Abstract
BACKGROUND Genetic testing frequently identifies variants of uncertain significance (VUSs). Providers, however, are often ill-prepared or too time-constrained to manage these findings, and insurers are concerned about impacts on clinical care and cost. Here we compared the contribution of panel-based and genomic (exome and genome) testing to the generation of inconclusive results due to VUSs.
METHODS Rates of inconclusive results due to VUS and diagnostic yield were determined from over 1.5 million sequencing test results from 19 clinical laboratories in North America from 2020 - 2021.
RESULTS We found a lower rate of inconclusive test results due to VUSs from genomic tests (22.5%) compared to multi-gene panel tests (32.6%; p<0.001) and a higher diagnostic yield (17.5% vs 10.3%; p<0.001). For panel tests, the rate of inconclusive results correlated with panel size. The use of trios improved yield (19.5% vs 15.2%; p<0.001) and reduced inconclusive rates (18.9% vs 27.6%; p<0.001). The use of genome sequencing compared to exome improved yield (25.1% vs 16.6%; p<0.001) without increasing the rate of inconclusive results (22.2% vs 22.6%).
CONCLUSION Genomic sequencing demonstrated reduced uncertainty and higher molecular diagnostic yield compared to panel testing. This difference is best explained by obligatory reporting of all VUSs in panel-based testing compared to genomic testing where correlation with phenotype is used to constrain variant reporting. These results may inform future genetic testing practices and heighten appreciation for the professional skills involved in genomic test interpretation.
A growing percentage of the population receives genetic testing, yet DNA sequencing often uncovers variants of uncertain significance (VUSs) that pose challenges for patients as well as providers who are sometimes ill-prepared or too time-constrained to manage these findings. Furthermore, insurers are concerned about downstream costs from possible follow-up of VUSs. The ClinVar database contains submissions from over 2000 submitters, with over 80% coming from clinical laboratories and the largest fraction classified as uncertain significance due to the high rate of extremely rare and unique variation in the human population. These data represent unique variants but do not account for case-level data to assess how many individuals are actually receiving results of uncertainty during genetic testing.
The likelihood of receiving a VUS is influenced by a variety of factors: 1) Indication for testing - tests with a high prior probability of identifying a causal variant (e.g., testing the single CFTR gene in a patient with cystic fibrosis) will yield less uncertainty than a test with a lower prior probability (e.g., a large panel test in an individual with muscle weakness). In addition, VUSs are generally not reported for tests performed for screening purposes, often performed on ostensibly healthy individuals, whereas VUSs are reported if uncovered in tests performed on symptomatic individuals for diagnostic purposes, consistent with professional practice guidelines (Miller et al. 2021); 2) Ancestry - individuals whose genomic DNA is more ancestrally distinct compared to the human genome reference and existing population databases will have higher rates of VUSs (Gudmundsson et al., 2022); 3) Testing platform - the size of the test (including the number of genes) and the type of variants detected will influence the rate of VUSs; 4) Laboratory reporting practices - inclusive of laboratory specific rubrics for VUS inclusion and reporting; 5) Family structure - the use of additional family members in support of the proband analysis can both rule in and out variants and affect VUS reporting. Additionally, while genomic sequencing interrogates the full genome and therefore has the potential to generate the highest rate of VUSs for any test, professional judgment is applied to decide which variants to report, potentially yielding a lower rate of VUSs. We sought to explore this question through analysis of testing data with the primary goal to understand which test type (panels versus genomic testing) leads to more uncertainty in results due to VUSs, while also examining the diagnostic yield. We were able to analyze the effects of testing platform, indication and family structure, though we did not collect individual level ancestry data to confirm the impact of ancestry on VUS rate. Furthermore, we did not directly analyze individual laboratory reporting practices, though by collecting a large enough fraction of testing data from a recent two year period we feel this data reflects actual practice in the US.
Methods
LABORATORIES AND TIME PERIOD
Data from diagnostic (symptomatic) testing was collected from 19 clinical laboratories in North America including Ambry Genetics (Aliso Viejo, CA), ARUP Laboratories (Salt Lake City, UT), Children’s Hospital of Philadelphia (Philadelphia, PA), Children’s Mercy Hospital (Kansas City, MO), Fulgent Genetics (Temple City, CA), GeneDx (Gaithersburg, MD), HudsonAlpha Clinical Services Lab (Huntsville, AL), Illumina Clinical Services Laboratory (San Diego, CA), Invitae (San Francisco, CA), Mass General Brigham Laboratory for Molecular Medicine (Cambridge, MA), Mayo Clinic (Rochester, MN), New York Genome Center (New York, NY), Quest Diagnostics (Secaucus, NJ), Rady Children’s Institute for Genomic Medicine (San Diego, CA), Stanford Medicine (Palo Alto, CA), The Hospital for Sick Children (SickKids) Genome Diagnostics Laboratory (Toronto, ON, Canada), The University of Chicago Genetic Services Laboratory (Chicago, IL), University of Washington (Seattle, WA), and Variantyx (Framingham, MA). Results were limited to a two-year period spanning from January 1st, 2020 through December 31st, 2021. Three laboratories reported all (panel, exome, and genome) test results, nine reported panel and exome, two reported panel and genome, four reported genome only, and one reported panel only, as their exome platform was strictly interpreted based on predefined panels of genes selected at the ordering step (Supplemental Table 1).
DATA COLLECTION
Deidentified summary data was collected (see Supplemental Material) and aggregate statistics were calculated for inconclusive cases with at least one VUS and positive cases in which causal variant(s) were identified. Inconclusive cases without a VUS (e.g., cases with a single heterozygous variant pathogenic for a phenotypically consistent recessive disease) were not included in the inconclusive rates and VUSs present in positive reports were also not included. Panel-based test results were grouped by panel size (2-10, 11-25, 26-50, 51-100, 101-200, >200 genes). Genomic sequencing tests were categorized by exome versus genome and by parent-child trio versus less-than-trio (e.g., proband only, single parent with proband, multiple siblings, etc.). For some laboratories, test results were further categorized by disease area across twelve broad indications (Cardiovascular, Hereditary Cancer/Cancer Syndrome, Neurodevelopmental/Intellectual Disability/Autism, Neurological/Muscular/Neuromuscular, Retinal Disease, Hearing Loss, Renal/Gastrointestinal, Endocrine, Metabolic, Dysmorphic/Skeletal, Hematologic/Rheumatologic/Immunologic, Dermatologic) or Other. Average panel size for each disease area was computed by taking the midpoint in the panel range (e.g., 6 for 2-10) or 201 for >200, and multiplying the midpoint panel size by the volume of each panel, summing across lab panels and then dividing the aggregate sum by the total volume of tests. Panel tests performed on an exome or genome backbone were categorized as panel tests as long as the test was limited to interpretation of the panel-defined gene content. Laboratories had the option to delineate VUS results by sub-tiers mapped to VUS-High, VUS-Mid, VUS-Low; however, only two laboratories used sub-tiers. Laboratories were asked to provide race/ethnicity/ancestry information if available. Detailed summary data can be found in Supplemental Table 3.
DATA QUALITY CONTROL AND EXCLUSIONS
Single-gene test data were collected but excluded from the analysis given that these tests are often performed as follow-up to carrier screening and not offered as diagnostic tests, and this distinction was not tracked in laboratory systems. One laboratory offered custom add-on options for their tests (i.e., a provider could select one or more genes to add to a panel) but these results were not aggregated with the primary panel result; therefore, this data was excluded from analysis. One laboratory offered customizable panel orders but did not track exact panel size and these data were excluded. Also, one laboratory allowed providers to opt-out of VUS reporting for neurodevelopmental/neurological panels and as such, only cases in which the provider did not opt out for VUS reporting were included. In some cases, if a patient had multiple tests ordered, only the largest panel was counted to avoid double-counting overlapping genes.
STATISTICAL ANALYSIS
All statistical analyses were performed using Chi-square analysis with Yates correction. A p-value below 0.001 was considered statistically significant.
Results
Data from 1,512,306 genetic and genomic sequencing tests were collected from 19 clinical laboratories in North America. The data spanned 1,463,812 multi-gene panel tests (96.8%), 42,165 exome tests (2.8%) and 6329 genome tests (0.4%); the latter two categories are collectively referred to as genomic tests. Race/ancestry/ethnicity data was provided for approximately half of the tests (770,403 individuals) with 59% White, 10% Hispanic, 8% Black, 4% Asian, 12% Mixed/Other and 8% Not Specified (see Supplemental Table 2).
The rate of inconclusive test results (due to the presence of at least one VUS in the absence of a causal etiology) was lower for genomic tests (22.5%; 10,933/48,494) than the rate from panel tests (32.6%; 477,617/1,463,812; p<0.001) (Fig. 1A). For panel tests, the rate of inconclusive results correlated with panel size (number of genes) ranging from 6.0% for 287,811 panel tests of 2-10 genes to 76.2% for 84,316 panel tests >200 genes (Fig. 1B,C). Diagnostic yield (17.5%) from genomic testing was higher than that from panel testing (10.3%; p<0.001; Fig. 3A).
All laboratories were able to differentiate genomic sequencing results by parent-child trio versus less-than-trio (one or both parents missing). When examining method (genome versus exome) and samples (trio versus less-than-trio) for genomic testing, the use of trios led to higher yield (19.5% vs 15.2%; p<0.001) and lower inconclusive rates (18.9% vs 27.6%; p<0.001) (Table 1). In contrast, the use of genome sequencing compared to exome led to higher yield (25.1% vs 16.6%; p<0.001) but did not increase the rate of inconclusive results (22.2% vs 22.6%) (Table 1).
One of twelve disease areas representing broad indications for testing were specified for 50.2% of panel tests and 13.4% of genomic tests. Only six disease areas had >25 cases and were used to compare differences in yield and VUS rates between panels and genomic tests (Fig. 2&3; Supp. Table 3). This analysis showed higher diagnostic yield and reduced VUS inconclusive rates for genomic tests as compared to panels in cardiovascular, neurologic/muscular, and neurodevelopmental/intellectual disability/autism. Metabolic disease showed no difference in yield and a reduction in VUS inconclusive rate. Hematologic/rheumatologic/immunologic showed a marginal increase in diagnostic yield and VUS inconclusive rate with genomic testing. The diagnostic yield for and dysmorphic/skeletal was higher for panels and showed no difference in VUS rate.
Two laboratories (Mass General Brigham LMM and Quest) reported VUSs in sub-tiers allowing analysis of the rate of inconclusive results by further subdivisions of VUS evidence. Sub-tier results were mapped to three categories: VUS-High, VUS-Mid, VUS-Low. In this dataset, the sub-tier rate was 22% VUS-High, 56% VUS-Mid, 22% VUS-Low for panels and 40% VUS-High, 60% VUS-Mid, 0% VUS-Low for genomic tests (Fig. 4).
Discussion
In this study, we observed that genomic (exome and genome) sequencing demonstrated reduced uncertainty and higher molecular diagnostic yield compared to panel testing. The increased yield was generally expected given the ability to interrogate all genes in genomic analyses. However, it is unknown what fraction of individuals who received genomic testing had already received a negative panel-based test and/or cytogenomic microarray, potentially biasing the population tested. Assuming an appreciable fraction, we anticipate genomic testing yield to be even higher than reported here as clinical practice moves towards genomic testing as a primary test.
The increased yield of genome sequencing compared to exome sequencing was also expected, based on improved quality and evenness of coding sequence coverage and increased detection of structural variation (Mattick et al., 2018). Nevertheless, we express caution in overinterpreting this result given that genome testing represented a much smaller fraction of testing (0.4%) and therefore the results could more easily be impacted by lab specific pipelines, ordering practices and patient populations.
In contrast, the observation of higher inconclusive test rates due to VUSs from multi-gene panel-based testing compared to genomic testing may be more of a surprise to many. However, when one considers the reporting practices of genetic testing laboratories based on current guidelines (Rehder et al., 2021), which is to report all VUSs when performing diagnostic (symptomatic) panel-based testing, such results are less surprising. The long-standing rationale has been that VUS reporting allows providers the ability to follow-up on VUS results with additional testing that can inform the pathogenicity of variants. For example, α-galactosidase enzyme activity testing in an individual with a VUS in the GLA gene can provide evidence to inform pathogenicity. Likewise, family member testing may identify de novo occurrence, cis/trans configurations of variants, or segregation with disease, all informing pathogenicity.
However, as panels have increased in gene content and therefore the range of included phenotypes broadened (e.g., pan-cardiac panel as opposed to a hypertrophic cardiomyopathy panel), the correlation between the phenotype in the patient and what is known about the disease associated with the gene harboring a VUS may range from weak to strong. In genomic testing, in which the same variants seen on panels would also be detected, but where the large quantity of identified VUSs cannot be reported, laboratories use different reporting rules where professional judgment is often required to determine which VUSs, if any, will be reported. These rules and judgments involve several factors, including: 1) if a clear genetic etiology is identified, VUSs in other genes are unlikely to be reported; 2) VUSs will be reported only if there is a plausible correlation of the gene’s associated phenotype(s) with the patient’s phenotype, with weaker correlation requiring more variant-level evidence of pathogenicity to report the variant; and 3) for variants identified in genes associated with recessive disease, lesser evidence for the VUS and the phenotype correlation is tolerated if variants are identified on both alleles (more informative through trio testing), whereas more evidence (at the phenotype and/or variant-level) is required if only a single heterozygous variant is identified on one allele. Some laboratories use sub-tiers of VUS to aid in this decision-making (Karbassi et al., 2017; Zouk et al., 2021), and this is reflected in the sub-tier results from the two laboratories reporting this data. No VUS-Low results were reported from genomic testing whereas 22% of VUSs were in this category for panel-based testing. In summary, detailed phenotyping is typically required for genomic analysis and allows for a comprehensive yet specific analysis of the genome and at the same time limits reporting of unrelated VUSs findings.
While the utility of reporting VUSs is clear in many circumstances, the high rate of VUSs observed in current multi-gene panel-based testing suggests that an examination of current practices is warranted, with a potential for changes in practice for both the laboratory and the ordering provider. For example, a general practice that could be altered is that for large multi-gene panels with broad indications, VUSs might only be reported if there is a strong correlation with phenotype, similar to how genomic testing reports are currently interpreted. This would necessitate improvements in the provision of patient phenotype information for panel-based testing by ordering providers, as well as more time for report generation given that clinical expertise would be required to review the patient’s phenotype and correlate it with the gene’s associated disease(s). Also, given the significant beneficial impact of trio-based genomic testing to both yield and a reduction of inconclusive results due to VUSs, it is worth considering whether trio-based testing should become standard practice for panel testing despite being rarely, if ever, offered currently (and not reimbursed), particularly for large panels with substantial genetic heterogeneity and phenotypic variability in disease presentation. And finally, the applications of algorithms to rank variants by gene-phenotype correlation, that are increasingly used in genomic analyses, may also prove beneficial for panel-based testing.
Another approach is to allow healthcare professionals ordering tests to opt-out of receiving any VUSs or only a sub-tier of VUS (VUS-High) at the time of ordering. This decision might be based on the ability to follow up on VUSs for a given patient, with providers considering: Do any genes on the test have clinical follow-up tests that can inform pathogenicity? Are parents or affected family members available for segregation testing? Does the patient want VUSs to be reported? What downstream costs might be incurred with follow-up? As an example of this practice, some laboratories now allow opt out of VUS reporting for large neurodevelopmental panels. It is also important to note that patients may have preferences about the receipt of VUSs, in part based upon the balance of distress versus future value, as experienced by patients (Skinner et al., 2018; Culver et al., 2013), and providers should ideally engage patients, where possible, in making these decisions.
Furthermore, wider use of VUS sub-tiers could allow more nuanced reporting practices and enable providers to direct follow-up to only those variants with the highest probability of pathogenicity. Given the amount of effort expended by physicians and potential costs to the healthcare system if physicians follow-up on all VUSs, restriction of VUS to only those variants with higher likelihood of becoming pathogenic would likely be welcomed by payors, providers and patients. Finally, VUS reporting guidance from professional organizations would likely improve the consistency of VUS reporting across laboratories.
In summary, this study identified the largest source of inconclusive results from multi-gene panel tests, whereas the use of genomic sequencing tests both reduced the inconclusive rate while improving diagnostic yield. This is best explained by the current laboratory practice for the reporting of all VUSs in panel-based testing, based on societal guidelines, compared to genomic testing where correlation with phenotype is used to constrain and limit the reporting of VUS. These results set the basis for future reporting practices and guidelines, may guide payor coverage in genetic and genomic testing, and provide a heightened appreciation for the professional skills deployed during genomic test interpretation.
Data Availability
All aggregate data produced in the present study are available upon reasonable request to the authors