Designing Oncology Trials That Use Genomic Alterations for Eligibility

Why use genomic alterations as inclusion criteria—and when?

Genomic inclusion criteria align the investigational therapy’s mechanism of action with patients most likely to benefit. Instead of enrolling “all‑comers,” you prospectively select participants with actionable alterations—EGFR exon 19 deletions, ALK/RET fusions, BRAF V600E, BRAFV600K, BRCA1/2 pathogenic variants, IDH1 R132H, NTRK fusions, and so on—so that the observed treatment effect reflects target engagement rather than chance. This approach increases biological signal, reduces sample size, and can support expedited pathways when effect sizes are large. That said, “genomics‑only” eligibility is not automatically optimal. In tumors with low alteration prevalence or uncertain predictive value, overly narrow criteria can cripple accrual, inflate screen‑fail rates, and introduce spectrum bias (you only study patients with extensive prior testing and access). A principled decision requires: (1) strong translational evidence that the alteration is predictive, not merely prognostic; (2) an analytical pipeline capable of reliably detecting the alteration; and (3) a trial design that preserves internal validity while remaining feasible across regions and labs.

Start from a target–biomarker hypothesis map. For a selective RET inhibitor, for example, a primary cohort might require confirmed RET fusions by RNA‑based NGS or IHC‑triage plus orthogonal RNA confirmation, with exploratory cohorts for high‑copy RET amplifications. For DNA damage response agents, you may specify pathogenic loss‑of‑function variants in BRCA1/2, PALB2, or ATM, and predefine how variants of unknown significance (VUS) are handled (usually excluded unless centrally adjudicated). “Eligibility ≠ diagnosis”: you must encode bioinformatics rules in the protocol—what variant callers are allowed, minimum read depth, and whether subclonal variants from circulating tumor DNA (ctDNA) count toward inclusion.

From biomarker idea to eligibility language: writing precise, auditable criteria

Eligibility language should be specific enough for monitors and inspectors to verify, yet feasible for sites to implement quickly. Replace vague phrases like “genomic evidence of target activation” with operational definitions. Example: “Presence of an ALK rearrangement detected by an RNA‑based NGS assay with (a) minimum 50,000 total mapped reads, (b) paired‑end strategy, (c) fusion junction coverage ≥10 reads, and (d) reporting by a CLIA‑certified/ISO‑15189 laboratory; FISH‑positive cases are eligible if the break‑apart signal proportion is ≥15% in ≥50 evaluable nuclei.” For ctDNA‑based inclusion, pre‑specify variant allele frequency (VAF) thresholds—e.g., “EGFR L858R with VAF ≥0.5% by validated digital PCR or hybrid‑capture NGS, limit of detection (LOD) ≤0.2%.”

To guide investigators, include a concise matrix linking tumor type, alteration, test method, and line of therapy. Also define time windows: “genomic result within 90 days of consent” and whether archived tissue is acceptable. If multiple platforms are permitted, add a comparability statement (e.g., concordance ≥90% in a bridging study) and a central confirmation workflow for discordant cases. A short “ineligible but interesting” pathway helps capture patients with near‑miss results (e.g., VAF 0.4%) into exploratory cohorts without contaminating the primary efficacy population. For reference SOP templates and checklists, many teams adapt materials similar to those found on PharmaSOP.in to keep site screening consistent and auditable.

Assay strategy and validation: LOD, LOQ, and practical cutoffs that survive inspection

Analytical performance drives who gets in. Before first‑patient‑in, document the assay’s sensitivity, specificity, and reportable range, and map those parameters to inclusion thresholds. Use a short, inspector‑friendly table like the one below to anchor your protocol and lab manual. Include illustrative values if proprietary data can’t be published verbatim in the protocol; keep full validation in the laboratory appendix/TMF.

Why mention MACO/PDE in a clinical protocol? Inspectors look for a complete chain of control when diagnostics interface with IMP prep or shared cleanrooms. Even when your CDx is external, a brief cross‑reference to cleaning validation and permissible daily exposure (PDE) helps show risk‑aware governance. Finally, predefine variant classification rules (ACMG/AMP), how tumor purity affects interpretation, and how copy‑number thresholds translate to “amplified” status—e.g., “ERBB2 copy number ≥6 by NGS or ratio ≥2.0 by FISH.”

Choosing the right design: enrichment, basket, umbrella, and platform options

Enrichment RCTs (biomarker‑positive only) maximize effect size and can power overall survival (OS) with fewer patients. They’re ideal when the biomarker is strongly predictive and prevalent (e.g., EGFR mutations in non‑smokers with NSCLC). Basket trials test one drug across multiple histologies with a shared alteration (e.g., NTRK fusions), using parallel cohorts and Bayesian borrowing to stabilize estimates in rare tumors. Umbrella trials test multiple drugs within a single tumor type, randomized by genomic subtype. Platform/master protocols maintain a permanent backbone with arms opening/closing as signals emerge—useful when the genomic landscape shifts rapidly.

Statistical planning hinges on alteration frequency and expected effect size. For a single‑arm basket cohort with historical control ORR 10% and expected ORR 30%, a Simon two‑stage design (α=0.05, 1‑β=0.8) might enroll 15 in stage 1 (stop if ≤2 responses), expanding to 35. For RCTs, stratify by key covariates (ECOG, disease burden) and enforce central confirmation of biomarker status before randomization. Multiplicity control is essential when testing several alterations; prespecify a hierarchical sequence or use alpha‑sharing across cohorts. Keep interim futility rules transparent—e.g., “stop a cohort if posterior P(ORR ≥25%) <10% after 12 evaluable patients.”

Operations: screening logistics, consent, data flow, and query resistance

Real‑world screening is the hardest part. Build a screening cascade: (1) prescreen with existing reports; (2) reflex NGS on archival tissue; (3) if inadequate, repeat biopsy or ctDNA; (4) central review/adjudication; (5) slot reservation. Encode turnaround time targets (e.g., tissue NGS ≤14 calendar days; ctDNA ≤7 days) and escalation if breached. Consent must explicitly address re‑biopsy risks, germline findings (for HRR pathways), and data sharing for variant reclassification. Include a “return of results” plan and a path for incidental actionable germline variants (e.g., referral to genetics).

Data collection: require upload of variant call files (VCF) or structured reports, not just PDFs. Capture bioinformatics pipeline versions to ensure analyses remain reproducible. To avoid endless queries, provide CRF fields for: sample type (tissue/ctDNA), tumor purity %, read depth, VAF, fusion junction reads, and assay platform. A small on‑protocol “bioinformatics glossary” (hotspot vs non‑hotspot, indels vs SVs) helps harmonize multi‑country sites. Build screen‑fail logs with reasons (no alteration, insufficient tissue, below VAF cutoff) to refine feasibility assumptions mid‑trial.

Regulatory expectations and real‑world examples

When a companion diagnostic (CDx) is intended, regulators expect a tightly coupled drug–diagnostic package: analytical validation, clinical validation, and bridging if multiple assays will be allowed commercially. For supportive context and up‑to‑date definitions, see the U.S. agency’s overview of CDx concepts at the FDA. Common real‑world patterns include: (1) tissue‑based CDx for initial approval with a post‑marketing commitment to add ctDNA; (2) centralized testing in pivotal studies followed by decentralization via a ring study; and (3) prespecified retesting rules for discordant local vs central results. In the EU, scientific advice often focuses on the clinical utility of the chosen cutoff (e.g., TMB ≥10 mut/Mb) and assay harmonization across notified bodies.

Case vignette (hypothetical but representative): a selective KRAS G12C inhibitor uses inclusion “KRAS p.G12C by tissue NGS or ctDNA VAF ≥0.5% with LOD ≤0.2%.” Early cohorts showed similar responses for VAF ≥1% and 0.5–1.0%, supporting the ctDNA path. However, false positives clustered around 0.2–0.3% VAF from fragmented samples, prompting a protocol amendment to require orthogonal confirmation (amplicon‑based ddPCR) for VAF 0.3–0.49%. This change cut screen‑fails due to discordance by half while preserving accrual velocity.

Equity, access, and bias mitigation in genomics‑based eligibility

Genomic eligibility can inadvertently exclude patients from under‑resourced settings or minority populations with lower test access. Bake equity into the design: reimburse molecular testing, allow ctDNA for patients without safe biopsy options, and include mobile phlebotomy or courier support. Stratify analyses by testing modality to ensure ctDNA‑included participants do not have systematically different outcomes due to lower sensitivity at low tumor burden. Provide translated consent forms and community‑site training to avoid “academic‑center‑only” recruitment. Finally, add sensitivity analyses that drop cases with borderline VAF or sub‑threshold depth; if conclusions hold, you’ll have stronger external validity.

Putting it all together: a step‑by‑step checklist and a mini‑case study

Checklist: (1) Define the predictive biomarker and clinical context; (2) Lock analytical specs (LOD/LOQ, depth, fusion reads) and write eligibility as auditable rules; (3) Choose design (enrichment, basket, umbrella/platform) and simulate power under realistic prevalence; (4) Stand up screening logistics with defined TATs and adjudication; (5) Predefine handling for VUS, borderline VAF, and discordant results; (6) Implement equity measures and track screen‑fail reasons; (7) Archive assay versions, pipelines, and central review decisions in the TMF;

Mini‑case (RET fusion basket): Multi‑tumor basket with primary endpoint ORR. Inclusion: RET fusions by RNA‑NGS, ≥10 junction reads, ctDNA allowed with confirmatory RNA‑NGS if VAF 0.3–0.49%. Stage 1 (n=14): stop if ≤2 responses. Results: 6 responses → expand to n=35. Subgroup ORR (illustrative): thyroid 60% (n=10), lung 53% (n=15), pancreas 22% (n=10). Safety acceptable; RP2D maintained. The protocol’s tight fusion criteria prevented misclassification from read‑through events and allowed a clean efficacy signal, enabling a registrational strategy with a confirmatory cohort.

Conclusion: precision eligibility that’s scientific, feasible, and inspection‑ready

Using genomic alterations as inclusion criteria isn’t merely adding an NGS line to the protocol—it’s a system of analytical rigor, operational discipline, and ethical foresight. Write eligibility that laboratories can execute reproducibly, anchor cutoffs in validated LOD/LOQ, select designs that respect prevalence and effect sizes, and build logistics that make testing accessible for all eligible patients. With those pieces in place—and transparent documentation that regulators can follow—you’ll deliver trials that are faster, fairer, and far more likely to reveal the true value of precision oncology.