Comparison of a 3-component genomic instability score to individual biomarker scores in classifying homologous recombination deficiency.

Abstract

e15165 Background: Ovarian tumors with defects in the homologous recombination (HR) pathway have improved treatment response to DNA-damaging agents. HR deficiency (HRD) status is calculated through biomarker scores like large scale state transitions (LST), telomeric allelic imbalances (TAI), loss of heterozygosity events (LOH), or genomic LOH (gLOH). We assessed how performance of individual biomarker scores compares to MyChoice, an FDA-approved, 3-component genomic instability score (GIS) (LST+TAI+LOH), for predicting HRD status. Methods: Patients with ovarian, fallopian tube, and peritoneal cancer were clinically tested at Myriad Genetics. BRCA status was determined as BRCA mutant (BRCAmut) or as BRCA wildtype (BRCAwt). Thresholds for GIS, LOH, TAI, and LST were set based on the 95th percentile of scores observed in tumors deficient in BRCA. The threshold for gLOH was based on a previously validated threshold. 1 HRD status was considered positive for: (1) all BRCAmut tumors regardless of score; and (2) BRCAwt tumors with scores above the pre-specified thresholds. HRD-status classification by individual biomarker thresholds was compared to the classification determined by the GIS threshold, including using positive- and negative-percent agreement (PPA and NPA). 95% confidence intervals (CI) were calculated via an exact binomial test. HRD status classification by LOH and gLOH were also compared. Results: We analyzed 14,968 cancer patients. Among BRCAwt tumors (n = 13,357), the HRD status based on individual component thresholds all had meaningful discordance with the HRD status based on GIS threshold (LOH: 14.7%, LST: 11.9%, gLOH: 11.6%, TAI: 11.1%). PPA was lowest for gLOH at 59.5%, followed by LOH at 82.6%, TAI at 93.3%, and LST at 98.3%. NPA was lowest for LST at 84.9%, followed by LOH at 86.2%, TAI at 87.6%, and gLOH at 97.6%. The percent of BRCAwt tumors that would qualify for treatment by a single component score, but not by GIS, was 11.5% by LST, 10.5% by LOH, 9.4% by TAI, and 1.8% by gLOH. The percent of BRCAwt tumors that would qualify for treatment by GIS, but not by a single component score, was 9.7% by gLOH, 4.2% by LOH, 1.6% by TAI, and 0.4% by LST. The proportion of BRCAmut tumors (n = 1,611) with scores above the score-specific HRD status threshold differed by score: 89.8% (CI 88.2-91.2%) for GIS; 92.4% (CI 91-93.7%) for LST; 92.5% (CI 91.1-93.7%) for TAI; 81.6% (CI 79.6-83.5%) for LOH; and only 64.1% (CI 61.7-66.5%) for gLOH. The HRD status based on LOH and gLOH had 14.9% discordance, indicating that the two scores are not equivalent. Conclusions: Single-component scores did not perform equivalently to the FDA-approved GIS. This has clinical implications for patients since clinicians rely on accurate HRD status to support treatment decisions. Inaccuracies could lead to inappropriately under- or over-treating patients. Reference: doi:10.1200/po.19.00345.