Integrating Companion Diagnostics into Precision Oncology Trials

What Are Companion Diagnostics and Why They Matter

Companion diagnostics (CDx) are in vitro diagnostic devices or imaging tools essential for the safe and effective use of a corresponding therapeutic product. In oncology, CDx testing is often the gateway to trial enrollment—patients must meet specific biomarker-defined eligibility criteria before receiving the investigational drug. For example, a HER2-targeted therapy requires HER2 amplification confirmation, an EGFR inhibitor needs exon 19 deletions or L858R mutations, and an ALK inhibitor demands ALK rearrangement detection.

The role of CDx is not only to identify patients most likely to benefit but also to exclude those at higher risk of adverse effects. Regulators like the FDA and EMA mandate that, when biomarker-based eligibility is critical, the diagnostic must be validated to the same standard of evidence as the drug itself. This concept is central to precision oncology: the therapy’s approval can be contingent on having an approved CDx available.

Real-world example: Trastuzumab deruxtecan was approved alongside a specific HER2 testing method with defined scoring cutoffs. Without an approved HER2 IHC or ISH assay, trial enrollment would not have been possible. Similarly, osimertinib’s label specifies that only EGFR T790M-positive patients by an FDA-approved test are eligible post-EGFR-TKI resistance.

Regulatory Expectations: FDA, EMA, and Global Considerations

From a regulatory standpoint, companion diagnostics are considered high-risk (Class III in the US, Class C under IVDR in the EU) because incorrect results can lead to inappropriate treatment. The FDA’s guidance “In Vitro Companion Diagnostic Devices” specifies that CDx must demonstrate both analytical and clinical validation. Analytical validation ensures that the assay reliably and reproducibly measures the biomarker; clinical validation confirms the biomarker’s predictive value in identifying patients who will benefit from the therapy.

In the EU, under the IVDR (Regulation (EU) 2017/746), companion diagnostics must be assessed by a notified body and involve consultation with a competent medicines authority, such as the EMA. This adds complexity and timelines, especially for global oncology trials seeking simultaneous approval in multiple jurisdictions. Countries like Japan, China, and Australia have their own specific regulatory frameworks, and harmonizing CDx approvals can be a major operational challenge.

One frequent pitfall in global trials is assuming that a US-approved CDx automatically meets EU or APAC requirements—it often does not. This requires early regulatory strategy alignment between drug and diagnostic development teams, ideally before pivotal trial protocol finalization.

Analytical Validation: Establishing Assay Performance (LOD, LOQ, and More)

Analytical validation parameters for CDx include sensitivity, specificity, limit of detection (LOD), limit of quantitation (LOQ), reproducibility, and robustness. For example, a ctDNA-based assay for detecting EGFR T790M may need an LOD of 0.2% variant allele frequency (VAF) with ≥95% confidence to ensure that eligible patients are not missed. LOQ might be set at 0.5% VAF to ensure reliable quantitation for therapy decision-making.

For cross-contamination risk in diagnostic reagent preparation, applying pharmaceutical cleaning validation concepts like MACO (Maximum Allowable Carryover) and PDE (Permitted Daily Exposure) ensures that no assay-to-assay contamination occurs in multi-test platforms.

Designing Clinical Trials with Companion Diagnostics

When integrating CDx into oncology trials, trial design must reflect the biomarker’s prevalence, predictive power, and the assay’s availability. In an enrichment design, only biomarker-positive patients are enrolled, maximizing effect size but potentially slowing accrual if prevalence is low. An all-comers design with biomarker-stratified analysis allows exploratory evaluation of biomarker-negative patients.

Adaptive designs can allow for mid-trial modifications based on interim biomarker prevalence data, while basket and umbrella trials can leverage a single assay to assign patients to multiple targeted therapies. For example, a comprehensive NGS panel could identify HER2 amplification, BRAF mutations, and RET fusions for allocation to different arms within the same master protocol.

Operationalizing CDx Testing in Trials

Operational success depends on fast turnaround times (TAT) and consistent assay performance across global sites. Establishing a central testing laboratory can standardize results but may increase logistical complexity for sample shipment. Alternatively, a decentralized model with harmonized local labs requires rigorous cross-validation (≥90% concordance with central lab results).

Consent forms must explicitly mention the use of a companion diagnostic, potential incidental findings (e.g., germline BRCA mutations), and data sharing for regulatory purposes. Clinical trial management systems should track test performance metrics, including invalid rates, re-testing frequency, and median TAT.

Reference operational SOPs, such as those available on PharmaGMP.in, to streamline documentation for audits and inspections.

Regulatory Submission and Approval Pathways

The drug and the CDx are often submitted concurrently in a coordinated regulatory package. The FDA requires a premarket approval (PMA) for most CDx devices, while the EMA mandates a CE marking under IVDR rules. Bridging studies may be required if the pivotal trial assay differs from the commercial version, with statistical comparability set at ≥90% concordance.

Post-approval, CDx manufacturers may need to expand the assay’s indications, such as adding ctDNA detection to a tissue-based test. These modifications typically require supplemental PMA submissions or revised technical documentation under IVDR.

Conclusion: Making CDx Work for Precision Oncology

Effective companion diagnostics require early and integrated planning between drug and diagnostic development teams. By aligning regulatory strategies, ensuring rigorous analytical validation, and building operational workflows that can deliver results rapidly and reproducibly, CDx can significantly increase the probability of trial success and regulatory approval. The reward is a therapy that reaches the right patients faster, with robust evidence that the biomarker truly guides treatment benefit.

How to Design and Run Bridging Studies for Global Diagnostic Approvals

What Is a Bridging Study and When Is It Needed?

In companion diagnostics (CDx) and other regulated in vitro diagnostics (IVDs), a bridging study demonstrates that results obtained with a new test system, cut-off, matrix, site, or population are clinically and analytically comparable to those used to generate pivotal evidence. Sponsors use bridging when moving from a development assay to a commercial kit, from one instrument or reagent lot to another, or when seeking approval in a new region where local conditions or populations differ. The goal is to show that the medical decisions derived from the “bridged” configuration are as safe and effective as those from the reference configuration.

Typical triggers include: (1) Assay changes (e.g., switching from RUO NGS panel to an IVD kit; reagent reformulation; algorithm update); (2) Specimen or matrix differences (FFPE tumor to plasma ctDNA; serum to whole blood); (3) Platform transfer (central lab to decentralized sites; new instrument generation); (4) Population/region expansion (U.S. to EU/JP/CN) and (5) Cut-off migration (re-optimized thresholds for sensitivity/specificity). Bridging may be analytical (equivalence of measurement) and/or clinical (equivalence of clinical classification and outcomes). The depth of work depends on risk: the more a change could alter clinical calls, the more robust the bridging must be. Internationally, the spirit aligns with ICH E17 on multi-regional trials—ensure data are applicable to the new region with appropriate concordance, bias, and precision analyses; see the ICH page for principles on regional acceptability and consistency of treatment effect. ICH guidance.

Regulatory Triggers and Expectations by Region

FDA (United States). For CDx, bridging is common when the clinical trial used an investigational assay but the marketed device differs. FDA typically expects positive/negative percent agreement (PPA/NPA), overall percent agreement (OPA), bias analyses, and where applicable, kappa for categorical results or Deming/Passing–Bablok regression and Bland–Altman plots for quantitative results. Changes to critical design elements (probe set, antibody clone, software algorithm) often require a PMA supplement with bridging data; if the device is used prospectively in a pivotal drug trial, an IDE may be needed.

EU (IVDR + EMA consultation for CDx). Under IVDR, most CDx are Class C and require a Notified Body conformity assessment with Performance Evaluation (scientific validity, analytical, and clinical performance). When any major element changes (platform, reagent, matrix, cut-off), the Performance Evaluation Report should include bridging demonstrating that clinical claims and IFU statements remain valid. For drug-linked CDx, Article 48(3) mandates EMA consultation; sponsors should pre-align on the bridging statistical plan to avoid rework.

Japan (PMDA/MHLW). Bridging to Japanese populations is frequently requested if the pivotal data were generated elsewhere. PMDA may accept ethnic sensitivity analyses plus a smaller local clinical performance sample set if analytical comparability is robust. Labeling changes typically proceed via Partial Change Application (PCA) supported by bridging.

China (NMPA). Class III CDx often require local clinical study data. When justified, NMPA may accept bridging using archived local specimens to establish concordance between the global and local workflows. Regardless of region, plan bridging early—slotting Notified Body or authority consultations can take months. Practical templates and checklists for aligning dossiers are available at PharmaValidation.in.

Designing a Fit-for-Purpose Bridging Plan

A sound plan starts with a change impact assessment and a risk-based strategy. For low-risk changes (e.g., label typography), documentation may suffice. For moderate/high-risk changes (e.g., antibody clone swap; algorithm re-train), you will need pre-specified acceptance criteria, appropriate sample size, and robust statistics that reflect intended use. At minimum, define: (1) Reference method/configuration, (2) Test (bridged) method/configuration, (3) Clinical decision boundary (cut-off), (4) Primary endpoint (agreement), and (5) Success criteria with 95% CIs.

For qualitative CDx (e.g., PD-L1 IHC), assess PPA, NPA, OPA, and weighted kappa. For quantitative CDx (e.g., TMB), assess Deming regression, correlation, Bland–Altman mean bias and limits of agreement, reclassification tables around cut-off, and total error. Include lot-to-lot, site-to-site, and operator components. A practical acceptance table might look like:

Note: While LOD and LOQ are central to IVDs, terms like PDE and MACO are typically used in cleaning validation for manufacturing; they are not bridging metrics for diagnostics, but teams sometimes cite them in broader product lifecycle risk registers. Keep bridging criteria clinically meaningful: prioritize agreement at or near the decision threshold used for therapy selection.

Statistical Methods and Acceptance Criteria

Bridging statistics must reflect how clinicians use results. For binary/categorical outcomes (e.g., “PD-L1 high vs low”), compute PPA, NPA, OPA with exact (Clopper–Pearson) 95% CIs and weighted kappa to account for ordered categories (e.g., TPS <1%, 1–49%, ≥50%). Include McNemar’s test for discordance symmetry. For quantitative markers (e.g., gene copy number, TMB), use Deming regression (accounts for error in both methods), Bland–Altman plots for mean bias and limits of agreement, and total allowable error tied to clinical risk. Around the cut-off, report reclassification (how many patients flip across the threshold) with 95% CIs.

Sample size. Power your study to bound the lower confidence limit above your acceptance threshold. Example: to show PPA ≥95% with the 95% CI lower bound ≥90% assuming true PPA=97%, you may need ~180–220 positives, depending on exact design and pairing rate. Include a discordant resolution plan (e.g., adjudication by orthogonal method) only to understand root causes—most regulators prefer primary analyses without post-hoc “fixes.” For multi-site bridging, include random effects for site in generalized linear mixed models to ensure agreement holds across locations.

Operational Execution: Specimens, Logistics, and Documentation

Good operations make or break bridging. Start with a specimen adequacy plan (minimum tumor content, RNA/DNA yield, pre-analytical controls). Lock down sample accessioning, blinding, and chain-of-custody. For matrix bridging (FFPE→plasma), specify paired draws, maximum time to processing, and shipping temperatures (e.g., plasma 2–8°C ≤48 hours; FFPE ambient ≤72 hours). Use identical cut-offs and reporting rules in both arms unless the goal is to validate a new threshold—then present side-by-side ROC/Youden analyses, with clinical rationale.

Document everything: Bridging Protocol, Statistical Analysis Plan, Reagent/lot history, instrument calibration, operator training, and deviation/CAPA logs. Align data transfers with EDC/LIMS specifications and audit trails (21 CFR Part 11). A simple shipping matrix helps sites comply:

Case Studies: EGFR, PD-L1, and TMB

EGFR ctDNA (China NMPA). A sponsor moved from a central RT-PCR to a commercial NGS kit for local registration. Using 320 archived Chinese plasma samples paired with tissue calls as clinical truth, PPA was 95.8% (95% CI 92.3–97.9) and NPA 98.1% (95% CI 95.9–99.2). Bias at the 1% VAF cut-off was negligible by Deming regression, enabling kit approval without a full prospective trial.

PD-L1 IHC (Japan PMDA). After changing the staining platform, a 3-lab round-robin (n=420 cases) showed category-weighted kappa=0.86 and OPA=93% at TPS≥50%. A small Japanese subset (n=120) confirmed ethnic applicability; PMDA accepted a PCA with labeling alignment to the drug’s SmPC.

TMB (EU IVDR). For an IVD transitioning from 1.5 Mb to 0.8 Mb panel, bridging used 400 FFPE samples. Agreement around the 10 mut/Mb cut-off: reclassification 3.8% (95% CI 2.1–5.9), mean bias −0.4 mut/Mb. Notified Body and EMA consultation endorsed the PER with updated IFU language.

Common Pitfalls and How to Fix Them (CAPA)

Cut-off drift. If the new method exhibits systematic bias near the threshold, pre-specify cut-off transfer via regression mapping, justify clinically, and validate stability across lots. Specimen bias. Excess archival positives can inflate PPA; maintain disease prevalence and include consecutive samples or adjust via re-weighted analyses. Over-fitting algorithms. Freeze the model prior to bridging; document training/validation splits and lock software under design control. Discordant handling. Do not purge outliers; investigate with orthogonal tests, summarize root causes, and implement CAPA (e.g., slide restaining criteria, ctDNA input QC).

Templates and Submission Packaging

Package bridging in a way reviewers can navigate quickly. Include: Change Impact Memo, Justification for Bridging vs New Study, Protocol/SAP, Specimen Accountability, Primary/Supportive Analyses, Risk–Benefit, and Labeling Redlines. Provide machine-readable data listings and annotated programs. For IVDR, make sure the Performance Evaluation Report explicitly references the bridging evidence; for FDA, craft a PMA Supplement or main PMA section with searchable tables/figures.

Conclusion

Effective bridging compresses timelines and avoids duplicative clinical trials while maintaining patient safety. By aligning statistics with clinical decisions, executing rigorous operations, and packaging results clearly for each region, sponsors can extend CDx indications and markets efficiently—and compliantly.