3077 Background: NGS-based profiling of circulating cell-free DNA (cfDNA) enables accurate and sensitive tumor genotyping from blood. Guardant360 Liquid CDx is intended as a companion diagnostic for comprehensive tumor mutation profiling across >700 genes in patients with malignant neoplasms. Analytical performance was evaluated per FDA IVD standards and guidelines across five variant classes: single-nucleotide variants (SNVs), insertions/deletions (indels), gene rearrangements, gene amplifications (CNA), and copy-number losses (CNLs) – with emphasis on accuracy, precision, specificity (limit of blank LoB), and limit of detection (LoD). Methods: Analytical Validation tested >4000 samples from nearly 2,000 unique patients across 29 cancer types, as well as contrived materials. Concordance was evaluated against FDA-approved Guardant360 CDx and an externally developed ctDNA assay, where applicable. Primary endpoints were positive percent agreement (PPA) and negative percent agreement (NPA) by variant class and category (clinically significant vs panel-wide). LoD was established at both low and high input levels. No minimum cfDNA ng input cutoff was required; low-input performance was evaluated at minimum molecule coverage (QC threshold). Precision was assessed across workflow components and blood collection tube (BCT) lots. Specificity was assessed using healthy donor cfDNA sequenced with multiple replicates at high input. Results: Clinical samples demonstrated a high QC pass rate of 99.1%. Clinically significant variants showed high concordance: SNV PPA 94.9% with NPA 99.86%, and indel PPA 96.24% with NPA 99.98%. Panel-wide specificity was near-perfect (SNV NPA 99.99966%; indel NPA 99.99991%). Rearrangements achieved PPA 94.64% and NPA 99.76%; CNA PPA 91.67% and NPA 99.97% ; CNL PPA 100% and NPA 97.56%. High-input LoD was established in silico and confirmed with clinical samples: 0.2% mutant allele frequency (MAF) for all clinically relevant SNVs, indels and most rearrangements, with selected variant LoD as low as 0.1% (KRAS G12C). At the minimum input, panel-wide LoD was 1.0% for clinically relevant SNVs and 0.9% for indels; 0.5-1.6% for rearrangements, 2.3-2.4 copies for CNAs, and 22.7% tumor fraction for CNL. Precision studies showed ≥98.6% agreement across conditions at near-LoD variant levels; within-lot and between-lot BCT precision yielded >97% analytical positive/negative agreement (targeted variants ≥1× LoD). In LoB, no false positive variants were detected in any healthy donor replicates (0% FPR per sample and per variant), validating in-silico CHIP detection methods. Conclusions: The assay demonstrates high accuracy, precision and specificity across all reportable variant classes, with LoD at or below 0.2% MAF for key variant types. These data support its use for comprehensive, pan-cancer tumor genomic profiling in routine oncology practice.
Kennedy et al. (Wed,) studied this question.