Elucigene FH20 and LIPOchip for the diagnosis of familial hypercholesterolaemia: a systematic review and economic evaluation

Clinical diagnosis

Different sets of clinical criteria have been developed for the diagnosis of FH. These criteria primarily include a combination of high cholesterol, presence of tendon xanthomata in the patient or first-degree relative and a family history of premature CHD or high cholesterol.

The most widely utilised and validated sets of clinical criteria are:

1. the UK Simon Broome Register criteria

2. the US MedPed (make early diagnosis, prevent early death) criteria

3. the Dutch Lipid Clinic Screening Network criteria.

Genetic diagnosis

DNA-based mutation screening methods provide a definitive diagnosis of FH by identifying a causative mutation and confirming the clinical diagnosis. 25 DNA testing adds clinical certainty to a diagnosis among relatives. Mutations associated with FH have been mostly found in the LDLR gene and rarely in the APOB and PCSK9 genes. 26 The LDLR gene is divided into 18 exons (coding regions in a gene) and 17 introns (non-coding regions in a gene). 27 There are different types of mutations. Large rearrangements or deletions in the LDLR gene have been reported in 5% of FH patients in the UK. 5 Different genetic screening systems are used to screen the entire coding region for the LDLR gene, such as single-strand conformation polymorphism (SSCP) analysis, denaturing gradient gel electrophoresis (DGGE), DNA sequencing and RNA analysis. 27 None of these techniques has been reported to be 100% accurate, with detection rates of 75–85%. Techniques such as Southern blot analysis27 or multiplex ligation-dependent probe amplification (MLPA) are used to identify larger rearrangements and deletions. MLPA analysis, being a simple and rapid method for detecting large rearrangements, has been recommended to be included in the comprehensive genetic analysis (CGA) testing strategy for FH. 28

Low-density lipoprotein cholesterol concentration measurement (Simon Broome criteria)

Low-density lipoprotein cholesterol is most commonly assessed using an estimated figure calculated from the TC and high-density lipoprotein (HDL) cholesterol values and the triglyceride level, the combination commonly being referred to as ‘lipids’. Because triglyceride measurements vary with fasting status, assessments are usually performed after an overnight fast. LDL-C by itself is neither fully sensitive nor specific for the diagnosis of FH, with considerable overlap between FH and non-FH individuals. LDL-C assessment would be recommended whether or not a genetic test is being undertaken, as other hyperlipidaemias (and the small proportion, perhaps around 5%, of patients with gene-negative FH) would have to be managed on the basis of lipid analysis, and the response to treatment would also be gauged by measuring lipids.

Comprehensive genetic analysis

Comprehensive genetic analysis is defined as the most complete genetic analysis generally available for FH within a diagnostic setting and is expected to detect almost all known FH-causing mutations. This analysis includes DNA sequence analysis of the promoter, all exons and the exon/intron boundaries and into the 3′ untranslated region of the LDLR gene, which will detect the majority (around 88%) of detectable FH mutations, MLPA for each exon and the promoter region of the LDLR gene to detect deletions and duplications (around 5% of detectable FH mutations) plus analysis for the common APOB p.Arg3527Gln gene mutation (around 5% of FH mutations) and the PCSK9 p.Asp374Tyr gene mutation (around 2% of FH mutations).

Targeted gene sequencing

Targeted gene sequencing is used to describe the genetic test for sequencing a specific part of the gene where the family mutation is found. Targeted gene sequencing may be used for cascade testing to identify FH in the biological relatives of index cases.

UK national audit of the management of familial hypercholesterolaemia

Following the publication of the NICE guideline for FH in 2008,1 a national clinical audit investigating the care received by individual patients with FH was undertaken by the Royal College of Physicians, with the results published in 2010. 18 A 2008 survey had shown that around 15,000 adults and 500 children were being managed in UK lipid clinics and the audit examined around 15% (n = 2324) of the adults and 30% (n = 147) of the children. 34

The results, key findings and recommendations of the audit in relation to cascade and DNA testing are detailed below.

Diagnostic technologies

With regards to genetic tests, two novel screening techniques have emerged (Elucigene FH20 and LIPOchip). Some reports suggest that DNA testing for FH costs approximately £400, whereas other work estimates that the process could cost between £500 and £1000 per test. The main reasons for the large variation in reported costs are (1) the definition of DNA testing has varied in previous reports with differences in the genes sequenced and whether or not genes were screened for deletions or duplications and (2) the cost of DNA sequencing has reduced over time as laboratories build up economies of scale and improve equipment allowing for faster processing and reporting times; as a result, previous cost estimates for testing for FH have varied greatly across reports and studies.

The Elucigene FH20 kit is available at a cost of £15 per test and LIPOchip is available at a cost of €250 or approximately £198. However, these costs do not account for staff time to process samples, consumables or overheads. Therefore, the costs of Elucigene FH20 and LIPOchip will be much greater in practice than just their unit test cost.

A standard NHS tariff does not exist per se for genetic tests; however, a recently developed system is now increasingly used by genetics laboratories across the UK to apportion costs to genetic testing services. This ‘MOLU’ (MOLecular Units) pricing system is the most commonly used costing mechanism for genetic testing of FH in laboratories across the UK. Genetic testing strategies vary in complexity depending on the type and volume of analysis required for different reports (genetic tests). The PCR amplicon or equivalent was chosen as a measure of complexity, which is transparent and easily counted. Reports are grouped into a total of six ‘bands’ (A–F). Bands are assigned and given a weighting according to the number of amplicons analysed to produce a report in that band. The number of reports multiplied by the appropriate band weight produces a final number of MOLUs. The total number of MOLUs derived from the exercise can be divided into the total laboratory budget to give an approximate monetary value to MOLUs. This in turn produces an indicative cost for the various testing strategies. Laboratories that can keep their budget constant or can reduce it but increase the number of MOLUs produced will have lower unit costs. It is estimated that the average cost per MOLU is between £30 and £35. Costs of all genetic tests including targeted gene sequencing for relatives can be estimated in this way.

Although the MOLU costing approach has been decided upon as the most appropriate and generally accepted method to cost these test strategies, it is far from ideal. The approach does not necessarily account for full economic costing or indeed opportunity costs of resources. The MOLU approach is basically a price banding agreed upon in collaboration between the laboratories from a UK Genetic Testing Network (UKGTN) group and the Clinical Molecular Genetics Society (CMGS). This has limitations in terms of the accuracy of the costs produced; however, in the absence of any more robust costing methods for these genetic tests, the MOLU classification system has been deemed the most appropriate method with which to compare these testing strategies.

Costs of LDL-C measurement will need to take into account the costs of resource use to retrieve samples and the costs of testing the samples by a laboratory. LDL-C testing is relatively inexpensive compared with genetic testing. These assays are performed routinely in most laboratories using current fully automated equipment (e.g. the laboratory in Aberdeen Royal Infirmary performs > 100,000 per annum) and the reagent cost is minimal (pence), so the overall cost of the procedure consists almost entirely of the general costs associated with processing any sample (around £3–10).

Ancillary costs

The genetic equipment required to process the tests is assumed to be readily available in UK laboratories. However, should this not be the case as standard, the costs of one-off purchases of this equipment will be included in the laboratory budget and thus indirectly accounted for using the MOLU system identified above.

Treatments

As per recommendations from NICE clinical guideline CG71,1 the recommended treatment for patients with FH is high-intensity statin therapy (usually atorvastatin 80 mg). For patients at risk of CHD based on high lipid levels but who do not have FH, the recommended treatment is low-intensity statin therapy (e.g. simvastatin 40 mg). The cost of atorvastatin is due to decrease during the course of this assessment and is likely to be equivalent to that of generic simvastatin. The implications of this are explored in the cost-effectiveness analysis. Costs of a number of other statin-based therapies such as rosuvastatin, pravastatin, etc. are considered. Other treatments include ezetimibe (evidence of efficacy uncertain) and bile acid sequestrants (costly).

Other costs

Other cost considerations include the cost of health-care professionals to identify family pedigree and the costs of initiating contact with relatives for cascade testing. Costs of annual follow-ups for patients diagnosed with FH are also considered in the analysis.

Index cases

The assessment investigates the use of diagnostic strategies including Elucigene FH20 and/or LIPOchip for providing an unequivocal diagnosis of FH for those with a clinical diagnosis based on the Simon Broome criteria.

Cascade testing of relatives

The assessment investigates the use of diagnostic strategies including Elucigene FH20 and LIPOchip for cascade testing to identify FH in the relatives of index cases. The use of Elucigene FH20 or LIPOchip for cascade testing depends on the mutation detected in the index case and the cost of targeted gene sequencing. (In index cases with an identified genetic mutation, targeted gene sequencing is also considered for cascade testing of relatives. In index cases without an identified genetic mutation, cascade testing using LDL-C concentration measurement is considered.)

A scenario encompassing a single test strategy (Elucigene FH20 or LIPOchip) that does not end in CGA for test-negatives may not detect all cases of FH. In such a scenario there may be implications for test-negative patients in terms of how their condition is managed.

Population

The population considered was adults and children with a clinical diagnosis of FH (the index cases/probands) based on the Simon Broome, Dutch or MedPed criteria and, for cascade testing, the first-, second- and third-degree biological relatives of the index case. (In the protocol for the review we stated that we would consider those with a clinical diagnosis based on the Simon Broome criteria as recommended for clinical diagnosis of FH in the UK. However, we also identified a few studies based on the Dutch and MedPed criteria and in consultation with our clinical advisers we relaxed our inclusion criteria to also include studies in which participants had received a clinical diagnosis of FH based on these criteria, as clinical advice suggested that these criteria were sufficiently similar to the Simon Broome criteria and if consistently applied would also provide potentially useful evidence.)

Given sufficient evidence, subgroup analysis was to be undertaken on the performance of Elucigene FH20 and LIPOchip in ethnic populations.

Setting

The settings considered were secondary or tertiary care.

Interventions and comparators

The interventions considered were Elucigene FH20 and LIPOchip for index cases and cascade testing of relatives. The comparators considered for testing in index cases were (1) CGA and (2) LDL-C concentration measurement (Simon Broome, Dutch or MedPed criteria). The comparators considered for cascade testing of relatives were (1) targeted gene sequencing and (2) LDL-C concentration measurement (age- and gender-specific criteria as recommended in NICE clinical guideline CG711).

Reference standard

The reference standard was CGA in combination with the Simon Broome, Dutch or MedPed criteria. CGA was defined as the ‘most complete genetic analysis’ generally available for FH within a diagnostic setting and is expected to detect almost all known FH-causing mutations. This analysis includes DNA sequence analysis of the promoter, all exons and the exon/intron boundaries and into the 3′ untranslated region of the LDLR gene, which will detect the majority (∼88%) of detectable FH mutations, MLPA for each exon and the promoter region of the LDLR gene to detect deletions and duplications (∼5% detectable FH mutations) plus analysis for the common APOB p.Arg3527Gln gene mutation (∼5% of FH mutations) and the PCSK9 p.Asp374Tyr gene mutation (∼2% of FH mutations).

During the screening process it was ascertained that some studies reporting genetic analysis did not fulfil all of the above criteria for CGA, for example:

• LDLR, APOB and PCSK9 gene analysis but testing for deletion/duplication was carried out using a process other than MLPA such as Southern blot analysis or quantitative multiplex PCR methodology (QMFSP)

• LDLR and APOB gene analysis, but no PCSK9 analysis.

Therefore, we took a pragmatic decision to still include studies reporting such an ‘incomplete CGA’ and to assess the quality of such a reference standard in terms of comprehensiveness and variations in test accuracy.

Studies reporting the following single genetic analyses were excluded:

• APOB gene analysis only

• PCSK9 gene analysis only

• test for deletion/duplication only.

In the event of a sequential mutational detection strategy used for the diagnosis of FH, for example Elucigene FH20 followed by gene sequencing for those negative on Elucigene FH20 and then followed by MLPA tests for those negative on gene sequencing, the combination of these sequences could be considered to be CGA.

Low-density lipoprotein cholesterol measurement as part of the clinical diagnosis was one of the comparators. Estimates of the accuracy of LDL-C using the reference standard of CGA plus a clinical diagnosis that includes LDL-C measurement are likely to be inflated compared with the estimates of accuracy of other index tests being evaluated. Therefore, for inclusion of studies reporting the diagnostic accuracy of LDL-C (which is a part of the clinical diagnosis), we considered the estimates of accuracy of LDL-C against a reference standard of CGA (either most complete or incomplete) only.

Outcomes

The following outcomes were considered:

• test accuracy: sensitivity, specificity, positive likelihood ratio and negative likelihood ratio.

In any studies reporting the above outcomes the following outcomes were also considered:

• proportion of cases with an unequivocal diagnosis identified by Elucigene FH20 and LIPOchip

• proportion requiring CGA after Elucigene FH20 and LIPOchip

• proportion of FH identified from cascade testing

• acceptability of the tests

• interpretability of the tests.

Test accuracy data on the absolute numbers of true-positives, false-positives, false-negatives and true-negatives were extracted or calculated from the information provided in the studies. We also considered studies in which derivation of a complete 2 × 2 diagnostic table was not possible but which reported data to allow derivation of one of the test accuracy measures, for example sensitivity but not specificity.

Study design

The following types of studies were considered:

• direct (head-to-head) studies in which the index test, comparator test and reference standard test were carried out independently in the same group of people

• randomised controlled trials (RCTs) in which people were randomised to the index and comparator test(s) and all received the reference standard test.

In case of insufficient evidence from direct and randomised studies, indirect (between-study) comparisons in the following types of study were also considered:

• diagnostic cross-sectional studies comparing the index test or comparator test against a reference standard test

• case–control studies in which two groups were created, one known to have the target disease and one known not to have the target disease, in which it was reasonable for all included to go through the tests.

Exclusion criteria

The following types of reports were excluded:

• preclinical and biological studies

• reviews, editorials and opinions

• case reports

• reports investigating technical aspects of a test.

Non-English-language reports were excluded.

Diagnostic accuracy metrics

For the purpose of this assessment, we define test-positive and test-negative as follows:

• Elucigene FH20/LIPOchip tests: those with a FH-causing mutation detected by Elucigene FH20 or LIPOchip were defined as ‘test-positive’ and those with no mutations detected were defined as ‘test-negative’.

• LDL-C tests (as a part of the Simon Broome criteria): we assumed that people with positive clinical criteria would have positive cut-offs of LDL-C as suggested in the definition of the criteria. A minimum LDL-C level of 4 mmol/l is required to diagnose index cases.

• Age- and gender-specific LDL-C test (as recommended in NICE guideline): those with LDL-C levels greater than the cut-offs were defined as ‘test-positive’ and those with LDL-C levels lower than the cut-offs were defined as ‘test-negative’.

• True-positives: people with clinical FH who are positive on tests (Elucigene FH20 or LIPOchip or LDL-C as part of the Simon Broome criteria or age- and gender-specific LDL-C) and positive on CGA.

• False-negatives: people with clinical FH who are negative on tests, but positive on CGA.

• False-positives: people with clinical FH who are positive on tests, but negative on CGA.

• True-negatives: people with clinical FH who are negative on tests and negative on CGA.

• Sensitivity = true-positive/(true-positive + false-negative) × 100.

• Specificity = true-negative/(true-negative + false-positive) × 100.

Quantity of studies identified

The searches identified 1529 records for the review of test performance. Following screening of titles and abstracts, 1296 articles were excluded and full-text reports of the remaining 233 articles were obtained for further assessment. Figure 3 shows a flow diagram outlining the screening process.

FIGURE 3.

Flow diagram outlining the screening process.

Appendix 6 lists the 15 studies (17 reports) that were included in the review of test performance (Table 6 lists the studies, tests evaluated, publication status and other linked reports). Of the 15 studies, three (four reports) reported Elucigene FH20,36–38 five (six reports) evaluated LIPOchip,39–43 four reported LDL-C compared with genetic analysis44–47 and three reported age- and gender-specific LDL-C for cascade testing of relatives. 48–50 We did not identify any studies reporting a combination of the index tests, that is Elucigene FH20 and LIPOchip.

Number and type of studies excluded

A list of the 221 potentially relevant studies identified by the search strategy for which full-text papers were obtained but which subsequently failed to meet the inclusion criteria is given in Appendix 7.

Study design

All of the studies were diagnostic cross-sectional studies evaluating the performance of Elucigene FH20,36–38 LIPOchip39–43 or LDL-C44–50 against a reference standard of genetic analysis (either incomplete or complete in terms of the definition of CGA as stated in Inclusion and exclusion criteria) in which all participants received a clinical diagnosis using Simon Broome, Dutch or MedPed criteria. No RCTs were identified that randomised participants to any of the tests of interest with all receiving a reference standard test.

Country and setting

Of the eight studies evaluating Elucigene FH20 or LIPOchip, four were conducted in the UK37,38,40,41 (two37,38 reporting Elucigene FH20 and two40,41 reporting LIPOchip) and one was conducted in Australia36 (evaluating Elucigene FH20), with the remaining three taking place in Spain39,42,43 (all of which reported LIPOchip). Of the seven studies reporting the performance of LDL-C (in index cases or for cascade testing of relatives), one each was conducted in the UK,48 Spain,44 Denmark,45 Austria,47 Japan46 and the Netherlands. 50 The study by Starr and colleagues49 included participants from the Netherlands, Denmark and Norway. When reported (seven studies), the clinical diagnosis was performed in lipid clinics.

Clinical diagnosis

The clinical diagnostic criteria used tended to differ according to the country where the study was carried out. The studies by Palacios and colleagues,41 Callaway and colleagues,40 Taylor and colleagues,37 Yarram38 and Lee and colleagues48 were conducted in the UK and their participants had a clinical diagnosis based on the Simon Broome criteria. Of three studies set in Spain, those by Stef and colleagues42 and Alonso and colleagues39 used the Dutch criteria, whereas the study by Tejedor and colleagues43 employed the MedPed criteria. The study by Hooper and colleagues,36 set in Australia, used the Dutch criteria whereas the study by Widhalm and colleagues,47 set in Austria, used the MedPed criteria.

In the studies by Civeira and colleagues44 and Damgaard and colleagues45 patients were given a clinical diagnosis followed by a genetic diagnosis and were then retrospectively classified by the Simon Broome, Dutch and MedPed criteria. Civeira and colleagues44 used an initial clinical diagnosis based on the MedPed criteria, whereas Damgaard and colleagues45 included participants who fulfilled two of the following three criteria: (1) LDL-C > 6 mmol/l, TC > 8 mmol/l and triglycerides < 2.5 mmol/l; (2) tendon xanthomata; and (3) a history of coronary artery disease before the age of 60 years in the patient and/or in a first-degree relative and/or hypercholesterolaemia in a first-degree relative.

In the studies by Starr and colleagues49 and Mabuchi and colleagues46 a genetically tested cohort of relatives was recruited to study the test performance of age- and gender-specific LDL-C cut-offs and a cut-off of 4 mmol/l, which is the minimum cut-off required by Simon Broome criteria respectively. In the study by Starr and colleagues, clinically diagnosed index cases based on the Dutch criteria (the Netherlands) and a combination of lipid levels, clinical characteristics and family history (Norway and Denmark) were included, whereas the study by Mabuchi and colleagues46 included clinically diagnosed index cases based on TC (≥ 5.9 mmol/l and < 12.9 mmol/l) with tendon xanthomata or primary hypercholesterolaemia with/without tendon xanthomata in a family with FH patients among first-degree relatives. The study by Wiegman and colleagues50 recruited relatives of index cases with a clinical diagnosis of FH based on the MedPed criteria or from a genetic diagnosis.

Participants

In the studies by Taylor and colleagues,37 Damgaard and colleagues,45 Mabuchi and colleagues,46 and Tejedor and colleagues43 the participants were all adults. In the study by Wiegman and colleagues50 the participants were children. In the studies by Starr and colleagues49 and Widhalm and colleagues47 the participants were a mixture of adults, adolescents and children, whereas in the study by Civeira and colleagues44 they were adults and adolescents. The remaining seven studies36,38–42,48 (six abstracts and one full text) did not specify whether the participants (index patients or relatives) were adults, children or adolescents.

Eight studies reported diagnostic accuracy in index cases only,36,39–44,46 whereas four37,38,45,47 reported this information both for index cases and for cascade testing of relatives, with the remaining three studies48–50 reporting test performance for cascade testing of relatives only. In studies reporting cascade testing of relatives these were all first-degree relatives apart from the two studies by Damgaard and colleagues45 and Lee and colleagues,48 in which this information was not specified.

For studies evaluating the test performance of Elucigene FH20 and LIPOchip, the sample size ranged from 22 patients40 to 2462 patients. 42 In studies reporting test performance of LDL-C, the sample size ranged from 26347 to 3294. 49 In studies reporting cascade tests through targeted sequencing the sample size of relatives ranged from 27 relatives (from 104 index cases)38 to 1034 relatives (from 591 index cases). 50

In the study by Lee and colleagues48 all included relatives were heterozygous FH (coming from homozygous FH index cases), whereas in the study by Mabuchi and colleagues46 none was homozygous. The rest of the studies did not report on the status of FH patients. Three studies43–45 reported the number of participants at baseline with coronary artery disease, which ranged from 15% to 20%, and xanthomata, which ranged from 16% to 56%. The mean LDL-C concentration of participants as reported in two studies ranged from 4.3 mmol/l to 5.7 mmol/l. 47,49

Only one study reported the proportion of participants by ethnic group. 37 In this study most of the patients were white British (85.4%), 5.8% were of European origin and very few were from ethnic minorities, including 1.7% of Middle Eastern origin, 4.5% of Indian-Asian origin, 1.3% of African or Afro-Caribbean origin and 0.8% from the Far East.

Characteristics of the tests reported by the included studies

Table 7 summarises the characteristics of the studies reporting Elucigene FH20 or LIPOchip, whereas Table 8 summarises the characteristics of the studies reporting LDL-C.

Overview

This section reports the performance of the index tests Elucigene FH20 and LIPOchip and comparator test LDL-C against a reference standard in the diagnosis of FH in index cases and for cascade testing of relatives. Elucigene FH20 and LIPOchip are designed to detect mutations that are most frequent in the European Caucasian population and which have already been identified using sequencing techniques, a ‘gold standard’ of genetic tests, in this population. Therefore, the mutational analysis of these techniques against the gold standard is most likely to give 100% sensitivity and 100% specificity in this population. Therefore, the results focus mainly on patient-level analysis at trial level. Studies evaluating Elucigene FH20 or LIPOchip comprised a sequence of genetic tests in which the participants received the index test as a pre-screen and test-negatives would then receive further genetic tests such as gene sequencing and MLPA. None of the studies directly compared Elucigene FH20 or LIPOchip with LDL-C, and none reported the acceptability or interpretability of the test or the clinical effectiveness outcomes resulting from use of the test. The results for each of the different tests are reported under the broad headings of Diagnosis of index cases and Cascade testing of relatives. This is followed by sections on other outcomes and subgroup analysis, including by ethnicity, region and type of gene, followed by a brief summary of the chapter. Individual study results are given in Appendix 10.

Diagnosis of index cases

Elucigene FH20

Table 9 shows the test performance results for the three studies that reported Elucigene FH20, by Taylor and colleagues,37 Hooper and colleagues36 and Yarram,38 involving 802 participants. Taylor and colleagues37 and Yarram38 reported the performance of Elucigene FH20 in detecting FH-causing mutations in patients with a clinical diagnosis of definite FH and possible FH and overall for both (Simon Broome criteria) against CGA, whereas Hooper and colleagues36 reported the performance of Elucigene FH20 for those with a clinical diagnosis of definite FH (Dutch criteria) against CGA. Sensitivity of Elucigene FH20 in detecting FH-causing mutations in overall clinical diagnosis ranged from 44.0% in the study by Taylor and colleagues37 to 52.0% in the study by Yarram. 38 Data were not pooled because there was no information on true- and false-positives and negatives in the study by Yarram38 to compute a CI.

TABLE 9 - Sensitivity: individual study results for Elucigene FH20

DFH, definite FH; PFH, possible FH; UFH, unclassified FH.

In Taylor 2010 initial testing was carried out with the FH13 kit and then later the FH20 kit was used, but detection rate data were reported as if all samples were tested using FH20.

Reference standard: Elucigene FH20 + MLPA for test-negative with Elucigene + sequencing for test-negative with MLPA (included LDLR, PCSK9, APOB genes).

Study, country	Diagnosis	Criteria	n	Sensitivity (%)
Hooper 200936 Australia	DFH	Dutch	63	28.6
Taylor 201037a England	DFH	Simon Broome	190	48.6
Taylor 201037a England	PFH	Simon Broome	394	40.2
Taylor 201037a England	UFH	Simon Broome	51	38.5
Taylor 201037a England	DFH/PFH/UFH	Simon Broome	635	44.0
Yarram 201038 England	DFH/PFH/UFH	Simon Broome	104	52.0

Taylor and colleagues37 reported sensitivities of 48.6% and 40.2% of Elucigene FH20 in detecting FH in patients with a clinical diagnosis of ‘definite’ and ‘possible’ FH respectively. Hooper and colleagues36 reported a lower sensitivity of Elucigene FH20 of 28.6% in detecting FH in those with a clinical diagnosis of definite FH. The difference may be explained at least in part by the fact that the two studies used different clinical diagnostic criteria and that the Elucigene FH20 kit was used in two different populations (UK and Australia), given that the kit is designed to screen for the most common mutations in the UK population. For the same reason, a pooled estimate was not calculated but the estimated sensitivities of Elucigene FH20 in confirming FH-causing mutations of definite FH are represented graphically to visualise the heterogeneity (Figure 7). The specificity of Elucigene FH20 in these studies could not be calculated as test-positives did not go on to receive a reference standard test.

FIGURE 7.

Forest plot of Elucigene FH20 sensitivity for patients with a clinical diagnosis of definite FH.

A previous report54 to Taylor and colleagues37 provided information on an earlier version of Elucigene (FH13). In this report the FH13 kit was validated against a reference standard of sequencing the LDLR gene in a patient population in which all patients were clinically diagnosed with definite or possible FH based on the Simon Broome criteria and all received testing with the kit and sequencing of the LDLR gene. The sensitivity of the kit was found to be 30% for patients with a clinical diagnosis of possible FH, 52% for patients with definite FH and 38% for those with a clinical diagnosis of definite or possible FH (Table 10).

TABLE 10 - Sensitivity: individual study results for Elucigene FH13

DFH, definite FH; PFH, possible FH.

Reference standard: Elucigene FH13 + SSCP/dHPLC (included LDLR, PCSK9, APOB genes but no MLPA).

Study	Diagnosis	Test(s) evaluated	n	Sensitivity (%)
Taylor 200754 (linked to Taylor 201037)	DFH	Elucigene FH13	400	52
	PFH	Elucigene FH13	400	30
	DFH/PFH	Elucigene FH13	400	38
	DFH/PFH	SSCP/dHPLC (LDLR only)	400	62

Mutation-level analysis

The previous report54 to the study by Taylor and colleagues37 reported that there were no false-positive and no false-negative results from the Elucigene FH13 kit for detection of FH-causing mutations in patients.

In the study by Taylor and colleagues,37 99 mutations plus eight different deletions and duplications were identified in total. Of the 20 mutations present in the Elucigene FH20 kit, three were not identified in any of the participants. Taylor and colleagues37 also reported the prevalence of each mutation that was present in the Elucigene FH20 kit and was identified in the study. The most frequently identified mutations were the mutation in the APOB gene, with a prevalence of 12%, and the three LDLR mutations, p.Gly218del, the intron 3 splice variant c.313+1G>A and the p.Pro685Leu exon 14 variant, with a prevalence of 5% of the total mutations detected.

LIPOchip

Table 11 shows the test performance results for the four studies that reported LIPOchip, by Alonso and colleagues,39 Callaway and colleagues,40 Palacios and colleagues41 and Stef and colleagues,42 and involving 3418 participants. The studies used different versions of LIPOchip. Palacios and colleagues41 and Callaway and colleagues40 reported results for LIPOchip version 8 (Spanish version but study conducted in the UK). Based on the sample reported in Palacios and colleagues,41 the manufacturer of LIPOchip provided further information on version 10, which contains mutations specific to the UK population. Alonso and colleagues39 and Stef and colleagues42 used Spanish versions but did not provide information on the version number. Although the Spanish versions are not specific to the UK, they cover the mutations that are more frequent in Western Europe including the UK.

TABLE 11 - Sensitivity: individual study results for LIPOchip

DFH, definite FH; NC, not calculable (test-positives on LIPOchip did not receive a reference standard test); NR, not reported.

Reference standard: (1) LIPOchip platform (DNA array + QMFSP for test-negative + sequencing for test-negative) (LDLR, APOB) (Alonso 200939); (2) LIPOchip platform (LIPOchip including copy number changes in the LDLR gene + sequencing of the LDLR gene for test-negatives on the chip) (LDLR, PCSK9, APOB) (Palacios 2010,41 Stef 200952); (3) Elucigene FH20 + dHPLC/sequencing + MLPA (LDLR, PCSK9, APOB) (Callaway 201040). Elucigene FH20 + SSCP/dHPLC/direct sequencing + MLPA against reference standard 2, sensitivity = 95% (n = 126).

Study, country	LIPOchip version	Diagnosis	Criteria	n	Sensitivity (%)	Specificity (%)
Palacios 201041 and data received from the manufacturer UK	Version 10, UK mutations	NR	Simon Broome	126	78.5	NC
Callaway 201040 UK	Version 8 (251 mutations)	DFH or possible FH	Simon Broome	22	33.3	93.8
Palacios 201041 UK	Version 8 (251 mutations)	NR	Simon Broome	120	56.9	NC
Stef 200942 Spain	247 mutations	NR	Dutch–MedPed	2462	94.5	NC
Alonso 200939 Spain	195 mutations	DFH or probable FH	Dutch criteria	808	78.0	NC

The sensitivity reported by the studies ranged from 33.3%40 (LIPOchip version 8, Simon Broome criteria) to 94.5%42 (Spanish version designed to detect 247 mutations, Dutch–MedPed criteria). Palacios and colleagues41 reported sensitivity of 56.9% for LIPOchip version 8, which was based on 126 samples (120 analysed). Based on the above sample, the sensitivity of LIPOchip UK version 10 would be 78.5% (51/65) (Progenika, 2011, personal communication) There was heterogeneity across the studies, particularly in relation to Palacios 201041 (version 8 LIPOchip) and Stef 2009,42 and therefore a pooled estimate was not calculated. The estimated sensitivities (with 95% CIs) of LIPOchip in confirming FH-causing mutations are graphically represented to visualise the heterogeneity (Figure 8). The heterogeneity may be explained at least in part by the fact that different versions of LIPOchip and different clinical diagnostic criteria were used. In addition, the difference may also be explained by the fact that the LIPOchip kit was used in two different populations (UK and Spain), given that the prevalence of the FH-causing mutations varies according to the country of origin. The specificity of FH detection by LIPOchip in three studies could not be calculated as test-positives did not go on to receive a reference standard test. 39,41,42 In the study by Callaway and colleagues40 the specificity of LIPOchip version 8 was reported to be 93.8% with one false-positive diagnosis.

FIGURE 8.

Forest plot of LIPOchip sensitivity for patients with a clinical diagnosis of FH.

None of the included LIPOchip studies reported accuracy data according to definite or possible FH.

Mutation-level analysis

In a mutational-level analysis, the studies by Alonso and colleagues,39 Palacios and colleagues41 and Tejedor and colleagues43 reported sensitivity and specificity of LIPOchip of around 100%. Results on validation with mutation-negative samples and LIPOchip-positive samples by sequencing and QMFSP against MLPA were reported in these studies. See Appendix 10 for the tabulated results.

In the study by Tejedor and colleagues,43 59/118 mutations were detected using this earliest version of LIPOchip. Palacios and colleagues41 reported that in 37 patients 17/251 mutations were picked up by LIPOchip and in 28 patients 25/251 mutations were picked up by sequencing. Overall, the mutation detection rate (by the LIPOchip platform) was 42/251 mutations in 65 patients. 41

Cascade testing of relatives

Discussion of included studies evaluating the cost-effectiveness of Elucigene FH20 and/or LIPOchip

One study57 was identified that met our inclusion criteria and evaluated the cost-effectiveness of one of the intervention tests. This study assessed the cost-effectiveness of LIPOchip in identifying and testing first-degree relatives of index cases identified with FH in a Spanish population. The analysis also included subsequent treatment with statins of test-positive individuals. Screening and treatment management was compared with a strategy of no screening and the perspective of the analysis was that of the national health system (payer). Cost-effectiveness outcome was measured as incremental cost per life-year gained. Clinical diagnosis of at-risk individuals was based on a uniform protocol for clinical diagnosis and genetic testing of index cases was carried out using the LIPOchip platform, which included the following diagnostic steps:

1. LIPOchip DNA array

2. multiplex quantitative PCR used to identify significant gene rearrangements (applied if DNA array was negative)

3. complete sequencing of the LDLR gene (applied if the previous two steps were negative).

Among confirmed cases, the DNA array had a specificity and sensitivity of 99.7% and 99.9%, respectively, for all 118 mutations tested. Once index patients were identified, first-degree relatives were tested using steps 1 and 2 above only. Effectiveness among relatives was based on relative risks adjusted for age and sex59 and applied to national mortality rates. Once identified and treated, it was assumed that mortality risk reduced relative to untreated patients. The total cost of detecting a positive case was €1447 based on the assumption that, to detect one positive case, 3.4 relatives would need to be tested. This was combined with treatment costs based on simvastatin 40 mg and costs of acute myocardial infarctions (MIs) avoided based on risks calculated from Wonderling and colleagues. 59 The cost-effectiveness was thus estimated based on cost per life-year gained as €3243 in the base-case analysis. Sensitivity analyses conducted varied the incremental cost-effectiveness ratio (ICER) from €1073 to €7235 per life-year gained. Probabilistic analyses indicated a 95% probability of cost-effectiveness at a societal willingness to pay for a life-year gained of > €7400 and a probability of 45% at a willingness to pay of €3450. The results suggest that genetic screening of first-degree relatives with LIPOchip in Spain is a cost-effective use of resources. The main limitation to this study in terms of this assessment is that there is no active comparator – it is assumed that no screening would take place in routine care. However, the study is useful and informative regarding the potential of LIPOchip. No studies were available reporting on cost-effectiveness for any of the other intervention tests.

Discussion of supplementary cost-effectiveness evidence

The remaining supplementary papers detailing cost-effectiveness of cascade testing among relatives using targeted cascade testing and other methods are briefly summarised and discussed below. Full data extraction pertaining to these reports is available from the NICE website as appendix D to the NICE clinical guideline document CG71. 1 None of these studies evaluated the cost-effectiveness of any of the tests specifically in index patients; however, all indicated cost-effectiveness of cascade testing for FH among relatives of known FH index patients. The five included studies are discussed briefly below.

Marang-van de Mheen and colleagues60 compared five screening options in the Dutch population compared with no screening: treating (1) all patients with cholesterol level > the 95th percentile for the general Dutch population; (2) individuals fulfilling treatment criteria based on Dutch Institute of Health Care Improvement guidelines on hypercholesterolaemia; (3) (1) above but only those untreated at screening; (4) (2) above but only those untreated at screening; and (5) all FH-positive patients. The Framingham equation61 was used to estimate risk, survival and costs and the economic outcome measure is cost per life-year gained. This is explicitly not recommended as part of CG711 for calculating risk in the Simon Broome population. The most cost-effective option is option (2) with an associated ICER of €24,376 per life-year gained. Discounting was not conducted and there are questions relating to generalisability to a NHS perspective.

Marks and colleagues62 completed a cost-effectiveness analysis of screening for FH patients aged 16–24 years from the perspective of the NHS. Strategies evaluated were universal screening, opportunistic screening (unrelated reasons), opportunistic screening (patients with premature MI) and full screening of all first-degree relatives diagnosed with FH. The main comparison for the analysis was no screening. The primary outcome measure was cost per life-year gained and the study showed that tracing family members (first-degree relatives) systematically was the most cost-effective strategy with an ICER of £3097 per life-year gained.

Marks and colleagues63 conducted additional work over a 10-year period estimating the cost-effectiveness of (1) family tracing of index cases and (2) systematically screening all 16-year-olds. Primary economic outcomes were cost per case detected and cost per death averted. The main comparison for the analysis was no screening and no incremental analyses were conducted between groups. Costs per case identified were £3505 (family tracing) and £13,141 (universal screening). Costs per death averted were £3187 and £1.6M for the family tracing and universal options respectively. Therefore, the authors conclude that a more targeted screening programme identifying relatives of index cases is more cost-effective.

Wonderling59 used data from the Dutch screening programme from year 2000 in a sample of 18- to 60-year-olds to estimate the cost-effectiveness of screening compared with no screening. Treatment was administered using statins and it was estimated that screening would prevent 26 MIs per 100 patients receiving statin therapy. Primary outcome measures for the economic analysis were cost per case detected and cost per life-year gained, which were $7500 and $8800 respectively. Results were sensitive to the price of statins and a worst-case scenario estimated that the ICER could increase to $38,300 per life-year gained.

The additional included study was an older version of the currently included Marks study. 64 Therefore, the up-to-date data have been reported. Other studies, including those by Leren,65 Humphries and colleagues66 and Hadfield and colleagues67 all suggest that genetic screening is a cost-effective use of NHS resources and should be implemented across the UK.

The main background for the economic modelling of these candidate tests comes from NICE clinical guideline CG71,1 in which an economic model was developed to compare DNA testing with LDL-C testing. The results showed that DNA testing was cost-effective with an associated ICER of £2676 per QALY gained. This model has been updated and integrated to account for the testing of Elucigene FH20 and LIPOchip and other plausible scenarios for the identification of FH and is described in more detail in the following sections.

We did not identify any other health economic models for the identification of FH that would be informative to the development of this assessment.

Identification of probabilities for the decision model

The probabilities used to populate this model were estimated using standard conventions of Bayes’ theorem. Basically, once we know the sensitivity and specificity of a test as well as the a priori probability of disease in the target population, we can calculate positive, negative, true-positive, true-negative and thus false-positive and false-negative values for the model. The formulae used for the calculation of each branch of the tree for single test strategies (e.g. CGA alone) are described in Table 19.

When tests are connected in series as add-ons to each other (i.e. the second test detects the same mutations as the first test plus additional FH-causing mutations), the theory is essentially the same but will be represented by the associated values of the second test. Taking Elucigene FH20 followed by CGA as an example, the positive rate will be [(proportion testing positive on Elucigene FH20 + proportion testing positive on CGA) – proportion testing positive on Elucigene FH20]. The proportions testing positive on Elucigene FH20 cancel each other out as they are incorporated in CGA and CGA detects all the mutations detected by Elucigene FH20 and more; therefore, the proportion testing positive on this example strategy is simply the value of the most comprehensive test in the strategy (i.e. CGA). A similar argument applies to Elucigene FH20 followed by LIPOchip.

For strategies in which MLPA is used as an add-on test to Elucigene FH20 or LIPOchip, the calculations are slightly different. As MLPA detects additional cases not detected using Elucigene FH20 or LIPOchip (we assume here that the detection of deletions and duplications on LIPOchip is inadequate and MLPA will still be needed to give a more robust estimate), the effect of the two tests in series is not as before. Therefore, for the calculation of true-negatives on Elucigene FH20 followed by MLPA, the effect will be multiplicative and can be calculated as [(1 – prevalence) × (specificity of Elucigene FH20) × (specificity of MLPA)]. The MLPA test has not been considered separately from CGA because, by definition, CGA will already include MLPA as part of the process.

Sensitivity and specificity values used in the calculations of the model are presented in Table 20 for information. More detailed information on sensitivity and specificity for all included studies is presented in Chapter 2. Studies chosen to inform the economic modelling fulfilled two main criteria: (1) they were based on patients with a Simon Broome definite FH or possible FH clinical diagnosis of FH (preferably in a UK population) where possible and (2) when tests were conducted in a number of different countries (outwith the UK) in a study, we have chosen the cohort with the largest sample size (unless some explicit reason existed why this would not be appropriate). These were assumed to offer the most robust estimates in the absence of UK data. When studies did not report clinical diagnosis based on the Simon Broome criteria or when evidence was of poor quality and limited usability, we obtained parameter values from Dutch and MedPed criteria instead. For reasons discussed in the statistical analysis, it has not been possible to pool estimates of sensitivity and specificity for a combination of definite FH and possible FH diagnoses across studies in a robust way because of study heterogeneity (see Chapter 2, Assessment of test performance). Therefore, single studies have been chosen based on the best available evidence and the most recent version of each test analysed. The impact of these choices on our base-case conclusions will be explored through the use of lowest and highest estimates available from all of the included studies, based on all clinical criteria (MedPed and Dutch criteria included), in sensitivity analysis.

It is important to note that there is likely to be some correlation between those patients detected on MLPA and those detected using LIPOchip. Clinical expert opinion (Dr Zosia Miedzybrodzka, University of Aberdeen, personal communication) suggests that the LIPOchip test may be inadequate to detect deletions and duplications and in practice MLPA may be required to give a more accurate diagnosis.

LIPOchip can be used within the model in two separate ways. First, the strategy ‘LIPOchip’ refers to the test purchased by a laboratory in the UK from the manufacturer and processed at the UK laboratory. Additionally, the manufacturer offers a service whereby blood samples can be sent to the manufacturer’s plant in Spain for analysis using a two-stage process, first testing with LIPOchip and then sequencing of the LDLR gene for those testing negative. This is referred to as LIPOchip platform (Spain). Because of its second stage, at an additional cost of €100, this test has a higher sensitivity. It is, however, not CGA as the process does not include MLPA. Therefore, the sensitivity is less than that of CGA. Clinical expert opinion in the UK suggests that, to be able to fully detect all deletions and duplications of the gene, the MLPA test would be required as LIPOchip’s own method of detecting these cases may be inadequate. Additional data presented at the spring meeting of the CMGS70 suggest that (using data from Bristol’s NHS Hospital Genetics Laboratory) LIPOchip version 10 may be inadequate to detect copy number changes compared with MLPA, with only two cases out of a sample of seven correctly identified using LIPOchip.

In addition, there is much debate about the true prevalence of detectable FH-causing mutations among patients testing positive (definite FH or possible FH) based on the Simon Broome criteria. There is also great variation in this number between laboratories and this is likely to be because of issues of ethnicity as some tests will have different detection rates based on different ethnic groups (see Chapter 2, Assessment of test performance for additional information). For the purposes of our base-case analysis, we have assumed that 36.5% of clinically diagnosed patients (Simon Broome definite FH or possible FH) will have an identifiable mutation. 37 Data from four regional Scottish genetics services (Aberdeen, Dundee, Edinburgh and Glasgow; Dr Zosia Miedzybrodzka, University of Aberdeen, 2010, personal communication) suggest that, between 2007 and 2010, this value was approximately 35% for the whole of Scotland based on data classifiable as definite FH or possible FH. This has been confirmed in personal communication with Dr Zosia Miedzybrodzka, who estimates that, for every three patients tested in Aberdeen using CGA, on average only one will have a detectable FH-causing mutation. NICE CG711 estimates, using data extracted from the UK FH Cascade Audit Project (FHCAP),70 that 80% of patients clinically diagnosed with definite FH will have a detectable FH-causing mutation and 30% of those diagnosed as possible FH will have a detectable mutation. Given that the FH audit 201018 identifies 36% as definite FH and 58% as possible FH (the remainder being homozygous or not stated), this would suggest that 46.2% of patients clinically diagnosed as definite FH or possible FH would have an identifiable genetic mutation using CGA. Other studies quote varying estimates of these values and so maximum and minimum values will be explored in the sensitivity analysis. It is estimated that 50% of first-degree relatives of an index case will have an inherited mutation. This evidence for first-degree relatives has been applied to second- and third-degree relatives in the model. The reason for this is that the process of cascade testing is an iterative approach. Second-degree relatives will not be tested using targeted sequencing unless a first-degree relative has an identified mutation. Therefore, it is assumed that the second-degree relative is in fact the first-degree relative of an individual with an identified FH-causing mutation and so will also have a 50% probability of having inherited that mutation.

Markov model

The Markov model for this assessment has been adapted from the model used for the estimation of treatment effect used to inform NICE CG71. 1 The model was developed by the Royal College of Physicians Guideline Development Group and is updated in this assessment. This model calculated the lifelong treatment costs and outcomes of high-intensity statin therapy for the management of FH and low-intensity statin therapy for the management of others at risk of CHD because of elevated lipid levels. In addition to those who were classed as well, a total of eight further health states were modelled [unstable angina, MI, peripheral arterial disease (PAD), stroke, heart failure, revascularisation, cardiovascular death and other death]. Baseline risks were sourced from NICE technology appraisal 9471 and relative risks were sourced from the Simon Broome register. Utility weights were sourced from the literature and validated by the health economist working on this assessment. Utility of the general population was taken from the Health Survey for England 1996,72 which is the most up-to-date data source for the UK general population, and was adjusted for age and sex differentials. Beneficial health outcomes were used to estimate QALYs based on reduced risks of cardiovascular incidents. These treatment effects were sourced from a meta-analysis of two RCTs, the Incremental Decrease in Clinical Endpoints Through Aggressive Lipid Lowering (IDEAL) and the Treating to New Targets (TNT) trials conducted as part of the NICE CG71 assessment. 1 Data from Versmissen and colleagues8 were checked against and found to be consistent with the assumptions and data used for CG71,1 in so far as they show the efficacy of statins in improving the clinical causes of cardiovascular disease and by extension the reduction in serious cardiovascular events such as MI. However, they do not describe the exact causal relationship between the improved clinical outcome and reduced events. The data from Versmissen and colleagues8 are consistent with those of the CG711 assessment in that they suggest efficacy of statins and by extension the reduction in serious cardiovascular events such as MI. Costs and outcome data have been updated to current values using the latest available literature in the field or inflated to current prices (2010/2011) if no updated literature was available. Further details of the model structure are available from the NICE website (appendix E to the clinical guideline document1). The perspective of this economic evaluation is that of the UK NHS and all costs and resource use are applied in accordance with NICE guidelines on the methods of technology appraisal. NICE recommends that, where possible, the desired economic outcome is cost per QALY gained. Treatment costs and QALYs gained are extrapolated to the patient’s lifetime horizon and discounted at a rate of 3.5% per annum in line with standard NICE methods. It was not deemed necessary to discount diagnostic costs for each individual as the time taken for diagnosis is < 1 year. Sensitivity analyses explore the impact of varying the discount rate for both costs and QALYs between 0% and 6%. All other follow-up clinical costs that are expected to occur annually once a diagnosis of FH has been made are discounted as described.

Clinical resource use

For the purposes of this evaluation, we have assumed that all index cases will have received a clinical diagnosis of FH based on the Simon Broome criteria. Resource use and costs associated with this diagnosis are common to all tests being evaluated and so are not included. This is standard economic evaluation practice to include only resource-use estimations which differ between tests under consideration. However, the resource use associated with tests after the initial diagnosis is important and has been considered in the analysis. It is assumed that, once the proband has a genetic test or LDL-C confirmation of FH, he or she will attend a lipid clinic to discuss treatment and lifestyle management of the condition. It is at this point that family pedigree will be identified and contact with relatives will be initiated. It is assumed that initially only first-degree relatives will be contacted as there would be no point in contacting second-degree relatives until a diagnosis was confirmed in first-degree relatives using a genetic screen. Table 21 details resource use and cost estimation for this process based on clinical expert opinion and Hadfield and colleagues. 70

Index cases or relatives diagnosed with FH are offered an annual follow-up appointment with a lipids specialist at an outpatient clinic. In the absence of a specific unit cost tariff for a lipids specialist, this service is assumed similar to a cardiologist appointment (Dr William Simpson, University of Aberdeen, 2011, personal communication) and is costed at £222 per outpatient consultation.

Diagnostic resource use

A new national activity unit has been developed for molecular genetics and cytogenetic tests in the UK. This is based on a weighted report and uses for molecular genetics an amplicon as the base unit. All molecular genetic tests are then assigned a relative number of units that slot into bands with some efficiency built in as the number of amplicons increases. This new methodology for measuring activity for molecular genetic tests was developed by collaboration between the CMGS and the UKGTN. The objective was to devise a transparent and consensus system for measuring molecular test activity that could be implemented by all laboratories. Tests are weighted by complexity so that, for example, simply booking in a sample has the lowest weight and sequencing a gene of over 100 exons, for example RYR2, the highest. All realisable costs of each laboratory are collated and a total cost of the service is then calculated including salaries, consumables, overheads, etc. Each laboratory can derive its own unit cost, based on dividing budget by activity, and thus in effect derive a cost per test. For example, a £1.2M service producing 30,000 MOLUs will have a unit cost of £40.00. This system of pricing has been modelled by most of the laboratories in the UK and has been accepted by the professional bodies and UKGTN as a suitable approach to establishing a national tariff for genetic tests. Details of the national MOLU bands are included in Appendix 11 for information. The MOLU system is not a perfect system of estimating costs, however, and the limitations are outlined in Chapters 1 and 5.

For the base-case analysis, transportation costs of samples (preferably blood samples) for DNA testing and blood samples for LDL-C testing are included. Based on clinical expert advice (Dr Gail Norbury, Guy’s Hospital, London, 2011, personal communication), an increasing number of genetics samples are tested by processing saliva samples. Saliva-based samples are less costly to transport as they are more stable and require only first-class postage; however, the kits to extract the DNA are substantially more expensive. These resource use differences, however, will be included in the MOLU consumables mentioned above based on 1 MOLU for DNA extraction. The majority of tests are carried out in the UK; however, LIPOchip may be processed by the manufacturer on site in Spain. The additional resource and transportation costs associated with sending a blood sample overseas via air are considered for the LIPOchip platform processed in Spain. This was assumed to take a cost of 1 MOLU, commonly applied in genetic testing to cost transferring samples to laboratories overseas. Therefore, a cost of £30 has been applied in the base case. Additionally, there may be extra costs associated with resampling an estimated 3% of samples (Progenika, 2011, personal communication). These costs are also incorporated.

Clinical costs

Costs of clinician time for treating patients, identifying a family pedigree, counselling relatives on the importance of their condition and contacting relatives themselves are estimated using Payment by Results (PBR) national tariffs where available (e.g. for a first appointment with a lipid specialist). For all other resource use, including clinical nurse specialist (to identify pedigree and counsel patients), GP time to confirm second LDL-C test and administrator time to contact relatives, costs are estimated using Personal Social Services Research Unit (PSSRU) unit costs of health and social care. 75 Costs are based on the median of the appropriate agenda for change pay scale and include overheads, training costs, insurance, annual leave, etc.

Diagnostic costs

Costs of genetic testing strategies vary greatly among laboratories, especially based on their area of expertise and also in relation to their size – the greater the laboratory size, the greater the throughput of samples tested and thus the lower the costs based on economies of scale through mass genetic testing. Laboratories that can keep their budget constant or can reduce it but increase the number of MOLUs produced will have lower unit costs. The incentive then is to reduce the total budget while maintaining or increasing output. This system is simplistic and transparent and is the method adopted by most laboratories in the UK in setting their genetic testing tariffs (Dr Zosia Miedzybrodzka, University of Aberdeen, and Dr Gail Norbury, Guy’s Hospital, London, 2011, personal communication). For the purposes of the base-case analysis, it is assumed that the MOLU cost is £30 per MOLU (Dr Kevin Kelly, University of Aberdeen, 2011, personal communication). The cost of each MOLU will be varied in sensitivity analysis provided by Dr Gail Norbury (£33 per MOLU). Unit cost estimation is adjusted within the model for strategies that have more than one test in order to account for the cost differentials associated with earlier positive test identification. The cost of DNA extraction is also incorporated into the analysis and receives a unit of 1 MOLU. Details of MOLU units applied and the associated costs for each test strategy are presented in Table 22. The cost of testing a hypothetical cohort of 1000 index cases with combination strategies is dependent on the numbers testing positive on the first test in that strategy. For example, in a strategy such as Elucigene FH20 followed by CGA for negatives, an index case who tests positive on Elucigene FH20 will not receive the second more comprehensive test.

In addition to the tests outlined above, the LIPOchip platform (Spain) as a genetic testing platform is a potential alternative to CGA. The test, which involves using the LIPOchip followed by sequencing of test-negative cases, is offered by the manufacturer (Progenika) at a cost of €250 for a LIPOchip test and €350 for the whole process. The associated costs are incorporated into the analysis using an exchange rate of €1 = £0.89. The LIPOchip platform processed in Spain is explained in Chapter 1. Briefly, this is a two-stage process whereby, if the sample is positive on LIPOchip, no further testing takes place. If the sample is negative on LIPOchip then the sample is sequenced for an additional €100. Therefore, assuming that the sensitivity of LIPOchip is the same regardless of where it is processed and using similar methodology to that in Table 22, we estimate the total cost of the strategy (before transportation of samples costs) as (1000 × 250 × 0.89) + (713 × 100 × 0.89) = £285,957.

The cost of targeted sequencing may also be estimated using the MOLU system. Targeted sequencing (including DNA extraction) is allocated a MOLU of 3. At a cost of £30 per MOLU, this would amount to £90 per targeted sequencing test. Based on the MOLU system, targeted sequencing is cheaper than Elucigene FH20 and is therefore the strategy of choice for cascading relatives.

Low-density lipoprotein cholesterol concentration measurements will be taken for all members of the study, regardless of testing strategy. Additional measures will, however, be carried out to confirm the diagnosis. Therefore, an additional two LDL-C tests will be required (at least one of which will be a fasting blood sample) to confirm the Simon Broome diagnosis if this is the method of diagnosis being adopted. It is assumed that, in order to get an extra blood test taken for the additional LDL-C measurement, an additional visit to a GP will be required. It is not expected that transportation costs of samples sent to laboratories for analysis will differ significantly between LDL-C and genetic tests as both require the transportation of potentially hazardous blood specimens.

Treatment costs

As discussed in Treatment options to be evaluated and as recommended by CG71,1 treatment will be of either high or low intensity, predominantly with statins. Should a patient be intolerant to statins, treatment may be administered using ezetimibe as per the NICE CG711 guideline. There is, however, some debate as to the relative effectiveness of ezetimibe monotherapy; therefore, only a small proportion of patients are likely to receive this treatment in practice (Dr William Simpson, NHS Grampian, personal communication). Also based on personal communication (Dr Anthony Wierzbicki), ezetimibe as monotherapy is ineffective and patients who have an inadequate response to statins may need to be treated with ezetimibe plus bile acid sequestrants. A number of FH patients will receive polypharmacy incorporating treatment with statins and ezetimibe. Table 23 details the unit costs per year of treatment with all of the potential drugs included in the modelling process with costs sourced from the British National Formulary (BNF). 76 To reflect differential treatment practice among clinicians, various combinations of these drugs (based on clinical expert opinions) are explored in the model. The most common combination therapies are included in Table 23.

For the base-case analysis, we used data from the FH audit 2010,18 the most up-to-date data source on FH treatment in practice. We also use data from clinical experts (Dr Anthony Wierzbicki, Guy’s and St Thomas’ Hospitals NHS Trust, 2011, personal communication, and Dr William Simpson, NHS Grampian, personal communication) to conduct sensitivity analysis surrounding the proportions of patients on each treatment as part of either a high- or a low-intensity statin therapy. The cost impact of atorvastatin, which is due to come off patent during the course of this assessment, will have implications for treatment costs in the model. This will be explored in sensitivity analyses.

Costs of cardiovascular events avoided as a result of treatment

Table 24 details the costs of cardiovascular events avoided. For the base-case analysis, these costs have been calculated using weighted averages of all Health Resources Group (HRG) codes pertaining to each cardiovascular event avoided. Elective and non-elective tariffs from PBR data for 2010–1174 are used and weighted for the numbers of elective and non-elective cases sourced from the Hospital Episodes Statistics online database (www.hesonline.nhs.uk/Ease/servlet/ContentServer?siteID=1937%26categoryID=192).

Data sourced from current NICE guidelines1 such as for subsequent MI are not available as part of PBR nor do any national tariff prices exist for these events. Therefore, values have been sourced from CG711 and inflated to current price levels for use in the model. Costs of cardiovascular death or other deaths have been assumed to be equal to £0 as it is not envisaged that this would have cost implications for the NHS. However, such deaths avoided would have great impact on the results from a societal perspective.

Base-case analysis

For the base-case analysis, we analyse an index patient of age 50 years, with an assumed average first-degree relative age of 50 years. The decision model is run on the basis of a hypothetical cohort of 1000 patients with a clinical diagnosis of FH based on the Simon Broome criteria (including both definite FH and possible FH). Cost and QALY values are estimated as described in the preceding sections and applied to the number of people passing through each branch of the decision tree illustrated in Figure 9. On the basis of test accuracy, a proportion of all 1000 index patients are positive (true-positive or false-positive) or negative (true-negative or false-negative). These patients are assigned the relevant cost and QALY values as described and total costs and QALYs are generated for the full cohort.

Test strategies are ranked in ascending order of cost. Those strategies that are more costly and less effective are excluded on the basis of simple dominance. Additional tests that are dominated by a combination or two or more alternative strategies are excluded by extended dominance. ICERs are calculated as incremental costs divided by incremental QALYs between non-dominated strategies. This is the most common method of presenting ICERs and relates the options sequentially ranked by costs. For the purposes of this assessment, the most relevant comparators are:

1. CGA, recommended indirectly by NICE guidance CG71. 1

2. LDL-C concentration measurement only. The reason for this is that, in practice, LDL-C is the main method of identification presently adopted in the UK (although genetic testing is more common in Scotland, Wales and Northern Ireland than in England).

Therefore, ICERs are presented as cost per QALY compared with the two suggested reference standards for this evaluation (LDL-C and CGA).

This process is applied to two distinct research questions. First, we investigate the cost-effectiveness of each of the 12 strategies for index cases alone. However, of greater importance and thus the primary focus of the analysis is to present cost-effectiveness estimates for the complete process of index case confirmation of clinical diagnosis but also for the identification and testing of relatives (i.e. the whole cascade testing process).

Subgroup and additional scenario analysis

The cost and QALY results for different age groups are explored in this section for the full cascading project only (i.e. index and relative cases). Results for index cases alone are presented in Appendix 12.

These subgroup analyses include a range of age profiles and also include the incorporation of any available evidence relating to the efficacy of statins in the treatment of children. To this end, we have completed a structured search of the literature, which has identified four systematic reviews of the efficacy of statins in children, the most recent of which is a Cochrane review of high quality that is used to inform the discussion and the model. 77 The data suggest that statins are efficacious in children in reducing cholesterol and have non-significantly different adverse events to placebo. Therefore, statins are likely to be safe in children with FH although long-term follow-up of this patient group is required. As data relating directly to CHD are lacking, treatment effect relative to CHD is assumed to be similar to that of a young adult (equivalent to a 30-year old index case in the economic model).

A number of age-specific subgroups were considered (probands aged 15, 30, 50, 65, 75 and 85 years). These age subgroups are similar to those used in previous economic modelling for FH1 and represent a good distribution of the ages of the population who may present for testing. Table 26 details the calculated number of relatives for each index case and their average age used in the model.

As discussed in Model structure, there may be alternative estimates of cost-effectiveness based on whether the index case is identified as definite FH or possible FH as their clinical diagnosis. It should be noted, however, that because of a lack of sensitivity data for each test separated into definite FH and possible FH subgroups, it was not possible to conduct robust analyses of FH cases split by clinical diagnosis subgroup. We have, however, conducted threshold analyses which show the combination of mutation prevalence and test sensitivity that would be required for the candidate test to be considered cost-effective as a pre-screen to CGA. The probabilistic sensitivity analysis accounts for the combined variation in all of the input parameters.

Sensitivity analyses

As many assumptions are made throughout the modelling process and selective data are chosen to inform the parameters, it is possible that the results generated will be sensitive to some of the judgement calls, assumptions and decisions made in the analysis. Therefore, we carry out a range of sensitivity analyses to determine the sensitivity of the base-case results to changes in our assumptions. A range of univariant deterministic analyses are presented in Appendix 14, the main results of which are reported and discussed in Analysis of uncertainty, including probabilistic sensitivity analysis. In addition, a probabilistic sensitivity analysis is also presented to explore uncertainty in the model.

Areas of uncertainty that are explored include:

1. Prevalence rates of FH-causing mutations among clinically diagnosed index cases and at-risk relatives.

2. Treatment differences for those with genetically confirmed FH and those without a genetic confirmation. The implication of forthcoming price reductions of atorvastatin is also explored.

3. Uncertainty surrounding the proportion of probands and relatives with a given test result receiving treatment (e.g. the proportion of those with a false-negative or true-negative test result receiving statin therapy).

4. The costs of diagnostic strategies, especially issues of uncertainty surrounding the MOLU pricing system and the likely cost of a 1-unit MOLU output.

5. Key assumptions relating to the model structure, including cascade testing only of first- and second-degree relatives, discount rates applied to costs and effects, the impact of not cascade testing negative index cases and the proportion of index and relative cases agreeing to participate in the identification and testing process.

6. Uncertainty associated with assumptions listed in Table 25 including structural assumptions regarding management of negative-testing index and relative cases.

Probabilistic sensitivity analysis

Deterministic one-way sensitivity analyses and point estimates of ICERs do not adequately provide information on the true impact of uncertainty surrounding the model parameters. Because of imperfect information on both the resource use and effectiveness of each treatment strategy, costs and QALYs are highly likely to be subject to at least some degree of uncertainty. Therefore, we conducted additional probabilistic sensitivity analysis using Monte Carlo simulation (5000 repetitions). Distributions were fitted to each of the parameters based on published studies (where available), CG71 data1 and a number of assumptions where no data were available. For example, where insufficient data existed in published sources to fit distributions to parameters, standard errors were assumed in order to calculate alpha and beta values. This may slightly under- or overestimate the variation in some of the parameters; however, it is not likely to impact greatly on resultant cost-effectiveness acceptability curves (CEACs). For sensitivity of test strategies (Elucigene FH20 and LIPOchip) the analysis was bounded by the highest and lowest reported mean values in all of the studies identified from the systematic review of the literature. Full details of probabilistic sensitivity analysis parameters are presented in Appendix 16.

The net benefit framework was used to estimate net monetary benefits for each simulation as described in Briggs. 78 The defining characteristic of this approach is that all strategies add to a probability of cost-effectiveness equal to 1. This uncertainty is illustrated in the form of CEACs for each of the non-dominated strategies of testing. CEACs for the base-case analysis are presented in the text, with supplementary analyses following the same approach for each age subgroup in the model presented in Appendix 15 for completeness. The analysis is presented for non-dominated test strategies only. The comparison for the calculation of incremental costs and QALYs for this analysis is LDL-C as this is current practice in the NHS. As the remit of this report is primarily to assess the cost-effectiveness for index cases and relatives, we have not conducted probabilistic sensitivity analysis for index cases alone. In addition, CEACs are presented for 5%, 10%, 20% and 50% mutation prevalence rates in order to reflect the uncertainty surrounding mutation detection rates in various subgroups of the population, primarily varying based on ethnic background.

Index (proband) familial hypercholesterolaemia patients

Table 28 presents total costs and total QALYs for each treatment strategy for index cases alone ranked according to cost with dominance or otherwise indicated. Patients without FH will have a slightly longer survival prognosis and will thus receive slightly greater QALY gains than those with FH. Such patients are clinically diagnosed as having FH, have high lipid levels and are at an increased risk of CHD and so will have a positive response to cholesterol-lowering therapy.

The Elucigene FH20 diagnostic strategy alone generates the lowest costs for identifying index patients for two reasons: first, it is the cheapest genetic diagnostic test available and, second, it detects the lowest number of true-positive index cases. Therefore, it confirms the clinical diagnosis in the fewest index cases with FH and for that reason is associated with the lowest QALYs of all of the tests included. LDL-C identifies the largest proportion of positives (not necessarily true-positives for FH – although all index cases are technically true-positives based on their clinical diagnosis) and has the highest QALY gain as it detects the greatest number of patients at increased risk of CHD. Of the non-dominated sequences, LIPOchip platform (Spain) and CGA are both associated with ICERs between £20,000 and £35,000 per QALY gained.

Index cases and relatives

However, where genetic testing has the greatest advantage over LDL-C is in the identification of relatives for cascade testing. Cascade testing using LDL-C alone is less likely to be cost-effective in this population because of the large number of false-positives (including relatives incorrectly identified as having FH) who may be treated using high-intensity statin therapy when low-intensity therapy would have sufficed to reduce their cholesterol levels. Assumptions regarding false-positive relatives are detailed in Table 25 and can be tested in sensitivity analysis within the model. Table 29 presents total costs and QALYs for index and relative cases combined (i.e. a whole integrated strategy for the identification and management of index cases and relatives with FH). Cascade testing of relatives of an index case with an identified mutation is by targeted sequencing. This is because targeted sequencing is less costly than both of the other candidate tests. Therefore, as all tests would detect the identified mutation they are supposed to in the relatives, targeted sequencing is the most cost-effective way to do this in relatives of a mutation-positive index case.

In the analysis presented in Table 29, LDL-C is the least effective of all tests. Elucigene FH20 is the least costly genetic testing strategy and is also the most cost-effective of all non-dominated genetic testing strategies relative to LDL-C, being less costly, more effective and thus dominant. CGA is the most effective non-dominated strategy in terms of QALYs gained, with an associated ICER of £2135 per QALY gained relative to the next most effective non-dominated strategy (LIPOchip platform, Spain). Combination genetic tests are dominated by single genetic test strategies. For example, Elucigene FH20 followed by LIPOchip is dominated by LIPOchip alone-meaning that the extra cost of pretesting with Elucigene FH20 does not add any additional QALYs over and above LIPOchip. The reason for this is that LIPOchip will detect the same mutations and cost more when added to Elucigene FH20. A similar argument can be made for tests used for pre-screening prior to CGA. The case for test strategies including MLPA as a component is slightly different in that MLPA detects deletions and duplications of the gene and so detects approximately an extra 5% of mutations that would not otherwise be detected by Elucigene FH20. MLPA is incorporated and included in the CGA process and has not been considered separately here. In relation to LIPOchip, there is some uncertainty in relation to the detection of deletions and duplications of the gene. Therefore, we have taken a pragmatic approach and included LIPOchip alone and LIPOchip followed by MLPA. This will allow the reader to decide, based on further investigation of LIPOchip, whether or not MLPA would be required to obtain a definitive diagnosis among positive test results (i.e. a specificity of 1) as assumed in the model.

Index cases

Tables 30 and 31 evaluate the non-dominated sequences compared with the relevant comparators for index cases. The scope and protocol for this assessment define two important comparators: (1) the comparator recommended as part of the NICE clinical guidelines – full genetic DNA testing (or CGA as defined in our protocol) and (2) LDL-C, which is currently the most commonly used method as part of the Simon Broome criteria to identify FH in practice in the UK. Currently, DNA testing is available only in 15% of UK primary care trusts (UK FH audit project 201018) and therefore LDL-C is deemed an appropriate comparator based on current clinical practice in the UK (NICE diagnostic advisory group, 2011, personal communication).

Of the non-dominated sequences, LDL-C is the most costly and most effective test overall (see Table 30). Elucigene FH20 is the least costly but also the least effective test in terms of QALYs. In fact, all of the non-dominated testing strategies are cheaper overall and generate fewer QALYs than LDL-C. Although diagnosis costs for LDL-C are lower than the alternatives presented, treatment costs are much higher. This is because as all index patients will technically have FH based on their clinical diagnosis on the Simon Broome criteria, they will benefit from statin therapy. Additionally, even if they were not true FH, they would still be at an increased risk of coronary artery disease based on their cholesterol levels and so would benefit from treatment. LDL-C is therefore also associated with the greatest number of QALYs gained for index cases. This is because, should a negative diagnosis be based on a genetic test, patients who test false-negative may be inappropriately treated and would thus gain fewer QALYs than if they were prescribed high-intensity treatment for their FH based on LDL-C levels.

Table 31 presents similar information for index cases alone when the comparator of interest is CGA.

When the comparison of interest for index cases alone is CGA, all other non-dominated genetic tests are less costly and less effective than CGA. The question for a decision-maker in this scenario would thus be whether or not the cost savings are worth the associated QALY loss. ICERs are not reported in informing such a question as there is lack of evidence regarding how much society is willing to accept in compensation (in the form of cost savings) for a QALY loss.

Figure 10 presents the cost-effectiveness plane comparing all tests for index cases. This confirms the results alluded to in the tables of results above.

FIGURE 10.

Cost-effectiveness plane (index cases).

There are two important things to note from this illustration. First, LDL-C is the most costly strategy (driven by high-intensity statin treatment costs). As all patients are at risk of cardiovascular disease, however, QALY gain is highest driven by the extra-intensive treatment based on false-positive diagnoses of FH by LDL-C. These patients benefit from the increased statin therapy as they are at increased risk of cardiovascular disease based on their cholesterol levels. Second, the graph illustrates the dominance of single test strategies over similar strategies preceded by less-sensitive screening tests such as Elucigene FH20. As all strategies ending in CGA generate the same QALY gains, dominance is due to greater costs amongst multiple test strategies.

Confirmation of clinical diagnosis in index cases and cascade testing of relatives

Tables 32 and 33 evaluate the non-dominated sequences compared with the relevant comparators for the full process of index case confirmation of the clinical diagnosis and cascade testing of relatives. The comparators are LDL-C and CGA as in the preceding section.

Multiple testing strategies are dominated by single testing strategies generating the same sensitivity and test-positive rate overall. All non-dominated genetic tests are highly cost-effective compared with LDL-C in the identification of index cases with FH and cascade testing of relatives (assuming that society’s willingness to pay for a QALY gain is £20,000). The Elucigene FH20 single test strategy is the most cost-effective, being less costly and more effective and thus dominant over LDL-C (see Table 32). However, should a decision-maker wish to have a DNA test with a definitive genetic diagnosis, (i.e. CGA) then this is more expensive but generates the most QALYs gained compared with LDL-C. Relative to LDL-C (current practice), CGA could be considered a cost-effective testing strategy with an associated ICER of only £1030 per QALY gained. This is also well below a willingness-to-pay value of £20,000 per QALY gained.

When cost and QALY pairs are compared with CGA (current NICE recommendations) for the whole process of identification of index cases and cascade testing of relatives, all non-dominated tests are less costly and less effective than CGA. Table 33 presents this comparison.

Again, as discussed previously, the reporting of ICERs for this scenario does not inform the question of what reduction in QALYs a decision-maker is willing to accept in order to achieve a predefined cost saving. All non-dominated testing strategies are less costly and also less effective than CGA.

Figure 11 presents the cost-effectiveness plane comparing all tests for index cases and cascade testing of first-, second- and third-degree biological relatives. This confirms the results alluded to in the tables of results above.

FIGURE 11.

Cost-effectiveness plane for index cases and cascade testing of relatives.

In this scenario (including cascade testing of relatives in the analysis), LDL-C used as a method of identification of relatives is less costly than all other tests (with the exception of Elucigene FH20) but does not generate the same QALY gains as any of the genetics-based tests. LDL-C is an inexpensive test to carry out (relative to other more costly genetic options); however, LDL-C alone will falsely diagnose many index cases as having FH and hence many relatives will be cascade tested unnecessarily for fewer QALYs gained. LDL-C is thus dominated by the lower-cost Elucigene FH20 test.

One-way deterministic sensitivity analyses

A range of one-way deterministic analyses are presented to investigate the sensitivity of the model to uncertainty in some of the key parameters and in relation to model structural assumptions as outlined in Table 25. All deterministic sensitivity analyses were carried out on the base case of an age 50 years index case.

A full range of sensitivity analyses in relation to treatment effect have been carried out previously in the NICE clinical guidance CG71. 1 The model was found to be insensitive to a range of sensitivity analyses including assumptions surrounding nurse and consultant time with patients, costs of cholesterol testing, costs of letters to relatives for cascading, cascading from alternative numbers of relatives (first and second degree), relative risks of non-cardiovascular disease deaths and treatment effect used in the model. As data from the CG711 assessment have been updated and used for the purposes of this report, it is highly unlikely that these sensitivity analyses will have any impact on the sequences of ICERs for this analysis. We have additionally explored the impact of including a cost of £80 (standard A&E tariff) for those patients who die in the model. This is to reflect any additional costs that may be involved over the £0 assumed in the base-case analysis. The results are not sensitive to this assumed value.

Therefore, the focus of sensitivity analyses for this assessment centres on parameters and assumptions that we hypothesise may have an impact on the sequence of ICERs or on the overall cost-effectiveness conclusions. Many parameters alter the cost-effectiveness of identifying index cases alone; however, as the remit for this report is primarily the detection and treatment of relatives with FH, we focus mainly on analyses that affect the overall outcome (i.e. the confirmation of index cases and the cascade testing of at-risk relatives). Full analyses for both groups are included in Appendix 14 for information. In the appendix, results for the index case analysis are presented first, followed by results for index cases and relatives together. The order of tables follows the sequence of results presented below.

The following discussion refers to index cases and relatives together.

Prevalence of familial hypercholesterolaemia-causing mutations among index cases and relatives

Prevalence of FH-causing mutations among index cases is varied between 28%79 and 52%. 41 For both low and high estimates of mutation prevalence, the order of the ICERs remains unchanged compared with the base case. Elucigene FH20 remains the most cost-effective strategy relative to LDL-C (associated ICERs = dominant and £395 per QALY gained for low and high estimates respectively). The next most cost-effective options after Elucigene FH20 are LIPOchip (platform processed in Spain) and CGA for both low and high mutation prevalence rates with all ICERs < £1300 per QALY gained. See Differential results for subgroups for a threshold analysis estimating the prevalence required for Elucigene FH20 or LIPOchip to be deemed a cost-effective pre-screen to CGA.

Prevalence of FH-causing mutations among relatives of index cases is an uncertain parameter that is generally held to be approximately 50%, based on the logic that one out of every two offspring will inherit a genetic mutation. Sensitivity analysis varied this assumption by ±20% to between 40% and 60% of first-degree relatives inheriting the culprit gene (author assumption). A low estimate suggests that Elucigene FH20 is dominant, being less costly and generating more QALYs than LDL-C. After that, as in the base-case analysis, LIPOchip platform (processed in Spain) and CGA remain the next most cost-effective testing strategies. A higher estimate of mutation prevalence among relatives of 60% suggests the same three non-dominated test strategies as in the base case, all with ICERs of < £1200 per QALY gained. Therefore, as similar tests are recommended as being cost-effective for all prevalence values considered, the base-case conclusions remain insensitive to any assumptions surrounding prevalence rates in either index cases or relatives with all ICERs for non-dominated strategies < £1300 per QALY gained relative to LDL-C.

Familial hypercholesterolaemia treatment

Analyses reducing the cost of atorvastatin did not change the base-case conclusions, with no difference in the sequence of the presented ICERs. Elucigene FH20 remains the most cost-effective option relative to LDL-C; LIPOchip (platform processed in Spain) is the next most cost-effective option followed by CGA, as was reported in the base-case analysis. ICERs for all three non-dominated tests are insensitive to changes in the cost of treatment used in the model (all reported ICERs are < £1100 per QALY gained relative to LDL-C).

Data in relation to the base-case model sourced treatment proportions for FH from the FH audit 201018 and assumed generic simvastatin treatment for those without confirmed FH. This assumption was tested using treatment proportions provided by Dr Anthony Wierzbicki (personal communication, Guy’s and St Thomas’ Hospitals NHS Trust, 2011). This assumed that both genetically confirmed FH and genetically non-confirmed FH patients would receive a range of treatments. This included polypharmacy for some patients including treatment with ezetimibe as well as statins. The order and magnitude of the ICERs relative to LDL-C remain similar to that in the base-case analysis.

The conclusions drawn are therefore not sensitive to changes in treatment pattern or to costs of treatment administered to patients.

The impact of the decision to treat negative-testing relatives or index cases

The base-case analysis assumes that 10% of negative-testing relatives will require treatment. However, it may be that 0% or at least no more than in the general population will require treatment. Therefore, sensitivity analysis investigates a scenario in which none of these relatives would receive statin therapy. In this scenario, the magnitude and order of the ICERs are very similar to those in the base-case analysis, with Elucigene FH20 remaining the most cost-effective strategy, dominating LDL-C. LIPOchip (platform processed in Spain) and CGA are the next most cost-effective options (ICERs of £902 and £1062 per QALY gained, respectively, relative to LDL-C). Hypothetically increasing this proportion to 50% does not lead to any significant change in the order or magnitude of the ICERs presented.

In an unlikely situation that negative index cases do not receive treatment or clinical follow-up, Elucigene FH20 is the only non-dominated genetic testing strategy and is less costly but less effective than LDL-C, the reason being that index cases testing negative for a FH-causing genetic mutation are still at significant risk of cardiovascular events and so not treating based on genetic mutation alone would lead to large numbers of at-risk individuals being missed, hence the reason LDL-C would be the most cost-effective strategy. It is important to note, however, that the above-mentioned analysis is for illustration only and is not necessarily a reflection of the true care pathway.

Costs of diagnostic strategies

Increasing or decreasing the MOLU costs associated with each test by ±£10 (varying cost per MOLU from £20 to £40) does not impact on the overall test order, with only minimal changes in the relevant ICERs. This is because the model is determined primarily around lifelong costs and health outcomes associated with treatment for FH or otherwise.

Sensitivity of key assumptions (model structure)

The assumption that cascade testing takes place of first-, second- and third-degree biological relatives of the index case is tested by assuming that the process stops after the second-degree relative regardless of test result. All genetic tests are even more cost-effective in this scenario. Elucigene FH20 is less costly and generates greater QALYs than LDL-C and is thus dominant. LIPOchip (Spain) and CGA are both associated with ICERs of < £800 per QALY gained.

The base-case analysis assumes that all index patients with a clinical diagnosis will have their family pedigree investigated, with cascade testing using targeted sequencing for relatives of genetically confirmed index cases. However, those that do not receive a genetic test or are test-negative will still be cascade tested using LDL-C. This is because, although a genetic mutation may not be detected, it is possible that such individuals have mutations or genes that have not yet been identified as causing FH. However, as a sensitivity analysis, we have explored the impact on the results of not cascade testing from such genetically test-negative index cases. In this scenario, all non-dominated genetic tests are actually less costly and less effective than LDL-C testing. Although the results are sensitive to this aspect of the model, clinical advice suggests that this would be highly unlikely in practice as cascade testing from negative index cases is a very important part of the cascade process. The results are not sensitive to assumptions regarding the proportion of index and/or relatives agreeing to have their family history investigated or agreeing for cascade testing to take place.

We varied the discount rate between 0% and 6% for costs and benefits as is standard practice in economic modelling to test our model to assumptions regarding uncertainty surrounding the value of future costs and health gains accrued over a lifetime horizon. For a discount of both 0% and 6% the order of the ICERs relative to LDL-C remained the same as in the base-case analysis. The magnitude of these ICERs showed no significant changes either. The results for the base-case analysis present estimates of cost-effectiveness based on current costs of CGA. However, the cost of genetic DNA sequencing will fall in the coming months and years with the development of next-generation (non-Sanger based) sequencing techniques. Therefore, we have explored the impact on the results of reducing the cost of sequencing by an estimated 40% (Dr Gail Norbury, Guy’s Hospital, London, 2011, personal communication). In this scenario, LIPOchip (platform processed in Spain) becomes extendedly dominated. Elucigene FH20 is dominant and CGA is associated with an ICER of £995 per QALY gained relative to LDL-C.

Sensitivity relating to diagnostic test accuracy

For each of the main tests we have investigated the cost-effectiveness based on studies reporting the highest and lowest sensitivity values for Elucigene FH20 and for LIPOchip. This gives a greater picture of the uncertainty across studies and the impact on associated cost-effectiveness results. It also reflects the sensitivity of our analyses to different population groups, some of whom may have greater sensitivity on Elucigene FH20, with others doing better with LIPOchip.

In relation to Elucigene FH20, the upper limit of the sensitivity analysis (0.5238) increases the ICER associated with Elucigene FH20 relative to LDL-C. This suggests higher proportionate increases in costs relative to proportionate increases in QALYs, thereby increasing the ICER between the two tests. Lower estimates (0.28636) work in the opposite direction and lead to Elucigene FH20 being dominant over LDL-C. Such findings are somewhat counterintuitive, with there usually being a positive relationship between higher test sensitivity and improvements in cost-effectiveness. The situation here, however, is more complex because of the clinical benefit (and QALY gain) of LDL-C at minimal cost as well as the addition of cascade testing. Higher sensitivity tests lead to a greater number of positive relatives being given a targeted sequencing test (which is more expensive). Although this test detects more true FH cases and generates greater QALY gain, this is offset somewhat by the advantages of LDL-C (individuals will gain improvements in QALYs regardless of whether or not they have FH, through statin-based therapy for their high cholesterol) that form part of the comparator testing. A similar situation arises with LIPOchip strategies relative to LDL-C. However, in all of these analyses, the rank ordering of Elucigene FH20, LIPOchip and CGA in terms of effectiveness and cost-effectiveness remains the same. As the sensitivity of Elucigene FH20 and LIPOchip increases, their associated ICERs approach that of CGA.

The sensitivity and specificity of LDL-C among relatives are taken from Starr and colleagues49 and varied according to the upper and lower bounds of the 95% CIs. In both scenarios, Elucigene is the most cost-effective option relative to LDL-C. ICERs tend to be slightly lower for genetic tests using the higher bound of the CI for sensitivity and slightly higher for the lower bound. These differences are, however, small in magnitude and the counterintuitive effect of test sensitivity in relation to the ICER can be explained as discussed above. Similar analysis of the specificity of LDL-C among relatives does not alter the sequences of the ICERs or the conclusions drawn from the relevant comparisons. Again, all non-dominated sequences are highly cost-effective relative to LDL-C.

In conclusion, based on the above analyses, the results show some sensitivity to changes in some parameters and structure for the confirmation of index cases alone, but are more robust to variations in key parameters when index cases and relatives are analysed together. In all scenarios presented, Elucigene FH20, LIPOchip (Spain) and CGA are cost-effective uses of NHS resources relative to LDL-C. There is some uncertainty surrounding the direction of movement of the ICER as a result of changes in the sensitivity of the tests that may seem counterintuitive. The results of the one-way sensitivity analyses should therefore be interpreted with caution and, for a more accurate measure of overall model uncertainty, probabilistic sensitivity analysis is likely to offer a better estimate.

Probabilistic sensitivity analysis

Probabilistic sensitivity analysis is carried out for the base case as described in the methods section. Figure 12 illustrates the results in the form of a CEAC.

FIGURE 12.

Cost-effectiveness acceptability curves for the base-case analysis.

This figure shows that, at low threshold values of willingness to pay for a QALY gain relative to LDL-C (< £2500), Elucigene FH20 has the highest probability of being cost-effective, but this reduces as the willingness to pay for an additional QALY increases. CGA is the most cost-effective option at threshold values > £2500 and is associated with a > 90% probability of being cost-effective at all threshold values > £3500 per QALY gained (Figure 12 is scaled down to aid discussion of low threshold values). The other non-dominated strategy, LIPOchip platform processed in Spain, is never associated with a probability of cost-effectiveness > 20%. Probabilistic analysis also generates similar results and conclusions for each age subgroup in the analysis (see Appendix 15). At threshold values of willingness to pay for a QALY gain approaching £20,000, CGA is always the most cost-effective option. This is an important point and confirms the generalisability of the base-case probabilistic results to other age groups.

In addition to deterministic analysis surrounding the mutation detection rate among clinically diagnosed FH patients, we considered some extra analysis based on input from Dr Anthony Wierzbicki (Guy’s and St Thomas’ Hospitals NHS Trust, 2011, personal communication), who states that, among his patient group, the majority of patients are possible FH and he estimates that the proportion likely to be detected by CGA is approximately 20–25% or may even fall to 5% in some population groups. With this in mind, we have conducted probabilistic sensitivity analysis for a range of potential mutation detection rates for CGA. The relevant CEACs are presented in Appendix 17 and show that the results are somewhat sensitive to this value in the model, especially at low threshold values and for lower rates of prevalence. For the lowest prevalence rate considered (5%), there is quite a bit of uncertainty at threshold values < £6000 per QALY gained. At very low values of willingness to pay (< £3000 per QALY), Elucigene FH20 is the strategy most likely to be cost-effective. LIPOchip platform processed in Spain is less likely to be cost effective except at very specific values of willingness to pay between £3000 and £4000 per QALY gained and at a low prevalence rate of 5%. However, this test is never associated with a probability of cost-effectiveness of > 50% regardless of prevalence rate or willingness-to-pay threshold. For higher estimates of prevalence (i.e. 10–50%), the results mirror those of the base-case analysis. However, of greater importance is that, for all prevalence rates of FH considered, CGA is the most cost-effective strategy at threshold values of > £5000 per QALY gained, increasing to 70% at the conventional value of willingness to pay of £20,000 per QALY gained for a prevalence of 5%. This probability increases to almost 100% for all other prevalence rates considered (i.e. 10%, 20% and 50%). Therefore, although there is some uncertainty surrounding the results based on varying mutation detection rates in clinically diagnosed index cases, probabilistic analysis shows CGA to be the most likely cost-effective use of NHS resources. The conclusion of the cost-effectiveness of CGA confirms the results of the previous NICE guidance in that the most comprehensive test for FH is cost-effective. NICE CG711 estimated that CGA was cost-effective with an associated ICER of £2676 per QALY gained versus LDL-C. Our results generate similar conclusions with a lower estimate of the ICER of £1030 per QALY gained relative to LDL-C. This is likely to be because of the cost reductions in CGA and in treatment over time.

Elucigene FH20 and LIPOchip studies

The included studies on Elucigene FH20 and LIPOchip reported a sequential genotyping test in which (1) the participants received a clinical diagnosis of FH followed by the index test (as a pre-screen) and then (2) those who tested negative received further genetic investigations such as gene sequencing and MLPA. CGA was the reference standard considered in the review. Assessment of the methodological quality of the included studies showed that overall the participants were representative of those who would receive the tests in practice.

Based on the data from the included studies we were able to deduce true-positive, true-negative and false-negative rates for each test. False-positive results could not be derived for any of the studies evaluating Elucigene FH20 or LIPOchip as only those who initially tested negative went on to receive the reference standard. Therefore, only sensitivity, and not specificity, of Elucigene FH20 or LIPOchip could be deduced and reported (sensitivity represented the percentage of cases with mutations found by CGA that are also detected by the candidate test).

Because of the sparse data on overall clinical diagnosis and variability in the LIPOchip versions used, sensitivity data could not be pooled. Therefore, we have provided a narrative overview and graphical presentation of the diagnostic performance of Elucigene FH20 and LIPOchip.

Low-density lipoprotein cholesterol studies

In general, the included studies on LDL-C (as part of the Simon Broome criteria for the diagnosis of probands or age-and gender-specific LDL-C cut-offs for the diagnosis of relatives as recommended by NICE guideline CG711) provided data on true- and false-positives and -negatives, allowing the calculation of both sensitivity and specificity. Again, because of the variability in both the clinical diagnosis and the comprehensiveness of the genetic tests used, sensitivities and specificities could not be pooled to provide a single estimate.

Diagnostic accuracy

The sensitivity of Elucigene FH20 and LIPOchip in detecting FH varied. Amongst UK populations with a clinical diagnosis based on the Simon Broome criteria, Elucigene FH20 was reported to detect 44% and 52% of those with FH-causing mutations that were detected by CGA. The UK has a population with a wide mutational spectrum and, as Elucigene FH20 is designed to detect a limited number of mutations, the sensitivity of the kit is largely dependent upon the prevalence of these specific FH-causing mutations in the population, hence resulting in variations. For example, predicted sensitivities of 32% in Wales48 and approximately 33% in Aberdeen, Scotland (prevalence with CGA 28%),79 were reported for Elucigene FH20 (by tallying the mutations that are covered in the Elucigene kit against the mutations that were picked up by CGA in those setting), which is lower than sensitivities reported by included studies. Moreover, it has been suggested that interpretation of the Simon Broome diagnostic criteria is not uniform throughout the lipid clinics in the UK and this may also lead to variation in the detection of FH. 70

Variation was observed across countries in the sensitivity of Elucigene FH20 in detecting FH in those with a clinical diagnosis of definite FH. Elucigene FH20 showed sensitivity of 49% in confirming FH in those with a clinical diagnosis of definite FH in a UK population (the prevalence of FH in the study was 28%), whereas sensitivity of only 29% was observed in an Australian population in which the prevalence of definite FH in the study was very high (78%). These differences could possibly be explained at least in part by the following two factors: (1) mutations included in the Elucigene FH20 kit were selected based on their frequencies in a sample of around 400 patients who were diagnosed as definite FH based on a Simon Broome diagnosis26 and (2) differences in the definitions of definite FH used in the two study populations (in the Dutch criteria a clinical diagnosis of definite FH does not require the presence of xanthomata, unlike the Simon Broome criteria).

The sensitivity of LIPOchip ranged from 33.3% (UK population) to 94.5% (Spanish population) using various versions. Using LIPOchip version 8, which contains 251 of the mutations most prevalent in a European population, the sensitivity observed in a UK population with a clinical diagnosis of FH based on the Simon Broome criteria ranged from 33.3%40 to 56.9%,41 while specificity was reported as 93.8%. 40 It should be noted that LIPOchip version 8 does not detect five mutations that are detected by Elucigene FH20 and which are common in the UK population; therefore, the sensitivity of the UK version of LIPOchip is likely to be higher. In the version of LIPOchip including frequent UK mutations (version 10), sensitivity would be improved to 78.5%41 (Progenika, personal communication) in detecting FH in a UK population. However, this was based on a very small sample size (n = 120) and only those with a confirmed genetic diagnosis were included and therefore the results should be interpreted with caution. None of the included LIPOchip studies reported accuracy data separated by definite FH or possible FH.

The sensitivity of LDL-C as part of the Simon Broome criteria compared with CGA was high (90–93%); however, specificity was low (28–29%) with a large number of false-positives observed. Nevertheless, only four of the LDL-C studies (three full text and one abstract) used the most complete CGA as defined in the assessment. The implications of this high false-positive rate are that potentially unnecessary additional tests or treatments may be given to those who do not have (genetically diagnosed) FH.

A risk of 10–20% of either incorrect diagnosis or misdiagnosis of FH has been reported. 81,82 In the UK, an overlap in LDL-C distributions amongst those with and without FH has been reported. It has been suggested that, because of the overlap in LDL-C levels, no cut-offs are 100% accurate. 22 Mabuchi and colleagues46 reported higher accuracy of LDL-C using a cut-off of 4.1 mmol/l (sensitivity and specificity of > 98%) among genetically diagnosed FH patients and unaffected relatives in Japan. The mean LDL-C levels amongst those with and without FH may differ from country to country, with some studies reporting an overlap in LDL-C distributions while others do not. 49 In the study by Mabuchi and colleagues46 the mean LDL-C level was 6.7 (SD 1.52) mmol/l in those with FH compared with 2.97 (SD 0.65) mmol/l in those without FH, with almost no overlap in LDL-C distributions. This could partly explain why using a LDL-C cut-off of 4.1 mmol/l was found to be highly sensitive in this population.

Cascade testing

In a family with FH, 50% of first-degree relatives are likely to have the condition. One of the advantages of genetic testing is that, if a mutation is identified in probands, targeted gene sequencing can be used in cascade testing of relatives to detect the culprit mutation and provide an unequivocal diagnosis of FH. Using targeted gene sequencing, the observed detection rate of FH in relatives ranged from 53% to 56% in two studies,37,45 which is broadly consistent with rates reported by others (37–56%; see Appendix 18). 11,19,67,83–85 A study by Wiegman and colleagues50 reported that a high detection rate (77%) in children from families in whom a mutation was identified in probands was observed through targeted sequencing. However, the authors of the study suggested that one possible reason for the high detection rate was that siblings with very low LDL-C levels were not taken to the clinic to undergo targeted sequencing. Moreover, children are present with monogenic causes of hypercholesterolaemia and are likely to have a higher detection rate of FH-causing mutations.

High sensitivity and specificity of age- and gender-specific LDL-C cut-offs compared with CGA were reported in cascade testing of relatives, suggesting the clinical utility of this approach in the absence of genetic diagnosis. In the study by Lee and colleagues,48 91% sensitivity and 93% specificity of cascade testing of relatives were reported using age- and gender-specific LDL-C cut-offs, although one explanation for the high values reported is that the included index participants were all homozygous for FH. In a subgroup analysis, Wiegman and colleagues50 reported sensitivity of 96% using LDL-C cut-offs of ≥ 3.5 mmol/l in children of parents with a clinical diagnosis of definite FH. 48,50 The authors of the study suggested that this sensitivity would apply to those children in a family with definite diagnosis of FH only.

Methodological quality of the included studies

We did not find any studies that directly compared Elucigene FH20 and/or LIPOchip with LDL-C (either as part of the Simon Broome criteria for the diagnosis of probands or age- and gender-specific LDL-C for the diagnosis of relatives) against a reference standard of CGA. A RCT86 (and its secondary report84) was identified, conducted in the UK, in which all participants received a clinical diagnosis based on the Simon Broome criteria and one group received a genetic test while the other received a LDL-C test using Simon Broome cut-offs. However, there were insufficient data to calculate the sensitivity and specificity of these tests in index cases, and also Simon Broome LDL-C cut-offs were used in testing relatives instead of age- and gender-specific LDL-C cut-offs; hence, this study was excluded from the assessment.

All of the included studies were cross-sectional in nature, with only two studies recruiting consecutive patients. 37,44 Abstracts were not quality assessed as they were not considered to contain sufficient information to allow for an adequate assessment of study methodology. In all but one study (LDL-C test46), patients were representative of the spectrum of those who would receive the test in practice. These were patients with a clinical diagnosis of FH based on the Simon Broome, Dutch or MedPed criteria or, for cascade testing, the relatives of those index cases with a confirmed clinical diagnosis of FH. The results from studies in which participants are clinically diagnosed based on the Simon Broome criteria would be of specific interest to UK practice.

An incomplete genetic testing strategy may result in mutations not being detected because of the limitations of the testing strategy. Only one study reporting Elucigene FH20,37 one study reporting LIPOchip39 and 50% (three out of six) of the LDL-C studies44,45,49 used genetic analysis that comprised DNA sequence analysis of the LDLR and APOB genes in conjunction with MLPA. Given that the FH-causing PCSK9 gene is rare and was discovered only fairly recently, those studies that otherwise met the definition of CGA but without assessing the PCSK9 gene were judged to include an acceptable reference standard in terms of classifying the target condition in this assessment. However, only two of the above studies that employed CGA did not perform PCSK9 analysis. 45,49 With respect to the reference standard used, all six abstracts (two reporting Elucigene FH20, three reporting LIPOchip and one reporting LDL-C) used adequately defined CGA, which comprised DNA sequence analysis of the LDLR and APOB genes (plus PCSK9 gene) in conjunction with MLPA that was likely to classify the target condition.

In the Elucigene FH20 and LIPOchip studies, only those who tested negative went on to receive further genetic investigations; thus, none of the test-positives received the reference standard (differential verification bias), giving rise to the possibility of overestimation of test performance. Because of the sequential nature of the tests used, these studies were also at risk of partial verification bias as neither the whole sample nor a random sample received verification with a reference standard. All of the LDL-C studies were free of partial verification bias and one was free of differential verification bias. 44 Test review bias (the results of the index test are interpreted with knowledge of the results of the reference standard test) was avoided in the Elucigene FH20 study, one of the LIPOchip studies and two of the LDL-C studies. It has been suggested that both test review bias and diagnostic review bias (in which the results of the reference standard test are interpreted with knowledge of the index test) may lead to higher values being reported for sensitivity. 87 However, these biases are of more importance in tests in which the results are based on subjective interpretation rather than automatically generated.

Psychological impact and acceptability of genetic testing

Evidence has suggested that genetic diagnosis of FH has no clinically relevant adverse psychological effects. 11,91 A RCT conducted in the UK that included probands with FH and their relatives found that there was no significant effect of genetic diagnosis on perceptions of control over FH, fatalism of FH, control over cholesterol or control over heart disease and adherence to risk-reducing behaviour; however, those with a confirmed genetic diagnosis had a strong belief in the efficacy of cholesterol-lowering drugs and a less strong belief in the efficacy of diet. 86 Another prospective comparative study conducted in the Netherlands among participants in a family-based genetic screening programme, however, found that those with an identified mutation perceived that they were at greater risk of heart disease than those with no mutation. The result was influenced by age, education, cholesterol level and cardiovascular disease in the family. 92

Cascade screening studies have reported high participation rates of 7311–90%. 19 In a UK cascade screening study in Oxfordshire, 97% of parents asked for their children to be screened. 93 In terms of approaches used, directly contacting relatives from clinics has reported higher participation rates19 than contacting relatives through probands. 93 Using both approaches resulted in a higher participation rate (73%) than contact through probands but was not higher than directly contacting relatives. However, there may also be concerns about the possible consequences of receiving a positive genetic test result. In a large genetic screening study in the Netherlands, 10% of individuals declined genetic testing because of the fear of negative effects on employment or insurance. 19

Although genetic diagnosis may be generally acceptable to patients, for clinicians making a diagnosis of FH still remains a challenge and dependent upon their judgement in circumstances in which patients have raised cholesterol levels but no identified FH-causing mutation. 94

Risk associated with different types of mutation

Based on the lipoprotein levels, mutations have been categorised as either ‘severe’ (functional null alleles and missense in exon 3/4; or functional null alleles plus splice variants) or ‘mild’ (missense outside exons 3/4 and splice; or any missense mutation). 84 A null mutation has been identified as one of the important risk factors associated with PCVD in FH patients. 43,95 A significantly higher risk of PCVD, recurrence of cardiovascular events and family history of PCVD was reported in patients carrying null mutations compared with patients with defective mutations. 95 The relative risk of PCVD in patients with a null mutation was 3.1 times higher than that in patients with a missense mutation. 43 The mean PCVD-free survival time in those with null mutations was 51–53 years, in those with missense mutations was 58 years and in those carrying defective mutations was 53 years (p < 0.01). 43,95

Taylor and colleagues37 reported a similar prevalence of severe mutations across study participants with a clinical diagnosis of definite FH, possible FH and also unclassified FH.

Index cases only

The base-case analysis refers to an index case aged 50 years and a mutation detection rate for FH equal to 36.5% of cases with a Simon Broome possible or definite FH diagnosis. With regards to the identification of index cases alone, Elucigene FH20 was the least costly test but also generated the least QALY gain because of the high number of false-negatives associated with this test. Accounting for the inclusion of both diagnostic and treatment costs (including clinical management), LDL-C was the most costly option for index cases alone but also generated the greatest number of QALYs gained. The reason for this is that all patients who meet the Simon Broome clinical diagnosis of FH will have elevated cholesterol levels as part of that diagnosis. The diagnosis is not definitive but patients with false-positive test results on LDL-C for FH will still gain from statin therapy on the basis of them having high cholesterol; the difficulty, however, with this strategy arises when cascade testing incorrectly takes place from false-positive index cases.

Index cases and relatives

Of greater relevance, however, to the decision problem is the identification of at-risk relatives of index cases in whom cascade testing should be carried out. Each 50-year-old index case will have on average five first-degree relatives, four second-degree relatives and four third-degree relatives still alive and eligible for contact for cascade testing. These numbers refer to the base-case scenario and will vary depending on the average age of the index case. For example, older people may have more second-degree relatives eligible for testing as they are likely to have living grandchildren. Similarly, younger people will have more grandparents alive than an index case aged 50 years. This variation is incorporated in the model results.

In the analysis of index cases and relatives, LDL-C as a stand-alone test is the least costly testing option. CGA dominates pre-screen tests also including CGA as part of the strategy of testing. This is because of the assumption that QALY gains for FH are not time sensitive and the extra time taken to deliver a tiered-strategy diagnostic test will have no implications for treatment or QALY impact. This suggests that, although pre-screen tests such as Elucigene FH20 and LIPOchip (Spain) are less costly in their own right, they do not offer overall cost savings as a pretest to CGA, suggesting that the extra costs associated with running negative samples through all tests in a sequence outweigh the cost savings of those detected as positive on either Elucigene FH20 or LIPOchip. Only at very specific prevalence and sensitivity combinations would Elucigene FH20 be a cost-effective pre-screen to CGA. As the cost of gene sequencing falls in the future, it is less likely that targeted tests (at current prices) will offer a cost-effective pre-screen strategy for the majority of the population in whom testing would be carried out.

Of greater interest, however, is the comparison of the non-dominated tests with the relevant comparators for this assessment. CG711 recommended that DNA testing in combination with LDL-C testing was the most cost-effective strategy to test index cases and identify relatives for cascade testing. However, in practice, uptake of DNA testing for FH has been very slow, especially in England, where the 2010 FH audit18 suggests that only 12% of trusts have access to a formal system of cascade testing, with a further 14% stating that such a system is in development. One issue for this may be a lack of funding in the area, and clinical advice suggests that, in reality, LDL-C testing in combination with the Simon Broome criteria is a more realistic reference standard for this assessment. With this in mind, the main comparison of non-dominated sequences is with LDL-C. However, we also report sequential results ordered by cost and a comparison of cost and QALY differences against CGA for completeness.

When compared with LDL-C, CGA is a highly cost-effective diagnostic test and is the only option that gives an unequivocal diagnosis of FH. This strategy is estimated to cost £1030 per QALY gained relative to LDL-C and thus at usual thresholds would appear to be a highly cost-effective use of NHS resources, and this is further confirmed through probabilistic sensitivity analysis. This finding is in agreement with similar findings from NICE clinical guideline CG71,1 which also found that DNA testing was highly cost-effective in the identification of relatives of index cases with FH. As reported in Chapter 2, there is always the potential that there exist undiscovered genetic mutations and culprit genes that may lead to FH. For this reason, cascade testing would be carried out from mutation-negative index cases using LDL-C testing as they are still technically FH positive based on their clinical diagnosis.

The LIPOchip manufacturer, Progenika, offers a service for testing samples in Spain. This platform offers LIPOchip as a pre-screen and a follow-up screen of all negative samples using sequencing of the LDLR gene. As LIPOchip tests for duplications of and deletions in the gene, this may be described by the manufacturer as CGA; however, there is much clinical uncertainty in relation to the accuracy of the LIPOchip method of detecting deletions and large rearrangements of the gene. Additional evidence suggests that LIPOchip will correctly detect only two out of seven exon copy number changes compared with MLPA. 69 The authors acknowledge that these data are based on small sample numbers; however, they raise further questions with regard to the accuracy of LIPOchip compared with MLPA. Therefore, MLPA would also be required to obtain a completely unequivocal diagnosis. The LIPOchip platform, processed in Spain, does not offer MLPA as part of the process and so it is assumed that the diagnosis obtained from this strategy would be inferior to that of CGA as described in this analysis. As this platform is slightly less sensitive, it is likely that there will be more uncertainty in the test result in about 5% of patients in whom the MLPA test would be required for additional confirmation. The LIPOchip platform processed in Spain is also found to be a cost-effective strategy, with an ICER of £871. Information provided by the manufacturer of LIPOchip suggests that 80.5% of index cases would be detected using the LIPOchip test and the remaining 19.5% would need gene sequencing to confirm the diagnosis (Progenika, 2011, personal communication to NICE). These figures refer to a definite FH sample being tested. However, it is estimated that of all the samples presenting for genetic testing, only 30–50% will have an identifiable mutation. Therefore, the estimates of cost provided by the manufacturer may have underestimated the number of samples testing negative on the LIPOchip test that would also require gene sequencing as the second stage of analysis. Equally, it may be that these costs would be reasonable because of manufacturer economies of scale. Should this strategy be recommended, it is imperative that the issue of price be confirmed before any decision is made.

The results do, however, suggest that if the recommendation to undertake CGA for all was considered impractical or too expensive (e.g. sufficiently large increases in CGA testing might lead to a requirement for extra laboratory space and associated infrastructure, not captured by the existing unit cost assumptions) and an alternative test (Elucigene FH20 or LIPOchip platform) is deemed appropriate, then either or both could be recommended. The test chosen for a particular population would pragmatically reflect the sensitivity of each of these tests within a particular clinic catchment area as it is found that tests may perform differently on groups of different ethnic origin. It may also reflect local resource conditions and clinical judgement, as there is a trade-off between costs and effects. The difficulty, however, in adopting such a strategy is that not all tests detect all individuals. Therefore, should any of these less-than-perfect testing strategies be accepted, they may miss out a number of potentially important cases. This may raise ethical and equity concerns as only a proportion of people will have the culprit gene identified and have their relatives followed up for genetic cascade testing.

The economic model results are more sensitive to changes in parameters for index cases alone than for index and relative cases together. As the latter is the main focus of the analysis of this project, these results are given the most weight in the discussion and reporting of results.

Relative to the comparator of LDL-C, probabilistic sensitivity analysis reveals that CGA is the most cost-effective diagnostic test strategy for the identification and testing of relatives with FH, with a probability of cost-effectiveness of > 90% for all age groups at all conventional values of willingness to pay for a QALY gain. Some variation is identified in the probabilistic analysis based on mutation prevalence rates, which may vary greatly in practice between different geographical areas. Although there is some variability in the results at low threshold values and at very low mutation prevalence rates, CGA does, however, remain the most likely strategy to be cost-effective relative to LDL-C (never dropping to < 50% probability at threshold values of societal willingness to pay of > £5000 per QALY gained).

Strengths

This work is important and builds on the health economics evidence generated as part of the NICE CG71 development process. 1 This economic modelling exercise identifies a range of potential diagnostic strategies that were not available for the CG711 assessment (namely LIPOchip and Elucigene FH20) and investigates their cost-effectiveness either as stand-alone tests or as pre-screens to reduce costs associated with CGA. However, one issue of importance is that the cost of CGA, and gene sequencing in particular, has fallen since the CG711 assessment and is likely to fall still further with the evolution of next-generation sequencing. This favours the CGA approach and means that, in the future, going forward, the costs of gene sequencing may well fall further with the evolution of new methods and new technologies. However, even at current prices (approximately £480 per test), CGA is still a cost-effective method of cascade testing. It is the most cost-effective strategy in terms of attaining an unequivocal genetic diagnosis of FH among index cases and for cascade testing those relatives at greatest risk of having the disease. The analysis considered all currently available and approved diagnostic tests and linked test accuracy with final treatment outcomes measured in terms of QALYs. As additional tests become available, it will be possible to incorporate these into the model and re-run the analysis incorporating newly available evidence and tests.

The model structure is a key strength of the analysis in that it presents a linked evidence approach linking intermediate outcomes (i.e. diagnostic accuracy of each testing strategy) to the associated lifelong costs and health outcomes associated with each test result and whether tests were true-positive, false-positive, true-negative or false-negative. This was explored for a total of 12 alternative testing strategies through decision tree and Markov model analysis.

A structured literature search was carried out to identify existing cost-effectiveness evaluations of these tests and of the cascade testing of relatives. No studies directly compared the interventions under consideration, and only one study detailed the cost-effectiveness of any of the new interventions relative with no testing. This referred to the LIPOchip test but was used in relatives. As targeted sequencing is a much lower cost method of testing identified relatives, this LIPOchip study was not relevant to this analysis. Our results, however, do confirm the findings from a number of other studies evaluating the cost-effectiveness of the cascade testing process more generally and also the findings of the previous NICE guideline. 1

Methods used to identify and obtain parameter estimates for the economic modelling sought to identify and utilise published sources and best available information; however, this was not possible in all cases. In such scenarios, for example the proportion of patients receiving various statin therapies and how treatment changes based on diagnosis, we have relied on clinical expert opinion from two or more clinical experts. Estimates of parameters are also sourced from the CG711 analysis where available and tested in sensitivity analyses.

Limitations

The model structure focused on the identification and treatment of index cases and relatives of index cases with FH as clinically diagnosed using the Simon Broome criteria. The costs and benefits of identifying other causes of similar symptoms have not been modelled in detail with the exception of the prescription of statin therapy for all index cases with high cholesterol levels. The impact of additional therapies such as diet, exercise, smoking cessation, etc. has not been included and is beyond the scope of this assessment. The effect of not including such detail in the model is uncertain; however, it is generally widely acknowledged that all patients with a clinical diagnosis of FH will require some form of active treatment, generally statin therapy.

Another challenge related to the analysis of subgroups of the population, especially in relation to FH in children. Relative risks associated with children aged 15 years are assumed to be similar to those of adults aged 30 years. There is little evidence linking the efficacy of statins directly to cardiovascular events avoided, and for this reason we have assumed risks similar to those of a 30-year-old in the model. This probably creates some uncertainty in the estimation of quality of life in this subgroup of the population. Although data do exist relating to clinical outcomes in children (i.e. TC level), these are not linked directly to cases avoided. As Avis and colleagues96 show, statins are efficacious in reducing cholesterol in children with FH and are not associated with significantly different adverse events to placebo; it is therefore assumed that health effects are similar to those of the next youngest age group in the model (aged 20–39 years). It is unlikely that this assumption greatly impacts on the cost-effectiveness results and associated conclusions drawn.

In terms of the estimation of costs for each testing strategy in the economic model, we have used the MOLU classification system to assign tests to bands and apply MOLUs to each band. This is an agreed costing mechanism devised by the UKGTN and CMGS with some built-in flexibility in pricing for each laboratory’s individual circumstances. This is not necessarily an accurate reflection of true economic costs or indeed opportunity costs associated with testing for FH. However, in the absence of price data for all combinations of tests considered, robust costing methods for genetic testing for FH or a NHS tariff for the tests as well as uncertainty surrounding variability from laboratory to laboratory, it was impossible to cost all testing strategies fairly using any other universally acceptable approach. Using the cost of the test alone would be insufficient as it would not account for staff time and consumables required. Therefore, although acknowledging its limitations, we have on the basis of expert opinion relied on the MOLU system for the estimation of diagnostic testing costs in the economic model. Studies retrieved from the cost-effectiveness searches showed great variability in the costs of testing for FH. The reason for this is that testing for FH and genetic testing more generally is a rapidly evolving discipline. Many studies presented alternative definitions of DNA testing and the costs of completing the tests have fallen almost yearly in recent years. Therefore, older estimates would be an overestimate of the true costs. This is another reason why the MOLU system was used for this assessment.

A number of other assumptions were made in relation to the clinical management of patients with and without disease. Much uncertainty exists among clinicians in the treatment of FH, with many advocating a start low approach followed by an increase in treatment intensity if a satisfactory response is not achieved. Others believe that, as statin therapy generates very few adverse events, it would be appropriate to treat everyone with a high-intensity statin (e.g. atorvastatin). The impact of this uncertainty is explored in sensitivity analyses and is found not to alter our base-case results and conclusions, with only very small differences in ICERs.

Further, in relation to the utility of diagnostic information, there is no published evidence that links the outcome from the results of a genetic test for FH (e.g. increased anxiety) to quality-of-life outcomes and hence QALYs gained. A number of plausible scenarios are possible, including reduction in QALYs emanating from the shock of knowing that one has a genetic disorder. Equally, however, people diagnosed with FH could gain some reassurance from the fact that they know what is causing their illness and can aim to develop a plan of action in dealing with this. Additionally, parents may place a positive value on knowing the source of a child’s illness. It is, however, likely that these factors would cancel each other out or favour genetic testing and would be much smaller in comparison with the QALY gains associated with being treated for a life-threatening condition such as FH.

An additional limitation of our analysis is that it is on an average patient level. We are aware that some mutations cause greater harm and are associated with greater risk of cardiovascular events. For example, mutations of the PCSK9 gene are associated with greater clinical risk. Also, there are differences between missense and nonsense genetic mutations. Although these are likely to have implications for the prognosis of individual patients, this has not been modelled as there is insufficient evidence available to estimate how this would impact on quality of life. Further, as cost and QALY differentials are based primarily around treatment decisions and the general insensitivity of the model to small changes, it is unlikely that analysing the data on a mutation level for the purposes of cost-effectiveness would generate great differences in results or conclusions. Further, such data would be available for only few if any mutations and the inclusion of these only would generate further ‘noise’ into the analysis.

There are limited data available for the sensitivity of the LIPOchip test, with wide variation in all of the reported studies for various versions of the test. This generates a lot of uncertainty surrounding the true sensitivity of LIPOchip as used to populate the economic model. As many of the tests were analysed at the manufacturer’s own laboratory, there is a lack of academic peer-reviewed information on this input for the economic model. To deal with the associated uncertainty we have conducted wide variation in the probabilistic sensitivity analysis, which should counter any biases that may have arisen as a result of a lack of critique of the estimate of sensitivity used for LIPOchip version 10 in the model.

A further limitation of the analysis refers to the accuracy of test sensitivity and specificity differentials between those relatives of genetically negative index cases and those relatives of genetically confirmed index cases. For the purposes of the economic modelling, we have assumed, because of a lack of relevant usable data, that sensitivity and specificity of LDL-C testing in both groups are similar. Although this may overestimate the sensitivity and specificity of LDL-C as a test for relatives of genetically negative index cases, it is justifiable on the basis that all index cases are clinically diagnosed with FH and so may well have genetic mutations that as yet may not have been discovered or have not been detected using any of the tests for this assessment. In terms of the cost-effectiveness results, this could represent a bias of uncertain magnitude in favour of CGA, although in the context of insensitivity to variations in this parameter in probabilistic analysis it is unlikely to alter overall results and conclusions.

American Association for Clinical Chemistry (December 2010)

URL: www.aacc.org

2010 annual meeting

2009 annual meeting

American Society of Human Genetics (December 2010)

URL: www.ashg.org

2010 meeting

2009 meeting

Atherosclerosis Supplements (December 2010)

URL: www.sciencedirect.com/science/journal/15675688

78th European Atherosclerosis Society Congress, Atherosclerosis Supplements 2010;11(2)

XII Brazilian Congress of Atherosclerosis, Brazilian Society of Cardiology, Atherosclerosis Supplements 2009;10(3)

XV International Symposium on Atherosclerosis, Atherosclerosis Supplements 2009;10(2)

European Society of Human Genetics (December 2010)

URL: www.eshg.org

European Human Genetics Conference 2010, European Journal of Human Genetics 2010;18(Suppl. 1)

European Human Genetics Conference 2009

https://www.eshg.org/eshg2009/abstracts.htm.

Fonazione Giovanni Lorenzini (December 2010)

URL: www.lorenzinifoundation.org

4th International Conference of Biomarkers in Chronic Diseases (Diabetes, Obesity and Cardiovascular Diseases) 2010

National Genetics Reference Laboratory (December 2010)

URL: www.ngrl.org.uk/Wessex/tech_meeting10.html

New and Developing Technologies for Genetic Diagnostics ‘10