Analytic and clinical validation of a negative prediction algorithm for actionable mutations utilizing genomic and epigenomic profiling in cfDNA.

Abstract

3584 Background: One challenge in cell-free DNA (cfDNA) profiling for genomic tumor profiling is the inability to confidently confirm the absence of actionable genomic mutations. This limitation arises from the challenge of determining whether key driver mutations are truly absent or if tumor levels are below the detection threshold of the assay. Accurate negative variant prediction could enable clinicians to expedite clinical decisions based on cfDNA results without relying on tissue biopsy sequencing when no actionable alterations are found. Here, we report analytic and clinical validation of a novel algorithm to enable negative prediction from liquid biopsy to address this critical clinical need. Methods: Using the Guardant Infinity platform, which simultaneously profiles genomic and epigenomic signals in a single sample, we integrated highly sensitive and precise tumor fraction estimates and developed a negative prediction algorithm to allow for confident reporting of samples that do not detect an actionable genomic finding. The algorithm estimates the post-test probability of a cfDNA sample harboring genomic biomarkers with FDA approved therapies relevant to treatment selection, based on population priors, epigenomic tumor fraction (TF), and the analysis of mutant and non-mutant coverage across variants of interest which was assessed for advanced colorectal (CRC) and lung cancer (NSCLC) patients. Results: In 3973 CRC and 7654 NSCLC analyzed patients, 41% of CRC and 22.6% of NSCLC were found to have an actionable mutation. Among the remaining samples, 66% of CRC and 56.3% of NSCLC had sufficient tumor fraction to assess the sample as variant negative with > 95% confidence. Reasons why the remaining samples could not be confidently assessed included low tumor shedding (including 15% with nondetectable tumor), low genomic coverage over loci of interest, and mutant allele support below the confident call threshold. An additional cohort of 237 CRC and 316 NSCLC patients with paired tissue and cfDNA results was used to clinically validate the negative prediction algorithm. All samples with sufficient tumor fraction for > 95% confidence and predicted to be negative by the algorithm for genomic biomarkers with FDA-approved therapies were confirmed to be negative in tissue results. Conclusions: A plasma-based epigenomics-based approach for confident negative prediction is feasible in CRC and NSCLC, as demonstrated by validation results. Confident negative prediction has the potential to enhance the utility of liquid biopsy and accelerate clinical decision-making in advanced solid tumors with biomarker-guided treatment pathways and should be validated in additional clinical datasets.