Companion diagnostics (CDx) have become the cornerstone of precision oncology — but beneath the regulatory frameworks and validated assays lies a set of genuinely contested clinical questions that the field has not yet resolved. As of early 2025, the FDA had cleared or approved more than 78 drug–CDx combinations,[1] and the pace of new approvals shows no sign of slowing. Yet the clinical application of CDx testing is marked by real, consequential disagreements — about which assay to trust, which biospecimen type to use, and whether drug-specific testing should give way to broader genomic interrogation.

This article examines three such debates. For each, we present the primary evidence on both sides, identify where sources agree, and highlight the specific points of disagreement that have real consequences for clinical trial design, patient selection, and treatment decisions. All claims are cited to primary academic sources; we have deliberately avoided relying on commentary, editorials, or news articles.

78+ FDA-approved drug–CDx pairs As of early 2025 — up from fewer than 20 in 2015[1]
4 FDA-approved PD-L1 assays 22C3, 28-8, SP142, SP263 — each validated with a different drug, each giving different results[2,3]
49% Liquid–tissue concordance ROME trial: only half of mutation findings agreed between paired liquid and tissue testing[4]

Debate 1: The PD-L1 Assay Problem — Which Test Should You Trust?

PD-L1 immunohistochemistry is the most widely used biomarker in oncology immunotherapy. Yet it is also the CDx with perhaps the most acute assay fragmentation problem in clinical medicine. Four distinct PD-L1 IHC assays have FDA approval for use with specific checkpoint inhibitors, each paired with a different drug and each scoring tumour samples differently:

Assay Antibody Clone Paired Drug Key Score Regulatory Status
PD-L1 IHC 22C3 pharmDx (Dako) 22C3 Pembrolizumab (Keytruda) TPS, CPS FDA CDx PMA
PD-L1 IHC 28-8 pharmDx (Dako) 28-8 Nivolumab (Opdivo) TPS FDA CDx PMA
VENTANA PD-L1 SP142 (Roche) SP142 Atezolizumab (Tecentriq) TC%, IC% FDA CDx PMA
VENTANA PD-L1 SP263 (Roche) SP263 Durvalumab (Imfinzi) TC% FDA CDx PMA

The problem is that most pathology labs cannot feasibly run all four assays on every tumor sample. In practice, clinicians often apply one assay's result to inform a treatment decision tied to a different assay's validation dataset. This is either an acceptable clinical approximation or an unvalidated shortcut — and the evidence supports both views.

Viewpoint A: The assays are not interchangeable — and treating them as such has clinical consequences

Primary Evidence A cross-sectional study of 362 gastric cancer samples scored using all three major assays (22C3, 28-8, SP142) using multiplex IHC found that at every clinically relevant CPS threshold, the 28-8 assay produced approximately double the positivity rate of the 22C3 assay: at CPS ≥1, 70.3% vs. 49.4% (p <0.001); at CPS ≥5, 29.1% vs. 13.4% (p <0.001); at CPS ≥10, 13.7% vs. 7.0% (p = 0.004). Mean CPS on 28-8 was 6.39 vs. 3.46 on 22C3 — a near-doubling. At the clinically pivotal CPS ≥5 cutoff, concordance between 22C3 and 28-8 was only moderate. The authors concluded: "Until stronger evidence of inter-assay concordance is found, we urge caution in treating the assays as equivalent."[2]

A separate study of SP263 vs. SP142 in muscle-invasive bladder cancer found that SP263 was statistically more sensitive for tumour cell (TC) scoring (p = 0.0009), while SP142 was more sensitive for immune cell (IC) scoring (p = 0.0067). The authors reported that harmonisation between the two clones could not be achieved.[3]

A 2025 evaluation of all four FDA-approved PD-L1 assays in clear cell renal cell carcinoma found immune cell positivity rates of 14.7% (22C3), 16.1% (28-8), 2.1% (SP142), and 15.0% (SP263) — with pairwise concordance coefficients ranging from only 0.16 to 0.52, indicating poor to moderate agreement at best.[5]

Viewpoint B: Practical clinical harmonisation is feasible and necessary

Primary Evidence A 2022 study in triple-negative breast cancer (TNBC) examining concordance between 22C3 and SP142 found substantial agreement (κ = 0.80) in primary tumors and moderate agreement (κ = 0.60) in metastatic tumors when tested at diagnostic cutoffs. This suggests that for some tumor types and some clinical settings, cross-assay use may be clinically acceptable — particularly for the primary tumor at the time of initial diagnosis.[6] Proponents of this view argue that requiring strict assay–drug pairing is logistically unworkable in most health systems and that single-assay strategies (typically 22C3, given pembrolizumab's breadth of approvals) would serve most patients adequately.
⚖️ Debate 1 · PD-L1 Assay Fragmentation — The Opposing Positions
Position A — Assay-specific use only
Each CDx assay is validated only for its paired drug in its specific clinical setting. Applying the wrong assay may declare patients eligible who should not be treated, or exclude patients who would benefit.

The 28-8 assay produces ~2× higher CPS scores than 22C3 in the same gastric tumor sample — if substituted, it would classify nearly twice as many patients as pembrolizumab-eligible (pembrolizumab's CDx uses 22C3), without the clinical trial validation to support that expanded eligibility.[2]

Yeong et al. Gastric Cancer 2022. DOI:10.1007/s10120-022-01301-0
Position B — Pragmatic harmonisation
Running four separate assays per patient is economically and logistically impractical in most oncology centres. For some tumor types, concordance between assays is sufficient to justify a unified testing strategy.

In TNBC, κ = 0.80 concordance between 22C3 and SP142 at diagnostic CPS cutoffs suggests that a single validated assay may be sufficient for many treatment decisions — provided the limitation (lower concordance at metastatic sites, κ = 0.60) is understood.[6]

Discordance in PD-L1 expression using 22C3 and SP142 between primary and metastatic TNBC. PubMed PMID: 37668667

Debate 2: Liquid vs. Tissue Biopsy as the Primary CDx Platform

For most of CDx history, tissue biopsy was unquestioned as the gold standard specimen type. A pathologist examines tumour cells; biomarker expression is assessed directly. But tissue biopsy has real limitations: it is invasive, it captures a single spatial snapshot of a heterogeneous tumour, it can fail due to insufficient material, and results take weeks in many settings. Plasma-based circulating tumour DNA (ctDNA) — "liquid biopsy" — offers an alternative. The question is whether it is yet equivalent.

Viewpoint A: Liquid biopsy is ready for primary CDx use and may outperform tissue

Primary Evidence — Raez et al., Clinical Lung Cancer 2023 (PMID: 36585341) A retrospective analysis of 170 patients with metastatic NSCLC who received both tissue and liquid biopsy NGS as standard of care found that physicians based the majority of treatment decisions on liquid biopsy results (73.5% vs. 25.9% tissue-based). Liquid biopsy returned results an average of 26.8 days faster than tissue. For guideline-recommended biomarkers, liquid biopsy was 94.8–100% concordant with tissue. A liquid-first approach identified actionable biomarkers in 76.5% of patients compared with 54.9% in a tissue-first approach. There was no significant difference in overall survival or progression-free survival between patients whose treatment was guided by liquid vs. tissue biopsy findings.[7]

A large multi-centre study of 421 NSCLC patients (the Korean Lung Liquid Versus Invasive Biopsy Program) found 77.6% concordance between ctDNA-based (Guardant360) and tissue-based (Oncomine Focus Assay) NGS testing. Crucially, plasma testing detected additional genomic alterations in 11 patients (4.2%) not found by tissue, and in 50 patients without tissue-based NGS results, ctDNA detected alterations in 40%. Upfront ctDNA-based testing alone identified actionable alterations in 60.4% of patients. The authors concluded that ctDNA-based testing should be used to complement tissue-based testing rather than replace it — but their data demonstrated substantial additive value that tissue testing alone cannot provide.[8]

A 2025 prospective study explored a third biospecimen type — pleural effusion cell-free DNA — and found that it outperformed both conventional tissue and plasma testing. In 50 patients with malignant pleural effusion, actionable mutations were detected in 88% by pleural effusion cfDNA NGS, compared with 78% by Sanger sequencing of pleural cell pellets and 66% by standard clinical tissue genetic testing. This suggests that the tissue vs. liquid binary framing undersells the full landscape of biospecimen options available to oncologists.[9]

Viewpoint B: Tissue remains the gold standard; liquid biopsy sensitivity is a critical limitation

Primary Evidence — Malapelle et al., ESMO Open 2026 (PMID: 41483627) A retrospective observational study using a US nationwide real-world database assessed 425 patients with stage III NSCLC who received both liquid and tissue EGFR mutation testing. Liquid biopsy showed high specificity (98%) but critically low sensitivity (45%). Only 5% of liquid samples detected EGFR mutations versus 8% of tissue samples. Importantly, concordance improved over time — from 87% in 2017–2019 to 96% in 2020–2023 — but the core finding held: liquid biopsy misses roughly half of EGFR-mutated stage III tumors. The authors concluded that "a negative liquid biopsy result should prompt tissue testing," explicitly rejecting a liquid-first-only strategy.[10]

The ROME trial (2025), which examined concordance between liquid and tissue biopsy for actionable mutations in patients eligible for targeted therapy, found overall concordance in only 49% of cases. Alterations were detected exclusively in tissue in 35% of cases and exclusively in liquid in 16%. The clinical significance was stark: patients in the concordant group who received matched targeted therapy had a median progression-free survival of 4.90 months, versus 2.86 months in the true discordant group and 2.53 months in the failure discordant group.[4] Low concordance translated directly into worse survival outcomes.

🧬 Debate 2 · Liquid vs. Tissue as Primary CDx Platform — The Opposing Positions
Position A — Liquid-first or co-testing
In metastatic NSCLC, liquid biopsy is 94–100% concordant with tissue for actionable biomarkers and returns results 27 days faster, with no difference in survival outcomes. It should be used first-line or co-currently with tissue.

Liquid biopsy also finds mutations missed by tissue (4.2% of cases in the Korean LUNG program[8]) and captures tumour heterogeneity across metastatic sites that a single tissue biopsy cannot represent.

Raez et al. Clinical Lung Cancer 2023. DOI:10.1016/j.cllc.2022.11.007
Position B — Tissue remains gold standard
In stage III NSCLC, liquid biopsy sensitivity is only 45% for EGFR mutations. In the ROME trial, overall concordance was 49%. A negative liquid biopsy result has a high false-negative rate; tissue must follow.

Sensitivity varies significantly by disease stage, tumour shedding, and mutation type. The faster turnaround does not justify the risk of false negatives when tissue is obtainable, especially in non-metastatic settings.[10]

Malapelle et al. ESMO Open 2026. DOI:10.1016/j.esmoop.2025.106029

Debate 3: Drug-Specific CDx vs. Comprehensive Genomic Profiling

The traditional CDx model — one drug, one biomarker, one validated assay — has served precision oncology well but creates practical constraints. Each new drug requires a new co-developed CDx, consuming time and resources. Tumour tissue is finite; sequential single-gene assays can exhaust a sample before the most relevant test is run. Comprehensive genomic profiling (CGP) panels like FoundationOne CDx and Guardant360 CDx test hundreds of genes simultaneously from a single specimen. The FDA has approved some of these panels as CDx for multiple drugs across multiple tumour types. The debate is whether this broader approach enhances or dilutes the precision of CDx.

Viewpoint A: Drug-specific CDx ensures analytical rigour and avoids variant overinterpretation

The FDA's foundational CDx framework — articulated in its 2014 guidance on In Vitro Companion Diagnostic Devices and refined in subsequent guidance documents — requires that each CDx device be validated analytically and clinically for its specific drug–biomarker–tumour type context.[1] A single-gene assay for EGFR exon 19 deletion has highly specific analytical validation for that alteration. Sensitivity, specificity, and cut-off thresholds are defined and documented for exactly the test being used.

CGP panels generate variants of uncertain significance (VUS) alongside actionable findings. A 2023 FDA pilot program explicitly sought to "provide greater transparency regarding performance characteristics that certain tests for oncology biomarkers used to select oncology drug treatments should meet"[1] — acknowledging that many NGS panels used in clinical practice lack the performance validation of formally approved CDx devices. Critics of broad CGP point to the complexity of communicating VUS results to patients and clinicians, the risk of off-label treatment decisions based on insufficiently validated findings, and the cost and reimbursement challenges of comprehensive panels (approximately $3,000–5,000 USD per test).

Viewpoint B: CGP preserves tissue and expands therapeutic options — and FDA has already adapted

Primary Evidence — FDA Regulatory Evolution FDA's 2020 guidance on "Developing and Labeling In Vitro Companion Diagnostic Devices for a Specific Group or Class of Oncology Therapeutic Products" explicitly facilitated class labelling — allowing a single CDx to be approved for use with a group of drugs or a group of tumour types, rather than drug-by-drug. FoundationOne CDx achieved this model: a single PMA-approved NGS panel covering 324 cancer-related genes, approved as a CDx for more than 20 drugs across six or more tumour types, detecting SNVs, indels, CNAs, fusions, TMB, and MSI from a single tissue specimen.[1]

A 2025 study of NSCLC patients examined the clinical utility of CGP for biomarker results reported below the limit of detection. It found that the majority of patients who received targeted therapy matched to a CDx biomarker — even at analyte levels formally below the assay's detection threshold — still demonstrated clinical benefit from that therapy. This suggests that CGP-guided treatment decisions retain clinical value even at the analytical margins where single-gene assays would return no result.[11]

CGP proponents also argue that tissue conservation is clinically decisive: for small biopsy specimens common in advanced lung cancer, exhausting tissue on serial single-gene tests leaves nothing for the assay that ultimately determines treatment. A single CGP run from limited material yields information that would require 10–15 separate single-gene assays to replicate — and a single-gene reflex testing cascade may consume the sample before reaching the most relevant target.

🔬 Debate 3 · Drug-Specific CDx vs. Comprehensive Genomic Profiling — The Opposing Positions
Position A — Rigorous single-biomarker CDx
Each drug–biomarker pair requires independent analytical and clinical validation. CGP generates variants of uncertain significance that create clinical noise, off-label treatment pressure, and reimbursement barriers.

FDA's pilot program (2023) exists precisely because many NGS tests used in clinical oncology do not meet CDx-level performance standards. The existence of the programme acknowledges a gap between what is approved and what is used in practice — a gap that may harm patients through under- or over-treatment.[1]

FDA Guidance: Companion Diagnostics. Last updated 2023. fda.gov/companion-diagnostics
Position B — Broad CGP as CDx platform
FoundationOne CDx is FDA-approved for 20+ drug–biomarker pairs from a single 324-gene panel. Broader testing prevents tissue exhaustion, uncovers actionable alterations missed by reflexive single-gene testing, and reflects FDA's own evolution toward class labelling.

CGP-detected biomarkers below the limit of detection still predict clinical benefit, suggesting the assay's clinical value exceeds its formally reportable sensitivity range — a finding that argues for broader, not narrower, genomic interrogation.[11]

Clinical utility of CGP biomarkers below the limit of detection, advanced NSCLC. PubMed PMID: 40549041

Where All Three Debates Agree

Despite the disagreements, the primary literature across these three debates converges on several core principles that all viewpoints accept:

Points of Consensus Across All Three CDx Debates

  • Biomarker testing is non-negotiable. No serious position in any of these debates argues against testing. The question is which test, which specimen, and which analytical standard — not whether to test at all.
  • Analytical and clinical validation are minimum requirements. Even proponents of CGP accept that analytical validity must be demonstrated. Even proponents of liquid biopsy accept that false-negative rates must be understood and disclosed.
  • Turnaround time matters clinically. Liquid biopsy's 27-day speed advantage over tissue[7] is clinically meaningful. Consensus holds that testing pathways that delay treatment beyond 3–4 weeks from diagnosis are clinically suboptimal.
  • Concordance studies are necessary but not sufficient. All parties agree that knowing the concordance rate between two testing modalities is not the same as knowing which is "right." A single reference standard (tumour behaviour under matched therapy) is the true arbiter.
  • Serial testing adds value that baseline testing cannot provide. Tumour evolution under therapy, acquired resistance mutations, and heterogeneity at progression are not captured by a single baseline CDx result — liquid biopsy proponents and tissue proponents alike accept this limitation of point-in-time testing.

The Specific Disagreements That Matter for Clinical Practice

PD-L1 Assay

Whether cross-assay substitution (using 28-8 results to guide pembrolizumab prescribing, for example) is an acceptable clinical practice or an unvalidated shortcut. No randomised trial has tested this directly; existing data show material scoring differences that could affect eligibility in up to 2× the number of patients in some tumour types.[2]

Liquid Biopsy

Whether a negative liquid biopsy result should reflexively trigger tissue biopsy (sensitivity-aware approach) or whether, in the right clinical context, a negative result can be accepted without tissue confirmation. The ROME trial's 51% discordance rate and the survival data associated with concordance vs. discordance make this clinically consequential.[4]

CGP vs. Single-Gene

Whether the variants of uncertain significance generated by CGP panels cause net benefit (through novel findings that eventually prove actionable) or net harm (through downstream off-label treatments, patient anxiety, and resource consumption without proportionate benefit). This question cannot yet be answered from existing prospective data.

What These Debates Mean for Clinical Trial Design

For oncology clinical research organisations and sponsors, these debates have direct operational implications. Protocol-specified CDx requirements must anticipate each of the issues above:

  • PD-L1 CDx specification: Protocols must specify not just the biomarker but the exact assay (clone, platform, scoring algorithm) and cut-off. A protocol that requires "PD-L1 positive" without specifying 22C3 CPS ≥1 vs. 28-8 CPS ≥1 will generate patient populations that differ by tumour type in unpredictable ways.
  • Liquid biopsy eligibility: If liquid biopsy is accepted as the CDx specimen, the protocol must pre-specify the handling of negative results — does a negative plasma result exclude the patient, or does it trigger mandatory tissue biopsy? Given sensitivity as low as 45% in some settings,[10] "negative by liquid = excluded" may miss half the patients who would benefit.
  • CGP vs. reflex single-gene: Protocols using CGP panels as CDx should pre-specify how VUS results are handled (not used for eligibility decisions) and should include provision for archiving tissue in the event that single-gene confirmatory testing is required by the central laboratory or regulator.
  • Concordance sub-studies: For trials that accept both liquid and tissue CDx results, a planned concordance analysis (liquid vs. tissue vs. clinical outcome) in all patients, not just discordant cases, generates data that directly inform future guideline development — a practical way for investigator sites to contribute to resolving these debates.

Conclusion

Companion diagnostics are not a solved problem. The regulatory framework is robust. The clinical utility of CDx-guided therapy is not in dispute. But the three debates examined here — which PD-L1 assay to trust, which biospecimen type to prioritise, and whether breadth or specificity better serves patients — are live, evidence-contested, and clinically consequential. The academic literature does not yet provide clean resolution to any of them.

What the evidence does show is that the right answer is almost certainly context-dependent: liquid biopsy may be first-line in metastatic settings and inadequate in stage III; SP142 may perform differently in bladder cancer than in breast cancer; CGP may be appropriate when tissue is limited and harmful when generating VUS in a population that lacks genetic counselling resources to interpret them.

For clinical research teams designing biomarker-selected trials or managing CDx decisions at investigator sites, these nuances are not academic — they determine which patients are enrolled, which patients receive treatment, and how outcomes data should be interpreted when trials report results. Understanding where the field agrees and where it genuinely does not is prerequisite to making defensible, evidence-based decisions in each specific clinical context.

References

  1. U.S. Food and Drug Administration. Companion Diagnostics. Updated June 2023. Includes guidance on the 2014 IVD CDx guidance, April 2020 class labelling guidance, and June 2023 Pilot Program guidance. fda.gov/companion-diagnostics FDA Regulatory
  2. Yeong J, Lum HYJ, Teo CB, et al. Choice of PD-L1 immunohistochemistry assay influences clinical eligibility for gastric cancer immunotherapy. Gastric Cancer. 2022;25(4):741–750. DOI:10.1007/s10120-022-01301-0 · PubMed PMID: 35661944 Peer-reviewed
  3. Gupta G, Pasricha S, Kamboj M, et al. PD-L1 expression in muscle invasive urothelial carcinoma: Comparison of SP142 and SP263 assay. Indian J Pathol Microbiol. 2022;65(4):839–843. DOI:10.4103/ijpm.ijpm_1472_20 · PubMed PMID: 36308190 Peer-reviewed
  4. The Impact of Concordance between Liquid and Tissue Biopsy for Actionable Mutations: Insights from the ROME Trial. PubMed PMID: 40833744. 2025. Concordance 49%; median PFS 4.90 months (concordant) vs. 2.86 months (true discordant) vs. 2.53 months (failure discordant). Clinical Trial
  5. Evaluation of the diagnostic concordance of FDA-approved PD-L1 assays in clear cell renal cell carcinoma. Scientific Reports (Nature). 2025. nature.com/articles/s41598-025-05697-4 IC positivity: 22C3 14.7%, 28-8 16.1%, SP142 2.1%, SP263 15.0%; κ 0.16–0.52. Peer-reviewed
  6. Discordance in PD-L1 expression using 22C3 and SP142 assays between primary and metastatic triple-negative breast cancer. PubMed PMID: 37668667. 2023. Primary tumor κ = 0.80; metastatic κ = 0.60; SP142 cross-site κ = -0.03. Peer-reviewed
  7. Raez LE, Brice K, Dumais K, et al. Liquid Biopsy Versus Tissue Biopsy to Determine Front Line Therapy in Metastatic Non-Small Cell Lung Cancer. Clinical Lung Cancer. 2023;24(2):120–129. DOI:10.1016/j.cllc.2022.11.007 · PubMed PMID: 36585341 Peer-reviewed
  8. Park S, Olsen S, Ku BM, et al. High concordance of actionable genomic alterations identified between ctDNA-based and tissue-based NGS testing in advanced NSCLC: The Korean Lung Liquid Versus Invasive Biopsy Program. Cancer. 2021;127(16):3019–3028. DOI:10.1002/cncr.33571 · PubMed PMID: 33826761 Peer-reviewed
  9. Wang HY, Liao WY, Ho CC, et al. Enhanced detection of actionable mutations in NSCLC through pleural effusion cell-free DNA sequencing: A prospective study. European Journal of Cancer. 2025;217:115224. DOI:10.1016/j.ejca.2025.115224 · PubMed PMID: 39799785 Prospective
  10. Malapelle U, Nasirova F, Wang A, et al. Liquid biopsy and tissue biopsy for the detection of EGFR mutations in patients with stage III NSCLC: an observational real-world study. ESMO Open. 2026;11(1):106029. DOI:10.1016/j.esmoop.2025.106029 · PubMed PMID: 41483627 Sensitivity 45%; specificity 99%; concordance 93% overall, 96% in 2020–2023 cohort. Real-world
  11. Clinical utility of companion diagnostic biomarker results below the limit of detection in comprehensive genomic profiling of patients with advanced NSCLC. PubMed PMID: 40549041. 2025. Majority of patients receiving matched targeted therapy at sub-LOD levels demonstrated clinical benefit. Peer-reviewed
  12. An analysis of FDA drug approvals for oncological hematological malignancies in relation to companion diagnostics. PubMed PMC12507643. 2025. By early 2025, >78 drug–CDx combinations approved. Kinase inhibitors, ADCs, and small-molecule drugs increasingly paired with CDx beyond antibody drugs. Regulatory analysis
Disclosure & Methodology This article was researched using PubMed, FDA.gov, and peer-reviewed journal databases. All citations are to primary sources: peer-reviewed original research articles, FDA guidance documents, and regulatory analyses. Editorial commentary, news articles, and industry white papers were excluded from the evidence base. According to PubMed records accessed July 2026, the cited articles have been published in journals including Gastric Cancer, Clinical Lung Cancer, Cancer, ESMO Open, and Scientific Reports. DOIs and PMIDs are provided for every claim.