Consistency Of Pathogenicity Determinations For Hereditary Cancer Gene Mutations

Background: As the number of laboratories offering genetic tests grows, the potential for inconsistent variant classifications increases. New resources can help address this: (a) ClinVar, a rapidly growing public database of clinical variants to which many (but not all) laboratories contribute; (b) the public release of thousands of BRCA1/2 reports from Myriad Genetics through the Sharing Clinical Reports Project (SCRP), the Free the Data (FTD) initiative, and recent publications; and (c) ExAC, a greatly improved database of population allele frequencies. These complement longstanding efforts, e.g. the ENIGMA consortium. In addition, the American College of Medical Genetics (ACMG) recently updated guidelines for the interpretation of sequence variants. Using these resources, we sought to investigate the consistency of variant classifications to help inform ongoing practice. Methods: Pathogenicity assessments for variants in hereditary cancer genes were collected from multiple sources. Among these were 15,364 BRCA1/2 submissions to ClinVar, including 5416 submissions from SCRP and 1062 from our prior work [1]. When 3 or more submissions for a variant were available, we determined a consensus interpretation requiring 2 of 3 submitters to agree (or 3 of 4, etc.) to identify outliers. For our own classifications, we established a point-based system based on the 2015 ACMG guidelines, and independently applied it to publicly available evidence of pathogenicity without regard to other labs9 classifications. Results: Initially, discordance among ClinVar submissions appears high (20-30%). However, upon investigation much of this discordance is a result of (i) research submissions to ClinVar, (ii) differences in confidence (e.g. benign vs. likely benign), (iii) older data in ClinVar, (iv) single lab outliers, and (v) nuances in the detailed structure of the ClinVar database. A careful comparison using the consensus methodology of objectively filtered ClinVar data shows high concordance between our interpretations and consensus: 95% were identical and 99% were similar (e.g. benign and likely benign were considered similar). Where consensus was not achieved (8% of variants with 3 or more independent sources) or not possible (any variant with only 2 sources), pairwise comparisons showed that few of these remaining differences (5%) were clinically significant. Also, many variants with significant discordances appeared to be particularly rare in the human population, and thus would be present in few patients. The rate of discordances with SCRP/FTD data was similar with that of other ClinVar submitters. Conclusions: Evaluations of inter-laboratory concordance need to be done carefully to avoid over-counting differences. Laboratories generally agree on the clinical significance of the vast majority of variants. Furthermore, the inter-laboratory consensus classification is often reached using a careful implementation of the ACMG guidelines and publicly available data. Thoroughly understanding the remaining differences is challenging when the evidence used by any laboratory is not available for peer review. Detailed data and methods from this study are available for review and alternate analyses. [1] Lincoln et al . SABCS 2014; JMD 2015. Citation Format: Lincoln S, Nykamp K, Kobayashi Y, Yang S, Powers M, Anderson M, Monzon F, Topper S. Consistency of pathogenicity determinations for hereditary cancer gene mutations. [abstract]. In: Proceedings of the Thirty-Eighth Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2015 Dec 8-12; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2016;76(4 Suppl):Abstract nr P2-09-11.