study to 68% in the fourth quarter of 2017) (32). Our review suggests that 22C3 can potentially be utilized as an alternative diagnostic to select responders to ICIs beyond pembrolizumab. In fact, the 22C3 assay is being increasingly utilized in various clinical trials involving atezolizumab to stratify PD-L1 positive patients (33,34).
Both Gadgeel et al. 2022 and Herbst 2020 noted subsets of patients with discordant PD-L1 status, suggesting differences in tumor heterogeneity, staining characteristics, and assay sensitivities (1,23). Gadgeel et al. 2022 found reduced OS benefit in the 22C3 positive group, indicating caution in solely relying on this assay for atezolizumab therapy selection due to the potential inclusion of patients with reduced OS.
Two additional analyses from the phase III IMpower010 trial have reported on the predictive value of alternative PD-L1 assays. This randomized, multicenter, open-label study evaluated adjuvant atezolizumab versus best standard of care (BSC) in patients with resected stage II–IIIA NSCLC. Although this population differs from the stage IIIB–IV population included in the current review, the findings offer insight into assay performance. The first pre-specified interim analysis used the SP263 assay and reported stratified OS HRs suggesting a positive trend favoring atezolizumab in PD-L1 subgroups: stage II–IIIA (n=882), HR 0.95 (95% CI: 0.74–1.24); stage II–IIIA PD-L1 TC ≥1% (n=476), HR 0.71 (95% CI: 0.49–1.03); and stage II–IIIA PD-L1 TC ≥50% (n=229), HR 0.43 (95% CI: 0.24–0.78) (35). A subsequent analysis comparing SP263 and 22C3 showed high concordance (83% for PD-L1-positive and 92% for PD-L1-high thresholds), with both assays demonstrating similar improvements in disease-free survival (DFS) in PD-L1-positive subgroups (36). These findings indicate that the SP263 or 22C3 assays may be used to identify patients with early-stage NSCLC who are more likely to experience benefit from adjuvant atezolizumab (35,36).
Future studies should also consider this difference in the study design. The two retrospective studies (Tamiya et al. 2020; Fujimoto et al. 2018), which evaluated nivolumab in the second-line setting or later, included a total of only 241 patients. While both studies examined PD-L1 expression using alternative assays, their findings should be interpreted with caution due to limitations in reported outcomes and small sample sizes. Tamiya et al. 2020 reported that the OS of patients with TPS ≥50% by 22C3 did not significantly differ from patients with TPS <50%. Fujimoto et al. 2018 reported that in patients with strongly positive PD-L1 staining by 22C3 and SP263 assays, PFS was significantly longer than that in patients with negative staining, which mirrored the results of the CDx 28-8. However, Tamiya et al. study did not report an analysis of PFS by assay, and OS data for the CDx 28-8 was missing (which limits comparison against 22C3) whereas the latter study did not report OS data. Given the absence of complete survival data, a small sample size, and the lack of a non-ICI control arm, the current evidence supporting the use of alternative diagnostics such as 22C3 and SP263 to predict the clinical benefit of nivolumab in NSCLC treatment is limited. Therefore, well-designed RCTs are needed to confirm this potential utility.
Real-world studies further support these findings. One study of NSCLC cases (serial sections of tissue microarrays from 198 cases of resected lung cancer) reported that the pathologists who normally use the 22C3 assay scored consistently higher at the 1% and 50% cut-offs across all assays compared to pathologists who typically use the SP263 assay (37). The authors concluded that there are discrepancies between the two assays when identifying PD-L1-positive cases at clinically relevant cutoffs, which may result in the underestimation of patients suitable for therapy (37). Another study of patients with NSCLC (n=211) reported that the concordance between 22C3 and SP263 assays was greater at higher cut-off levels (TC cut-off levels of 1%, 10%, 25%, and 50% had a concordance of 76.8%, 81.5%, 90.5%, and 94.3%, respectively) (38). However, a third study of surgical specimens from patients with stage IA–IIIA NSCLC (n=420) reported discordance in the positive percent agreement between 22C3 as a standard assay and 28-8 as a comparator assay at cut-off levels of 25% and 50% (39). The authors concluded that the results of the 28-8 assay could be translated to those of 22C3 assay, but not vice versa (39). Taken together, these results highlight the discrepancies between PD-L1 assays and the need to identify whether alternative PD-L1 assays may be used to predict the clinical efficacy of atezolizumab monotherapy in patients with advanced NSCLC.
This current review was limited by the different patient populations, low patient numbers, possible inter-pathological discordance, and heterogenous designs of included studies, which together precluded the conduct of a meta-analysis. Importantly, none of the studies were specifically designed to confirm the utility of alternative assays in predicting the clinical benefit of ICIs in patients with advanced NSCLC. Nevertheless, our analysis serves to highlight the importance of prospectively designing clinical trials to evaluate the use of alternative assays in predicting clinical benefit. Inconsistent data exists for alternative assays with nivolumab, while clinical concordance evidence is lacking for pembrolizumab and durvalumab despite good analytical concordance data (20,28). Well-designed RCTs may help overcome this data gap, particularly if they prospectively include biomarker detection to develop personalized treatment approaches (40).

Conclusions
Our review suggests that alternative PD-L1 assays, 22C3 and SP263, may be used to predict the clinical efficacy of atezolizumab monotherapy in patients with advanced NSCLC. With growing evidence for interchangeability of PD-L1 assays, in terms of both staining of PD-L1 for patient selection and clinical outcomes, physicians treating NSCLC will have greater confidence and flexibility to prescribe their ICI of choice, regardless of the assay used. This will significantly benefit patients with advanced NSCLC by broadening the choice of ICI-containing therapeutic options in these patient populations.

Supplementary
The article’s supplementary files as