Background
Objectives
-
What types of tests, platforms, and scoring algorithms are currently being used?
-
How has the validity of these tests, platforms, and scoring algorithms been tested?
Methods
Criteria | Inclusion Criteria | Exclusion Criteria | Rationale |
---|---|---|---|
Population(s) | Patients with bladder/urothelial cell, lung, gastric, or ovarian cancer | • Ongoing studies • Interim analyses • Studies of other oncology indications | Population criteria were designed to reflect cancer populations that are candidates for PD-L1 expression testing |
Interventions/Comparators | Diagnostic tests targeting the PD-1/PD-L1 pathway | N/A | To survey the range of tests currently in use, all PD-L1 tests or studies looking at diagnostics used in PD-1/PD-L1 immunotherapy trials were considered |
Outcomes | PD-L1 test-validation metrics. Information on PD-L1 test-scoring algorithms or cutoffs was also captured from those studies that reported on test performance | Studies that did not report outcomes of interest for the study population | As the review aimed to evaluate how well different tests performed against validation criteria, studies reporting outcomes relating to validation metrics were prioritized. It was also considered important to capture data relating to the tests’ characteristics (scoring algorithms and test cutoffs) in order to determine the comparability of different tests |
Time | Indexed databases: January 1 2010 to September 15, 2016) Grey literature: Three most recent meetings (January 2013 up to November 2016). | Studies published prior to 2010 or after the final search date in 2016 | Date limits were applied to reflect the very recent/current nature of this field of research |
Study Design | • Randomized trials • Observational studies • Diagnostic or clinical validation studies | • Animal studies • Case reports • Editorials | Study design criteria reflected the nature of the studies reporting on test-validation metrics of PD-L1 tests for use in human population. |
Other | PD-L1 tests required to be commercially available • English language only • Geographic emphasis on the US, EU5, and Japan | • Articles that were either not published in English or outside the geographic locations of interest • Publications on noncommercially available tests | Commercially available tests were prioritized to ensure that review was relevant to current practice. The geographic emphasis reflects the countries in which PD-1/PD-L1 immunotherapies are currently licensed. Most evidence in this field is published in English so language limits were designed to reflect this. |
Results
Search results
Lung cancer
Types of PD-L1 antibody tests identified in the SLR
-
One study looked at IHC-28-8 (Dako), developed alongside nivolumab [10]
Test-scoring algorithms and thresholds used among the PD-L1 tests
Antibody (developer) [drug against which the study validated the test] | Cutoff/Threshold |
---|---|
22C3 (Dako) [pembrolizumab] | 1% (used in training group): 1 study [4]a 50% (determined as optimal cutoff): 1 study [4]a |
28–8 (Dako) [nivolumab] | 1%: 3 studies [7, 10, 12]a 5%: 2 studies [10, 11]a 10%: 1 study [10]a 50%: 1 study [11]a |
SP263 (Roche) [durvalumab] | 25%: 3 studies [7, 14, 28]a,b |
SP142 (Roche) [atezolizumab] | 1%: 2 studies [19, 20]a 5%: 2 studies [19, 20]a,c 50%: 1 study [20]a |
E1L3N (Cell Signaling Technology; reagent provider) [not applicable] | 1%: 2 studies [20, 21]a 5%: 4 studies [11, 20, 21, 23]a,c 50%: 3 studies [11, 20, 21]a |
Types of cells tested for PD-L1 expression
Test validation metrics
Study Information | Inter-Observer Concordance % (95% CI) | Intra-Observer Concordance % (95% CI) | Inter-Site Concordance % (95% CI) | Intra-Site Concordance % (95% CI) |
---|---|---|---|---|
Antibody (developer): 22C3 (Dako) Roach et al. (2016) [Trial name: KEYNOTE-001] [5] Indication: NSCLC Drug against which the study validated the test: pembrolizumab | ANA: 92.6% (87.8%–96.7%) APA: 92.8% (88.1%–96.8%) OA: 92.7% (88.1–96.8%) | ANA: 96.4% (94.0%–98.5%) APA: 96.5% (94.3%–98.6%) OA: 96.4% (94.3%–98.6%) | ANA: 90.3% (84.4%–95.2%) APA: 85.2% (75.6%–92.9%) OA: 88.3% (81.4%–94.3%) | ANA: 91.9% (88.8%–94.8%) APA: 87.6% (82.5%–92.2%) OA: 90.2% (86.3%–93.7%) |
Antibody (developer): 28–8 (Dako) Phillips et al. (2015) [10] Indication: NSCLC Drug against which the study validated the test: nivolumab | 1% Cutoff ANA: 96.1% (94.7%–97.4%) APA: 96.5% (95.2%–97.7%) OA: 96.3% (94.9%–97.5%) 5% Cutoff ANA: 91.3% (89.2%–93.2%) APA: 89.3% (86.8%–91.6%) OA: 90.4% (88.3%–92.3%) | 1% Cutoff ANA: 97.7% (95.6%–99.3%) APA: 97.9% (96.0%–99.3%) OA: 97.8% (95.9%–99.3%) 5% Cutoff ANA: 98.7% (97.2%–99.7%) APA: 98.3% (96.5%–99.6%) OA: 98.5% (97.0%–99.6%) | 1% Cutoff ANA: 93.9% (92.4%–95.3%) APA: 95.5% (94.4%–96.5%) OA: 94.8% (93.6%–95.9%) 5% Cutoff ANA: 90.2% (88.5%–91.8%) APA: 90.2% (88.5%–91.8%) OA: 90.2% (88.6%–91.8%) | 1% Cutoff ANA: 97.0% (95.2%–98.4%) APA: 97.7% (96.5%–98.9%) OA: 97.4% (95.9%–98.7%) 5% Cutoff ANA: 94.8% (92.8%–96.7%) APA: 94.8% (92.8%–96.7%) OA: 94.8% (92.8%–96.7%) |
Antibody (developer): SP263 (Roche) Rebelatto et al. (2016) [14] Indication: NSCLC Drug against which the study validated the test: durvalumab | APA: 96.6% (93.8%–98.8%) ANA: 96.8% (93.9%–98.9%) OPA: 96.7% (94.2%–98.9%) | APA: 96.2% (92.7%–98.8%) ANA: 96.4% (93.0%–98.8%) OPA: 96.3% (93.3%–98.8%) | PPA: 93.3% (89.0%–95.9%) NPA: 79.5% (73.6%–84.4%) OPA: 86.4% (82.7%–89.3%) | NR |
Antibody (developer): SP142 (Roche) Boyd et al. (2015) [3] Indication: NSCLC Drug against which the study validated the test: atezolizumab | Met predefined acceptance criteria including > 90% inter-reader concordance. | NR | NR | NR |
Antibody (developer): E1L3N (CSTa) Gainor et al. 2016 [21] Indication: NSCLC (EGFR-mutant and ALK-positive) Drug against which the study validated the test: PD-1/PD-L1 inhibitors | Concordance between the two pathologists: 1%: 0.88 (κ = 0.75) 5%: 0.92 (κ = 0.80) 50%: 0.97 (κ = 0.89) | NR | NR | NR |
Antibody (developer): E1L3N (CSTa) Inamura et al. (2016) [23] Indication: lung (adenocarcinoma, surgically resected) Drug against which the study validated the test: N/A | Agreement between two pathologists: 5%: κ = 0.70 (95% CI 0.55–0.86), indicating substantial agreement | NR | NR | NR |
Antibody (developer): E1L3N (CSTa) Huynh et al. (2016) [22] Indication: lung (adenocarcinoma, surgically resected) Drug against which the study validated the test: N/A | Agreement between two pathologists: 1%: 0.84 (κ = 0.69) 5%: 0.91 (κ = 0.81) 50%: 0.91 (κ = 0.78) | NR | NR | NR |
Study Information | Type of Test (developer) | Overall Concordance/Discordance Between Tests | Other Comparisons Between Tests | Authors’ Conclusions |
---|---|---|---|---|
Anderson et al. (2016) [7] Drug against which the study validated the test: N/A | 28–8 (Dako) and SP263 (Roche) | Overall concordance between antibodies was 90.3%, but was only 66.7% for specimens considered positive for PD-L1 expression | There was considerable variation in the percentage of TC staining positive as determined by the two methods, which along with the different test cutoffs contributed to discordant results | This study points to the importance of methodological and interpretation variation, as well as other considerations such as tumor heterogeneity and dynamics of expression, when evaluating the use of PD-L1 as a biomarker of potential therapeutic response to checkpoint blockade inhibitors |
McLaughlin et al. (2016) [20] Drug against which the study validated the test: N/A | E1L3N (CSTa) and SP142 (Roche) | PD-L1 Comparison Using Different PD-L1 Antibodies and IHC κ concordance between antibodies was low, irrespective of the cutoff used: • 1% tumor PD-L1 cutoff: 0.340 • 5% tumor PD-L1 cutoff: 0.286 • 5% stroma PD-L1 cutoff: 0.124 • 50% tumor PD-L1 cutoff: 0.189 | Concordance between the two rigorously validated antibodies was fair to poor.While both E1L3N and SP142 reportedly bind to the intracellular domain of PD-L1, the difference between the two antibodies raises concerns and suggests antibody-validation data should be shown in future clinical trial reports | |
Rivalland et al. (2016) [11] Drug against which the study validated the test: N/A | E1L3N (CST
a
) and 28–8 (Dako) | The concordance between antibodies was 75.0% and 86.2% at 5% and 50% cutoffs, respectively | • E1L3N stained a significantly higher proportion of tumors at both cutoffs (P < 0.001), and in almost all clinic-pathologic subgroups • A significant correlation was observed in staining between antibodies (R2 = 0.40, P < 0.0001) • Small-cell lung cancer stained significantly more frequently than adenocarcinoma with Dako 288 (35.7% vs. 17.4%, P < 0.001) but not with E1L3N (44% vs. 35.1%, P < 0.08) | Overall PD-L1 positivity was correlated between these two antibodies, however the CSTa antibody stained significantly more samples |
Scheel et al. (2016) [6] Drug against which the study validated the test: N/A | 28–8, 22C3, SP142, and SP263 (NR) | • NR | • The tests 28–8 and 22C3 stained comparable TC proportions • In some cases, SP142 stained fewer TCs but more ICs and SP263 stained more TCs and ICs compared with the other tests • The differences in TC proportions would translate into different classifications by any of the dichotomous cutoffs | • The data indicate that unified PD-L1 IHC scoring criteria for TCs are feasible, while scoring of ICs requires detailed training • The four tested PD-L1 tests did not show comparable staining patterns in all cases of NSCLC • The results obtained by each test are not interchangeable. Thus, more studies are required to archive a harmonized “PD-L1 status” in NSCLC • In particular, more data on the predictive value of one test for multiple substances are needed |
Smith et al. (2016) [15] Drug against which the study validated the test: N/A | SP263 (Roche) and E1L3N (CSTa) | NR | Inter-pathologist correlation Membrane tumor staining scores • SP263: R2 > 0.87 • E1L3N: R2 > 0.82 Positively staining cells in the immune infiltrate • SP263: R2 > 0.66 • E1L3N: R2 > 0.80 | Due to its staining intensity, scoring range, and pathologist preference, the SP263 IHC test has been deemed superior to the E1L3N IHC test |
Ilie et al. (2016a) [13] Drug against which the study validated the test: N/A | SP142 and SP263 (Roche); and 28–8 (Antibodycam) | Inter-reader precision in determining the PD-L1 expression in TCs: • OA: 92% (κ = 0.910), 98% (κ = 0.976) and 96% (κ = 0.935) for SP142, SP263 and 28–8 tests Inter-reader precision in determining the PD-L1 expression in ICs: • OA: 81% (κ = 0.786), 87% (κ = 0.832), and 86% (κ = 0.817) for SP142, SP263 and 28–8 tests, respectively | Concordance analysis on TCs: • Poor correlation between the SP142 and SP263 antibodies (ρ = 0.852, κ = 0.362), and the SP142 and 28–8 antibodies (ρ = 0.860, κ = 0.412), while a good correlation was observed between the SP263 and 28–8 antibodies (ρ = 0.996, κ = 0.883) Concordance analysis on ICs: • Poor agreement between the SP142 and SP263 antibodies (ρ = 0.568, κ = 0.018) and the SP142 and 28–8 antibodies (ρ = 0.590, κ = 0.134), while a good correlation was noted between the SP263 and 28–8 antibodies (ρ = 0.880, κ = 0.721) | Our results suggest that PD-L1 protein expression is heterogeneous and that different antibody tests may yield variable results. The anti-PD-L1 antibodies SP142 vs. SP263, and SP142 vs. 28–8 showed fair to poor concordance, while the 28–8 and SP263 antibodies demonstrated a strong correlation for both the TC and IC compartments |
Schildhaus et al. (2016) [12] Drug against which the study validated the test: N/A | IHC: 28–8 (Dako) and FISH: ZytoLight SPEC CD274, PDCDILG2/CEN 9 Dual Color Probe | The correlation between IHC and FISH was statistically significant (χ2: P < 0.001) | NR | PD-L1/2 FISH could contribute to our understanding of PD-L1 expression and could therefore be a valuable adjunct biomarker in upcoming trials with PD-1/PD-L1 inhibitors |