Half of germline pathogenic and likely pathogenic variants found on panel tests do not fulfil NHS testing criteria

Genetic testing for cancer predisposition has been curtailed by the cost of sequencing, and testing has been restricted by eligibility criteria. As the cost of sequencing decreases, the question of expanding multi-gene cancer panels to a broader population arises. We evaluated how many additional actionable genetic variants are returned by unrestricted panel testing in the private sector compared to those which would be returned by adhering to current NHS eligibility criteria. We reviewed 152 patients referred for multi-gene cancer panels in the private sector between 2014 and 2016. Genetic counselling and disclosure of all results was standard of care provided by the Consultant. Every panel conducted was compared to current eligibility criteria. A germline pathogenic / likely pathogenic variant (P/LP), in a gene relevant to the personal or family history of cancer, was detected in 15 patients (detection rate of 10%). 46.7% of those found to have the P/LP variants (7 of 15), or 4.6% of the entire set (7 of 152), did not fulfil NHS eligibility criteria. 46.7% of P/LP variants in this study would have been missed by national testing guidelines, all of which were actionable. However, patients who do not fulfil eligibility criteria have a higher Variant of Uncertain Significance (VUS) burden. We demonstrated that the current England NHS threshold for genetic testing is missing pathogenic variants which would alter management in 4.6%, nearly 1 in 20 individuals. However, the clinical service burden that would ensue is a detection of VUS of 34%.

www.nature.com/scientificreports/ Studies that have forgone family history-based testing criteria. Researchers and healthcare professionals have explored the ramifications of simplifying access to testing, and eligibility criteria across a range of populations (Table 1).
Our Aim. Our aim was to determine the extent to which more actionable genetic variants are returned by panel testing compared to those that would be returned by criteria-dependent NHS testing of the same genes.

Methods
Participants. This study was conducted at a private oncogenetics clinic in London. Patients were included in the study if they underwent cancer predisposition genetic testing between 2014 and 2016. Genetic pre and post-test counselling was provided to all patients as part of standard clinical care. During clinical consultations, discussion of ancestry, particularly Ashkenazi Jewish ancestry, was highlighted as this influenced targeted testing for pathogenic founder variants. Based on the suspected cancer predisposition, specific gene tests were requested for patients. All patients gave their informed consent to have the panel test undertaken by the clinician, and no additional blood samples were taken for this study. Illumina sequencing was conducted in a UKAS-accredited private laboratory. Variant classification and interpretation were delivered to patients as part of their standard clinical care. The ACGS/ACGM classification was used for this period, prior to the availability of CanVig classification guidelines. Approval for this study was obtained from the Royal Marsden Hospital audit committee. All methods were performed in accordance with hospital guidelines.
Data collection. Electronic medical records and pedigrees were retrospectively reviewed for information on demographics, and oncological history including age at diagnosis. All patients had an associated family pedigree where data on cancer family history was gathered at the time of the consultation. Box 4 shows information collected on all patients.
Every panel undertaken was compared to the 2020/2021 National Genomic Test Directory Testing Criteria for Rare and Inherited Disease 39 using the patient information available at the time of the clinical consultation.

I. Individuals Affected by Cancer vs. Unaffected Individuals
The VUS rate was similar (34% affected, 35% unaffected). In unaffected individuals undergoing testing, there was a higher P/LP rate (13%) compared to affected patients (8%), however this was not statistically significant with a Chi-square test p-value of 0.58.

II. Age of Cancer Onset in Affected Individuals
The 98 patients undergoing diagnostic testing were divided into 2 groups: age of onset of cancer ≤ 50, and > 50, and the variants in each category were assessed. Associations between age and the following two variables were analysed: the number of P/LP variants observed, and the distribution of specific genes detected. Chi-square tests were both not statistically significant: P = 0.32 for the number of P/LP detected, and P = 0.55 for the distribution of the specific genes across age. In conclusion, there were no significant differences in gene panel results based on age at cancer onset in this study.

III. Fulfilment of Eligibility Criteria
All individuals in this study who underwent panel testing in the private sector were compared against eligibility criteria utilised by the NHS. 37% of patients (56 of 152) fulfilled NHS eligibility criteria. Summary statistics are displayed in Fig. 4. Comparing these two groups, there was a statistically significant difference in variant classification rates between those who were eligible for testing and those who were not, as expected.
Focusing on the pathogenic/likely pathogenic variants. Of the 15 P/LP variants detected by panel testing, 7 fulfilled criteria, 7 did not fulfil criteria, and 1 person was adopted. 46.7% of patients who received positive results in this study did not fulfil NHS testing criteria (Fig. 5  www.nature.com/scientificreports/ A closer look at the families of the 7 patients with P/LP variants who were not eligible for testing: In 2 cases where an APC pathogenic variant was detected in patients who did not fulfil criteria, a different member of their family did. Conversely, the remaining 5 families, although there was a history of cancer, also did not fulfil testing criteria.

Management of VUS outcomes. For the 52 individuals whose gene panel test yielded a VUS, all patients
were informed and received pre and post-test counselling. The subject's records were placed under annual review of VUS status. It is worth noting that some patients with VUS required additional cancer surveillance based on family history criteria, not the detection of a VUS. Over the 3-year period, only 1 VUS in BRCA1 was re-classified to benign in a patient with triple negative breast cancer. The VUS rate in individuals from a non-European origin was 39%, versus 33% in Europeans.
No P/LP variants were unexpected. Patients in whom P/LP variants were detected, the findings were consistent with either the personal or family history. There were 2 patients with breast cancer and heterozygous variants in MLH1 and MUTYH. This, however, was consistent with their family histories of colorectal cancer.
Clinical utility-Effect on the tested individual. 13 of 15 P/LP results changed management for the patient (Fig. 6). The first exception was a woman with breast cancer with a heterozygous pathogenic variant in MUTYH. Due to the autosomal recessive inheritance of the syndrome, neither prophylactic surgery nor cancer screening was recommended. However, the genetic testing did trigger predictive testing of her partner, due to implications for their child. The second exception was a patient in the palliative stages of advanced colorectal cancer.
Clinical utility-Effect on the family. 15  All VUS detected in our study were reviewed yearly and patients were contacted by letter if any management was altered. Limitations of generalising this practice in the UK are that this degree of follow-up is not currently possible on the NHS. The additional resource burden associated with VUS may also entail familial segregation studies, tumour studies and functional testing. International academic institutions are working to define a standardised approach in clinical practice and policy 45 . Research groups are also endeavouring to assign degrees of risk of deleteriousness to VUS, which may aid clinical decision-making. Following the ACGS variant classification guidance published in 2020, the threshold for reporting VUS is now higher, and only 'hot' VUS should be reported after MDT discussion 46 . Current reporting policy offers promise in reducing the extent of the burden of VUS outcomes. Many VUS transpire to be benign on further follow-up.
Eligibility criteria. Current tests offered in the NHS are restricted to patients who fulfil eligibility criteria.
From our study, unrestricted testing of patients returns more actionable genetic variants than current eligibility criteria allow. Our study returned double the number of actionable genetic variants than NHS criteria. Overall, 7 additional P/LP variants were detected in the private clinic, in addition to the 7 that would have been detected on the NHS. Missing this proportion of P/LP variants (46.7%) is supported by prior research 47 .
Given that these eligibility criteria are evidence-based and designed to identify high-risk patients 48 , the higher P/LP detection rate and lower VUS rate in the group who fulfilled criteria was expected.
Number needed to screen and risk reduction. Number needed to treat is a common statistical method used for assessing robustness of interventions, from drugs to screening programmes. A similar statistical method applicable to screening is the Number Needed to Screen (NNS) 49,50 . We applied this statistic to our data to identify how many patients need to undergo gene panel testing to find 1 P/LP variant. Results and calculations are shown in Fig. 7.
Although the same number of P/LP variants were found in both patients who fulfilled and did not fulfil NHS criteria, 6 additional patients were screened in the ineligible cohort. Overall, only 3 additional patients needed to be screened to find a P/LP variant.
We also tried to model to what extent additional testing is likely to improve survival. A large-scale metaanalysis conducted by Rebbeck et al. investigated the risk reduction of prophylactic salpingo-oophorectomy in carriers of P/LP variants in BRCA1 and BRCA2 51 . Prophylactic surgery conferred a reduction in breast cancer (hazard ratio = 0.49) and ovarian and fallopian tube cancers (hazard ratio = 0.21) 51 . In other words, there is a 79% risk reduction in ovarian and fallopian tube cancers in pathogenic BRCA1/2 carriers with risk-reducing surgery. Utilising this data, as well as mortality figures of ovarian cancer in pathogenic BRCA1/2 carriers, mortality is reduced 4.7-fold for BRCA1, and 46-fold for BRCA2. All calculations are demonstrated in Tables 3 and  4 for BRCA1 and BRCA2 respectively, and were based on the following assumptions: everyone will have had a prophylactic oophorectomy, patients have the same survival benefit as the documented literature, and all P/LP variants behave the same way.
In this study, there were four P/LP BRCA1/2 variants detected, one of whom was not eligible for testing and tested positive for a pathogenic BRCA2 variant which would have otherwise been undetected. Her risk of death from ovarian cancer was reduced 46-fold according to the above calculation (Table 4).

Unexpected findings.
In two patients with breast cancer, heterozygous P/LP variants were detected in MLH1 and MUTYH. Although a pattern of colorectal cancer was apparent in their family history, recent evi- www.nature.com/scientificreports/ dence does not define an association with breast cancer, except possibly in MUTYH carriers of two pathogenic alleles 54,55 . Both variants prompted cancer surveillance and/or family screening. An important note is that both of these genes are on the UK consensus of genes for medically-actionable conditions 1 . By and large, UK experts, drawing on their experience of the 100,000 Genomes Project, urge for clearer policy on the interpretation and reporting of secondary findings to patients 56 and this is an avenue for future research.
Clinical utility. The question posed here was as follows: will the variant alter clinical management compared with management based on family history alone? It is important to recognise that even without a detected variant, patients can be at an increased risk of cancer due to family history 27,28 . For example, if a 20-year-old woman has a first degree relative with breast cancer and an ATM pathogenic variant, and her own test is negative, her risk is still raised above the population 28 . In our study, 87% (13 of 15) of results changed management for the patient. Even the two patients who had already received prophylactic surgery were additionally enrolled into cancer surveillance programmes. As aforementioned, the patients in which the results were not clinically actionable were a patient undergoing colorectal cancer treatment with a P/LP MSH6 variant, and a heterozygous carrier of MUTYH (an autosomal recessive condition). However, 100% (all 15) of P/LP variants changed management for the families. This high rate of actionability is not discovered in all cancer gene panel studies 57 . Limitations. Demographics. As with any scientific study, this study too is not without limitations. These limitations are important to understand to enable critical review so that future studies may tackle what is left unanswered. The demographics are not representative of a general UK population: of 152 patients, 75.7% were female, and some ethnic minorities were over-represented. This is compounded by the small sample size. The skewed demographic may explain why common European P/LP variants, such as CHEK2 1100delC 30,58 , were not detected.
Additionally, we only included private patents from a single centre, and as such, local socioeconomic factors such as income and educational attainment could influence behaviour. Unlike on the NHS, all patients attending the centre were actively seeking testing. We also recognise that NHS eligibility criteria serve as guidance to healthcare professionals, and eligibility is also reviewed on a case-by-case basis 30 . Other differences in genetic testing between the NHS and private care include patient refusal, and not attending clinic appointments.
Data Collection and Analysis. We were limited by our reliance on family history and medical records which are not always complete. To evaluate psychosocial burdens, future studies including patient concern and satisfaction levels would be of value.
Cancer gene panels: promise and pitfalls. The promise of cancer gene panel testing. Panels tests for germline testing have generated enthusiasm amongst clinical genetics services and increased our knowledge base of cancer genotype-phenotype associations 21,78 .
Gene panels offer the promise of time-efficiency by testing multiple genes simultaneously 59 , which is especially useful for genetically heterogeneous conditions 48,60 and in patients who have previously tested negative in genetic tests 21 . Massively parallel sequencing is also cost-efficient utilising current advances in technology and requiring a smaller amount of DNA 1,6,21,61 . Table 3. Calculating risk of death from ovarian cancer in P/LP BRCA1 variant carriers. 9.4% / 2% = 4.7-fold reduction in risk of death from ovarian cancer in P/LP BRCA1 carrier.
Before risk-reducing bilateral salpingo-oophorectomy After risk-reducing bilateral salpingo-oophorectomy  www.nature.com/scientificreports/ Panel testing is paving a path towards the future of personalised medicine. Detecting clinically-relevant information through gene panels leads to early cancer detection and possible changes in management 45 , more so than conventional gene testing 30,62 . Risk stratification 32 and clinical outcome prediction 31 could be more accurate through the use of panel tests. Identification of variant status also has prognostic and therapeutic benefits. For example, carriers of P/LP BRCA1/2 variants with breast cancer display sensitivity to platinum chemotherapy and PARP inhibitors 7 .
Pitfalls of cancer gene panels. The major challenge to gene panels is that our ability to interpret lags far behind our ability to sequence. The detection of VUS both in predictive and diagnostic testing renders the process more complex, both in terms of clinical management and counselling 45 . VUS have ramifications for clinical genetics services, with need for pre and post-test counselling, and our inability to explain ambiguous results of unknown deleteriousness to patients may cause more patient concern 45,63,64 . Studies on patient perspectives regarding gene panels have demonstrated their concerns about VUS rates, while recognising the potential positive impact on their health 65 , and the trust they place in healthcare professionals communicating genetic testing information 66 . The latter highlights the need for a more genome-educated medical community 64,66,67 , and dedicating resources to pre and post-test genetic counselling to help cope with the practical and psychological burdens associated with gene testing 21 .
Although standardisation of variant classification has proven difficult in the past 34,68,69 The Cancer Variant Interpretation Group UK (CanVIG-UK) as well as the 2020 ACGS 46 approach facilitate a standardised approach to classification and data sharing within NHS diagnostic labs. Internationally, the Clinical Genome (ClinGen) is striving to standardise approaches to variant classification, with gene-specific expert panels and a forum to resolve discrepant classifications.
There is also a need to standardise diagnostic reports on VUS to clarify to healthcare professionals the medical guidance associated with that variant 64 . Due to disparities in access to both research and genetic testing 12 , nonwhite ancestries have higher VUS rates 70 as their reference panels are less well-defined, and ancestral branches differ in their genetic variation. Another area of ongoing research is the contribution of multiple variants to predisposition 29 .
VUS aside, detection of a P/LP variant is not always straightforward. If the variant is moderate risk, or is in a moderate-risk gene, parameters including risk estimates, phenotypic features, and actionability are not well-defined 29 . A discrepancy between genotype and phenotype may be detected, such as a moderate-risk gene in a high-risk family, or a highly pathogenic variant in a patient with unrelated symptoms. Such incidental or secondary findings are well described in the literature 6,30,31 . A well-known example is demonstrated by the Ohio Colorectal Cancer Prevention Initiative, where a high number of individuals with colorectal cancer had BRCA1/2 variants 18 , despite evidence showing BRCA1/2 does not confer a colorectal cancer risk.
Finally, there is concern that use of gene panels will raise ethical tensions, and large-scale implementation of gene panels needs to be both fair and appropriate. Factors to consider are financial sustainability, regulatory factors, and issues of equity 71,72 . At the rate at which gene tests are becoming available, it can be difficult to evaluate the genetic tests by the ACCE Framework, and their economic sustainability in universal healthcare systems such as the NHS 73 .
Key points. Eligibility criteria attempt to strike a balance between identifying high-risk patients and minimising the level of uncertainty. We have demonstrated that cancer predisposition variants are not always associated with a significant family or personal history. Almost half of all patients with a P/LP result would not have fulfilled eligibility criteria for genetic testing and would therefore have been missed. Our results support a genetic testing policy that is less stringent, but testing for genes with a high degree of association with personal or family histories.
Looking to the future: proceed, but with caution. There are multiple arguments to be made in favour of a broad approach to gene panel testing. Sequencing a larger number of affected and unaffected patients, then collating variants and associated features into a database will, in the long term, lead to enhanced classification of variants and understanding of their associated phenotypes across a multi-ethnic population 1,30,45 . Laboratories contribute coded germline findings to the Public Health England cancer registry, which is becoming an increasingly useful resource in practice, informing both prospective variant classification and assisting with review of existing classifications. The full potential of linking genetic data to the cancer registry has yet to be realised.
Limiting sequencing to patients with a strong personal or family history biases risk estimates, whereas a broad approach may reduce oversampling of severe cases and allow more reliable quantification of risks 32,79 . As our knowledge of individual variants and ability to call phenotypic consequences progresses, eligibility criteria may become more of a hindrance to both clinicians and patients 32 . Finally, we have yet to tap into the full scope of actionability of such variants. Development of targeted therapies 74,75 and cancer screening algorithms based on genetic profiling 76 are active areas of research.
There is disagreement on how to broaden our scope; how we go about simplifying access to genetic testing. This is an important area of future research. One approach is enhancing current policies on diagnostic testing of patients with cancer to enable more efficient and cost-effective cascade testing 26 . Delivering cancer predisposition gene testing through mainstreaming into routine oncological care would require remodelling of educational curricula and diagnostic clinical procedures, but can simplify access to genetic testing and benefit more patients 77 . The UK cancer genetics community is broadly in support of mainstreaming as specialist cancer genetics services do not have the capacity for the increasing demand 33 . Another approach is population-based www.nature.com/scientificreports/ screening to identify a cancer susceptibility gene, which may be a costly process, but it is argued to realise the true potential of screening 47 . Our major bottleneck is accurate variant interpretation in the new era of big data. This bottleneck is the root of healthcare professionals' hesitancy in embracing gene panels: ambiguous variant classification may trigger anxiety in patients, and at worst, lead to inappropriate tests and interventions 21 . The powerbrokers are bioinformaticians with an understanding of biology, or healthcare professionals and researchers equipped with the tools for understanding big data, who can transform raw primary sequences into variants with nuanced interpretations. This is another important area of future research.
To build a UK community of experts to deal with the influx in big genomic data, the Chief Medical Officer Professor Dame Sally Davies published her report 'Generation Genome' , proposing some policies to lead the UK into the genomics era. These include centralised data analysis, harmonising genomics information across healthcare institutions, focusing on patient partnerships in patient-centric trials, and a focus on genomic education 67 . Some of these policies have sprouted with the 100,000 Genomes Project, helping us tap into the benefits of our single point-of-access system. Future local and regional UK-based research would provide a better idea of the current state, and where changes may be implemented.
There remain concerns about a less-restrictive policy which would need to be validated and reproduced prior to integration into clinical care. However, the only way to ensure the robustness of our data is to undertake more testing and validate this data. If validated, we have shown that for every 11 people tested, we find one P/LP variant which would alter management.

Data availability
The datasets generated during and analysed during the current study are not publicly available as although anonymised, they contain information that could in theory be identifiable. They are available from the corresponding author on reasonable request.