A novel clinical tool to estimate risk of false negative KRAS mutation in circulating tumor DNA testing.

Abstract

3594 Background: Recently, in metastatic colorectal cancer (mCRC), the detection of RAS mutations by circulating tumor (ct) DNA has recently emerged as a valid and non-invasive alternative approach, overall showing a high concordance with the standard tissue genotyping, giving information on response to EGFRi treatment and resistant mechanisms. However, RAS mutations may be missed due to low levels of any ctDNA in the blood (false-negative), and it has been difficult to distinguish this from patients without a RAS mutation in the tumor (true-negative). We propose a methodology that can be applied to multi-gene ctDNA testing panels to accurately distinguish true- and false-negative tests. Methods: 357 subjects with tissue and multi-panel ctDNA testing from MD Anderson (MDACC) were used as a training dataset and 295 subjects from Massachusetts General Hospital (MGH) dataset as the testing dataset. CtDNA panels contained between 65 and 70 genes, allowing evaluation of tumor ctDNA shedding from variant allele fraction (VAF) levels in the plasma from other genes (such as APC and TP53). Based on the relationship between KRAS and the VAFs of other gene, we established a Bayesian model providing a posterior probability of false negative in the ctDNA test, using thresholds of < 5% (low), 5-15% (medium), and > 15% (high). This model was validated on the MGH database. Results: Across both cohorts, 431 patients were ctDNA wild type for KRAS. Of those, 29 had tissue documenting a KRAS mutation for a false negative rate of 8%. The model provides the posterior probability that a KRAS mutation is indeed present in the tissue given the observed values of allele frequencies for other mutated genes in the plasma. In the validation cohort, a predicted low false negative had no false negatives (0/62, 95% CI 0%-5.8%), while a predicted medium false negative rate was associated with 3% false negative (1/32, 95% CI 0%-16%). In contrast, a high predicted false negative rate was associated with 5% false negative (5/100, 95% CI 1.6%-11%). The results demonstrate the ability of our tool to discriminate between subjects with true negative and false negatives, as a higher proportion of false negatives are observed at higher posterior probabilities. Conclusions: In conclusion, our approach provides increased confidence in KRAS ctDNA mutation testing in clinical practice, thereby facilitating the identification patients who will benefit from EGFR inhibition while reducing the risk of false negative tests. Extension of this methodology to NRAS and BRAF is possible, with clinical application enabled by a freely available online tool.