Results

Using a traditional review approach

This traditional review approach covers a stratified random sample of 5 of the 200 rare diseases included in GEL’s Generation Study6 to establish the evidence base per condition and to provide a reference case for comparison with alternative review approaches.

Workload and implications for scaling up to more than five conditions

Extensive scoping, refinement, testing and running of the electronic searches for the five conditions took a total of 6 weeks. We cannot envisage any way in which this could be simplified or shortened in duration. A traditional review for 200 conditions would require 200 individual searches to be developed.

The average time required to sift 100 abstracts was 40 minutes across five reviewers with varying levels of systematic review experience. A single full-text sift took an average of 5–10 minutes, depending on the question and the condition. Table 6 reports the approximate time required for searches and sifting activities for the 5 conditions, with extrapolation to 200 reviews of 200 conditions. This does not include time for double sifting, data extraction and quality assessment, discussion of disagreements and categorisation of studies for synthesis purposes, that is, by population for genetic testing (review question 2) and by timing of treatment initiation (review question 4). These time estimates, therefore, only illustrate the extrapolation of two reviewing tasks and do not reflect the time needed for a complete review process. The complete review process took a team of three full-time and two part-time reviewers 7 months to complete. We anticipate that a single, similarly sized review team would need about 280 months or 23 years to review and synthesise the evidence for 200 conditions. The approximate time estimates demonstrate the scale of the effort required to evaluate WGS on a condition-by-condition basis, although some learning may be transferable to conditions similar to those reviewed here.

Volume of evidence for the six review questions for five individual reviews

We sifted a total of 19,689 title and abstracts, of which 1348 were selected for full-text assessment (range 55–449 per condition). A total of 268 studies were eligible for inclusion across the five reviews (range 31–78). Table 7 summarises the search and eligibility results across conditions, with flow diagrams per condition presented in Appendix 4, Figures 2–6). The excluded studies and reasons for exclusion are listed in Report Supplementary Material 2. Extrapolating to a review of 200 conditions, we could expect as many as 787,560 titles and abstracts and 53,920 full texts to be sifted.

No evidence was identified for four of the pre-specified review questions for any of the five selected conditions (Table 7). All four questions were those for which studies were required to be conducted in newborns:

• penetrance of the condition in a newborn cohort where sequencing was the first line screening test (Q1)
• test accuracy of WGS in the newborn screening setting (Q3)
• clinical effectiveness of WGS in newborns (Q5)
• harms or additional benefits from WGS in newborns (Q6).

Evidence was identified only for the two review questions focused on individuals with the conditions of interest (Table 7), that is:

• the prevalence of genetic variants and genetic spectrum in a paediatric population with the condition of interest (range 29–89 studies) (Q2)
• evidence for early versus late treatment, where ‘early’ was the closest approximation to management of screen-detected cases that were identified (range 1–9 studies).

Considering that the five conditions are rare diseases, the number of identified records was larger than expected. For some conditions such as PDE, data for the same study participants are likely to have been represented in multiple publications by different authors, but this was not always clearly identifiable within the study reports. Overall, however, the five individual reviews yielded very little evidence of the sort required by the UK NSC.

Findings from the traditional review approach for the five conditions

Gene/variant frequency in patients with condition(s) of interest

Review question 2 looks to identify the proportion of infants/young people up to 18 years with either (1) biochemical or (2) biochemical and clinical features of the five conditions (PDE, hRB, XLHR, fHLH and MCADD) that carry the gene variants known for the conditions. This question can be considered in multiple parts. First and foremost, the idea was to compare the genetic spectrum in children with disease to that in sequencing positive newborns (question 1) to assess the difference of clinical and genetic diseases. Secondly, to identify the proportion of individuals with the conditions who have an underlying genetic cause (that can be identified by WGS) and the proportion of patients who may be missed because of non-genetic cause of the condition or because of non-specific symptoms caused by variation in a different gene. Thirdly, consideration of the genetic spectrum in those with the condition to identify whether recurrent (or ‘reported’) genetic variants are responsible for large proportions of cases or whether novel variants that can be more analytically intensive to identify are frequently responsible. Finally, any reported patterns of expressivity associated with different variants or types of variants were considered.

The extent to which identified studies can inform these questions depends on the type of genetic test used and the inclusion of specific genes or genetic variants in the studies as well as the nature of the included target population. The target populations identified did not easily fit into the two pre-specified categories of clinically defined and clinically/biochemically defined disease and extended to genetically defined disease. Disease-specific categories were therefore used, which ranged from clinical symptoms, with or without biochemical markers, to study inclusion based on the presence of a genetic variant (Table 8). The categories were based on the expectation that the differences in the definition of disease may have an impact on the genetic spectrum reported.

Table 8 summarises the volume of evidence by disease definition for the five selected conditions. These definitions were condition-specific according to the availability of biochemical markers (not available for hRB), the number of umbrella terms of disease with overlapping symptoms (XLHR, hereditary hypophosphataemia and hypophosphataemia) and whether conditions can only be defined by genetic information (four conditions of fHLH).

Below, we report the largest most representative study from each subgroup per condition; some subgroups have more than one study where results are complementary. Details per study regarding gene/gene variant frequency are reported in Appendix 5. An overview of all included studies is included in Report Supplementary Material 3. The detailed quality assessment of extracted studies is presented in Appendix 3, Tables 17, 18 and 20 and summarised below for each condition. Given the differences in populations and variant frequencies in the selected studies resulting from the different disease definitions, synthesising across studies was challenging. We therefore start with a short summary of results across the disease definitions and then describe results in detail per disease definition beginning with the broadest, with a particular emphasis on the disease category that is most relevant to WGS in newborns. Because the frequency of genes and variants identified in a population does not depend on the sequencing method, we did not prioritise studies based on sequencing method used.

Pyridoxine-dependent epilepsy

The three subgroups and studies for PDE were: childhood-onset, pharmacoresistant seizures,30 clinically or biochemically defined PDE31 and genetically confirmed PDE (all with variants in ALDH7A1).32,33 Details per study are provided in Appendix 5, Table 21. Three of the four studies were retrospective,30–32 including one multicentre registry-based study.32 Two studies exclusively used sequencing to identify variants, two used targeted panels31,34 and one used WES followed by short- or long-read genome sequencing (GS) for cases not solved by WES.30 The registry-based study did not report the genetic tests used to identify variants.32

Quality assessment

Of the four extracted studies, only Boonsimma et al.30 was considered to have met all quality assessment criteria. This was a single-centre study with clearly defined population and period of recruitment, and the sequencing methods were described in detail and cases and variants were adequately ascertained. The study also broadly indicates the population of patients with unexplained infantile-onset seizures who may benefit from newborn WGS. We had several concerns about the other three studies extracted based on lack of information about the selection of participants, level of detail provided concerning the participants33 or testing methods used.31,32 Two of the four studies were considered to report the genetic spectrum of PDE–ALDH7A1 based only on sequencing techniques.30,33 It is worth noting, however, that the majority of PDE presents in the first 4 weeks of life such that the window of opportunity for WGS to benefit infants prior to developing seizures is small.

Summary of results

Using the broadest definition of a target population who might be suspected of having PDE, up to 6% could be confirmed as having ALDH7A1 variants following genetic testing (based on a sample of 103 children with infantile-onset (< 12 months of age) pharmacoresistant seizures.30 Restricting the population to those with a pyridoxine response increased the percentage detected by WES to 11% (4/35), 75% (3/4) of whom also had elevated biochemical markers for PDE.31 Studies of patients with clinical and biochemical indicators of PDE have reported higher percentages with biallelic ALDH7A1 variants of up to 86% (18/21).35

The international registry-based study of patients with clinical suspicion of PDE and at least one ALDH7A1 variant allele32 provides the most comprehensive current picture of the genetic spectrum of patients with PDE, with data regarding Chinese patients with PDE reported in Jiao et al.33 The majority of patients in both studies had compound heterozygous ALDH7A1 variants (58.9%, 109/18532 and 83.9%, 26/3133) compared to homozygous variants. Missense variants were the most commonly observed, accounting for 57% of variant alleles32 and 65% of all variants.33 Although four individual (missense) variants accounted for 38% (140/367) of all variants identified in Coughlin et al.,32 the majority of identified variants each occurred in a single individual. The occurrence of novel and often ‘private’ variants was a recurrent phenomenon across the included studies (17/26 variants occurred in a single individual in Jiao et al.33), with sequencing of one or both parents to confirm pathogenicity of identified variants commonly reported (e.g. in two of six children reported in Boonsimma et al.).30

Results for each subgroup by definition of disease

Childhood-onset pharmacoresistant seizures

The broadest category defined for PDE is a clinically defined population of children with early-onset (neonatal or at latest infantile-onset) seizures that have been shown to be resistant to one or more standard pharmacological treatments for epilepsy (i.e. refractory or pharmacoresistant). This is the group in whom PDE might be ultimately suspected on clinical presentation. Potentially, hundreds of genes can be associated with epilepsy; however, only a relatively small number (including ALDH7A1) have therapeutic implications (i.e. identification of the gene directly informs treatment options). Other (treatable) pyridoxine responsive seizures include pyridoxal phosphate-responsive seizures (resulting from variants in PNPO) or pyridoxal phosphate-binding protein (PLPB) deficiency (resulting from variants in PLPB).

Boonsimma et al. included 103 cases with infantile-onset (age ≤ 12 months) pharmacoresistant epilepsy that was seen or referred for genetic testing at a Thai tertiary care centre.30 Genes associated with genetic epilepsy syndrome (n = 728) were initially targeted using WES, followed by sequencing for additional selected candidate pathogenic variants in unsolved cases (only one heterozygous variant was identified).

Of 103 cases, 6 (5.8%) were identified as having biallelic variants in ALDH7A1; 4 (66%) were identified on initial WES and 2 (33%) required short- (n = 1) and long-read (n = 1) sequencing to identify two novel variants (all cases and both parents underwent sequencing). Two additional patients were identified as having ‘treatable’ disorders, one with a biallelic PNPO variant and one with a BTD variant. An additional 36 patients were identified as having genetic variants that could ‘inform’ treatment decisions.36

A total of eight ALDH7A1 variants were identified in the six patients, five recurrent and three novel variants [two-thirds were copy number variants (CNVs)]. Of the five recurrent variants, two occurred in more than one patient within the study sample, suggesting a possible founder effect (one was identified in four patients, and one in two patients).

Clinical–biochemically defined pyridoxine-dependent epilepsy

Traditionally, PDE was clinically defined based on seizure recurrence (increase in number or severity) following pyridoxine withdrawal. Increasingly, urinary- or blood-based biomarkers [elevated alpha aminoadipic semialdehyde (α‐AASA), piperidine-6-carboxylic acid and pipecolic acid concentrations] are used to define PDE and may help to distinguish it from other pyridoxine responsive seizures such as pyridoxal phosphate-responsive seizures (PNPO gene) or PLPB deficiency (PLPB gene). As PDE is an autosomal recessive disorder, those affected are usually children of unaffected carriers of the ALDH7A1 variant such that FH is not a strong indicator of the likelihood of PDE, although more than one sibling in a generation can be affected. Further, there is an increased risk in consanguineous families. It is important to note that this category includes those who have either clinically, or biochemically defined PDE, or both, and this difference in disease definition may have an impact on results obtained from sequencing.

Koul et al. included all children (n = 35) from a single centre with refractory neonatal or infantile seizures that did not respond to antiepileptic drugs, but later responded to pyridoxine.31 Only 7.9% (3/35) had elevated biochemical markers for PDE (pipecolic acid and AASA). All probands received ‘targeted variant testing in ALDH7A1’, but no further detail was given to describe the nature of this test. If the initial test was negative, WES was conducted on multiple genes, including PLPBP, PRRT2 and ALDH7A1. The full spectrum of genes considered for WES was not reported.

Of the 35 cases, 4 (11%) were identified as having a variant in ALDH7A1, including the 3 with biochemical indicators of PDE, but it is unclear whether these variants were detected by the first-line testing strategy or by WES. The number of patients with homozygous and compound heterozygous variants was not reported. Twelve (34%) patients were identified as having a PLPHP variant, and 2 (6%) had a variant in the PRRT2 gene, detected through WES. Specific variants were not reported.

Age at seizure onset and developmental delay were reported separately for patients positive for ALDH7A1 variant (n = 4), those positive for a PLPBP variant (n = 12) and those with neither of these variants (n = 19). Age at seizure onset was lower for those with ALDH7A1 variants (30 minutes–1 hour), and delayed development was reported in a higher proportion [75% (three-fourths)] compared to the other groups. For those with PLPBP variant, seizure onset was between 1 hour and 10 days old and 17% had developmental delay; for those with neither variant, seizure onset was reported at 4 hours–29 months and 37% (7/10) had developmental delay. It is not clear to what extent these differences are a result of selection into the study, and furthermore, based on the small numbers reported, results cannot be considered to be representative of the whole population of patients with clinically and biochemically defined PDE.

Gene (ALDH7A1)-positive pyridoxine-dependent epilepsy

The most tightly defined populations are those defined genetically. PDE–ALDH7A1 refers to PDE resulting specifically from variants in ALDH7A1. The primary focus for studies in this group is to characterise the genetic spectrum associated with ALDH7A1 variants.

Two studies are reported. One reported an international study, including 185 patients with clinical suspicion of PDE and at least one confirmed pathogenic variant in ALDH7A1.32 Participants were recruited from four clinical genetics laboratories that perform clinical testing of ALDH7A1, the international registry (which appears to cover North America, Europe, the Middle East and Australia) for PDE37 and the pyridoxine-dependent seizures patient registry.38 Specific details of the genetic tests used to identify genetic variants was not reported. The second study, based in China, included 33 participants from 31 families, 31 with PDE–ALDH7A1 and 2 with PLPB variants.33 Selection into the study appears to have been based on genetic testing results. A targeted polymerase chain reaction (PCR) panel was used to sequence each exon (1–18) and exon–intron boundary of the ALDH7A1 gene.

Biallelic variants in ALDH7A1 were identified in 98% (182/185) of patients,32 resulting in 367 alleles with variants; 3 patients had only 1 variant allele identified (1.6%). The percentage of patients with compound heterozygous variants was 58.9% (109/185),32 and 84% (26/31)33 of the remaining patients were homozygous. Jiao et al. reported two (6.5%) patients as homozygous for ALDH7A1; however, the supplementary table to the report clearly reports five patients as having the same genetic variant on both alleles, indicating homozygosity (16% of total).33 The reason for this discrepancy is not clear.

Types of variants were reported slightly differently between the studies; however, some similarities can be observed. In the registry-based study,32 209 (57%) of 367 variant alleles were missense [compared to 65% (17/26) of variants identified in Jiao et al.33], 66 (18%) were splicing errors [compared to 12% (n = 3) of variants33], 29 (8%) were inDel (insertion and/or deletion of nucleotides), 29 (8%) were single nucleotide variant (SNV) terminations, 18 (5%) were synonymous SNVs and 15 (4%) were CNVs. The remaining variants in Jiao et al.33 were nonsense (8%; 2/26) or deletions (15%; 4/26).

Of the 26 different variants identified in Jiao et al.,33 9 were recurrent within the study population (7/17 missense and 2/3 splicing site), and 17 (65%) occurred in only 1 patient each. The two most commonly identified variants were observed in 23% (7/31) of all PDE patients (a missense variant) and 19% (6/31) of patients (a splicing site variant), respectively; remaining recurrent variants were observed in between 2 and 4 patients each.

The total number of recurrent variants identified in the study population was not reported by Coughlin et al.;32 however, 4 individual variants accounted for 38% (140/367) of all variant alleles, 1 of which was identified in a quarter of all alleles (94/367, 25.6%). Forty-nine missense variants accounted for the 209 alleles with missense variants; however, the majority (65.3%; 32/49) of these were only identified in a single individual and 17 were recurrent (responsible for 177 of all variant alleles identified).

Heritable retinoblastoma

The three subgroups and studies for hRB were: any RB,39 sporadic RB40 and a combined category of familial, bilateral or germline RB.41 Details per study are provided in Appendix 5, Table 22. All three studies were retrospective and were conducted at a single centre and all exclusively considered the RB1 gene. None of the selected studies exclusively used sequencing to identify variants.

Quality assessment

Quality assessment raised similar concerns about all three extracted studies for hRB.39–41 Insufficient detail about the genetic testing methods used led to concerns about the ascertainment of variants by sequencing and about replication or application of results beyond the study. It was not possible to determine the applicability of study results to the review question, as techniques other than sequencing may have been used to determine the genetic status of the patients.

Summary of results across disease definitions

The percentage of RB cases with an identified germline RB1 variant (i.e. hRB) using various testing strategies varied from 20.5%40 for the most narrowly defined population (sporadic unilateral RB) to 44.2%39 for the most broadly defined population (any RB). Both of these studies included participants with germline mosaicism (4.6%39 and 27.5%40 of germline cases), which is not always detectable in peripheral blood. Mosaicism is where a percentage of cells in the body carry the variant allele, but others carry a normal copy. Sequencing is often done on DNA extracted from peripheral blood samples, which means patients with mosaicism (especially low-level mosaicism) may not be detected by WGS strategies if the DNA extracted from the blood does not carry the variant allele.42 Specific techniques, such as use of unique molecular identifiers and NGS are of interest for detecting mosaicism.43 Results from the largest, most inclusive study39 demonstrate that, without genetic testing, as few as 10% of RB cases might have been identified for ophthalmologic surveillance from birth based on known FH of the condition. This study did, however, employ a stringent definition of familial RB.39

Salviat et al.39 further demonstrated that as much as 38% of the total population (n = 517) had no RB variant identified (negative on germline screening and no tumour tissue available for testing). While it is likely that the majority of the 517 patients had somatic RB, a small proportion may have germline RB that was not detected by the genetic testing strategy. In Hulsenbeck et al.,41 for example, 3 of the 821 patients at the centre were excluded as genetic data revealed no RB1 variant and high MYCN (a different gene) amplification, suggesting a different genetic pathway to the development of RB that may not be identified using WGS.

In two studies that reported the variant types, the most common were nonsense and frameshift variants.39,41 The percentage of those with bilateral RB was reported as higher in those with nonsense variants in both of these studies.39,41

Results for each subgroup by definition of disease

Any retinoblastoma

Retinoblastoma is a childhood-onset cancer of the eye caused by biallelic variants in the RB1 gene. RB can be either heritable (hRB) or somatic. hRB occurs where a variant on one of the alleles is present from conception (either inherited from a parent or occurring sporadically) and is therefore present in every cell of the body (germline), and the second variant occurs within the cells of the eye at some point after conception (this can occur prenatally or at any time point after birth). Around 40% of all RB are heritable. Somatic RB occurs when both variants occur within the cells of the eye such that the variants can only be identified by genetic testing of the tumour tissue as opposed to testing peripheral blood samples. Patients with bilateral RB are frequently assumed to have hRB, the majority of which occurs sporadically; only approximately 10% of all cases of RB have a known FH of the disease (familial RB). Studies that include ‘any RB’ provide the best estimate of the percentage of RB patients who would benefit from newborn WGS.

Salviat et al.39 included 1371 consecutive RB cases (including bilateral and unilateral, familial and non-familial) who successfully completed genetic counselling at a single centre. Multiple genetic screening methods were used dependent on the year of testing, but these included various combinations of: denaturing HPLC, quantitative multiplex PCR of short fluorescent fragments, methylation-restriction PCR, multiplex ligation-dependent probe amplification (MLPA), comparative genomic hybridization, Sanger sequencing and NGS. The promoter region and all exons with their flanking intronic sequences were screened. The identified pathogenic variants were classified as germline (identifiable in blood and therefore present from the point of conception) or somatic (occurring at any point following conception and therefore only identifiable in tumour tissue) and then as associated with the presence or absence (complete loss) of RB protein. Where tumour tissue was available, this was screened first to identify the RB1 variants, with peripheral blood then tested for the identified variants to determine the germline (heritable) status (n = 293). The remaining patients with no tumour tissue available underwent germline screening (of peripheral blood) only.

Of the original 1404 eligible participants, 118 (8.4%) had a known FH of RB (defined as families with at least two germline carriers of a RB1 pathogenic variant). Of those who completed the study, 44.2% (606/1371) were found to have a germline RB1 variant (hRB), including 497 with bilateral RB and 109 with unilateral RB. Germline mosaicism was identified in 28 of 606 (4.6%). Of the 765 patients with no germline variant identified, 248 (32.4%) were identified as having somatic RB (biallelic variants in tumour tissue only) and 517 (67.6%) were negative on germline screening (no tumour tissue available). Of those with germline variants identified, the majority (561/606, 92.6%) were identified by the germline screening strategy, and 7.4% (n = 45) were identified through the first-line tumour screening strategy. The most common variants (comprising 77.2% of identified germline variants) were nonsense (222, 36.6%), frameshift (140, 23.0%) or out-of-frame splice variants (110, 18.2%), all of which are associated with a loss of RB protein (total of 537/606 identified germline variants were associated with loss of RB protein).

Among those with hRB (germline variants) (n = 606), the incidence of bilateral RB was higher in patients with variants associated with a loss of RB protein (84.2%; 452/537) compared to those with variants with no loss of RB protein (65.2%; 45/69) (p = 0.01). Germline variants associated with a loss of RB protein were also associated with earlier mean age at diagnosis of RB (p < 0.001), and later stage at diagnosis (p = 0.047) compared to variants not associated with the complete loss of RB protein.

Sporadic retinoblastoma

Sporadic RB occurs where there is no FH of disease, and it can be either heritable (the variant is present in the germline and can therefore be passed on) or somatic (occurring only in the tumour). Children who develop sporadic germline RB (hRB) are the population with the greatest potential to benefit from newborn screening with WGS, as regular intensive surveillance can be initiated to allow earlier detection and treatment. This disease definition identifies the additional patients who might undergo surveillance for RB as a result of WGS. It is worth noting, however, that WGS is still of benefit to those with a known FH of RB because genetic testing will identify those who do not need to undergo such intensive surveillance as the RB1 variant has not been passed on.

The study by Temming et al.40 included 195 patients who presented with unilateral sporadic RB and underwent genetic testing at the request of patients or their legal guardians. Only those with ophthalmological follow-up until age 5 years were eligible for inclusion. One or more of the following methods was used for genetic testing of blood or tumour tissue: analysis of allele loss in tumours, cytogenetic analysis, denaturing HPLC, exon-by-exon sequencing, MLPA, methylation-sensitive PCR, quantitative fluorescent multiplex PCR, quantitative real-time PCR, real-time PCR and single-strand conformation polymorphism. Forty (20.5%) patients were identified as having a germline RB1 variant (hRB), 29 (72.5%) of whom had a heterozygous RB1 variant and 11 (27.5%) had germline mosaicism (which can be passed on to offspring of the proband if it is found to be present in the particular germ cell that forms the embryo).44 Of those with heterozygous germline variants (hRB), 10 (34%) were classified as whole gene deletions, 13 (45%) as premature terminations and 6 (21%) as ‘mild’ variants.

Of the 195, 9 (4.6%) developed bilateral RB during 5-year follow-up, 8 of whom had a heterozygous RB1 variant (3 with whole gene deletions and 5 with premature terminations) and 1 had germline mosaicism.

Heritable/bilateral retinoblastoma

Heritable RB includes cases who have a FH of RB (familial or inherited germline variant), and those whose germline variant in RB1 occurs sporadically at the time of conception (sporadic germline variant). Most patients with bilateral RB have hRB such that studies frequently defined study eligibility as familial or bilateral RB as a proxy for identifying cases of germline hRB. Consideration of studies in this category provides an indication of the most commonly found types of variant; however, in many studies, multiple genetic tests are used to characterise the genetic spectrum and it is not always possible to identify those variants that would be most easily identified on WGS.

Hulsenbeck et al.41 included 287 cases with RB from a total population of 815 patients (342, 42% of whom were identified as having a germline variant). Patients with confirmed heterozygous pathogenic constitutional RB1 variants (hRB) who had not previously undergone ophthalmological screening for familial RB were included in the study report. Those with mosaicism and non-RB1 variants were excluded. Multiple genetic screening methods were used, some of which included sequencing or sequential testing. A retinoblastoma variant effect class (REC) was developed to classify the identified pathogenic variants according to their effect on RB protein structure and quantity (i.e. extent of loss in RB protein), from REC-I (largest effect) to REC-V (smallest effect).

The most common variants (comprising 98.6% of identified variants) were nonsense or frameshift variants (REC-I) (199, 69.3%), whole RB1 gene deletions (REC-II) (39, 13.6%) and missense or in-frame SNVs (REC-III) (45, 15.7%). Of those with whole RB1 gene deletions (n = 39), 27 (69%) were identified as also having deletion of the MED4 gene (hypothesised as being associated with lower penetrance of RB).45

The percentage of bilateral RB was highest in those with nonsense or frameshift variants (REC-I) (186, 93.5%) compared to those with REC-II (30, 76.9%) and REC-III (36, 80.0%). Age at diagnosis was lowest in patients with REC-I variants [median: 7.3 months (range: 0.2–48.0)], followed by REC-II [10.3 months, (0.4–40.9)] and REC-III [11.6 months (0.9–45.1)] variants.

X-linked hypophosphataemic rickets

The XLHR is one form of hereditary rickets (see Report Supplementary Material 1). Clinical symptoms of rickets are not specific to XLHR but are characteristic for a much broader range of conditions. Some conditions can be distinguished based on biochemical markers, while others require information on the inheritance pattern or genetic testing. The proportion of rickets patients identified by sequencing the PHEX gene will therefore depend on whether disease is defined clinically, biochemically or genetically. The findings of studies on PHEX frequency in children with the disease defined in four different ways are summarised in Appendix 5, Table 23.

Overall, six studies were selected across four different categories of hereditary rickets. One study was identified and described for the broad hereditary rickets category,46 and two studies were included for each of hypophosphataemic rickets47,48 and hereditary hypophosphataemic rickets49,50. The largest study in these two categories only tested for the PHEX gene, while the second largest study tested for additional genes, thus providing information about the PHEX-negative cases, which could either be undetected PHEX cases or caused by different genes. One study was selected for the XLHR category.51

Four studies were conducted at a single centre,46–49 while one study included patients from all paediatric hospitals in Norway50 and one study was an international registry study.51 Three studies were prospective46–48 and three were retrospective.49–51 Sequencing techniques were employed to identify variants in the three prospective studies, while both sequencing and MLPA were used in two retrospective studies. The registry study of XLHR patients did not specify the genetic test(s) used.51

Quality assessment

The quality assessment using the Murad tool raised concerns across the different dimensions assessed. For selection of patients in both Gaucher et al. and Jacob et al., it is unclear whether the included participants were representative of all eligible patients.46,48 The ascertainment of patients in Rafaelsen et al. is unclear.50 Except for Marik et al., all studies did not describe the cases in sufficient detail to enable other investigators to replicate the research.46–51

In three studies, concerns regarding the applicability of study findings to WGS was low because the reported genetic spectrum was based on sequencing that more closely resembles WGS in the screening application,46–48 while concerns were high in the two studies that used MLPA in addition to sequencing49,50 and unclear in one study where ‘genetic testing’ was not further specified.51

Summary of results

The range of genes identified varied across the different categories of rickets. The frequency of PHEX varied from 33% in the most broadly defined population46 to 89.7% in the most narrowly defined population.51 ‘Pre-screening’ of the population using biochemical testing resulted in a greater proportion of patients with PHEX in both the hypophosphataemic rickets category48 and the hereditary hypophosphataemic rickets category.49

The extent of test negatives (no confirmed pathogenic variant identified) varied from 3%49 to 21.1% in Rafaelsen et al.,50 which was largely due to the testing strategy used [number of genes considered, extent of sequencing, sequencing method (e.g. WGS vs. Sanger sequencing) and additional testing]. The same study sequenced the untranslated 3-prime region as well as exons and intronic regions in recognition that XLHR may be caused by variants in the regulatory region of the messenger ribonucleic acid.50 MLPA was used in both Del Pino et al.49 and Rafaelsen et al.50 to overcome limitations of sequencing to detect deletions and duplications. Neither study reported the proportion of variants detected by MLPA in addition to those detected by sequencing.

The XLHR is characterised by many different variants precluding the investigation of expressivity of specific variants, with one study reporting that there is no clear genotype–phenotype link.50 Furthermore, a great proportion of variants in each study was novel, which presents challenges for the confirmation of variant pathogenicity.

Results for each subgroup by definition of disease

Hereditary rickets

This category is an umbrella term of any type of hereditary rickets (excluding nutritional rickets). It consists of two main types: (1) vitamin-D-dependent rickets (low phosphate levels secondary to vitamin D deficiency) and (2) hypophosphataemic rickets (low phosphate is the primary defect and rickets is therefore vitamin-D-resistant). Each type consists of several conditions caused by a series of genes with similar clinical symptoms. Some have distinct biochemical markers.

Jacob et al.46 conducted a comprehensive assessment to identify the genotypic spectrum of rickets in 10 Indian families, with 10 patients suspected of having hereditary rickets. All 10 patients had symptom onset in childhood; however, 2 patients did not receive a diagnosis until early adulthood. Exome sequencing identified variants in six different genes, including 3 patients (3/10, 33%) with a PHEX variant. The results revealed three known truncating variants, c1482 + 5G > C, c1586_1586 + 1del and c.58C > T. The c1586_1586 + 1del variant resulted in a more severe phenotype compared to the other two variants. Other implicated genes included CYP27B1 (n = 3 patients), CYP2R1, VDR, SLC34A3 and SLC2A2 (all one patient each). No cases had an unidentified genetic cause.

Hypophosphataemic rickets

In hypophosphataemic rickets, low serum phosphate due to renal losses is the primary defect, which is either mediated by the hormone fibroblast growth factor 23 (FGF-23) (raised FGF-23 levels) or is independent of FGF-23 (normal FGF-23 levels). Measuring levels of FGF-23 can aid the distinction between the two types. There are 15 genetically distinct disorders that are grouped into FGF-23-dependent and FGF-23-independent hypophosphataemic rickets.52

Gaucher et al. analysed the PHEX gene in 118 families, including 56 familial and 62 sporadic cases using classical sequencing. Sequencing covered all 22 exons, intronic regions and the region at the 3-prime end, which is not translated into a protein but serves regulatory processes.48 The study was conducted at a single centre but encompassed a multiethnic population, comprising individuals of European, North African, Caribbean and Asian backgrounds. The inclusion criteria were based on low serum phosphate and ratio of tubular maximum reabsorption rate of phosphate to glomerular filtration rate (TmP/GFR) levels, bone deformities and radiological evidence of rickets. PHEX variants were found in 78% (93/118) of probands. The 93 variants comprised 78 different variants of which 60 (77%) were novel. Variant types included nonsense (28%), frameshift (30%), splice-site variant (23%) and missense variant (19%).

Some uncertainty regarding the pathogenicity of novel PHEX variants was noted. One patient with a novel c.1206A > G variant (a single nucleotide replacement) also exhibited a second missense variant. In two other cases, both patients with the c.505G > A variant (another single nucleotide replacement) harboured additional variants, one involving an insertion leading to frameshift and the other a deletion leading to frameshift. The findings emphasise the significant role of PHEX in X-linked dominant hypophosphataemic rickets and suggest that family members should be screened when a PHEX variant is found in a sporadic case. Additionally, when a missense variant is detected, a search for another PHEX variant should be conducted.

The study’s approach to sequence only the PHEX gene meant that a large proportion of PHEX negative cases remained unexplained. For 3/25 negative cases, a reason was proposed. Missing PCR samples for two exons in one case led to the conclusion that one patient had a large deletion. One patient was subsequently diagnosed with a different type of hypophosphataemic rickets, and one patient was diagnosed with tumour-induced osteomalacia with secondary hypophosphataemia.

Marik et al.47 screened 66 consecutive Indian patients with refractory hypophosphataemic rickets using WES. Patients were characterised by a lack of healing despite treatment with cholecalciferol and had lower-than-normal phosphate levels for their age. The mean age of onset of symptoms was 22.5 ± 14.3 months; 24/66 (26.4%) patients had a confirmed PHEX variant, and 40/66 (60.6%) patients had a confirmed variant in a different gene. The remaining two cases had a variant of unknown significance (VUS) which was classified as negative and for whom genetic testing could not confirm the diagnosis. All 24 PHEX variants were different and 13/24 were novel.

In the context of expressivity, one patient carried two PHEX variations [c.2048T > A; p.(Leu683His) and c.2071-1G > C], whereas her mother, who was clinically mildly affected, had only one PHEX variation [c.2048T > A; p.(Leu683His)]. This difference may explain the variability in disease severity between them.

Hereditary hypophosphataemic rickets

Hereditary hypophosphataemic rickets is marked by increased FGF23 activity, leading to hypophosphataemia due to renal phosphate wasting. Genes associated with hereditary hypophosphataemic rickets include PHEX, FGF23, ENPP1, DMP1 and FAM20C, which are clinically and biochemically similar but follow different inheritance patterns.

Del Pino et al. included 96 patients diagnosed with hereditary hypophosphataemic rickets, of whom 42 underwent molecular testing of PHEX by Sanger sequencing and MLPA to detect gene deletions and duplications.49 The condition was characterised by the typical presence of combination of clinical, laboratory and radiographic findings. Deleterious sequence alterations or large deletions in the PHEX gene were identified in 85.7% (36/42) of patients. The remaining six patients were not genetically confirmed.

Rafaelsen et al.50 investigated 28 Norwegian children with hereditary hypophosphataemic rickets from 19 families. Inclusion was based on serum phosphate levels below the age-dependent reference range combined with tubular reabsorption rate of phosphate (not due to hyperparathyroidism). Sanger sequencing and MLPA analysis (to look for deletions and insertions in sequencing negative patients) of the PHEX gene was conducted. This was followed by Sanger sequencing of FGF23, DMP1, ENPP1KL and FAM20C successively in PHEX-negative patients. Overall, 13/19 (68.4%) probands were identified with PHEX variants. The 13 variants were all different and none were novel. There was one variant each in FAM20C and SLC34A3; 4/19 (21.1%) had no confirmed variant. Exploration of the effect of different types of variants (missense vs. nonsense) revealed no difference in clinical outcomes.

X-linked hypophosphataemic rickets

X-linked hypophosphataemic rickets is the most common form of hereditary hypophosphataemic rickets. It is caused by variants in the PHEX gene, which is inherited in an X-linked dominant fashion. Genetic testing or knowledge of a FH with a typical X-linked dominant inheritance pattern can support the diagnosis of XLHR.

Ariceta et al.51 is a registry-based study, including multinational data of 579 participants with XLHR. XLHR diagnosis was based on the clinical judgement of an XLH-treating expert physician (FH, clinical, radiological and biochemical findings) and/or by genetic testing. However, genetic testing was not required for patient registration; 282 of children underwent genetic testing, which was not further described. A total of 89.7% (253/282) children tested were identified with a variant in the PHEX gene. There was a small number of patients with different genetic disease who were incorrectly registered, including four with a variant in FGF23, one in SLC34A3 and seven where the gene was not specified; 17/282 (6.0%) had no variant confirmed. The study only reported genetic findings at the gene level. While symptoms were not reported by PHEX variants, the study concluded overall that children with XLHR diagnosis despite early detection and treatment did not do too well.

Familial haemophagocytic lymphohistiocytosis

Haemophagocytic lymphohistiocytosis (HLH) is not a single disease but a syndrome that is associated with several heritable and non-heritable conditions. Symptoms and biochemical markers are, therefore, non-specific and do not aid in the differential diagnosis, making a population of patients with fHLH difficult to define based on the clinical and biochemical characteristics. Historically, HLH was divided into primary (early onset, genetic condition) and secondary HLH (later onset, secondary to underlying medical condition such as cancers, infections or autoimmune disorders) using the main underlying trigger of symptomatic disease to define subgroups. More recently, the boundary between primary and secondary HLH has blurred with a better understanding of the complexity of the syndrome (discovery of new genes involved, differences in severity, genetic involvement in secondary HLH and detection of digenic disease). It is more accepted now that primary HLH is an artificial and ill-defined category, which was reflected in the published studies and could not be adopted here.

The fHLH is used to describe a subset of primary HLH disorders caused by biallelic variants in the four genes PRF1, UNC13D, STX11 and STXBP2. Distinguishing between the four conditions is not feasible using clinical symptoms or biochemical markers, and this relies on genetic testing. The three categories and representative studies were, therefore, HLH, more broadly,53 fHLH encompassing the four genes PRF1, UNC13D, STX11 and STXBP254 and any one of the four conditions, of which we identified studies for three of the four genes.55–57 The findings of the studies are summarised in Appendix 5, Table 24.

Four of the studies were retrospective and multicentre studies,53,54,56,57 while Amirifar et al. was a systematic review.55 The test was well defined in one of the five studies, which specified different sequencing methods (Sanger sequencing, NGS and WES) for patients with different indications based on biochemical assays.54

Quality assessment

The quality assessment of Amirifar et al.55 was conducted using the ROBIS-2 tool for systematic reviews,19 while other studies were evaluated using a modified Murad tool. We observed that the four primary studies lacked clear or sufficient information on patient selection.53,54,56,57 The studies provided limited details regarding the inclusion of patients from different centres and did not report the time frames of recruitment. Amirifar et al.55 presented an unclear risk of bias due to insufficient details on the review’s conduct, study selection and data synthesis.

Summary of results

In patients with HLH symptoms that met the HLH diagnostic criteria of the Histiocyte Society, a genetic diagnosis of fHLH was established in one-third of the patients.53 This means, a third of clinically defined HLH patients could be detected by sequencing the four fHLH genes; 7% of patients would be missed because they carried variants in different genes, 3.5% would be missed because the genetic cause could not be identified, 10% would be missed because they had a monoallelic disease and 46% would be missed because they had a non-genetic disease. The number of patients with monoallelic disease identified in three out of five studies led to the suggestion in one study that there is a gene–dosage effect, which means that fHLH can no longer be regarded as a simple recessive disease.53 This needs to be considered in the interpretation of sequencing results and is further complicated by the occurrence of digenic disease (monoallelic variants in two of the four genes). However, the risk of disease in patients with monoallelic disease is currently unknown. Across all four conditions, there is some indication that homozygosity is associated with earlier onset and more severe disease54 and severity is linked to the type of variant55 but that there is no clear genotype–phenotype link, as siblings with the same genotype displayed different phenotypes and different ages of disease onset.53 Overall, the studies suggest that there are a few common variants that are linked to particular ethnicities, highlighting the need to understand the genetic disease in a broad spectrum of patients, which is applicable to the screening setting before considering the implementation of sequencing as a screening tool in a diverse population.

Results for each subgroup by definition of disease

Haemophagocytic lymphohistiocytosis

Haemophagocytic lymphohistiocytosis is a hyperinflammatory syndrome generally defined by the diagnostic criteria recommended by the Histiocyte Society including symptoms of fever, splenomegaly, cytopenia, elevated cytokines and haemophagocytosis.58

Cetica et al.53 is an Italian registry-based study that analysed 500 HLH patients over 25 years. The multiethnic study included patients of southern European, Eastern European, African, Asian and Hispanic origin. In 426 patients who underwent sequencing, a genetic diagnosis was possible in 171/426 (40.1%) of patients, while 43/426 (10.1%) had a monoallelic disease and 197/426 (46.2%) were believed to have a non-genetic disease; 15/426 (3.5%) cases with assumed genetic disease were missed. Of 171 patients with a genetic diagnosis, 141 (82.5%) had fHLH with PRF1 and UNC13D variants, accounting for 131/141 (92.9%) of fHLH cases. The 69 patients with PRF1 and the 62 patients with UNC13D biallelic disease carried 34 and 37 different variants, respectively. A small number of variants were common occurring in up to 19 patients. The data on disease onset in sibling pairs revealed that there is no clear genotype–phenotype relationship. In 9/26 sibling pairs, disease onset varied up to 17 years, while in 1 pair, one sibling developed the disease at 6.7 years, while the other remained unaffected at 25 years. The number of patients with monoallelic disease indicates that HLH likely results from both genetic predisposition and exogenous triggers.

Familial haemophagocytic lymphohistiocytosis

Familial HLH is an artificial category of four conditions caused by variations in four distinct genes. Patients with different fHLH subtypes are clinically and biochemically similar.

Shabrish et al.54 investigated 101 molecularly confirmed fHLH patients, of whom 98 were under the age of 18 years, over 10 years, from 20 referral centres in India. 86/98 (87.8%) had biallelic disease; 12/98 (12.2%) patients had a monoallelic disease and would be considered test negative on sequencing. The PRF1 and UNC13D variants accounted for 70/86 (81.4%) of cases with biallelic disease. Molecular analysis revealed that missense variations were the most common type of variation in all four genes. The number of different variants was significant (25 different variants in 34 patients with PRF1 and 28 different variants in 23 patients with UNC13D fHLH). Patients with homozygous variants across all four genes had an earlier disease onset (median 10 months) compared to those with compound heterozygous variants (median 3 years).

Single gene/variant

The narrowest category consists of patient populations that had confirmed fHLH caused by one specific gene, while patients with confirmed variants in other genes were excluded. The studies’ aim was to characterise the variant spectrum in this tightly defined patient population.

Trizzino et al.57 included 124 patients with confirmed biallelic PRF1 disease and a median age of disease onset of 3 months from six different international centres. They detected 63 different variants: 11 nonsense, 10 frameshift, 38 missense and 4 in-frame deletions; 15/63 variants were novel. The most common single variant was a missense variant in 32 patients. Specific PRF1 variants were strongly linked to Turkish, African American and Japanese ethnic groups. Patients with two disruptive variants had a younger age at onset than patients with missense variants only.

The systematic review by Amirifar et al.55 analysed clinical features, immunologic data and genetic findings from 57 articles covering 322 patients with UNC13D variant with a median age of onset of 6 months; 269/322 (83.5%) had biallelic disease, 50/322 (15.5%) had monoallelic disease, and for 3 patients, this information was not reported. Missense variations were the most common type of variation. Severe features appeared to be associated with a homozygous genotype and missense variants. Splice-site errors and compound heterozygosity were more prevalent in patients with mild features.

Pagel et al.,56 a multinational study with patients included mainly from Germany and Turkey, included 37 patients with confirmed biallelic STXBP2 variants. One of the 37 patients was 19 years of age at the time of diagnosis. Nine novel variants were reported. Variants included nine different missense variants, four different splice-site variants and several small deletions or insertions. Three variants were seen in more than 5 patients; 13/37 patients carried one of two splice-site variants affecting exon 15. The exon 15 splice-site variant was associated with mild disease and an atypical disease course. These patients often experienced chronic, recurrent episodes with long periods without HLH symptoms, and their reactivations typically responded to steroids-only treatment or underwent spontaneous remission.

Medium-chain acyl-CoA dehydrogenase deficiency

The MCADD is part of a group of conditions called inborn errors of metabolism (IEM). MCADD symptoms overlap with those of multiple IEM. Differential diagnosis can be achieved by determining the disease-specific biomarker profile, relevant enzyme activity levels and the underlying genetic variant. Because MCADD is on the current NBS screening panel, MCADD can be defined based on screening outcomes as positive on the initial MS/MS screening test and as positive on confirmatory serum and urine tests.

The five disease definition categories for MCADD (from broadest to narrowest) and representative studies were: IEM/inherited metabolic disease (IMD),59 fatty acid oxidation disorders (FAOD),60 ACAD deficiencies,61 clinical–biochemical MCADD62,63 and gene (ACADM)-positive MCADD.64 The two studies in the clinical–biochemical MCADD category included one that detected cases through NBS screening (genetic spectrum in MS/MS positives and in follow-up positives)62 and one that diagnosed patients clinically or biochemically following family screening.63

Details per study for each of the five disease definitions considered are provided in Appendix 5, Table 25. The genetic testing method was not reported in two studies,59,63 three used a sequencing technique60,61,64 and one used sequential testing.62 All six studies were conducted retrospectively.

Quality assessment

Three of the studies met all the quality assessment criteria.60–62 We had concerns regarding the selection criteria for sequencing for the study by Touw et al.,64 and reporting was considered inadequate in the study by Mesbah et al.63 Applicability to the review question was unclear in two studies, because the testing method was not reported.59,63

Summary of results across disease definitions

Five of the six studies described above reported the experience of using genetic confirmatory testing in national newborn screening programmes. The percentage of patients with a genetic MCADD diagnosis varied from 15%59,60 to 19%61 in populations with a broad disease definition and to 85% in a population of patients with biochemically confirmed MCADD.62 As biochemical testing is readily available for MCADD, the latter category is the most relevant to consider. Genetically confirmed disease was less common in those with only an initial positive NBS screening test for MCADD compared to those confirmed on second-tier or follow-up biochemical testing.61 Testing for only two common ACADM variants had a low yield in initially NBS test MCADD-positive babies [8/511 (1.6%) biallelic and 157/511 (30.7%) monoallelic],62 however, this strategy could not be fully evaluated because results were not reported for those with biochemical MCADD that was confirmed on follow-up testing.

Results from the study by Mesbah et al.,63 which included patients with a clinical diagnosis of MCADD, suggest that 18.0% (almost one in five) of MCADD cases cannot be detected through genetic screening. However, it is important to note that the sample size was small (with only 17 cases) and the genetic testing method used was not reported, so it is not clear how applicable these results are to WGS.

Of the studies that presented variant frequency, the 985A > G variant was found to be the most common.59,62–64 In the study restricted to genetically confirmed cases of MCADD, 61.8% were homozygous with this variant.64 However, all four studies were from countries with populations of mainly Caucasian origin. The only study from Asia did not report this variant in any of their three genetic MCADD cases. ACADM variants present little heterogeneity in the studies (17 genotypes in 68 genetically confirmed MCADD cases); however, this may be misleading as ethnicity appears to affect the genotypes detected.64

Little information regarding expressivity was presented in the included studies. Generally, the evidence points towards some variants causing ‘milder’ disease based on residual enzyme activity studies, and it appears that those were newly detected in the screening context,64 highlighting that detecting genetic variants in a screening context is not the same as finding them in a diagnostic context.

Results for each subgroup by definition of disease

Inborn errors of immunity/inherited metabolic disease

Inborn errors of metabolism, also known as IMD, includes a group of approximately 600 conditions that are individually rare but collectively common. They are hard to diagnose, given the non-specific symptoms that many affected patients’ experience.65 IEM can be caused by variants in different genes that affect the same metabolic pathway at different stages, resulting in different conditions with similar symptoms. Overall, IEM are very heterogeneous, resulting in groups of disorders affecting different metabolic pathways with different epidemiology, presentation and heritability.

Martin-Rivada et al.59 reported details for 224 Spanish newborns who underwent genetic testing following biochemical indication of an IEM as part of the national screening programme. The original cohort included 902 consecutive newborns with an initial abnormal NBS test result. The molecular genetic testing method was not reported; however, 30 different genes were included. Of the 224 babies, 222 (99.1%) were diagnosed with a genetically confirmed IEM; 2 participants were considered to have biochemical hyperphenylalaninemia (no variants identified). In the wider group of initial NBS-positive babies, this equates to 24.6% (222/902) with genetically confirmed IEM. Of those with a genetic variant, 19.3% (43/222) were identified with a variant in the ACADM gene. Among these 43 children, 14 different genotypes were identified. The 985A > G variant made up 70% of all alleles (60/86) in 43 newborns; 22 newborns were homozygous and 16 were compound heterozygous for this variant. The remaining five cases were compound heterozygous for other variants. Only one patient (homozygous for 985A > G) showed symptoms of MCADD before newborn screening results were available.

Fatty acid oxidation disorder

Fatty acid oxidation disorders are a particular group of IEM caused by variants of the genes associated with the metabolic pathway of fatty acids in the mitochondria. Symptoms overlap and different FAOD are identified by their specific acylcarnitine (fatty acid metabolites) profiles and can be confirmed by gene sequencing.

Maguolo et al.60 included 30 Italian patients with FAOD; 20 were infants diagnosed following NBS screening and 10 were clinically diagnosed with a mean age at onset of 29 years and therefore excluded from review. Five of the 20 infants were biochemically positive for MCADD. Sequencing was conducted using a custom-designed FAOD panel, including 15 genes. Biallelic MCADD was identified in 15% of newborns (3/20) or 60% (3/5) infants with biochemical MCADD; while 2 infants carried monoallelic ACADM variants and would be classed as negative on genetic testing. Fifteen patients (75%) had different FAOD subtypes (5/15 with genetically confirmed disease; 1/15 with monoallelic disease and 10/15 without genetic information). Of six ACADM variants in those with biallelic disease, five were different. One infant was homozygous for the 985A > G variant and two infants were compound heterozygous carrying a total of four different variants. All three infants with biallelic ACADM variants had a residual MCAD enzyme activity of < 5% associated with severe disease.

Acyl-CoA dehydrogenase deficiencies

The FAOD subgroup of ACAD deficiencies are caused by variants in 11 genes, with ACADM being one of them.

Wang et al.61 reported genetic testing results for 20 newborns with ACAD confirmed on diagnostic biochemical testing from a cohort of 83 newborns with an initial positive screening result for ACAD deficiency on NBS screening. Out of the 20 newborns, 4 had biochemically confirmed MCADD. High-throughput sequencing and Sanger sequencing was used with a wider IEM panel of 306 genes. Three of the 20 (15%) newborns with confirmed ACAD were identified with biallelic compound heterozygous variants in the ACADM gene. In the wider group of babies with an initial positive screening test, this would be 3.6% (3/83). One newborn was test negative on sequencing due to a monoallelic variant. Therefore, genetic testing identified three-fourths of patients with biochemical MCADD. The three biallelic MCADD patients carried five different variants, of which two were novel. None of the Chinese newborns carried the 985A > G variant common in European cohorts.

Clinical–biochemical medium-chain acyl-CoA dehydrogenase deficiency

This disease definition includes those with a diagnosis of MCADD, either confirmed by clinical characteristics or by biochemical testing. Ideally, we would have considered the two categories separately to investigate the impact of different disease definitions on the genetic spectrum of disease. However, studies tended to include a mix of patients or poorly defined the populations included.

We included two studies for this disease definition category, one of which included patients identified with MCADD through a NBS programme.62 The other study in this subgroup included children who were clinically diagnosed with MCADD or detected through family screening and presumably confirmed biochemically.63

Nichols et al.62 included 511 newborns with NBS octanoylcarnitine (C8) levels ≥ 0.3 µmol/l who were subsequently referred for molecular genetic testing using a sequential sequencing method. First-tier testing specifically aimed to identify two of the most common variants prevalent in the USA c.985A > G and c.199T > C, second-tier testing used full ACADM sequencing in those with at least one variant or those without variant detected on first-tier testing but repeat C8 levels of at least 0.4 µmol/l. Mesbah et al.63 included 17 children younger than 18 years of age with clinically diagnosed MCADD; 4 were diagnosed via family screening and 2 via post-mortem. The genetic testing method was not reported. Both studies solely considered the ACADM gene.62,63

The percentage of patients positive (biallelic ACADM variants) on first-tier screening for two variants initially was 1.6% (8/511 with MS/MS-positive screening test) in Nichols et al.62 A further 157/511 (30.7%) of newborns were monoallelic, and 83/511 (16.2%) had neither variant on first-tier testing. In 20 newborns with positive clinical follow-up, sequencing of the full ACADM gene revealed 17 (85%) with biallelic ACADM variants and 3 with monoallelic variants classified as test negative. Mesbah et al.63 reported genetic test results of 14 of 17 included patients, 11 of whom had biallelic ACADM variants; and 3 cases were missed by sequencing as only one variant was identified. The most common variant in these two studies from the USA and Ireland was 985A > G (12/17 and 11/11,62 either homozygote or compound heterozygote, respectively).62,63 Nichols et al.62 reported 13 different variants of which 5 were novel. They reported that the c.199Y  > C/c.134A > G genotype resulted in ‘mild’ MCADD.

Gene (ACADM)-positive medium chain acyl-CoA dehydrogenase deficiency

This disease definition includes those with confirmed ACADM variants and provides useful information on variant frequency.

The study by Touw et al.64 included 68 children from the Dutch newborn screening programme with confirmed variants in the ACADM gene by sequencing all exons and adjacent intron regions. Seventeen different genotypes were reported, of which 7 were novel. Most of the children (42/68, 61.8%) were homozygous for the 985A > G variant; a further 20 were compound heterozygous, including the 985A > G variant. The authors categorised genotypes into ‘classic’ (previously recognised in clinically confirmed cases, n = 53) and ‘variant’ (genotypes not previously recognised in clinically confirmed cases, n = 15), and they reported median residual MCAD enzyme activity of 0% for the former and 25% for the latter group. This may support the theory that screening identifies disease with milder MCADD due to genotypes not recognised in clinical cases.

Evidence on early versus late treatment

Question 4: What is the evidence on early [following screen detection or sibling detection (cascade testing)] versus late (following clinical presentation) treatment?

In the absence of RCTs investigating the outcomes of pre-symptomatic versus symptomatic treatment, we defined early versus late treatment for the five selected conditions based on the natural history of the disease (e.g. age of symptom onset and progressive vs. relapsing/remitting conditions) and the type of available treatment/management (e.g. preventative vs. symptom management). Table 9 summarises the natural history for the five conditions that informed our definitions of early versus late. Table 10 summarises the management strategies for the five conditions, and Table 11 details out our definitions of early versus late based on the available treatment studies. The process illustrates that the different condition–treatment pairs require individual definitions of early versus late.

Appendix 5, Tables 26–30, summarises the outcomes from studies investigating early versus late treatment as defined by the studies for the five conditions. Case studies comparing outcomes in early versus late-treated siblings are presented in Appendix 5, Table 31. The detailed quality assessment is included in Appendix 3.

Pyridoxine-dependent epilepsy

There were five studies that compared relevant outcomes in children ‘treated early’ for PDE and children ‘treated late’ for PDE (see Appendix 5, Table 26).68,76–79 Two studies included a series of families76,77 and three were single-arm studies where patients were recruited retrospectively.68,78,79 Treatments included pyridoxine monotherapy,76,77 LRT78 or a combination of the two.68,79 The definition of ‘early’ and ‘late’ treatments varied between the studies. Jiao et al.77 and Tseng et al.79 defined ‘early-treated’ as the sibling with the shortest delay in receiving treatment after seizures began. In Bok et al.,76 the impact of treatment given antenatally (or asymptomatically) was compared to symptomatic treatment initiation, and the remaining two studies used age at treatment initiation to differentiate early and late treatments.68,78 The number of PDE cases in the included studies were low and ranged from 4 (from 2 families)76 to 60.68

None of the studies reported seizure control as an outcome but instead focused on developmental outcomes. Three studies reported intelligence quotient (IQ) scores;76,78,79 one reported a ‘standardised developmental assessment’ score (similar to IQ score),68 and one reported psychomotor development.77 Other reported outcomes included motor performance76 and a standardised neurological outcome.78 The timing of the outcome assessment was not reported in two studies77,78 and ranged between 4 years (age of one child)76 and a mean of 15 years of age79 in the other studies.

Quality assessment

We had methodological concerns regarding the selection of patients for all studies76–79 apart from the one by Coughlin et al.68 The selection process was not clearly reported in these studies and no recruitment dates were given. The follow-up period was not reported in the study by Strijker et al.78 and, in Jiao et al.,77 one child was only 6 months old at the last follow-up. In addition, the description of the assessment processes was inadequate in the latter study.77

Findings

There is some indication of improved developmental outcomes in ‘early-treated’ children. In studies that reported IQ scores76,78,79 or a quantitative measure of developmental assessment,68 scores were higher (suggesting improvement) in the ‘early-treated’ groups, apart from those in the study by Tseng et al.79 When treated with pyridoxine and LRT, the full-scale IQ score was slightly lower in the ‘early-treated’ group (76.0 vs. 77.4). There were suggestions of improved psychomotor development77 and motor performance76 in ‘early-treated’ groups; all three children in the ‘early-treated’ group were assessed as normal compared to only one child assessed as normal in the ‘late-treated’ group in Jiao et al.,77 and one ‘early-treated’ child showed a slightly better outcome than their ‘late-treated’ sibling in the study by Bok et al.76 (walking 4 months earlier).

There was insufficient evidence to determine whether there were any apparent differences as a result of the type of treatment (pyridoxine monotherapy, LRT or a combination of the two).

No studies that looked at outcomes in ‘early-treated’ patients alone were identified.

Heritable retinoblastoma

Five studies compared relevant outcomes in children who underwent pre-emptive surveillance to allow an early detection of RB (‘treated early’) compared to those who presented clinically (‘treated late’) (see Appendix 5, Table 27).80–84 All five studies were retrospective and single-armed.80–84 Two studies81,82 were multicentre and three were single-centre studies.80,83,84 The number of included RB cases ranged from 1384 to 264.80 Four81–84 of the five studies included children with a FH of RB, and the fifth study80 included all cases of RB, but the main comparisons of relevance that were reported were in those with a FH of RB (with or without surveillance).

Four studies80,82–84 compared outcomes in ‘screened’ versus ‘not screened’ at the participant level, one of which83 further categorised the ‘screened’ group into intensively screened and screened. In the remaining study by Chantada et al.,81 the income status of the country was used as a proxy for ‘screened’ versus ‘not screened’, since it was assumed that more children in a high-income country will be detected by screening than their counterparts in low- and middle-income countries.

Quality assessment

We had concerns regarding whether the length of follow-up was long enough for outcomes to occur in the study by Moll et al.82 In all studies apart from Rothschild et al.,83 reporting was not sufficient to allow others to replicate the research.

Findings

There was some variability in reported outcomes between studies. In three studies, the mean or median age at diagnosis was lower in screen-detected versus clinically detected RB, suggesting that screening does allow an earlier detection of RB than would happen otherwise (mean 4.9 months vs. 17.2 months,82 mean 4.7 months vs. 16.7 months84 and median 0 months for intensively screened, 4 months for screened and 9 months for not screened.)83 In terms of patient health outcomes, some measure of both ocular survival and survival was reported in three studies on a per participant level.80,83,84 The largest study80 using Kaplan–Meier analysis demonstrated considerably higher rates of ocular survival at 1 year for the screened FH group (n = 86) compared to the not screened FH group (n = 178) (83.2% compared to 47.5%), with a smaller difference at 5 years (67.7% compared to 58.2%). In contrast, however, a marginal difference in survival at 1 year (100% vs. 97.3%) had increased slightly by 5 years (93.2% vs. 87.4%), potentially suggesting a benefit from earlier detection.80

Similar results were reported by Rothschild et al.,83 with enucleation rates of 0% (0/16) for those intensively screened, 8.7% (2/23) for those ‘screened’ and 65% (13/20) for those not screened, and mortality rates of 0% for both screened groups and of 5% (1/20) for the not screened group (follow-up time point not reported). The smallest study reported enucleation rates of 0% for those screened (0/5) and 75% for those not screened (6/8), with no deaths (median follow-up of 4.8 years).84

The final study reported considerably lower enucleation rates in the USA (25%; 8/32) compared to those in developing countries (71.7%; 43/60) and a considerably higher probability of event-free survival at 5 years {0.92 [standard deviation (SD) 0.05] in the USA compared to 0.81 (SD 0.07) in developing countries}.81 It is likely that other differences in the delivery of care have contributed to the observed differences in outcomes, and it is not possible to properly attribute these differences to the effect of screening alone.

‘Early-only’ studies

In addition, we identified three studies reporting on ‘early treatment’ only, that is identification of RB via surveillance.85–87 Results from two of the three studies identified high proportions of infants with RB present in the first 1–2 weeks after birth [70% (12/17)85 and 50% (4/8)86]. The third study reported screening of 23 asymptomatic siblings of probands with RB; 13% (3/23) were identified as having active RB on screening at a median of 6 months of age.87 It is not possible to determine whether the use of surveillance in these studies resulted in ‘better’ outcomes than would have occurred if they had presented clinically.

X-linked hypophosphataemic rickets

There were three studies that compared the outcomes in children with XHLR who underwent early treatment with those who received treatment later.50,88,89 All studies implemented standard treatments for rickets {i.e. daily oral phosphate, vitamin D or an analogue of vitamin D (alfacalcidol) [oral phosphate and calcitriol (Pi/D)]} as opposed to the more recently licensed burosomab. No studies using burosumab as first-line treatment were identified. Appendix 5, Table 28 provides the summary of their findings.

All three studies were retrospective and were conducted without control groups. Two studies involved multiple centres50,89 and one88 was conducted at a single centre. All three studies used the same definition of early (before 1 year of age) versus late treatment (at or after 1 year of age). In one study, patients in the early treatment group were diagnosed prior to the onset of clinical signs of rickets, while those in the late treatment group were diagnosed after the appearance of clinical symptoms.88

Study outcomes included measurements of height and various biochemical parameters such as serum calcium, phosphate, ALP, creatinine, parathyroid hormone and vitamin D3 levels. Makitie et al.88 and Quinlan et al.89 assessed the activity of rickets through radiographic examination. Additionally, Makitie et al.88 made predictions regarding adult height. The evaluation of results occurred at different time intervals across the three studies. Makitie et al.88 measured outcomes at the end of the first year of treatment and before puberty. Rafaelsen et al.50 conducted measurements at each clinic visit and assessed the results at the last recorded consultation. The mean age at the last recorded consultation of the early group was documented as 11.1 years, while the late group had a mean age of 8.4 years. Quinlan et al.89 analysed outcomes at medium treatment durations of 8.5 years and 11.9 years for early and late treatment groups, respectively.

Quality assessment

The evaluation of study quality revealed that Quinlan et al.89 lacked sufficient information on patient selection, while it remained uncertain whether the follow-up duration in Rafaelsen et al.50 was adequate to assess outcomes. The study by Makitie et al.88 satisfactorily addressed all aspects of the quality assessment.

Findings

Study findings suggest that initiating treatment during the early stages of growth moderately enhances outcomes for patients with XLHR. Both Makitie et al. and Quinlan et al. reported median height closer to the expected average in the early treatment compared to the late treatment groups [SD scores of −0.7 (n = 8) vs. −1.8 (n = 11), p = 0.009;88 and −0.7 (n = 10) vs. −2.0 (n = 13), p = 0.00989]. Additionally, Makitie et al. reported pre-pubertal height closer to the expected average in the early (−1.3 SDS; n = 8) compared to the late treatment group (−2.0 SDS; n = 11) (p = 0.054). Treatment effect in the early-treated group (z-score −0.2), compared to the late-treated group (z-score −1.2), was within that expected by chance (p = 0.06) that was observed by Makitie et al.88 By contrast, results from Rafaelsen et al.50 showed the same trend as Makitie et al.88 but were within that expected by chance.

Regarding biochemical parameters, Makitie et al.88 found a similar degree of hypophosphataemia between groups; however, serum alkaline phosphatase levels remained elevated in the late treatment group throughout childhood. Quinlan et al.89 reported no difference in the median levels of serum phosphate or serum alkaline phosphatase between the early and late treatment groups.

In terms of rickets activity, Makitie et al.88 observed more pronounced radiographic signs of rickets in the late treatment group, whereas patients receiving early treatment still displayed significant skeletal rickets changes. Quinlan et al.89 reported similar ricket severity scores in both groups.

‘Early-only’ studies

In addition, we identified three studies only reporting on early treatment.73,90,91 Exploration on early treatment indicated that early treatment could enhance metabolism, promote growth and mitigate deformities. However, these findings are not entirely consistent with the findings of the studies reporting a comparison of early versus late treatment.

Familial haemophagocytic lymphohistiocytosis

One retrospective, multicentre study that evaluated outcomes for children who received treatment for fHLH following asymptomatic detection before activation (treated ‘early’) and children who received treatment following clinical symptomatic detection or activation of fHLH (treated ‘late’) was identified75 (see Appendix 5, Table 29). The study included 32 genetically confirmed sibling pairs/triplets with fHLH. The asymptomatic children were diagnosed following the diagnosis of their sibling. Outcomes included mortality, cause of death and number of patients in complete remission at the end of follow-up. The follow-up period for each patient was different.

Quality assessment

The primary quality concern in Lucchini et al.75 was the lack of detail regarding which international centres provided the data and whether they sent in details for all eligible patients in their care. The patient characteristics, treatment and outcomes were reported adequately, and the follow-up time was sufficient to assess treatment outcomes.

Findings

We present results separately for per-protocol and intention-to-treat populations using a slightly different definition for ‘early’ and ‘late’ treatment.

Intention-to-treat population

For the intention-to-treat population, we defined ‘early’ and ‘late’ treatments as cases that were asymptomatic and symptomatic at diagnosis regardless of activation before or after treatment initiation. In other words, the ‘early’-treated group in the intention-to-treat population included patients who were asymptomatic at diagnosis but with some subsequently experiencing activation either before or shortly after commencing treatment, and this may confound the true effect of ‘early’ treatment on outcomes.

Mortality was lower among the ‘early’-treated group (15% vs. 38%). Six patients [two (8%) in the ‘early’-treated group and four (15%) in the ‘late’-treated group] died due to transplant complications, which was the most common cause of death. The proportion in complete remission was higher in the ‘early’-treated group (81% vs. 62%).

Per-protocol population

Ideally, ‘early’ treatment would begin in individuals before their first activation. However, individuals who present as asymptomatic may have experienced disease activation before or shortly after the initiation of treatment. Therefore, for the per-protocol population, we defined ‘early’ and ‘late’ treatment using four distinct groups, dependent on the activation status. The ‘early’-treated group can be split into two categories: asymptomatic patients who were treated with HSCT ± prophylactic treatment and did not activate (group 1, n = 15) and asymptomatic patients who were treated with prophylactic treatment and subsequently activated (group 2, n = 3). The ‘late’-treated group was separated into those who were symptomatic and were subsequently treated with active disease protocol ± HSCT (group 3, n = 26) and those who were asymptomatic at diagnosis but experienced disease activation before the start of treatment and thus were treated with active disease protocol ± HSCT (group 4, n = 7). Defining the treatment groups in this way helps to reduce any potential confounding effects that activation may have had on outcomes and allows us to better isolate the impact of ‘early’ treatment of truly asymptomatic individuals on clinical outcomes.

Mortality was lowest among group 1 (7%) and was similar between the remaining three groups (group 2 33%, group 3 38% and group 4 27%). Note that group 1 reflects those with less severe disease than group 2, since these patients did not activate after treatment initiation. Also, group 2 only included three patients, one of whom died. The most common cause of death was disease progression. Three patients died following transplant complications (four in group 3 and two in group 4). The number of patients in complete remission was highest in group 1 (93%). Proportions were considerably lower in the other three groups (group 2 33%, group 3 10% and group 4 7%). However, group 1 had the shortest median follow-up time, so it is possible that patients in this group subsequently experienced symptoms or other events. One patient in group 2 was lost to follow-up.

Group 3, who were symptomatic at diagnosis, and likely represent those with most severe disease, had the highest proportion of deaths and lowest proportion of patients in complete remission at the end of follow-up. Outcomes in the groups who had experienced activation following an asymptomatic diagnosis either before (group 4) or after (group 2) the start of treatment were similar to group 3. The opposite was true, however, for those in group 1, which indicates that ‘early’ treatment before activation has been experienced may be effective. Further, only three asymptomatic patients with no previous activation subsequently activated after commencing ‘early’ treatment. As previously mentioned, this reduces our confidence in results from group 2, but does, however, increase confidence in the conclusion that treatment in asymptomatic individuals with no previous activation may be effective.

Medium-chain acyl-CoA dehydrogenase deficiency

There were nine studies that looked at relevant outcomes in children with MCADD who received early management following asymptomatic detection through screening (treated ‘early’) and those who received management following clinical symptomatic detection of MCADD (treated ‘late’) (see Appendix 5, Table 30).92–100 All studies, apart from the study by Abdenur et al.,92 were single-arm and all were conducted retrospectively. Three Australian studies had overlapping patient cohorts.96,98,99 The sample sizes in the included studies ranged from 292 to 90.94,101

The definitions of ‘early’ and ‘late’ treatment varied slightly between the studies. In eight of the nine studies (all apart from Wilson et al.100), ‘early’ treatment was defined as management following detection of MCADD through NBS screening. One of the eight studies also included those detected through family screening.93 In six of the eight studies,92,94–96,98,99 patients were all asymptomatic, but in two studies, this group also included symptomatic patients.93,97 In the study with data for only two patients, one patient received dietary management from 5 months following diagnosis by NBS screening, and the other patient received management following symptom presentation.92 In Wilson et al.,100 ‘early’ treatment was management following asymptomatic screening due to an affected sibling.

In all nine studies, ‘late’-treated was defined as management following clinical presentation of MCADD.92–100 The study by Anderson et al.94 also included those detected through family screening in the ‘late’ group, and it is unclear how many patients this group included and whether they were symptomatic or asymptomatic. It is important to bear these differences in the definitions of ‘early’ and ‘late’ treatments in mind when interpreting results. Where reported, management strategies included avoidance of fasting,93,94,97,98,100 carnitine supplementation,93–95,97,100 various diets92,94,95,97,98 and sick-day regimens.98,100

The reported outcomes were heterogeneous and included mortality/severe episodes, descriptions of patients’ clinical statuses, various measures of physical and psychological development, healthcare use and biomarker levels. Results for each of these outcomes are described separately below. Follow-up length was varied, but two studies reported outcomes within the first 4 years of life.96,98

Quality assessment

In four studies, we found that cases were not described in sufficient detail to allow for replication of the study or to make inferences, and this was mostly due to the paucity of treatment definition and description.95,96,98,102 In four studies, the method for selection of patients was unclear.92,93,95,102 The follow-up times were variable, and this was unknown in one study.93

Findings

Mortality/severe episodes

Four studies reported mortality as an outcome.95,96,98,100 This outcome was assessed at 4 years of age in two studies,96,98 and follow-up was variable in the other two studies.96,100 In two studies, the proportion of patients who died by the age of 4 years was lower in the ‘early’-treated groups (4% vs. 17% in Haas et al.96 and 4% vs. 19% in Wilcken et al.98 in the early- and late-treated groups, respectively). In the study by Wilson et al.,100 21% and 17% of children had died by the age of 6 years in the early-treated and late-treated groups, respectively. However, it is important to note that in this study, the definition of early treatment was not those detected by NBS screening, and instead, this group included those diagnosed due to an affected older sibling. No patients died in the study by Gong et al.95 Wilcken et al.98 reported the number of severe episodes by the ages of 2 and 4 years. At both ages, the percentage who had experienced a severe episode was lower in the ‘early’-treated groups.

Description of clinical status

Four studies reported the clinical status of patients after varying follow-up periods.92,93,95,102 Where reported, age at assessment ranged between 24 months92 and 11 years.102 No studies assessed patients after a standardised follow-up period. In the study by Li et al.,102 all six children (four ‘early’-treated and two ‘late’-treated) were assessed as normal. In two studies, the ‘early’-treated groups were assessed as normal, but some children in the ‘late’-treated groups showed poorer outcomes.92,95 One of the two studies had only two patients; the late-treated child showed clinical abnormalities, including severe seizure disorder and cerebral palsy, and nasogastric feeding in one child.92 In the second study, one late-treated patient reported intermittent fasting hypoglycaemia and one experienced hemiplegia due to disease episode.95 In the study by Alcaide et al.,93 nine (29%) patients were symptomatic in the ‘early’-treated group, and both children in the ‘late’-treated group were symptomatic.

Physical and psychological outcomes

Various measures of physical and psychological outcomes were reported, and overall, there appeared to be few differences between the ‘early’- and ‘late’-treated groups. There were no differences in terms of height, weight or neuropsychological function (within the first 4 years of life) in one study,96 or in terms of intellectual ability score (assessed at > 4 years of age) in another study.98 In Wilcken et al.,99 one child (4%) in the late-treated group had a mild intellectual handicap, and two (8%) required extra assistance at school, whereas the children in the ‘early’-treated group were all assessed as normal (at 6 years of age or last follow-up).

Healthcare use

Four studies reported some measure of number of hospital/emergency room (ER) visits.94,96,98,100 The mean [95% confidence interval (CI)] number of hypoglycaemia-related hospital days and ER visits per patient-years was slightly lower in the ‘early’-treated group [0.09 (0.03 to 0.15) vs. 0.11 (0.04 to 0.19)] in the study by Anderson et al.94 However, in this study, it is important to note that the ‘late’-treated group may have included asymptomatic, screen-detected patients. In two studies, the percentage of patients who had previously been admitted to hospital was lower in the ‘early’-treated groups [42% vs. 85% (by age 4)98] and 25% versus 36% (variable follow-up).100 Haas et al.96 reported whether the hospital visits were inpatient, emergency or outpatient within the first 4 years of life. The percentage of children with inpatient stays and emergency room visits was lower in the ‘early’-treated group, but the number of outpatient visits was higher in this group. Length of inpatient stay in those admitted was also similar between the ‘early’- and ‘late’-treated groups.

Biomarker levels

Biomarker levels at diagnosis were reported in Gong et al.95 Mean levels of C6, C8 and C10 were lower in the ‘early’-treated children, but there were no differences in the mean C8: C2 or C8: C10 ratios. However, it should be noted that these measures were reported inconsistently.

‘Early-only’ studies

In addition, we identified nine studies only reporting on early treatment in which MCADD patients were detected through NBS screening or family studies and no patients were detected and treated following symptom onset.60,103–109 One further study identified five cases admitted to hospital before their NBS test results became available, but outcomes were not reported separately for the NBS screening group.64

Conclusions and learning from the review of five conditions

The five traditional reviews yielded insufficient evidence in populations relevant to a screening context to inform a UK NSC decision about WGS in newborns. The lack of evidence on the penetrance and expressivity of the genetic variants for all five conditions does not appear to support the identification of specific conditions or groups of individuals for which WGS may be beneficial in a screening context.

Available evidence about the genetic spectrum of patients with symptomatic or biochemical disease frequently pointed to small numbers of recurrent variants accounting for considerable proportions of cases; nevertheless, large numbers of variants were novel or occurred only in a very small number of cases. The pathogenicity and penetrance of such variants are unclear. We also saw that genetic spectrum can be strongly affected by factors such as ethnicity and the prevalence of consanguinity. Studies often employed a suite of genetic testing methods rather than relying on sequencing alone such that we could not determine the proportion of cases that could have been identified with WGS alone. As such, the results observed are not necessarily transferable to the screening context of apparently healthy newborns.

In terms of identifying benefit from earlier treatment, our results generally reflect acknowledged difficulties in evaluating the effectiveness of interventions for rare conditions. Studies were generally small, with variable definitions of ‘early’ and ‘late’ both within and between conditions and with a frequent reliance on an earlier intervention in siblings or where there was a known FH. Only a few examples of asymptomatic or very early initiation of treatment were identified. Outcome measures were often short term, with limited follow-up to identify longer-term patient-relevant outcomes such that any benefits from an earlier diagnosis resulting from newborn screening will be difficult to quantify.

A few aspects were identified that would render certain conditions less likely candidates for a newborn screening programme using WGS:

• The genetic heterogeneity is large and novel potentially pathogenic variants are common, causing a lot of uncertainty which has downstream implications in terms of time needed to determine pathogenicity.
• The types of frequent variations (mosaicism, large deletions, etc.) are not sufficiently captured by WGS and require additional genetic testing.
• The pre-symptomatic phase is likely to be shorter than the time until test results are available for diagnosis.
• An early intervention phase cannot be defined.
• The preferred curative treatment option is not licensed for newborns.
• The curative treatment option carries a high risk of adverse events.
• The available treatment option is for symptom management only.

A single approach to reviewing 5 conditions was not feasible, and a review of 200 conditions would require 200 individual reviews. The five reviews were undertaken by three full-time and two part-time reviewers, and it took 7 months to complete without writing up the findings. A review team of similar size could take as much as 280 months (23 years) to undertake 200 consecutive reviews for 200 conditions. Some learning may be transferable between reviews of similar conditions, shortening certain review processes. For instance, IEM are a group of related conditions that are generally managed with a specific diet or dietary advice, therefore some thinking and decisions may be applicable more widely across several conditions. We could not explore this with the five conditions reviewed, which we selected for a range of treatment scenarios. The five conditions were highly varied in their characteristics, treatment and aim of screening and the review process had to be tailored for each condition. Firstly, each search was developed individually, requiring an understanding of the condition in terms of:

• the presence of a non-genetic version in addition to a genetic version of the condition
• alternative names and aliases
• relevant umbrella terms.

Secondly, categorisation of sequencing studies by the studies’ disease definition required condition-specific categories, depending on:

• the availability of biochemical tests (i.e. the definition of disease in biochemical terms is possible)
• the number of disease groups with overlapping symptoms
• whether conditions were only defined genetically
• whether the condition is already screened for.

Thirdly, the definition of early versus late was specific for each condition and depended on:

• whether the relevant intervention is earlier detection or therapeutic and whether the treatment is preventative, curative or management of symptoms
• whether an early intervention phase could be defined
• whether conditions are progressive or present following a trigger
• whether early could be defined in other terms than pre-symptomatic (e.g. early stage of progressive disease).

Exploring Clinical Genome Resource as an evidence source for the United Kingdom National Screening Committee

The ClinGen is an evidence review resource that could be potentially used for the evaluation of WGS for 200 paediatric conditions. We, therefore, assessed the evidence base provided in ClinGen for the five conditions reviewed in our traditional review and compared ClinGen scoring dimensions and scores to the UK NSC criteria and our decisions. We also assessed the alignment of the dimensions and UK NSC criteria to the four principles used by GEL for decisions on gene inclusion. High agreement between criteria from the three resources would mean that decisions from ClinGen and/or GEL could inform UK NSC recommendations in the future.

Clinical Genome Resource as an evidence source for the five conditions in our review

We considered the gene–disease validity, variant classification (level of pathogenicity) and the actionability scores reported by ClinGen. The gene–disease validity has been confirmed in ClinGen for the eight genes of the five conditions considered in our review, and four of the five conditions reviewed (PDE, MCADD, hRB and XLHR) had a paediatric actionability report available on ClinGen in March 2024. However, no information on variant classification in terms of pathogenicity was available for any of the genes. Therefore, the resource has got limited informative value for decision-making on the variant level. The evidence provided on the actionability of the four conditions with a paediatric actionability report is explored in the next chapter (Comparison of the paediatric reports for pyridoxine-dependent epilepsy, medium-chain acyl-CoA dehydrogenase deficiency, heritable retinoblastoma and X-linked hypophosphataemic rickets from Clinical Genome Resource with our review findings).

Comparison of the paediatric reports for pyridoxine-dependent epilepsy, medium-chain acyl-CoA dehydrogenase deficiency, heritable retinoblastoma and X-linked hypophosphataemic rickets from Clinical Genome Resource with our review findings

The paediatric actionability reports for PDE, MCADD, hRB and XLHR provide information and scores on four dimensions: severity of disease, penetrance, treatment effectiveness and burden of intervention. These are the most relevant aspects in determining the medical actionability for genetic conditions identified as incidental findings according to ClinGen. Table 12 summarises the ClinGen scores for the four conditions.

According to ClinGen, PDE, hRB and MCADD are conditions with high actionability (overall score 10 or 11 out of 12) and XLHR is pending an actionability assertion. XLHR has got an overall score of 9, suggesting moderate actionability based on the score alone; however, the Paediatric Actionability Working Group may over-ride the score based on their clinical expertise. At the time of review, the evidence searches for the paediatric reports were 3.5–6 years old (see Appendix 6, Table 32).

The severity of disease for all four conditions was scored 2 out of 3 (a reasonable possibility of death or major morbidity) and penetrance was scored 3 out of 3 (> 40% chance of serious outcome). A summary of the penetrance findings is provided in Appendix 7. The score for penetrance was based on minimal evidence (level C evidence) for PDE, XLHR and hRB (see Appendix 7, Table 33) and on a single published study that was identified non-systematically through an expert review for MCADD. The evidence on penetrance for PDE and XLHR was provided by GeneReviews without further references to a primary data source (see Appendix 6, Table 32).113 For hRB, the penetrance information came from a non-systematic review and a published guideline, in addition to GeneReviews. The penetrance information for MCADD was based on a published study of 81 NBS test-positive newborns who were identified through the Danish NBS screening prgramme.119 Overall, 7/12 references that form the evidence base for penetrance and expressivity for these four conditions were websites; no evidence cited was based on sequencing data of an unselected newborn population and no additional references to our four systematic reviews were identified (see Appendix 6, Table 32).

The treatment available for PDE and MCADD was rated as highly effective (score 3 of 3) with minimal risk (score 3 of 3) (Table 12). Surveillance for hRB, while highly effective (score of 3), was judged to be associated with moderate risk (score of 2). The treatment for XLHR was classed as moderately effective with moderate risk (both score of 2). The treatments considered for the conditions are summarised based on the ClinGen information given in Appendix 7, Table 33. The evidence base for the effectiveness of treatment and surveillance was classed as moderate for hRB, MCADD and PDR, where at least one guideline or treatment recommendation was identified (5/8 as websites only, 3/8 published in peer-reviewed journals), and as minimal for XLHR in the absence of a clinical guideline or treatment recommendation. The evidence on the treatment effectiveness of Pi/D for XLHR was based on one non-systematic review, information from the OMIM website and GeneReviews. The main information that ClinGen included from GeneReviews was the findings from the retrospective study looking at Pi/D treatment in patients with XLHR < 1 year of age versus > 1 year of age by Makitie et al.88 included in our review above. One more reference to early versus late management was available for hRB, where the information provided was for screened (family members of probands with hRB) versus probands. This study concluded that surveillance resulted in earlier diagnosis and better outcomes measured by the likelihood of enucleation, eye radiation and visual acuity; however, less than half of the early group actually received surveillance.120 This study formed part of the treatment recommendation by Skalet et al.,121 cited in the ClinGen paediatric actionability report for hRB. This study was missed by our searches but would not have been included in our review because the definition of early treatment did not meet our inclusion criteria.

Comparison of the ClinGen scores of the five conditions to our review findings using the UK NSC criteria and GEL’s decisions for gene inclusion is shown in Table 12. This was complicated by three aspects. Firstly, the ClinGen scores out of 3 (3 being best for actionability) for the four dimensions do not provide a definitive yes/no assessment precluding an evaluation of the level of agreement for the individual criteria.

Secondly, the criteria for assessment do not map well across ClinGen, the UK NSC and GEL. While ClinGen assessed the severity of disease in the diagnostic context of an incidental finding in an individual, the UK NSC criterion 1 requires the assessor to consider the importance of disease from a public health perspective, considering whether screening for the condition is worth it, whether the natural history of the disease is understood and the link between the risk factor (here a genetic variant) and the disease is known. GEL’s principle A focuses more simplistically on the level of evidence that proves the gene–disease link.

The criterion around penetrance differs between ClinGen and GEL’s principle B versus UK NSC criterion 3, because the UK NSC criterion focuses on screen-detected variants rather than variants detected in patients with confirmed disease. This means that the studies identified for question 2 in our traditional review only partially address this criterion and do not provide the evidence needed to understand the penetrance in the complete spectrum of genetic disease identified through screening. This explains at least partly the different outcomes of assessment for this criterion.

There are also some important differences in the evaluation of the evidence of treatment effectiveness. While ClinGen requires an available and effective treatment that can be used in pre-symptomatic individuals, the UK NSC and GEL also require evidence on the benefits of earlier treatment compared to treating symptomatic disease. And, while ClinGen considers the burden of the treatment separately in their fourth dimension, GEL’s principle D is focused around equity in access to treatment, and UK NSC criterion 10 focuses on the available evidence of a treatment pathway for early treatment. While some criteria map reasonably well across GEL and the NSC, the evidence bar appears to be higher for the UK NSC, which has led to different decisions for five/five conditions concerning the treatment effectiveness. In contrast to the UK NSC, which requires a systematic and comprehensive approach to evidence synthesis, GEL’s approach was similar to that used by ClinGen in that a single supporting reference was sufficient to meet GEL’s principles for gene inclusion, and the evidence was often based on information provided by ClinGen or GeneReviews. This means that all five conditions met the GEL principles, while the current evidence base for none of the conditions meets all UK NSC criteria. Furthermore, ClinGen and GEL assessed gene–condition pairs against 4 criteria, while the UK NSC requires evidence to a further 16 criteria which are not part of this comparison.

Thirdly, the subjective approach to reach a final decision is not transparent and decisions may not follow from the evidence. GEL and ClinGen heavily rely on the expertise and opinion of an expert clinical panel. The UK NSC, on the other hand, relies on a balanced interpretation of the amount and quality of evidence, which is reviewed during stakeholder consultations.

Overall, it appears that different outcomes in the assessment of the five conditions are due to a combination of (1) different focus in the assessment criteria used (particularly, the individual patient vs. the public health perspective), (2) the expected limited evidence base for these rare conditions and a higher evidence bar used by the UK NSC and (3) the subjective interpretation of the evidence base to reach a final decision. This means that neither the ClinGen scores nor the GEL decisions would be an appropriate proxy for a UK NSC recommendation for the five conditions. And, using ClinGen as an evidence source for a future review of genomic screening of 200 conditions may not be feasible nor appropriate.

Conclusions and learnings from Clinical Genome Resource

The ClinGen is an excellent and important resource of information on medical actionability for incidental findings of monogenic diseases. It produces reports separately for the paediatric context, uses a transparent and standardised approach of evidence collection and a semi-quantitative metric to score the actionability of diseases across four domains. An expert group reviews the decisions and the evidence and formulates an overall assertion on actionability. The process allows an early rule out of conditions that do not meet the evidence threshold. Although this is a rich source of information and potentially a good starting point for discussion around reporting of genetic findings, the resource has got some limitations as an evidence source for the UK NSC:

1. The protocol that the paediatric actionability working groups follow is a generic one, and there is no information on what each group did in their evidence review and how consistent the approach was across groups.
2. The focus is on individuals with incidental findings in the diagnostic setting rather than the public health perspective for a screening programme decision.
3. The reports tend to be quite dated and would require updating considering the fast-moving field of genetics.
4. The evidence bar for actionability in the diagnostic setting is lower than what the UK NSC requires in terms of quantity and quality of evidence for population screening.
5. Knowledge of penetrance is not a requirement for actionability as early rule out is not triggered in the absence of information on penetrance.
6. Treatment effectiveness is judged by the availability of a treatment guideline with an existing intervention for an undiagnosed paediatric population without the requirement of evidence on benefits of early versus late treatment.
7. The number of conditions covered is limited as of March 2024, and no information on variant pathogenicity was available for any of the five conditions of interest.

Considering the relatively low evidence requirement, it may be feasible to explore using ClinGen as a tool to rule out conditions for reporting, which did not meet the requirements for full review, but it is inappropriate for the UK NSC to base decisions on potential screening programs on the actionability reported in ClinGen without further assessment.

Using existing genomic studies of newborn screening cohorts reporting penetrance as an evidence source

The purpose of this review was to explore the feasibility of identifying pathogenic variants of paediatric conditions that are known to have high penetrance and expressivity in an unselected newborn population to be considered for an initial newborn screening programme that maximises benefit and minimises harm. The review focused on studies using sequencing in newborns for childhood-onset disease that reported penetrance or an approximation.

Out of 4970 articles identified, 105 were taken through to full-text sifting. The majority were excluded because the test did not meet the inclusion criteria (indirect sequencing or sequencing was second- or third-line test), or the outcome was irrelevant (focus on carrier frequency). See Appendix 4, Figure 7 for the study flow diagram and Report Supplementary Material 2 for the list of excluded studies and their reason for exclusion. There were 14 studies that reported experiences with gene sequencing in newborns which were reported in 16 references.122–137 Green et al.,122 Green et al.123 and Ceyhan-Birsoy et al.124 all report on the BabySeq project, which will be cited as Green et al. from hereon.122 Full data extractions for the 14 studies are reported in Report Supplementary Material 4. Four studies sequenced genes for a single condition134–137 and were not considered for synthesis, and 10 studies sequenced genes for ≥ 74 conditions.122,125–133 Five of the 10 studies did not report any follow-up time and therefore could not be used to approximate penetrance. These five studies compared sequencing data to follow-up test results,126 conventional newborn screening with clinical review,130 categorised sequencing outcomes into levels of penetrance based on existing literature132 or categorised sequencing outcomes into pathogenic/likely pathogenic based on existing classification systems with131 or without clinical review.128,133 Information on penetrance could be inferred for the remaining 5 of the 10 studies because they reported some clinical follow-up after a positive sequencing test.122,125,127–129 These five studies are synthesised below and summarised in Tables 13 and 14, with one study reporting results for a prospective and a retrospective cohort.125 The five studies originated from two countries, the USA (n = 2122,127) and China (n = 3122,127). While the US studies used WGS,122,127 the Chinese studies used gene panel tests125,128,129 The number of included genes ranged from 134 to 954 and the number of newborns sequenced ranged from 127 to 29,989. Gene selection and variant interpretation varied across studies. In four out of five studies, the country’s newborn screening programme acted as a starting point,125,127–129 and two out of three Chinese studies stated that they considered conditions on the recommended unified screening panel, which details the mandatory conditions for newborn screening in the USA in addition to their own.128,129 However, the final scope of included conditions differed. Green et al. (2023) had the widest scope, including a small number of adult-onset conditions.122 The level of detail provided on variant selection also varied. In general, as a minimum, studies reported consideration of the American College of Medical Genetics and Genomics guidelines and/or ClinGen/clinical variation (ClinVar) for the classification of variants from pathogenic to benign. Four of the five studies clearly stated that only pathogenic and likely pathogenic variants were reported.122,125,128,129

The findings of the five studies are reported in Table 14. The proportion of sequencing positive babies ranged from 1.7%127 to 9.7%.125 Comparison to conventional newborn screening revealed that genetic screening and conventional screening are complementary, each detecting and missing different cases. Genetic screening has the potential to identify a significant proportion of additional conditions that are not on the conventional screening panel in the USA and/or China. Two of the three studies with a comparison to conventional screening reported that over half of the positive screens were made up of conditions not available on the conventional screening panel.127,128 However, 83.3–100% of these could not be confirmed clinically within the studies’ follow-up, so we do not know if detecting these was an overdiagnosis of clinically insignificant disease or misdiagnosis of disease or early detection of later-onset disease. In Chen et al.,128 359/395 (90.9%) of these unconfirmed additional cases were female babies with the X-linked inherited condition G6PD. This represents an example where sequencing performs differently in males and females.

The great proportion of unconfirmed cases can be partially explained by the short follow-up, which ranged within a retrospective study from 2 months to 29 months125 and from a median of 1.2 years128 to > 5 years122 in the prospective studies. Overall follow-up was insufficiently described, variable within studies and too short to capture the disease onset for all included conditions. The short follow-up had an impact on the estimates of penetrance because cases developing clinical symptoms after follow-up will be unaccounted for. This impact would differ for early-onset and later-onset childhood conditions. Therefore, the identification of variants in healthy newborns does not exclude a pathogenic role for these variants. Considering the short follow-up, penetrance, approximated by the number of confirmed cases after clinical follow-up for all genes considered, ranged from 1.6% after a follow-up of 24–48 months125 to 58.4% after up to 3 years of follow-up.129 While some of the studies were large enough to report a significant number of sequencing positive cases, these large studies did not report the findings on clinical outcome by gene variant for all cases to enable an estimate of penetrance on the variant level. One study reported 16 cases of PKU, all of which were compound heterozygotes with c.158G > A.129 All were weakly positive on conventional NBS testing due to slightly elevated Public Health England levels at regular intervals. However, none of the infants received any intervention, and the study authors concluded that the variant may be causing a mild phenotype. More such data are needed from sufficiently large studies to enable penetrance estimates to be derived for individual variants. However, the majority of variants reported in the studies appear to be single occurrences precluding the estimation of penetrance.

Clinical management was considered where appropriate in all five studies following a confirmed diagnosis including for ‘mild’ and subclinical cases on confirmatory testing. While clinical management in the studies was considered confirmatory of the positive sequencing result, early intervention also means that penetrance cannot be estimated for all cases from these studies. It is unknown whether symptoms would have developed even without treatment in confirmatory testing of positive cases.

All five studies were single arm, either retrospective or prospective cohorts. One study used an enriched study design; however, this was not reflected in the proportion of sequencing positive cases.129 Applicability may be a concern as conditions and variant frequencies vary across geographical regions (even within countries) and three out of five studies were undertaken in Chinese populations. Sequencing negatives were not followed up apart from one study, which reported that 82/445 newborns who received both genetic and conventional screening were missed by targeted gene panel sequencing.128 While this information is needed to evaluate the test accuracy of WGS, it does not contribute to penetrance estimates.

Conclusions and learnings from existing newborn sequencing studies

In comparison to our traditional review of individual conditions, the focused review of studies reporting penetrance of any gene/condition pair in the newborn screening setting took around 4 weeks to develop, and it produced a robust search strategy which can be re-run in the future without consideration of different conditions. The search picked up all publications from current genomics projects in newborns that were known to us.

Overall, the studies identified highlight that GS for newborn screening is still in its infancy and cannot be implemented without further research. There was 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 within the studies’ follow-up resulting in interventions, routine surveillance and regular follow-up of uncertain benefit. The studies do not lend themselves to determine a variant threshold for individual genes, that is, the number and types of variants to include in a screening programme that focuses on detecting highly pathogenic and highly penetrant variants with low risk of producing harm. This is because:

• The number of infants with a specific condition displaying a range of variants is too low even in large studies so variant frequencies cannot be estimated.
• Infants with confirmed genetic disease received management, which precludes estimation of penetrance and expressivity for cases without symptomatic confirmation of disease.
• Clinical follow-up was not sufficiently long to include all childhood-onset cases.

The studies, however, give some insight into aspects of WGS not identified from our review of five conditions, including challenges and promises of WGS of newborns, such as:

• The potential complementary role of WGS to traditional newborn screening programmes.
• Levels of agreement with conventional newborn screening.
• Conditions for which conventional screening may be better (e.g. 4/4 NBS test-positive congenital hypothyroidism infants missed by sequencing129).
• Conditions where regional differences may exist (e.g. G6PDD and PKU in China126).
• Potential differences in PPV between sexes (e.g. G6PDD128).
• Proportion of infants positive on sequencing with two recessive variants, where variants are in cisgender (i.e. in the same copy of the gene leaving the other gene copy intact) [7/55 (12.7%)128], therefore highlighting the potential role of parental testing in cases with two heterozygote variants to determine the phase.
• The impact of including analysis of CNVs [in one study, sequence analysis detected 33/47 (70.2%) at-risk genotypes, and CNV analysis contributed an additional 14/47 (29.8%),132 highlighting potential limitations of WGS].
• Some indication of turnaround times for WGS [mean turnaround time 1.5 times longer for WGS vs. exome gene panel testing (56 days vs. 37 days),132 sequencing to issuing formal report was within 11 days,125 turnaround time of newborn WGS 16 weeks–24 weeks130].

In addition to the included studies described above, the review also identified two studies that reported findings that approximate penetrance.138,139 In these studies, variants of a group of childhood-onset conditions were identified, and the number of healthy adults with disease-associated genotypes with these variants in a general population database were determined.138,139 Gold et al. concluded that the false-positive rate of genomic screening of newborns for diseases treatable with HSCT was 0.04% based on 59/141,456 healthy adults, with implicated genotypes using variants of 127 genes of severe childhood-onset conditions treatable with HSCT.138 Breilyn et al. found that clinically relevant variants according to ClinVar in ACADS (gene associated with short-chain acyl-CoA dehydrogenase deficiency) were not associated with the evidence of metabolic disease in a large and ancestrally diverse adult population (2035/30,000 healthy adults with implicated genotypes).139 These findings could imply that the variants of the 127 genes of severe childhood-onset conditions treatable with HSCT may be suitable for newborn screening, while variants in ACADS investigated may not be suitable. These studies may contribute useful insights on the penetrance and expressivity in a future review if the studies meet the following inclusion criteria:

• population database should be of healthy adults rather than newborns
• variants considered should be of genes associated with childhood-onset disease
• outcomes should be number of adults with implicated genotype (not simply allele frequency)
• outcomes should be reported on variant level.

Results for review of cost-effectiveness evaluations of whole genome sequencing and whole exome sequencing

Search results

Following the searches of databases and registries, the titles and abstracts of 2325 records were screened, of which 226 records were identified as potentially meeting the eligibility criteria and were flagged for full-text review [see the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) diagram in Appendix 4, Figure 8]. Seventy-one of these studies were judged to meet the eligibility criteria.

Nineteen of the records identified at the full-text stage process were literature reviews. The reference lists of all 19 reviews were checked to identify whether there were any potentially relevant records missed by our search or the original Schwarze et al. review.27 All included records in 13 of the 19 reviews had already been identified, but there were 6 reviews that included 9 additional studies that had not been identified. Of these, two studies were eligible for inclusion.

Of the 28 costing studies included in the Schwarze et al. systematic review,27 13 met our pre-defined eligibility criteria. Four studies were excluded because they were conference abstracts.140–143 One study focused on the use of WGS for a communicable disease.144 Six studies were included in the Schwarze et al. review as partial economic evaluations, but cost was not included as an outcome in the methods or results sections of the studies; cost was just mentioned briefly at some point in the paper.145–150 One study was categorised as a partial economic evaluation, but we could not find any mention of costs in the paper.151 A further study was excluded because the focus of the analysis was on the cost associated with identifying incidental findings, and the main analysis excluded the costs associated with WES and WGS.152

Two literature reviews were included in the Schwarze et al. review; we did not find any additional papers to assess for eligibility in these reviews.27,153

No additional papers were eligible from the Nurchis et al. scoping review.28

In total, there were 86 studies included in the review.

Study characteristics

Table 15 provides an overview of the characteristics of the studies included in this review. The full data extraction is included in Report Supplementary Material 5. Most of the included studies were conducted in high-income settings such as Europe (n = 27/86, 31%), North America (n = 28/86, 33%) or Australia (n = 20/86, 23%). Of those conducted in Europe, around half (n = 13/27) were from the Netherlands and four from the UK. The earliest study included in our review was published in 2014, and there has been an increase in the number of studies published on this topic since 2017 (Figure 1).

FIGURE 1 Number of studies included in the review by publication year. (Figure not reproduced here; see the original article.)

Population

None of the included studies focused on the use of WGS or WES specifically in a screening context; all studies focused on symptomatic populations or the cost of WGS or WES more generally. Over three-quarters of the included studies included newborns or children in their target population (n = 70/86, 81%), reflecting the early onset and presentation for most of the conditions targeted by WES and WGS.

The studies evaluated the use of WES and WGS for a wide range of conditions with a genetic component. The description of the target diseases was very broad for many of the papers, particularly those that focused on newborns or children. Terms including ‘genetic disorders’, ‘monogenic disorders’, ‘variety of conditions’, ‘rare diseases’, ‘mendelian disorders’ and ‘mitochondrial disorders’ were used, demonstrating the broad spectrum of possible genetic diseases potentially discoverable by WES and WGS in contexts where there is very little indication of a clear diagnosis for an individual. There were some common recurring themes such as neurological or neurodevelopmental disorders, intellectual disability or developmental delay (n = 20/86, 23%), cancer (n = 12/86, 14%), congenital anomalies (n = 6/86, 7%), different types of epilepsy (n = 4/86, 5%) and autism spectrum disorder (n = 3/86, 3%).

Intervention

Over half of the studies focused on WES as the intervention (n = 48, 56%), with just over a quarter focusing on WGS (n = 22/86, 26%) and 16 studies (19%) focusing on both tests. Around a third of the studies explored the impact of putting WES or WGS testing earlier or later in the diagnostic pathway (n = 29/86, 34%). In some of these studies, WES was already an established part of the diagnostic pathway under evaluation, and the question focused on whether it should be moved to later in the pathway (i.e. more targeted screening to reduce costs) or moved to earlier in the pathway to reduce the need for other diagnostic tests and potentially arrive at a diagnosis earlier.

There were also some studies (n = 10/86, 12%) which focused specifically on rapid WES or WGS, which is different to standard sequencing in that it has a much shorter turnaround time, potentially increasing the clinical utility of the test by returning actionable results quicker. All these studies focused on children and newborns and were looking for rare diseases, suspected genetic diseases or ‘diseases of an unknown cause’.

Comparator(s)

Of the studies that included a comparator (n = 78/86, 91%), 44 (56%) explicitly stated that the comparator was current standard of care testing. This typically consisted of a broad range of tests, some focusing specifically on more targeted genetic testing, and others also encompassing a wide range of imaging, biochemical and biopsy tests. In some studies, these were very clearly broken down by the individual test. Different assumptions were made in terms of which tests would no longer be needed following the incorporation of WES or WGS in the diagnostic pathway, typically dependent on the proposed timing of the different tests.

Many of these studies used cohort data to underpin their analyses (n = 58/78, 74%), capturing the healthcare resource use and, where applicable, outcomes for the cohort. Of these, over two-thirds (n = 39/58, 67%) used the same cohort of individuals to inform both their intervention and comparator(s). These studies relied on a single cohort who had received either the intervention or comparator, making assumptions about what would have happened to those individuals had they received the opposite. For example, the cohort may have received current standard of care and assumptions were made about what would have happened (usually which diagnostic tests could have been avoided) had the intervention been available. Some studies (n = 18, 23%) had two distinct cohorts, where one had received the intervention and one had received the comparator. Sixteen studies were modelling exercises, so the cohorts for the intervention and comparator(s) had been simulated based on aggregate data obtained from the literature.

Outcome

Of the 86 studies included in this review, only 10 were cost–utility studies and defined the outcome of their analysis as cost per QALY. Over a third of the included studies were cost-effectiveness evaluations (n = 29/86, 34%), that is, they reported the cost for a change in a specific health outcome. All of these studies focused on the number of diagnoses, additional diagnoses or diagnostic yield as the health outcome of interest. The costing studies included (n = 18, 21%) tended to report either the total cost per patient or the cost per test.

Methodology

The costing perspective was not specified in 33 of the included studies (38%). Where it was reported, many adopted a broad healthcare system perspective (n = 29/86, 36%) or a more specific health system perspective such as a hospital perspective, a clinical genetics service perspective or a laboratory perspective (n = 15). Some studies considered costs to the patient or payer (n = 8) or a broader, societal perspective (n = 5). The costing perspective was largely driven by whether the cost of WES or WGS testing would fall to a public healthcare system or to the individual. Of those adopting a societal perspective, one study included costs associated with the caregiver time spent in neonatal intensive care unit (NICU); post NICU informal care and direct non-medical costs, for example caretaking, modifications to home and special education.154 Another study estimated the parental time lost from accompanying their child to medical visits, which was valued using the human capital approach using hourly adult wage.155

Over half of the included studies (n = 50, 58%) did not state the time horizon of their analyses. Only seven studies (8%) adopted a lifetime horizon, with eight (9%) studies including costs over a 1-year time period or less, typically to only capture costs associated with the diagnostic testing pathway. Where time horizons of > 1 year were explicitly implemented (n = 24, 28%), 17 studies (71%) stated that they applied a discount rate to costs beyond the first year.

Only two-thirds of the included studies (n = 54/86, 63%) conducted scenario or sensitivity analyses to explore the impact of parameter or structural uncertainty in the economic evaluation.

Cost–utility studies

Cost–utility studies, that is, economic evaluations which produce estimates of the incremental cost per QALY, are typically the preferred economic evaluation method for HTA in the UK. Of the 86 studies included in this review, only 10 were cost–utility studies. All of these studies have been published since 2019. Three of these evaluations focused on individuals with non-small cell lung cancer.156–158 One study focused on children with prenatally diagnosed non-immune hydrops fetalis and fetal effusions.159 One study focused on adults and children with suspected monogenic kidney disease.160

Crawford et al.154 evaluated the cost-effectiveness of early exome sequencing in critically ill babies who are admitted to NICUs and are suspected to have mitochondrial disorders compared to current typical care. They used a decision tree–Markov hybrid model to estimate the costs and QALYs over a lifetime horizon. The model was populated using data from the published literature, expert opinion and the Paediatric Health Information System database in the USA. The model simplified the health outcome to two broad health states: within the NICU and post-NICU. Given the unavailability of quality of life metrics from the patient perspective, published utility values, based on standard gamble responses from parents with experience of having a child in the intensive care unit, were used as a proxy for the utility of a newborn with severe mitochondrial disease in a level III/IV NICU and post-NICU.

Sanford Kobayashi et al.161 evaluated the cost–utility of rapid WGS in critically ill children admitted to paediatric (not neonatal) intensive care units. Their analysis was based on a cohort of 38 children ranging from 4 months to 17 years old, where rapid WGS had resulted in a molecular diagnosis in 17 of the 38 children. Eight of these 17 children had a change in clinical care, and QALY savings were estimated by Delphi Consensus for 2 of these children. One of these children had autoimmune polyendocrinopathy syndrome, and vaccination for encapsulated organisms was assumed to reduce the risk of mortality. The other child was diagnosed with Factor XIIIA deficiency, and the initiation of prophylactic Factor XIII replacement decreased the risk of repeat central nervous system bleeds and associated mortality and neurologic complications. Cost savings were estimated for four of the eight children based on the change in clinical management.

Stark et al.162 report a cost–utility analysis based on the same cohort of individuals used to underpin a previously published CEA, which has also been included in our review. They collected data from a cohort of 80 infants aged 0–2 years, presenting with multiple congenital abnormalities and dysmorphic features, or other features strongly suggestive of monogenic disorders, from a single tertiary paediatric centre in Australia.163 In their cost–utility analysis, they analysed the cost-effectiveness of informative and uninformative results on continuing diagnostic investigation, changes in management, cascade testing in first-degree relatives and parental reproductive planning and outcomes. The time horizon over which costs and outcomes are estimated is unclear, but the longest time frame mentioned is 18 months. Schofield et al.164 build on this analysis, adopting a 20-year time horizon, which the authors justify as being adequate for examining the long-term cost-effectiveness of WES without projecting beyond reasonable certainty in the outcomes of rare genetic diseases. Across both studies, health utility values were assigned based on parent-reported preferences for health states using utility values derived from another study.

Lavelle et al.165 evaluated the use of WGS and WES in children with suspected genetic conditions. They modelled two separate patient populations, infants (< 1 year old) who are critically ill and children (aged < 18 years) who were not critically ill but with suspected genetic conditions. They conducted two analyses, one which produced the estimates of cost per additional diagnosis and another which produced the cost per QALY over a lifetime horizon. To achieve the latter analysis, they made a range of assumptions about the proportion of children who receive a change in clinical management following a diagnosis and the proportion who improve following this change in management. Children were assigned costs and QALYs based on their assumed level of long-term disability (none, mild, moderate and severe). The parameters for the model were taken from the literature.

Cost-effectiveness evaluations

Over a third of the included studies were cost-effectiveness evaluations (n = 29/86, 34%), that is, they reported the cost for a change in a specific health outcome. Twelve of these studies did not report the time horizon over which costs and outcomes were estimated. Seven studies estimated costs and outcomes over ≤ 3 years, or the time horizon was described as the diagnostic trajectory. Most studies focused only on the costs associated with arriving at a diagnosis. Two studies had a lifetime horizon.166,167 These two studies were led by the same researchers in Italy, and both explored the cost-effectiveness of WGS versus WES and standard testing in paediatric patients with suspected genetic disorders. The key difference between the studies was that one used a decision tree modelling approach, and the later study used a Markov model approach. Both models were developed using the same data from a cohort of 870 paediatric patients who underwent testing in Rome. Clinical effectiveness was measured in terms of the number of diagnoses. Costs included diagnostic tests as well as management and therapeutic procedures. The authors explain that, although a cost–utility analysis is the preferred economic evaluation method in Italy, there were no follow-up data for the cohort and therefore QALY impacts from change in clinical management were not available. Both studies concluded that WGS sequencing would be cost-effective compared to WES and conventional testing.

Cost–consequence analyses

Around a third of the included studies (n = 29/86, 34%) were cost–consequence analyses, that is, they reported cost and health outcomes separately and did not compare results to a cost-effectiveness decision threshold based on the acceptable amount of incremental cost per improvement in a health outcome. Most of these studies also focused on outcomes relating to the number of diagnoses, but some also focused on outcomes such as time to diagnosis and change in clinical management.

Costing studies

Eighteen were costing studies and did not report health outcomes. Nine of these studies were micro-costing studies, where each individual resource use associated with conducting WGS or WES has been measured and reported. These types of studies are particularly useful in the absence of a standard tariff cost for a test in public healthcare settings.

One of these studies was conducted as part of the Scottish Genomes Partnership (SGP) study, where each step of conducting a trio-based WGS is described from clinical assessment and recruitment to participant feedback of the WGS result. Data on testing histories from 259 probands referred to 1 of Scotland’s 4 regional genetics centres underpinned the analysis. Activities at regional genetics clinics, the regional genetics laboratories and the sequencing centre were all micro-costed. Costs to GEL were estimated based on the charges. For tests performed in Scotland, unit costs provided by NHS Scotland’s genetics laboratories were assigned. For tests conducted in other UK laboratories, unit costs were sourced from these laboratories. The authors identified which section of the standard genetic testing pathway WGS is intended to replace, a section of the pathway estimated to cost £1841 (2018 prices) on average. This cost varied considerably by and within disease, however, and the authors provide a useful breakdown of the costs by condition. The total cost of WGS was estimated to be £6625 per trio.

There was one other UK-based micro-costing study, Schwarze et al.,27 who estimated the cost of Illumina-based WGS in a UK National Health Service laboratory for a cancer case and a rare disease trio case. Cost data were collected for all steps in the sequencing pathway, and sensitivity analyses were conducted to identify key cost drivers. The estimated average cost for a rare disease WGS trio was £7050.11 (2016 values), with sequencing alone costing £4659 per case and bioinformatics and reporting costing £677. Interestingly, the cost for WGS for a cancer case and a rare disease case were quite similar, and the small difference was largely because the cancer case was based on two samples (tumour and germline), whereas the rare disease case was a trio (parents and proband).