The importance of algorithms when determining homologous recombination deficiency status in ovarian tumors.

Abstract

e15166 Background: Ovarian tumors with homologous recombination deficiency (HRD) respond better to PARP inhibitor and platinum-based chemotherapy treatment compared to those that are HR proficient. Markers of genomic instability can help predict HRD status. MyChoice is an FDA-approved genomic instability score (GIS) based on a proprietary algorithm with an established threshold of 42 for distinguishing HR-proficient from HR-deficient tumors. Attempts have been made to apply different algorithms to calculate GIS, raising concern for different HRD status results that could negatively impact patient treatment. Methods: An exploratory analysis was conducted by comparing GIS scores calculated using the clinically validated algorithm and an alternative algorithm with scores published in Marquard et al (2015) 1 , here termed the Marquard summed HRD score (MSHS). A tumor was considered BRCA deficient if it had no functional copy of BRCA1 and/or BRCA2 , and considered BRCA intact otherwise. Pearson correlations of continuous scores, discordance between HRD status, and negative and positive percent agreement (NPA and PPA, respectively) were calculated. Logistic regression models were used to predict BRCA deficiency status. Overall survival (OS) was predicted with Cox proportional hazards models. P-values from likelihood ratio tests and Kaplan-Meier survival estimates are reported. Results: 663 breast and ovarian tumor samples from The Cancer Genome Atlas 2 were analyzed, of which 395 ovarian tumors had OS data. Correlation between GIS and MSHS was 0.864. For BRCA intact tumors (n = 495), discordance of HRD status was 34.5%, PPA was 99.0%, and NPA was 56.9%. The percent of BRCA intact tumors that would qualify for treatment by MSHS but not by GIS was 34.3%. In a bivariable logistic regression model, GIS was significantly better at predicting BRCA deficiency status than MSHS (p < 0.001 and p = 0.23, respectively). The OS at 5-years was 22.4% for tumors with a GIS < 42 versus 41.9% for tumors ≥42 (p < 0.001); meanwhile, OS was 21.6% for tumors with MSHS < 42 versus 33.8% for tumors ≥42 (p = 0.058). Conclusions: Scores and thresholds to measure the same biological phenomena but calculated with different algorithms can vary substantially, which may result in over- or under-identification of HR-deficient tumors and therefore over- and under-treatment for patients. References: Marquard AM, Eklund AC, Joshi T, et al. Pan-cancer analysis of genomic scar signatures associated with homologous recombination deficiency suggests novel indications for existing cancer drugs. Biomark Res . 2015;3:9. Published 2015 May 1. doi:10.1186/s40364-015-0033-4 Cancer Genome Atlas Research Network. Integrated genomic analyses of ovarian carcinoma [published correction appears in Nature. 2012 Oct 11;490(7419):298]. Nature . 2011;474(7353):609-615. Published 2011 Jun 29. doi:10.1038/nature10166.