Sources of discordance among germ-line variant classifications in ClinVar

Abstract

PurposeClinVar is increasingly used as a resource for both genetic variant interpretation and clinical practice. However, controversies exist regarding the consistency of classifications in ClinVar, and questions remain about how best to use these data. Our study systematically examined ClinVar to identify common sources of discordance and thus inform ongoing practices.MethodsWe analyzed variants that had multiple classifications in ClinVar, excluding benign polymorphisms. Classifications were categorized by potential actionability and pathogenicity. Consensus interpretations were calculated for each variant, and the properties of the discordant outlier classifications were summarized.ResultsOur study included 74,065 classifications of 27,224 unique variants in 1,713 genes. We found that (i) concordance rates differed among clinical areas and variant types; (ii) clinical testing methods had much higher concordance than basic literature curation and research efforts; (iii) older classifications had greater discordance than newer ones; and (iv) low-penetrance variants had particularly high discordance.ConclusionRecent variant classifications from clinical testing laboratories have high overall concordance in many (but not all) clinical areas. ClinVar can be a reliable resource supporting variant interpretation, quality assessment, and clinical practice when factors uncovered in this study are taken into account. Ongoing improvements to ClinVar may make it easier to use, particularly for nonexpert users.

Figure 1

ClinVar variant NM_007294.3(BRCA1):c.4327C>G (p.Arg1443Gly). Key data fields used in our analyses are indicated. Where common names differ from ClinVar terminology, both are included. From https://www.ncbi.nlm.nih.gov/clinvar, accessed 6 March 2017.

Figure 2

Concordance by clinical area. Variant classification concordance measured as a fraction of variants for all genes (leftmost pair) and for genes in each clinical area (other pairs). The left bar in each pair (labeled ACT) represents our actionability analysis, and the right bar (PATH) indicates our pathogenicity analysis (see text for details). Note that the y axis starts at 50%.

Figure 3

Outlier rate by submitter. (a) Outlier rate (fraction of all classifications discordant with the majority consensus) calculated on an actionability basis for submissions of each major type. The left bar in each pair represents all ClinVar variants, and the right bar is restricted to the 23 hereditary cancer genes listed in the National Comprehensive Cancer Network guidelines (see text and Supplementary Table S1). (b) Outlier rate for specific submitters with the largest number of classifications in the 23 cancer genes. In the current data set, these clinical testing submitters each contributed more than 1,000 variants; submitters of other types provided more than 100 variants. More than 87% of all clinical testing submissions and more than 90% of all literature-only, curation, and research submissions are represented by the submitters in (b). BIC, Breast Cancer Information Core; ENIGMA, Enhancing NeuroImaging Genetics through Meta-Analysis; NIH, National Institutes of Health; OMIM, Online Mendelian Inheritance in Man; SCRP, Sharing Clinical Reports Project.

Figure 4

Outlier rate by classification date. Lines show the outlier rate for variants in each date range. Bars show the number of variant classifications for each range. Solid bars and lines show the complete ClinVar data set, whereas the hatched bars and dotted lines show only data for the 23 hereditary cancer genes listed in the National Comprehensive Cancer Network guidelines (see text and Supplementary Table S1).

Figure 5

Concordance for ClinVar and subsets. Variant classification concordance measured as a fraction of variants for all of ClinVar and for subsets of ClinVar filtered by submission type and classification date. Concordance is calculated on an actionability basis (see text).