Main findings
The five traditional reviews yielded little of the evidence necessary for decision-making by the UK NSC. This is belied by the large scale of the five reviews which took 7 months for the identification, selection and review of published studies and comprised sifting nearly 20,000 titles and abstracts and 1348 full texts. The 221 included studies addressed only two of the six review questions. No studies addressed the questions on penetrance or test accuracy. No studies addressed the questions on the clinical effectiveness and harms of genetic screening. In fact, we did not identify any studies that reported on WGS in the newborn screening setting for any of the five conditions. Between one and nine studies across each of the five reviews reported observations on the early versus late treatment, where ‘early’ tended to only approximate screen-detected children and no study was designed to compare the treatment in screen-detected versus symptomatically detected children. There was some indication that earlier is better across all conditions, but studies were observational, small and we had concerns over their quality mainly around reporting and insufficiently clear definitions of disease.
Between 26 and 89 studies across the 5 reviews reported the genetic make-up of patients with suspected or confirmed disease for the 5 conditions. These studies were of interest for two reasons. Firstly, we wanted to compare the variant spectrum in children with clinical disease with that in asymptomatic children detected by genetic screening to gauge how applicable variant information from children with disease is to the screening context. The example of MCADD illustrated that the variant spectrum and, consequently, the variant frequency shifts when moving from sequencing symptomatic children to sequencing those with biochemical risk factors (positive NBS test), suggesting that studies in children with disease have limited applicability to screening168 and genetic variants believed to be pathogenic can be very poorly predictive of symptomatic disease when detected by population screening. However, no studies were identified that used sequencing in unselected newborns. We were, therefore, unable to compare the variant spectrum in unselected and symptomatic newborns. This means that the review provided no information on the variant frequency, variant interpretation or penetrance in screening populations or on how much this differs from populations with confirmed disease (additional to the shift from symptomatic to biochemical disease).
Secondly, we wanted to determine the proportion of children with disease, which can be detected by sequencing to assess how many patients may be missed by newborn sequencing. This was complicated by the fact that the studies included different populations of patients, illustrating that disease can be defined in different ways. Before we can determine the detection rate of WGS in screening (i.e. the proportion of patients with disease identified through sequencing), a clear definition of disease is needed. This is not straightforward as the disease can be defined symptomatically or biochemically using various thresholds. Many diseases have similar clinical and biochemical characteristics and can only be diagnosed genetically. For instance, determining the proportion of children with XLHR caused by variants in the PHEX gene is complicated by the fact that XLHR has no disease-specific symptoms or biochemical markers. Defining XLHR symptomatically includes additional conditions caused by different genes, which affects the proportion detectable by sequencing the PHEX gene. This illustrates that measuring the detection rate of WGS in this way is impractical.
Our review identified other challenges for WGS for the screening of newborns. Firstly, the type of test had an impact on the genetic outcomes reported. This included, for instance, the scope of sequencing, that is, the number of exons covered and whether intron/exon boundaries were considered; the use of additional tests (e.g. MLPA to detect deletions and insertions), as studies generally used exhaustive genetic testing to identify genetic causes in clinically affected patients, and year of testing (since there is a rapid change in test development with the addition of genes and variants over time). This highlights potential concerns over the applicability of some of the study findings to WGS in the screening setting and limitations of WGS as a stand-alone test. Secondly, ethnicity had an impact on the number and type of variants identified.168 This limits the generalisability of some study findings to the UK context and highlights the importance of considering an ethnically diverse population in the evaluation of WGS for newborn screening and an evaluation of the potential impact of WGS on inequity and inequality in screening. Thirdly, we excluded several studies that focused on specific subgroups of patients. While unrepresentative to the review question, they identified particular challenges. For instance, patients with mosaicism are not identifiable by WGS, patients with digenic disease would be classified as negative and patients with atypical symptoms may not be identified if knowledge of their variants is limited. Finally, PDE and XLHR had a high number of novel and private variants for which pathogenicity can be difficult to ascertain. These and conditions with similarly high numbers of private variants may pose a challenge to WGS in screening because of potentially high numbers of variants of uncertain (or unknown) significance.
The ClinGen is a useful resource for information on the medical actionability of rare monogenetic diseases. Four of the five conditions reviewed have currently got an actionability report available on ClinGen. When compared to the UK NSC criteria, we identified two main limitations. Firstly, the availability of information on penetrance is not a requirement for the inclusion of a condition on ClinGen. Secondly, the benefits of early versus late treatment are not a focus of the ClinGen review. In addition, the evidence bar for a positive inclusion decision is lower than that for the UK NSC recommendations. As a result, it would be inappropriate for the UK NSC to base decisions for potential screening programs on the actionability reported in ClinGen without further assessment.
Our review of genomic studies of newborn screening populations provided information on international experiences with whole genome and panel testing for screening and illustrated that genetic screening is still in its infancy. Synthesis of the evidence revealed a lack of consensus on which genes to include, no indication of how to interpret discordant results from NBS programmes and genetic screening, uncertainty over which test is most appropriate for which condition, evidence of overdiagnosis and a large number of cases that could not be confirmed by the studies’ follow-up processes resulting in interventions, routine surveillance and regular follow-up of uncertain benefit. All of these would cause problems if WGS was implemented prematurely. The studies did not provide the information on variant penetrance and expressivity, which would be needed to determine a variant threshold for individual genes (number and types of pathogenic variants) to select variants for the evaluation for a screening programme that maximises benefits and minimises harm from overdiagnosis. One study reported 31 cases of PKU, of which 15 were caused by 13 different variants which were all confirmed by clinical presentation. This heterogeneity precluded penetrance estimates to be determined for individual variants even in this large study. The remaining 16 cases were clinically unconfirmed and involved a common variant that was believed to cause mild disease. This may represent one example where study information could be used to exclude a variant from consideration for reporting.
The three main approaches explored do not offer an immediate solution to the evaluation of WGS for newborn screening in the future. Our reviews highlight many evidence gaps and challenges for implementation, which may be a contributing factor to the wide variation between studies in decisions of which variants are reported to parents as potentially clinically significant. We conclude that information on the penetrance of pathogenic variants selected for inclusion in genomic studies in the screening context should be one focus of future research efforts.
Our review, unsurprisingly, identified no economic evaluations of WGS in a screening context. All the evaluations focused on individuals with suspected genetic diseases and explored the cost, cost consequences or cost-effectiveness of WES or WGS compared to the standard of care diagnostic testing or to different types of sequencing (rapid vs. non-rapid) or positions in the diagnostic pathway. Many of the studies based their evaluations on data collected from reasonably small (all < 1500) cohorts of individuals, either receiving the intervention or current standard of care or both. Most evaluations did not adopt a lifetime horizon for capturing costs and health outcomes, as recommended, and most were limited to outcomes and costs for the diagnostic trajectory rather than including costs and outcomes associated with changes in clinical management and the consequences of this change in management. Very few studies conducted a cost–utility analysis. This is likely due to a number of reasons. Firstly, WGS can produce a wide range of genetic information, affecting multiple conditions and health outcomes simultaneously. Where this was an issue in the included cost–utility studies, very broad classifications of health states and assumptions around outcomes were made to assign utility values. However, these broad classifications may oversimplify the complexity of the impact of WGS on HRQoL. Secondly, the long-term nature of the impacts of WGS on health outcomes means that the HRQoL needs to be measured or estimated over extended periods of time. Collecting these more detailed data over long periods of time can be challenging and resource-intensive. Finally, there is also the issue of estimating the causal effect of WGS on health outcomes, which would typically require a RCT design, which can be difficult to implement effectively for rarer diseases due to the large sample sizes needed to achieve an adequate statistical power.
Our update of the Schwarze et al.27 systematic review did identify a number of high-quality micro-costing studies, where a granular account of all the resource use associated with WGS or WES had been recorded and costed. These types of studies are incredibly useful as they can be updated as unit costs change over time, and they also help to understand which aspect of the testing pathway is driving the cost. One of the studies identified in the review was published as part of the SGP study, and we have been made aware that similar activity is underway but not yet published by GEL. We would encourage future UK micro-costing studies to be undertaken, particularly when conducted in a routine clinical practice setting, rather than research setting, to help underpin future UK-based cost-effectiveness evaluations of WGS. It would be useful to have these available across a range of clinical contexts to understand more fully when and why costs may differ.
In this review, standard of care testing that frequently included some sort of genetic testing was often listed as a comparator. Costs associated with standard of care varied widely between individuals, demonstrating the importance of developing large, heterogeneous cohorts of individuals to truly capture the breadth of resource use. Standard of care will be very difficult to define when evaluating WGS in a screening context unless there are clear comparative data on which conditions would have been detected later, when and using which diagnostics. Standard of care in terms of clinical management was rarely included in the economic evaluations in this review, and it is likely to add another layer of complexity, especially when trying to elucidate the benefits and harms associated with earlier diagnosis.
Contribution to existing knowledge
There are no previous systematic reviews of the benefits and harms of WGS of newborns and so our review is novel. We have not identified any other group who is approaching the question from the same perspective. The International Consortium on Newborn Sequencing is a consortium of international genomic projects aiming to bring together, exchange and harmonise ideas and efforts to responsibly implement newborn sequencing.169 This is a useful source of information, but the focus of the consortium is implementation rather than evidence synthesis for policy advisors. While the consortium website portrays certainty of health benefits from newborn sequencing, our review highlights the uncertainties and evidence gaps that could result in harms if WGS is implemented prematurely.
There are currently at least 15 genomic projects underway internationally aiming to sequence over 400,000 newborns.169 The study by Downie et al.170 compared gene lists across six of these genomics projects and reported substantial differences in genes’ lists due to different starting points, with some prioritising the clinical validity of the gene–disease association, while others prioritise treatability. We have observed similar disagreement in gene lists due to different processes and priorities used across the genomic studies included in our review of penetrance in newborn screening populations. This lack of consensus may reflect the uncertainty of what to report to parents as potentially significant, which is based on current evidence. The gene selection process is complex and multifactorial. It requires evidence to assess whether the condition is monogenic, whether the genotype/phenotype link is established by identifying pathogenic variants and the penetrance and expressivity of these variants. Evidence on pathogenicity, penetrance and expressivity is limited for general population cohorts. Because there is less proof of pathogenicity for likely pathogenic variants, there is no consensus on reporting likely pathogenic variants in screening, while VUSs are generally not reported. It is important to understand and evaluate the underlying gene and variant selection processes that underlie the decision on what to report to parents, and our work on this could be expanded in the future to a full systematic review to understand the different approaches to gene and variant curation and to synthesise the commonalities and differences to inform a robust and unified process.
The ClinGen resource represents one open access effort to develop and implement standardised, evidence-based methods to characterise the clinical actionability of genetic variations. However, our review found that the evidence bar for inclusion is low, the focus around treatment is clinical and therefore on treatability rather than benefits of early versus late treatment; and, while, some aspects are relevant to the screening context (e.g. treatment for pre-symptomatic children), focus is on the secondary findings in individuals in the diagnostic setting where the definition of an important condition may be different to the public health perspective of the screening setting.
Strengths, challenges and limitations
Strengths
To the best of our knowledge, this is the first attempt to identify an approach to assess the benefits and harms of WGS for newborn screening for policy advisors. This was a huge systematic review applying both traditional and novel approaches, and it has the potential to guide policy-makers towards a new approach for both synthesising and producing the evidence required to make evidence-based policy decisions in this complex area. We consulted widely with policy advisors and clinicians during protocol development to frame the right RQs for this review. We used a stratified random sample of conditions for our traditional review and published our protocol on PROSPERO. We consulted with clinicians specialising in the five disease areas, geneticists, clinical and laboratory advisors to the UK NBS programme and members of the UK NSC throughout the review process to ensure we deliver what is required by decision-makers and undertook an extensive piece of work to consider the public and patient views on this ethically challenging topic. While all these aspects ensured a high-quality review, we learned from several challenges that can inform subsequent reviews.
Challenges
Reviewing five complex, very different conditions at the same time was challenging. It is impossible to gain the knowledge and insight for five conditions in a similar time frame, as we would normally have to review a single condition. We attempted to work as one team for consistency in decision-making but had to focus on two to three conditions each. We attempted to use one review approach across all conditions, but we quickly recognised that the conditions required individual considerations. A review of 200 conditions would ideally require 200 review teams or an extensive time for a smaller number of teams to ensure that the rigour and subject knowledge can be developed for each condition. Issues around consistency would need to be factored in and addressed if several review teams undertake the evidence syntheses.
The outcomes presented here represent a snapshot of the evidence in this fast-progressing field. Gene and variant lists are constantly changing and developing in light of new evidence. New variants are identified, and variant classification changes based on new insights from clinical practice and research. Evaluations and policy decisions of 200 conditions once completed will not be up to date for long and will require a system that allows continuous evaluation.
The review highlights the challenge of defining disease because we were thinking across genetic, biochemical and clinical diseases. It is important to be clear about what aspects of disease we would aim to detect and prevent with WGS. This is linked to the challenging concept of penetrance, which depends on the definition of disease and at what time point we measure the disease. This will determine the required follow-up time of screen-positive newborns for the estimation of penetrance and is likely to be different for different conditions.
A further challenge was the integration of expert clinical advice into the review, and this will be a challenge in the future on how to incorporate clinical knowledge into the evidence base to inform a policy decision. Clinical knowledge was important and nuanced and was complementary to the review findings. The issue is that the information comes from individuals; however, in these rare diseases, it would be challenging to identify a group of experts large enough to establish a consensus. We were advised that (1) targeted gene panels would be more economical than WGS (and others share that view171), but this is not currently a question that is being addressed; (2) meaningful time scales from testing to reporting are unlikely to be achievable in the overstretched NHS where current waiting times for WGS results for the diagnosis of symptomatic children often exceed 1 year172 and (3) storage of huge amounts of data is expensive and may not be accessible in the future, while re-sequencing may be more appropriate than accessing stored data in the future. These are important issues that our current review did not ask or identify from the published evidence. Furthermore, one clinical advisor said that they would never initiate asymptomatic vitamin B6 (pyridoxine) because of treatment-induced seizures, yet this is the early intervention of interest here. This was an important finding but was not backed up by published studies. It was impossible for us to know how common this view is.
Limitations
Our approach of selecting the five conditions aimed to include a range of scenarios that would allow us to explore different challenges that the NHS would have to face with newborn sequencing. This meant that the conditions, the aim of screening for the conditions, the definitions of early versus late treatment and their underlying genetics were very different. Consequently, we were unable to use one approach to reviewing as planned and concluded that it is unfeasible to use one review approach for 200 conditions. However, we do not know whether this applies to a group of similar conditions, for example, metabolic conditions, where learnings from one review may be transferable to other conditions speeding up the review process. Repeating this approach with a group of similar conditions could shed light on this.
The number of studies reporting the genetic spectrum in children with confirmed disease was unexpectedly high for all five conditions. The studies were heterogeneous in terms of population, test and outcomes and would have been difficult to synthesise. The most appropriate population with disease to determine the detection rate for WGS was difficult to define for XLHR and fHLH because of non-specific symptoms. We, therefore, decided to categorise the studies by definition of the disease and extracted data from the largest study only (or more than one where results were complementary) to present the breadth of the evidence rather than aiming for completeness. The aim was to provide examples of outcomes depending on disease definition that could inform the inclusion and exclusion criteria of a future review for this question. However, this meant that we were unable to draw on all reported findings on variant frequency and expressivity or to highlight aspects where studies may have agreed or disagreed. We are uncertain how useful synthesising all the included studies would have been.
We were unable to undertake a number of explorations detailed in the protocol because the shortfall and type of data meant that they were unfeasible. We wanted to explore reporting of penetrance information from the genomic studies of newborn screening cohorts (1) by subgroups of conditions pre-defined by the studies, (2) by the GEL category the conditions would fall into and (3) combined for the conditions with a top quintile ClinGen score. The aim was to explore the feasibility of determining the variant threshold that corresponds with the most severe phenotype for conditions. However, this was not feasible because (1) categories are at condition/gene level and penetrance is reported on variant level, (2) the penetrance information reported in studies was inadequate numerically and qualitatively, (3) the ClinGen score was not reported on variant level and (4) the number of conditions covered in ClinGen was limited. This means that we were unsuccessful in exploring whether a variant threshold for individual conditions could be used to develop a restricted but safe screening programme.
Studies comparing genetic screening with traditional NBS screening suggest that tests are complementary in terms of cases they detect and miss, but they do not tell us anything about the place of WGS in an existing screening program; that is, how both tests could be integrated, how to interpret contradictory outcomes, whether the sequence of tests should be different for different conditions or whether newborn sequencing should be a completely separate programme. In general, the combination of metabolomics alongside genomics, in whichever order, is likely to add to our understanding of genetic variants and their significance, and considering all research in a future review will help to identify the best screening strategy for individual conditions.
Furthermore, there are several questions that we did not address or where there was insufficient evidence available from the studies we reviewed, but those are fundamental for the evaluation of newborn genetic screening.
- The differences in variant frequency and penetrance in screened (unselected) and symptomatic cohorts is important to understand, but data are lacking for most conditions. It is widely accepted that risk estimates for genetic disease from high-risk groups do not translate to the general population as has been shown for cancer predisposition genes in individuals with and without FH.173 And, in MCADD, a condition that is already screened for by NBS screening with genetic confirmatory testing, the shift from symptomatic disease to biochemical disease definition resulted in a different variant spectrum.119 In this study, a lower proportion of screen-detected newborns were homozygous for the c.985A > G variant, which is very common in symptomatic children. A significant number of the newborns had genotypes with variants that had not been observed in patients detected clinically and some, like c.199T > C and c.127G > A, were associated with a milder biochemical phenotype. This knowledge is important for all conditions considered for genetic screening and needs to be extended to understanding the variant spectrum in infants designated screen-positive based on genetic testing only.
- There was insufficient evidence to explore the differences in variant frequency and penetrance by sex, geographical region and ethnicity. Variants of uncertain (or unknown) significance are more common in non-European ethnic groups because of an under-representation in reference population databases used for variant frequency annotation.8 Consequently, there is a lack of genotype–phenotype correlations in populations, such as the UK and USA, with significant ethnic diversity. This is a particular problem for whole-population screening and means that it will be difficult to establish an equitable screening programme if ongoing genomic studies do not generate the evidence from an ethnically diverse population.
- The review did not address the question of whether WGS is the best test for expanded newborn screening compared to WES, panel testing, sequencing combined with other tests or biochemical assays. Many conditions could be tested for using biochemical assays or bespoke genetic tests, which could be superior to WGS.172 WGS has got limitations and is generally more expensive. It is unable to detect large deletions and needs to be adapted, for instance, to identify copy number diseases like spinal muscular atrophy (SMA) to ensure they are reliably detected. Furthermore, it can also produce inconclusive results as in cystic fibrosis screening.174 The gene-by-gene and exon-by-exon performance of WGS is improving all the time, but this remains a factor at the time of writing.
- Test failure of WGS was insufficiently reported by the included studies.
- We did not explore the impact of sequencing results on clinician behaviour and clinical care.
- We did not investigate workforce challenges, which will include the number of genetic counsellors needed, and training of staff to correctly interpret WGS results.
- We did not identify evidence-based pathways for those with rare asymptomatic genetic disease.
- More work is needed to explore ethical issues around knowing but not reporting certain variants.
- More work is needed to explore ethical issues of storing blood samples/genetic information, their re-use and who decides what can be done with the data.
- Existing resource use data associated with WGS need to be adapted/adjusted to account for the additional infrastructure and staff required to deliver WGS at scale, that is, in the context of a national screening programme.
- Long-term follow-up on cohorts of, ideally, asymptomatic newborns undergoing WGS is needed to understand the implications of testing on patient management in terms of costs and patient outcomes.