Clinical and cost-effectiveness of clopidogrel resistance genotype testing after ischaemic stroke or transient ischaemic attack: a systematic review and economic model

Article history

The research reported in this issue of the journal was funded by the HTA programme as award number NIHR135620. The contractual start date was in August 2022. The draft manuscript began editorial review in March 2023 and was accepted for publication in July 2023. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ manuscript and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this article.

Copyright statement

Copyright © 2024 Carroll et al. This work was produced by Carroll et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This is an Open Access publication distributed under the terms of the Creative Commons Attribution CC BY 4.0 licence, which permits unrestricted use, distribution, reproduction and adaptation in any medium and for any purpose provided that it is properly attributed. See: https://creativecommons.org/licenses/by/4.0/. For attribution the title, original author(s), the publication source – NIHR Journals Library, and the DOI of the publication must be cited.

2024

Population

The population of interest for this appraisal is people who have had non-cardioembolic ischaemic stroke, minor stroke or transient ischaemic attack (TIA), and for whom clopidogrel treatment is being considered. Approximately 100,000 strokes occur every year in the UK and between 46,000 and 65,000 people experience a TIA. 2 Around 85% of strokes are ischaemic, occurring when the supply of blood to a part of the brain is interrupted, usually by a blocked artery. 2 It has been suggested that a TIA is not a separate pathological entity, but exists on an ischaemic stroke spectrum, constituting the mildest form. 3 Symptoms of stroke often occur suddenly and vary depending on the part of the brain being compromised. Symptoms tend to include issues with movement, speech, facial drooping and vision.

The median age for stroke in the UK is 77 years and a quarter of strokes in the UK happen in people of working age. 4 Lifestyle factors associated with stroke and TIA include smoking, alcohol and drug abuse, physical inactivity and poor diet. The presence of cardiovascular diseases, and medical conditions including diabetes mellitus, atrial fibrillation, chronic kidney disease and migraine, are also risk factors for stroke. 2 Other risk factors include previous stroke/TIA, family history of stroke, lower education and genetic or hereditary factors. Strokes are more common in people with African Caribbean or South Asian background (Stroke Association; Kings Fund). 5,6

People who have experienced a stroke or TIA are at an increased risk of further occlusive vascular events [e.g. ischaemic stroke, TIA and myocardial infarction (MI)]. 7 TIA precedes stroke in 15% of cases, providing a crucial opportunity to prevent more severe stroke. 8 Risk of stroke after TIA has been found to be approximately 8% at 7 days, 11.5% at 1 month and 17.3% at 3 months. Risk of recurrent stroke after a minor stroke has been suggested to be 11.5, 15 and 18.5%, respectively. 9 National Institute for Health and Care Excellence (NICE) TA210 recommends the use of antiplatelet medications as a preventative treatment for people who have had an ischaemic stroke or TIA. 10 This includes clopidogrel treatment and is discussed further in Place of the technology in the treatment pathway.

Target condition: clopidogrel resistance

Clopidogrel is an irreversible adenosine diphosphate (ADP)-receptor antagonist with antiplatelet properties. It is available as branded and generic preparations and has marketing authorisation for patients who have recently had an ischaemic stroke or TIA. 11

Clopidogrel is a prodrug, which needs to be converted (metabolised) into an active form by P450 CYP enzymes. 12 Inactive clopidogrel has no effect on platelet aggregation, and therefore does not prevent occlusive vascular events. A substantial proportion of the population are less able to metabolise clopidogrel to its active form, and clopidogrel does not achieve its full pharmacological effect in these patients. One of the main causes of impairment in the metabolisation process are genetic variants, mainly in the CYP2C19 gene, which encodes P450 CYP enzymes. This is known as ‘clopidogrel resistance’. In addition to genetic variations in the CYP2C19 gene, other factors that may cause or exacerbate clopidogrel resistance include taking drugs such as omeprazole, which compete for metabolism by the CYP450 system,13 and factors such as obesity, diabetes and hypertension. 14 There is also a potential role of other rare genetic changes. Thus, both genetic and clinical factors need to be considered when determining whether an individual will respond to clopidogrel treatment. Inhibition of platelet function can be measured in laboratories to assess the impact of these factors in each individual and their potential response to clopidogrel or other antiplatelet drugs, but the applicability of these tests is limited because of technical limitations and a lack of standardisation in the definitions of non-responders. 11

Diagnostic test

This review focuses on two categories of CYP2C19 genetic testing: point-of-care tests (POCTs) and laboratory-based tests. POCTs include any analytical test carried out by a healthcare professional outside of the laboratory, although it is also possible to install near patient testing equipment in local laboratories, which may overcome challenges associated with storage of reagents. 20 These tests have the potential to deliver results more quickly than standard laboratory-based tests. The two POCTs in scope are the Genomadix Cube CYP2C19 System and the Genedrive CYP2C19 ID Kit. The Genomadix Cube test was previously known as the ‘Spartan Cube’, which is a successor to the ‘Spartan RX CYP2C 19 System’. The two Spartan tests are very similar but there are some differences: the three reaction tubes have been integrated into a single test cartridge, the swabs and test cartridges are packaged separately and the DNA analyser device is smaller. 21 There are also differences in the mechanisms used to heat and cool the samples; the storage, use and stability of the specimens on the swab; the optical system; and the test workflow. 22

Laboratory-based tests are conducted by technicians in the laboratory. In the NHS, genomic testing is generally delivered by a network of seven Genomic Laboratory Hubs. Testing for CYP2C19 is not currently included in the National Genomic Test Directory of tests commissioned by the NHS in England. Table 2 provides an overview of some of the available CYP2C19 genetic tests. The POCTs only target specific LOF alleles. Laboratory-based tests have the potential to target all LOF alleles; however, commercial kits are likely to only test for the most common variants or those with established clinical utility. However, lab-based testing would have greater flexibility to alter variants screened for as new evidence emerges.

Place of the technology in the treatment pathway

Guidelines on appropriate antiplatelet therapy for the secondary prevention of stroke vary. The two main guidance documents of relevance are NICE guidance NG128 on stroke and TIA4 and guidance from the Royal College of Physicians (RCP) on therapy for secondary prevention for people with stroke. 23 The treatment pathway is shown in Figure 1 for (1) adults with non-minor ischaemic stroke and (2) adults with minor stroke or TIA. Pathways are different in children and for patients with atrial fibrillation. In children, aspirin rather than clopidogrel is currently recommended to prevent recurrence. Other antiplatelets, including clopidogrel, should only be considered when there are other risk factors for cerebrovascular disease. 24 People who have disabling ischaemic stroke, and who are in atrial fibrillation, should be treated with aspirin for 2 weeks after which anticoagulation treatment should be considered. 4

FIGURE 1.

Treatment pathway for current NHS practice: (i) non-minor ischaemic stroke, (ii) minor ischaemic stroke (with NIHSS < 3) or TIA. LAMP, loop-mediated isothermal amplification; NIHSS, National Institutes of Health Stroke Scale; PCR, polymerase chain reaction; SNP, single-nucleotide polymorphisms. Doses: a, Clopidogrel 75 mg daily (after loading dose of 300 mg). b, Aspirin 300 mg daily. c, Aspirin 75 mg daily plus clopidogrel 75 mg daily (after loading dose of 300 mg).

Everyone with a suspected stroke should be admitted to a specialist acute stroke unit following assessment by first responders. NICE guidance NG128 states that within 24 hours of ischaemic stroke onset, daily aspirin 300 mg should be offered unless the individual is intolerant to aspirin. 4 Aspirin should be continued until 2 weeks after stroke symptoms begin or until discharged.

For people with high-risk TIA [often defined as patients with a risk of stroke scoring system after a suspected TIA (ABCD²) score of ≥ 4]25 or minor stroke, dual antiplatelet therapy (DAPT) of aspirin and clopidogrel is often used in line with guidance from the European Stroke Organisation, beginning with 2 weeks acute dual therapy. 26 After 2 weeks of acute treatment, NICE guidance recommends long-term antiplatelet treatment with clopidogrel monotherapy. 4 However, in practice, patients are often given dual treatment with aspirin and clopidogrel before moving to longer-term clopidogrel monotherapy. The recommended duration of dual therapy varies according to guidance from up to 21 days,9 21–90 days27 or up to 90 days. 28 This is consistent with the NICE clinical knowledge summary on secondary prevention following stroke and TIA, updated in 2022, which states that

Dual therapy with aspirin plus clopidogrel (for up to 90 days) or aspirin plus ticagrelor (for 30 days) may be initiated in secondary care for some people (for example, people at high risk of TIA, or those with intracranial stenosis) followed by antiplatelet monotherapy. 28

In those who are intolerant of aspirin, the RCP guidelines suggest that clopidogrel could be considered as initial treatment. 23

For patients with TIA that is not high risk (ABCD² score of < 4), NICE guidance (TA210) recommends urgent treatment with modified-release dipyridamole in combination with aspirin in the first instance. 10 However, the NICE clinical knowledge summary advises clopidogrel monotherapy following acute 2-week treatment with aspirin,28 and the RCP Guidelines recommend clopidogrel regardless of stroke risk score for TIA patients. 23

Currently, genetic testing for clopidogrel resistance is not routinely performed in the NHS before using clopidogrel in ischaemic stroke or TIA patients. If genetic testing to inform preventative treatment is introduced in the NHS in people with stroke, it could take place in hospital before long-term antiplatelet treatment is started 2 weeks post ischaemic stroke, or sooner in the case of TIA. People with an allele suggesting poor or intermediate metabolism of clopidogrel could be treated with an alternative to clopidogrel, while those without these alleles would receive standard clopidogrel treatment. Alternative treatments could include the following:

• aspirin

• aspirin combined with dipyridamole

• clopidogrel dose escalation (Unlicensed)

• ticagrelor (Unlicensed).

We heard from our clinical advisors that, of these, the most likely to be used in NHS practice would be aspirin combined with dipyridamole, with a potential treatment pathway shown in Figure 2 for people with (1) non-minor ischaemic stroke and (2) minor stroke or TIA. Ticagrelor does not have marketing authorisation in the UK for secondary prevention after ischaemic stroke or TIA. However, we have heard from clinicians that it is sometimes used in high-risk patients, although it is not considered in those at high risk of bleeding due to an elevated bleeding risk. There is a suspended NICE technology appraisal on ticagrelor for preventing stroke after previous ischaemic stroke or high-risk TIA. 29 This was suspended by the company on 11 May 2021, who also withdrew their application for marketing authorisation for stroke to the European Medicines Agency (EMA) in December 2021. 29 Ticagrelor in combination with aspirin for up to 30 days is however included as a potential treatment for secondary prevention for some people (e.g. people at high risk of TIA or those with intracranial stenosis) in the 2022 NICE clinical knowledge summary on secondary prevention following stroke and TIA. 28

FIGURE 2.

Potential treatment pathway for people with CYP2C19 LOF alleles: (i) non-minor ischaemic stroke, (ii) minor stroke (with NIHSS < 3) or TIA. LAMP, Loop-mediated isothermal amplification; NIHSS, National Institutes of Health Stroke Scale. Doses: a, Modified-release dipyridamole 200 mg twice daily. b, Aspirin 300 mg daily. c, Aspirin 75 mg daily plus modified-release dipyridamole 200 mg twice daily.

Inclusion and exclusion criteria

Study identification

Studies were identified using bibliographic and non-bibliographic search methods following the guidance in the NICE handbook. 32 We carried out two searches:

1. Search 1, undertaken on 10 August 2022, aimed to address Objectives 1, 2 and 3, taking the following form: [(search terms for Clopidogrel) AND (search terms for CYP2C19)]

2. Search 2, undertaken on 11 August 2022, aimed to address Objectives 4 and 5, taking the following form: [(terms for POCTs OR Genomadix OR Genedrive) AND (terms for CYP2C19 OR terms for Clopidogrel)]

The search strategies are reported in Appendix 1: Literature search strategies, using a search narrative. 35 They were developed by one researcher (CC) and checked by another (ET) using the peer review of electronic search strategies (PRESS) checklist. 36

Review strategy

Two reviewers independently screened titles and abstracts identified by the searches. Full copies of all reports considered potentially relevant were obtained and two reviewers independently assessed these for inclusion. Any disagreements were resolved by consensus or discussion with a third reviewer.

Data were extracted using standardised data extraction forms developed in Microsoft Access. Data extraction forms were piloted on a small sample of papers and adapted as necessary. Data were extracted by one reviewer and checked in detail by a second reviewer. Any disagreements were resolved by consensus or discussion with a third reviewer.

Risk-of-bias assessment

The risk of bias (RoB) in included RCTs and controlled clinical trials (CCTs) was assessed using the RoB 2 tool. 37 Observational studies of exposure were assessed using the ROBINS-E tool. 38 Diagnostic accuracy studies were assessed using a modified version of QUADAS-2. 39 We omitted two signalling questions – ‘If a threshold was used, was it pre-specified’ in the Index Test domain, and ‘Was there an appropriate interval between index test and reference standard’ in the Flow and Timing domain. Genetic tests do not have a threshold in the standard test accuracy sense – they identify the presence or absence of certain alleles and so we considered that this question did not apply to this review. Similarly, the question on timing is not relevant for genetic tests as the allele would either be present or not and this would not change over time: therefore, the time interval between tests does not matter. We did not formally assess applicability as our research question was broad and all studies were applicable; instead, we extracted data on potential sources of variation such as population and considered these in our synthesis. Details of the tools are provided in Report Supplementary Material 1. Quality assessment was undertaken by one reviewer and checked by a second reviewer. Disagreements were resolved by consensus or discussion with a third reviewer.

Survey of laboratories

We conducted a web-based survey to gather data on the technical performance characteristics of CYP2C19 genetic tests (Objective 5). The survey was sent to seven genomic laboratory hubs who are responsible for delivering genomic testing in the NHS in England and to genomic laboratory hubs in Wales, Northern Ireland and Scotland. The survey collected information on:

• Platforms capable of performing CYP2C19 testing available in the lab.

• Preferred test platform for running CYP2C19.

• Reason for preference.

• For each platform or genetic test, we ask for information on:

  • alleles that would be tested for

  • impact of having to test for additional alleles

  • time to results

  • resources for running tests: staff time, staff grade, cost per test to run, maintenance of machines/quality assurance, additional administrative resources

  • ease of use

  • test failure rate

  • current testing capacity

  • whether faster turnaround would be possible with additional resources and what these would be

  • whether additional testing capacity would be possible with additional resources and what these would be

  • whether the test could be performed in local testing laboratories.

• Facilitators and barriers to implementing testing, and what platform would be most likely to be implemented.

• How feasible would it be to install POCTs in local laboratories and extra resources required.

Synthesis methods

For each objective, a narrative summary of all the included studies is presented. This includes a summary of the study characteristics and study quality.

Results of the searches

The process of study identification and selection is summarised in Figure 3 (Objectives 1–3) and Figure 4 (Objectives 4–5). Studies included, stratified by objective, and studies excluded at full text are reported in Appendix 2: Tables of included, ongoing or excluded studies.

FIGURE 3.

PRISMA flow chart: Objectives 1–3.

FIGURE 4.

PRISMA flow chart: Objectives 4–5. a, Three included studies include a pre-trial and a main trial, we are therefore treating these as separate studies.

Objective 1

Two controlled trials from China were included for Objective 1. 46,47 Full details on these studies are reported in Report Supplementary Material 1. Both studies were small (80 and 190 patients) and did not provide sample size or power calculations. Duration of follow-up was 90 days in one study and 1 year in the other. One of these studies was reported in Chinese and was extracted with help from a native Chinese speaker and using Google Translate. 46 Both studies used laboratory-based testing to determine the presence of LOF alleles.

Xia et al. 46 allocated 80 patients to two groups:

• Group A: all received clopidogrel 75 mg/day.

• Group B: genotyped for the *1, *2, *3 and *17 alleles.

  • No LOF alleles: clopidogrel 75 mg/day (same as control).

  • One LOF allele: clopidogrel 150 mg.

  • Two LOF alleles: ticagrelor (this was recorded as ‘tigrillo’ in the English abstract but translation of the Chinese term suggested that this was ticagrelor).

Lan et al. 47 genotyped all participants for the *1, *2, *3 and *17 alleles. Participants were then divided into two groups (groups A and B with 90 patients in each) so that equal numbers with each potential genotype were included in each group. All patients were initially treated with clopidogrel (300 mg loading dose followed by 75 mg/day) and aspirin 100 mg day for 21 days. Treatment after this varied by intervention group and presence of LOF alleles:

• Group A: clopidogrel 75 mg/day.

• Group B:

  • normal metaboliser (no LOF alleles) and extensive metaboliser (EM) (1 or 2 *17 alleles): clopidogrel 75 mg/day

  • PM (1 or 2 LOF alleles): aspirin 100 mg/day.

This study did not technically meet inclusion criteria for Objective 1, as all patients were tested, however, as half of the tested patients were treated as if they had not been tested (i.e. standard treatment), we considered it appropriate to include this study for this objective.

Both studies enrolled patients with a stroke as a primary event. Mean age was 69 years and percentage of female participants was 38% in both studies. One study was funded by non-industry47 and the other did not report funding sources. 46

Results

Incidence of secondary vascular events

Both studies46,47 presented data on incidence of secondary ischaemic stroke and MI. Xia et al. 46 reported the incidence of TIA, vascular death and a composite outcome (including stroke, TIA, MI and death). Lan et al. 47 reported data on haemorrhagic stroke. We additionally calculated a composite outcome for Lan et al., adding events for all outcomes reported. Report Supplementary Material 1 (see Table 3) provides an overview of results for incidence of secondary vascular events in these studies. Hazard ratios were N/R in these studies, so estimates were calculated from event frequencies and follow-up time. We did not meta-analyse results from these two studies due to the differences in interventions. In general, HRs suggested a reduction in composite outcomes, secondary ischaemic stroke, haemorrhagic stroke and TIA in patients tested for LOF alleles and treated accordingly, but CIs were wide and included the null (HR = 1) in all cases (Figure 5). There was no evidence of benefit in either group for vascular death or MI, although incidence of these outcomes was low (< 5%). Full details on results are presented in Report Supplementary Material 1.

FIGURE 5.

Forest plot showing HRs (95% CI) for secondary vascular events in patients treated with clopidogrel compared with patients tested for LOF alleles and offered personalised treatment.

Objective 2

Seven trials, reported in 23 full report publications, were included for Objective 2. 48–54 All studies were published in English. Two trials were restricted to patients with LOF alleles who were then randomised to different antiplatelet therapies. The other five studies were not restricted based on LOF alleles – patients were randomised to different antiplatelet strategies, and a subgroup analysis was then performed restricted to those with LOF alleles. Table 6 shows an overview of the studies included for Objective 2. Full details on the studies are reported in Report Supplementary Material 1.

Three studies included patients who presented with stroke as their primary event, and four included patients with either stroke or TIA. Five studies took place in China and recruited patients predominantly of Chinese origin, one was done in South Korea including mostly patients of South Korean heritage and one took place in an international setting, with a majority white (67%) ethnicity. Mean age ranged from 60.8 years [standard deviation (SD) 8.7] to 64.8 years (SD N/R). The percentage of females ranged from 24% to 45%. Sample size ranged between 154 and 6412.

Three studies compared clopidogrel plus aspirin with aspirin alone. In the clopidogrel arm, one of these studies gave a one-off 300 mg loading dose of clopidogrel and aspirin only for an initial 21-day period, the second did not offer a loading dose of clopidogrel and stopped aspirin after 30 days, and the third gave a 600 mg loading dose of clopidogrel and continued the aspirin in combination with clopidogrel longer term. Two studies compared clopidogrel with ticagrelor – both studies included a 300 mg clopidogrel loading dose and an initial 21-day period when aspirin was given in addition to the clopidogrel or ticagrelor. One study compared clopidogrel with triflusal, without a loading dose in either arm. The final study compared a standard dose of clopidogrel (75 mg) with a higher dose of clopidogrel (150 mg). In this study, all patients received a 300 mg loading dose of clopidogrel and 150 mg aspirin for the first 21 days; after this, clopidogrel was stopped and patients continued treatment with 150 mg aspirin alone. One study was funded by industry organisations (drug manufacturer), one was funded by non-industry but drugs and genetic tests were supplied by industry and five were funded by non-industry organisations.

Four studies used laboratory-based genotyping tests [Seeplex CYP2C19 angiotensin-converting enzyme (ACE) genotyping system and Real-Q CYP2C19 genotyping kit, Drug Metabolism Enzyme TaqMan Allelic Discrimination Assay, and Sequenom MassARRAY increased plexing efficiency and flexibility (iPLEX) platform (Sequenom)], one used a POCT (GMEX point-of-care genotyping system) and two did not report the type of test that was used. Six studies investigated the two main LOF alleles (*2 and *3) and one study only genotyped one LOF allele (*2).

Duration of follow-up ranged across studies: five studies had a follow-up time of 90 days, one followed patients up between 2 and 3 years and one between 4 and 5 years.

Results

Included studies presented data on incidence of secondary vascular occlusive events, adverse events and mortality, in people who had LOF alleles associated with clopidogrel resistance and were treated with alternative interventions compared to standard treatment with clopidogrel. There were no studies reporting data on other outcomes of interest for Objective 2.

Secondary vascular occlusive events

Six studies reported data on the incidence of a composite outcome of secondary vascular occlusive events (including stroke, TIA, MI and vascular death), five studies on incidence of secondary stroke, six studies on incidence of secondary ischaemic stroke, one on incidence of secondary TIA, two on secondary MI, two on secondary vascular death and two studies presented data on mortality of any cause. Figure 6 shows the network of intervention comparisons for each outcome. These are all seen to be disconnected, with no loops of evidence, so network meta-analysis was performed. As there were a maximum of two studies making any one comparison between treatments, only fixed-effect meta-analyses were performed.

FIGURE 6.

Network plots showing drug comparisons for main outcomes in Objective 2. (a) Composite outcome, (c) any stroke, (c) ischaemic stroke, (d) any bleeding. a, Drug given on a temporary basis (21–31 days).

Composite outcome of secondary vascular occlusive events

There was some evidence that treatment with alternatives to clopidogrel reduced the risk of secondary vascular events in those with LOF alleles (see Appendix 4, Figure 15). Ticagrelor was associated with a reduced incidence of secondary vascular events compared to clopidogrel [summary HR 0.76, 95% CI 0.65 to 0.90; two studies]. There was a suggestion that high-dose (HD) clopidogrel plus aspirin was associated with a reduced incidence of secondary vascular occlusive events compared to standard-dose clopidogrel plus aspirin, but CIs were wide (HR 0.18, 95% CI 0.02 to 1.52; one study). There was no difference in the incidence of vascular events amongst those taking clopidogrel alone compared to aspirin, although one other study suggested that the risk of secondary vascular events was higher for those taking aspirin alone compared to clopidogrel plus aspirin. However, this was a small study with very few events (all corresponding to ischaemic strokes), and CIs were wide (HR 3.03, 95% CI 0.83 to 11.11). There was no evidence of heterogeneity for any comparison. All summary estimates are from fixed-effects meta-analysis.

Stroke

The risk of stroke and ischaemic stroke was also reduced in those with LOF alleles taking ticagrelor compared to those taking clopidogrel (HR 0.76, 95% CI 0.63 to 0.92 for any stroke; HR 0.77, 95% CI 0.65 to 0.93 for ischaemic stroke; two studies) (see Figure 7 and Appendix 4, Figure 16). There was no evidence of a difference in stroke risk between clopidogrel and triflusal, or between clopidogrel alone and aspirin. As with the composite clinical outcome, the study that compared clopidogrel plus aspirin (vs. aspirin alone) for the duration of the study suggested that the risk of stroke was higher for aspirin alone compared to clopidogrel plus aspirin (HR 3.03, 95% CI 0.83 to 11.11). There was no evidence of heterogeneity for any comparison.

FIGURE 7.

Forest plot showing HRs (95% CI) for incidence of ischaemic stroke in carriers of LOF alleles receiving standard therapy with clopidogrel (or clopidogrel + aspirin) compared with an alternative antiplatelet regimen.

Other secondary efficacy outcomes

Other secondary outcomes evaluated included TIA, MI, vascular death and mortality. There were very few events for these outcomes and no statistical evidence of a difference between any of the antiplatelet strategies evaluated (see Appendix 4, Figure 17).

Adverse events

Seven studies reported data on incidence of bleeding events in those with LOF alleles treated with different antiplatelet therapies. Details on how bleeding outcomes were defined and categorised are summarised in (see Appendix 3, Table 47).

One study reported an increased risk of bleeding with ticagrelor compared to clopidogrel, while the other study that compared ticagrelor with clopidogrel found no difference in the risk of bleeding. There was no statistical evidence for differences between antiplatelet treatment strategies for any of the other comparisons or bleeding outcomes (see Appendix 4, Figures 18 and 19).

Objective 3

Twenty-five studies reported in 45 publications were included for Objective 3. 48,49,52–74 All studies were published in English. Five of the studies included for Objective 2 also provided data for Objective 3. 48,49,52–54

Table 8 provides an overview of the studies included for Objective 3. Full details on the studies are reported in Report Supplementary Material 1.

Twenty studies used a cohort design – 13 enrolled participants prospectively and 7 used a retrospective design. The five RCTs also included for Objective 2 compared standard clopidogrel therapy against an alternative and provided data for participants with and without LOF alleles. Data were extracted from these studies for the clopidogrel treatment arm only, effectively giving a cohort of patients treated with clopidogrel in whom results could be compared between those with and without LOF alleles. All studies administered clopidogrel to all patients, and outcomes were compared between those with and without LOF alleles. In 15 studies, patients received clopidogrel alone, 7 studies gave clopidogrel plus transitory aspirin (14–30 days), 3 studies administered both clopidogrel and aspirin for the duration of the study and the other two included patients taking clopidogrel, with or without other antiplatelets. In four studies, an initial loading dose of clopidogrel was given to all participants, in two studies some patients had been given an initial loading dose, and 21 studies did not give a loading dose.

Four studies had a follow-up time of 90 days, 5 followed up patients for 180 days, 2 for 365 days, four studies from 1 to 2 years, one study from 2 to 3 years, one study from 3 to 4 years and two from 4 to 5 years. Eight studies did not report follow-up time.

Most studies enrolled patients who had experienced a stroke as their primary event (14 studies), 1 study only enrolled patients who had experienced a TIA and 10 studies enrolled patients who had experienced a stroke or TIA. Most studies were conducted in Asia (13 in China, 2 in Japan and 1 in Korea), 4 studies were conducted in the USA with single studies from other countries. One study had drugs and tests provided by industry, one was sponsored by a commercial company, other studies either did not report on funding source or were funded by non-commercial organisations. A variety of different laboratory tests were used to determine CYP2C19 status – none of the studies used POCTs. The majority of studies tested for both *2 and *3 LOF alleles, five studies only tested for *2 and two did not report on which LOF alleles were tested for. Two studies tested for additional alleles as well as *2 and *3 – *8 in one study and *5, *6, *7 and *8 in the other.

Results

Secondary occlusive events

There was strong evidence that people with LOF alleles treated with clopidogrel (or clopidogrel plus aspirin) have a greater incidence of secondary vascular events (HR 1.72, 95% CI 1.43 to 2.08; 18 studies) (see Appendix 4, Figure 20), stroke (HR 1.46, 95% CI 1.09 to 1.95; 5 studies) (see Appendix 4, Figure 21) and ischaemic stroke (HR 1.99, 95% CI 1.49 to 2.65; 12 studies) (Figure 8), than those without LOF alleles (estimates from random-effects meta-analysis). There was some evidence of heterogeneity for the composite outcome of secondary vascular events (I² = 33%; τ² = 0.027); there was little or no evidence of heterogeneity for other outcomes. Fixed-effect meta-analysis estimates were very similar to pooled results from random-effects analyses.

FIGURE 8.

Forest plot showing HRs (95% CI) for incidence of ischaemic stroke in carriers of LOF alleles compared with non-carriers of LOF alleles receiving standard therapy with clopidogrel (or clopidogrel + aspirin).

Secondary efficacy outcomes

There was little evidence to suggest any association between LOF alleles and secondary outcomes of mortality and TIA (see Appendix 4, Figure 22). However, these were evaluated in very few studies and there were very few events. There was evidence that the risk of vascular death is increased in patients with LOF alleles treated with clopidogrel compared to those without LOF alleles (HR 5.07, 95% CI 1.26 to 20.39).

Investigation of heterogeneity

Within studies that evaluated multiple vascular occlusive event outcomes, estimates of HR were very similar for composite outcome, stroke and ischaemic stroke (ischaemic stroke accounted for most of the secondary vascular outcomes reported in all studies). This is shown in (see Appendix 4, Figure 23). As described in the Methods, a post hoc decision was therefore made to combine data across different types of vascular event when exploring heterogeneity.

Results of univariable metaregressions are show in (see Appendix 4, Table 49). Forest plots stratified for each of these variables are provided (see Appendix 4, Figure 31). There was evidence of a reduced effect of LOF alleles in patients given a loading dose of clopidogrel relative to those who were not [ratio of hazard ratios (RHR) 0.64, 95% CI 0.43 to 0.96], in patients taking clopidogrel plus long-term aspirin relative to those taking only clopidogrel or clopidogrel plus short-term aspirin (RHR 0.47, 95% CI 0.22 to 0.96), and in studies that included patients with stroke and TIA as primary events compared with only patients with stroke (RHR 0.62, 95% CI 0.44 to 0.86). The stratified analysis based on clopidogrel regimen suggested that there was no evidence of a difference in the risk of secondary vascular events between those with and without LOF alleles when taking clopidogrel plus aspirin (HR 0.74; 95% CI 0.23 to 2.38) stratified analyses results shown in (see Appendix 4, Figures 24–31). There was no evidence of a difference between studies which included patients with TIA as primary event and those including patients with stroke, but only one study investigated TIA patients exclusively.

There was some suggestion from subgroup analyses that effects of LOF alleles may vary by ethnicity, with a possibly reduced effect in studies in mixed and Hispanic populations compared with white. However, there was considerable uncertainty in these stratified estimates, resulting in no statistical evidence for differences between LOF effect by ethnicity in the metaregression.

There was no evidence for a difference in LOF alleles effect on secondary vascular occlusive outcomes based on RoB, proton pump inhibitor (PPI) use, study location or duration of follow-up.

Investigation of small-study effects

The funnel plot showing HRs for incidence of secondary vascular occlusive outcomes in carriers of LOF alleles compared with non-carriers of LOF alleles appears symmetrical (see Appendix 4, Figure 33). This suggests that there is no evidence of small-study effects.

Adverse events

Eight studies reported data on incidence of bleeding events in those with LOF alleles treated with different antiplatelet therapies. Details on outcome definitions for each study and categories assigned for this analysis are presented in (see Appendix 3, Table 48). There was no evidence of a difference in the risk of bleeding among those with and without LOF alleles for each category of bleeding assessed: any bleeding (HR 0.98, 95% CI 0.68 to 1.40; five studies), severe bleeding (HR 0.97, 95% CI 0.42 to 2.20; five studies), haemorrhagic stroke (HR 1.34, 95% CI 0.29 to 6.20; two studies) or mild bleeding (HR 0.68, 95% CI 0.30 to 1.54; two studies). This is shown in (see Appendix 4, Figure 32). There was no evidence of heterogeneity for any of these outcomes (I² = 0). For this reason, subgroup analyses and metaregression was not performed. Fixed-effect meta-analysis estimates were identical to pooled results from random-effects analyses.

Objective 4

Nine studies, reported in 12 publications, reported data on test accuracy of the POCTs in scope. Two studies reported separate accuracy data for a pre-trial and the main trial – these are treated as separate studies giving a total of 11 studies. 76,77 Three studies were only available as clinical trial registrations, all others were published as full reports. All studies available only as clinical trial registrations were conducted by Spartan (Genomadix), who provided additional information when requested for two of the studies. 78,79 All studies were reported in English.

All studies evaluated Spartan versions of the test. Two evaluated Spartan Cube,78,79 eight evaluated Spartan RX76,77,80–83 and one evaluated Spartan FRX. 84 These tests are considered broadly equivalent to the Genomadix Cube and so were evaluated as a single group referred to from here as ‘Genomadix (Spartan) CYP2C19 tests’, unless referring to specific tests. There were no studies on the accuracy of Genedrive.

Table 10 provides an overview of the studies included for Objective 4. Full details of the studies are reported in Report Supplementary Material 1. Five studies were funded by the test manufacturer. One study was funded by other industry organisations and one by both industry and non-industry.

Six studies recruited patients undergoing percutaneous coronary intervention (PCI). The two pre-trials included healthy volunteers as they were pre-trial validations of the test. Three studies did not report details on the population studied – all were only available as clinical trial registrations. None of the studies were conducted in our population of interest – stroke patients.

The number of participants ranged from 879 to 2587. 76

Two studies took place in Europe, six studies in Canada, one in South Korea and two studies (reported in the same publication) were multinational conducted in USA/Canada/South Korea/Mexico.

Studies targeted different combinations of the three alleles that can be detected using Genomadix (Spartan) CYP2C19 tests (*2, *3, *17). Seven studies targeted all three LOF alleles, one targeted *2 and *17 and the remainder targeted only *2. We dichotomised results into presence of LOF alleles or no LOF alleles so that those with at least one *2 or *3 LOF allele were considered to have LOF alleles; we categorised *17 as normal function, as described in the methods. The reference standard (standard laboratory test) was bidirectional sequencing in three studies, direct DNA sequencing in two studies and Sanger sequencing in one study (all these methods can detect the presence of any LOF allele). The remaining four studies used Taqman, which can be set up with different probes to detect different LOF alleles. One of the studies used Sanger sequencing as an additional reference standard where there were discrepancies between the Genomadix Cube and Taqman results. In all studies, even those that used a reference standard that could detect any LOF alleles, the laboratory tests only targeted the same alleles as were targeted by the Genomadix Cube. Estimates of accuracy from these studies therefore show the accuracy in detecting only those variants that Genomadix (Spartan) CYP2C19 tests can detect (and in four studies only the *2 LOF allele), rather than the accuracy for the detection of any variant associated with LOF.

Objective 5

Review of economic evaluations of CYP2C19 genetic tests for clopidogrel resistance in non-cardioembolic ischaemic stroke and transient ischaemic attack patients

Model structure and methods of economic evaluation

We developed a decision-analytic model to estimate the incremental costs and QALYs for CYP2C19 genetic testing for clopidogrel resistance in patients in England and Wales who have had a non-cardioembolic ischaemic stroke or TIA where treatment with clopidogrel is being considered, compared with no genetic testing. We refer to LOF carriers as LOF patients, and LOF non-carriers as loss-of-function non-carrier (NoLOF) patients.

Model structure

The model structure was developed to capture the short- and long-term costs and benefits of CYP2C19 genetic testing, based on our review of previous cost-effectiveness models (see Results of the review of cost-effectiveness studies for CYP2C19 testing strategies and Results of the review of cost-effectiveness studies of secondary prevention of ischaemic stroke), results from the Chapter 3, Survey of laboratories, information provided by Genedrive and Genomadix, and in discussion with specialist members of the Diagnostic Appraisal Committee (DAC).

The model utilises a hybrid decision-tree and Markov structure. Diagnostic decisions and short-term 90-day outcomes were modelled using a decision-tree structure, and long-term outcomes were modelled using a five state Markov model. The model is split into short-term (90-day) and long-term outcomes to reflect the elevated risk of a subsequent stroke in the short term following an event, which is particularly relevant for patients who have had a TIA.

Decision tree

The decision tree (Figures 9–11) differs by test type according to the diagnostic outcomes for the test. For POCTs, there are four branches of the tree for patients who receive a true-positive (TP), false-negative (FN), true-negative (TN) and false-positive (FP) result. We assume that the lab test is a gold standard test with perfect sensitivity and specificity, so there are just two branches of the tree for LOF patients and NoLOF patients who receive appropriate test results and corresponding treatment. For the no-testing strategy, there are also two branches of the tree for LOF patients and NoLOF patients, but the event rates differ because all the LOF patients will receive clopidogrel rather than dipyridamole + aspirin. The proportion of the modelled population that are of LOF carriers is the same regardless of test.

FIGURE 9.

Point-of-care testing decision-tree branch. The timing of clopidogrel treatment will depend on the indication. Those patients who have had a TIA/minor stroke may begin dual clopidogrel–aspirin treatment immediately. Those patients who have had a major stroke may initiate with 2 weeks of aspirin before clopidogrel treatment. Alternative treatment (Alt Tx) is aspirin combined with dipyridamole in the base case, with scenarios for low-dose aspirin and ticagrelor.

FIGURE 10.

Laboratory testing decision-tree branch. The timing of clopidogrel treatment will depend on the indication. Those patients who have had a TIA/minor stroke may begin dual clopidogrel–aspirin treatment immediately. Those patients who have had a major stroke may initiate with 2 weeks of aspirin before clopidogrel treatment. Alternative treatment (Alt Tx) is aspirin combined with dipyridamole in the base case, with scenarios for low-dose aspirin and ticagrelor.

FIGURE 11.

No test decision-tree branch. The timing of clopidogrel treatment will depend on the indication. Those patients who have had a TIA/minor stroke may begin dual clopidogrel–aspirin treatment immediately. Those patients who have had a major stroke may initiate with 2 weeks of aspirin before clopidogrel treatment. Alternative treatment (Alt Tx) is aspirin combined with dipyridamole in the base case, with scenarios for low-dose aspirin and ticagrelor

The second part of the decision tree captures the 90-day outcomes, which are the same for each diagnostic outcome branch of the tree but with different event rates depending on treatment and LOF status. The 90-day health outcomes are:

• no further event

• further minor stroke

• major bleed/ICH

• further moderate stroke

• further major stroke

• death.

The health outcomes are defined according to stroke severity, which correspond to disability states. Advice from clinical members of the DAC suggested the following mRS scores for the different stroke severity states:

• TIA: mRS range 0

• minor stroke: mRS range 0–1

• moderate stroke: mRS range 2–3

• major stroke: mRS range 4–5.

These are largely in line with the categories used in previous economic evaluations of CYP2C19 testing. 104,109,121 We place major bleed/ICH between further minor and further moderate stroke in terms of severity, based on clinical advice and utility estimates (see Model parameters and inputs).

Under lab-based testing, there may be a delay in receiving test results after which those identified as LOF carriers switch from clopidogrel to appropriate alternative treatment (Alt Tx). We assume that POCT results are available within a day, so that there is no delay in starting appropriate treatment for LOF carriers. Treatment switches were modelled by averaging the event rates during the short-term (90-day) part of the model according to the time spent on different treatments.

For all strategies, patients may discontinue treatment as a result of treatment-related side effects, and it is assumed that patients discontinuing would switch to low-dose aspirin monotherapy.

The model structure is the same for the non-minor ischaemic stroke and TIA/minor stroke subpopulations, but the model inputs and transition probabilities differ between populations.

As the decision tree models the first 90 days after the patient’s initial ischaemic stroke or TIA, no discount rate is applied to costs and QALYs accrued in this period.

Markov model

Health states

Following the initial decision tree, a Markov model (Figure 12) was used to model long-term patient outcomes for a cohort of patients. In the model, patients move between five possible health states: no recurrent stroke, post minor stroke, post major bleed/ICH, post moderate stroke or post major stroke. Health states were chosen based on discussion with clinical experts and inspection of the clinical and health economic literature on the most impactful and frequent events experienced by ischaemic stroke survivors. Health states differ in costs, health-related quality of life (HRQoL), mortality rate and recurrent event rates. The ordering of the health states by severity was motivated by consultation with patient and clinical experts and inspection of long-term HRQoL measurements from studies which had measured this by health state (see Model parameters and inputs).

FIGURE 12.

Long-term Markov model structure.

We assume that patients can progress to a more severe disease health state, but they cannot move from a more severe health state to a less severe state. This was based on the nature of the long-term outcomes associated with the events included in the model (strokes, major bleeds) and the chronic nature of the disease. Patients are categorised into the most severe category that they have experienced.

The proportion of patients in each health state in the first Markov time cycle is determined by the proportion in each health state at the end of the 90-day decision tree, according to true LOF status and treatment allocation.

Cohorts evaluated

The transitions in the Markov model depend on treatment and LOF status, and so four cohorts are evaluated corresponding to the paths in the decision tree:

1. patients with LOF alleles undergoing clopidogrel treatment

2. patients with LOF alleles undergoing a non-clopidogrel alternative treatment

3. patients without LOF alleles undergoing clopidogrel treatment

4. patients without LOF alleles undergoing a non-clopidogrel alternative treatment.

We assume that patients stay on the same treatment they were on at the end of the decision-tree period, and only switch treatment as a consequence of a bleeding event, which is modelled with a discontinuation rate set equal to the rate of new major bleeds/ICH each year for each treatment.

Time cycles, time horizon and discounting

The initial time cycle in the Markov model takes the length of 275¼ days, calculated as a 1 year minus the 90-day period of the decision-tree period. The second and subsequent time cycles take the length of a year (365¼ days). The Markov model utilises a lifetime time horizon and so patient outcomes in the Markov model are followed until the general population life tables end (age 100 years).

Costs and QALYs in the long-term Markov model are discounted at a rate of 3.5% per annum. Discounting begins in the second Markov time cycle which models the second year after the patient’s initial stroke or TIA.

Model outcomes

Costs are accrued in the model through health state-specific costs and treatment costs, which when summed over time spent in health states and time on treatment give total (discounted) expected costs over the model time horizon for each path in the decision tree. QALYs accrue in the model through the health state utilities summed over time spent in each health state to give (discounted) total expected QALYs over the model time horizon for each path in the decision tree. The total expected costs and QALYs of the decision-tree pathways are then averaged to calculate the total expected costs and QALYs associated with the POCTs, laboratory testing and no testing, respectively. The cost-effectiveness results are summarised with ICERs and expected net benefit.

Model results

All results are reported separately for (1) the TIA/minor stroke population and (2) the non-minor ischaemic stroke population. Key summary results are also reported for a mixed TIA/ischaemic stroke population using a weighted average using the proportions of the population in each group in Appendix 9, Figures 36–42, Tables 60–63. Due to the paucity of clinical efficacy data for the Genedrive system, we assumed that sensitivity, specificity and test failure rates are set equivalent to those for the Genomadix Cube. For this reason, the results for Genedrive should be considered illustrative only, and only key summary results are reported for Genedrive. Deterministic base-case results are outlined in the section Deterministic base-case analyses, with deterministic sensitivity analyses reported in the section Deterministic sensitivity analyses. PSA, scenario analyses and diagnostic test cost and accuracy threshold analyses are reported in sections Probabilistic analysis, Scenario analyses and Threshold analyses.

Probabilistic analysis

The sensitivity of results to uncertainty in model inputs was explored using a probabilistic analysis where uncertainty around model parameters is described by the distributions outlined in Table 30. The PSAs were run for 5000 iterations in each population (non-minor ischaemic stroke, TIA/minor stroke and mixed). In the mixed population at each iteration, the population was sampled with a probability 0.68 for non-minor ischaemic stroke 0.32 and the TIA/minor stroke population otherwise.

Results are reported in incremental cost-effectiveness planes where the incremental costs are plotted against incremental QALYs for each probabilistic iteration, with a willingness-to-pay threshold of £20,000 per QALY indicated by a dotted line. Points that are south-east of the cost-effectiveness threshold indicate that the intervention is cost-effective against the comparator at that willingness-to-pay threshold. Cost-effectiveness acceptability curves are also presented which show the probability that each testing strategy is cost-effective at each willingness-to-pay threshold assessed.

Incremental cost-effectiveness planes for the laboratory test and Genomadix Cube are reported for the non-minor ischaemic stroke population in Appendix 9, Figures 36 and 37, TIA/minor stroke population (see Figures 38 and 39) and mixed population in (see Appendix 9, Figures 40 and 41). In all cases, the spread of points lies below the willingness-to-pay threshold indicating a high certainty that both tests are cost-effective compared with no testing. Cost-effectiveness acceptability curves are reported for the non-minor ischaemic stroke population (Figure 13), TIA (Figure 14) and mixed populations (see Appendix 9, Figure 42), where Genedrive is excluded due to insufficient data. In all cases, there is a very high probability that one of the testing strategies is cost-effective, with the probability of no testing being cost-effective close to zero across all willingness-to-pay thresholds. There is uncertainty as to whether the laboratory test or Genomadix Cube CYP2C19 test is most cost-effective, but in the willingness-to-pay threshold range of £20,000–30,000, the probably of being most cost-effective is higher for laboratory tests for the non-minor stroke population, and for the Genomadix Cube CYP2C19 test for the TIA/minor stroke population. There is more certainty that the Genomadix Cube CYP2C19 test is most cost-effective in the TIA/minor stroke population, due to impact of the delay in receiving results with laboratory testing.

FIGURE 13.

Cost-effectiveness acceptability curve for the non-minor ischaemic stroke population (excluding Genedrive due to insufficient data).

FIGURE 14.

Cost-effectiveness acceptability curve for the TIA/minor stroke population (excluding Genedrive due to insufficient data).

Cost-effectiveness results from the probabilistic analysis are reported as costs and QALYs averaged over the iterations. Table 33 reports the probabilistic fully incremental analysis for the three populations. As for the deterministic analysis, all laboratory and point-of-care CYP2C19 testing strategies dominated no testing in all populations (i.e. cost less and have higher QALYs). Omitting Genedrive, due to insufficient data, the ICER for the Genomadix Cube CYP2C19 test relative to laboratory testing was £86,272, £10,797 and £46,446 in the non-minor ischaemic stroke population, TIA/minor stroke population and mixed population, respectively. This suggests that the Genomadix Cube CYP2C19 test is the most cost-effective option in the TIA/minor stroke population where clopidogrel is started sooner, whereas laboratory testing is more cost-effective in the non-minor stroke population. Note however that the differences between the tests in QALYs were very small, making the interpretation of the ICERs challenging, so we prefer to compare the tests using expected net monetary benefit.

TABLE 33 - Probabilistic fully incremental analysis for the non-minor ischaemic stroke population

Treatments	Total costs £ (discounted)	Total QALYs (discounted)	Strictly dominated	Extendedly dominated	ICER (£)
					vs. Lab test	vs. Genedrive	vs. Genomadix Cube
Non-minor ischaemic stroke
Laboratory genetic test	98,517	6.538
POCT: Genedrive	98,523	6.539	No	No	5530
POCT: Genomadix Cube	98,616	6.539	No	No	86,272	2,985,232
No test	100,450	6.324	Yes	N/A	Dominated	Dominated	Dominated
Transient ischaemic attack/minor stroke
POCT: Genedrive	45,016	8.688
Laboratory genetic test	45,086	8.685	Yes	N/A	Dominated
POCT: Genomadix Cube	45,118	8.688	No	No	3,1432,806	10,797
No test	46,155	8.598	Yes	N/A	Dominated	Dominated	Dominated

Table 34 reports the pairwise comparisons from the probabilistic analysis of each CYP2C19 testing strategy compared with no testing. In the non-minor ischaemic stroke population, the expected net monetary benefits were £6230 for Genedrive, £6138 for the Genomadix Cube and £6214 for the laboratory test. In the TIA/minor stroke population, the expected net monetary benefits were £2932 for Genedrive, £2829 for the Genomadix Cube and £2802 for the laboratory test. In the combined TIA/ischaemic stroke population, the expected net monetary benefits were £5211 for Genedrive, £5119 for the Genomadix Cube and £5163 for the laboratory test. In all populations, net monetary benefit is similar, suggesting little difference between the tests. Omitting Genedrive, where there are insufficient data, the highest expected net benefit is for the laboratory test in the non-minor ischaemic stroke population, and the Genomadix Cube CYP2C19 test in the TIA/minor stroke population.

TABLE 34 - Probabilistic pairwise comparisons vs. no testing for the non-minor ischaemic stroke population

Comparator vs. no test	Incremental costs (discounted)	Incremental QALYs (discounted)	ICER	Net monetary benefit
Non-minor ischaemic stroke
POCT: Genedrive vs. no test	−£1927	0.215	−£9118	£6230
POCT: Genomadix Cube vs. no test	−£1835	0.215	−£8650	£6138
Laboratory genetic test vs. no test	−£1933	0.214	−£9214	£6214
Transient ischaemic attack/minor stroke
POCT: Genedrive vs. no test	−£1139	0.090	−£12,843	£2932
POCT: Genomadix Cube vs. no test	−£1036	0.090	−£11,592	£2829
Laboratory genetic test vs. no test	−£1069	0.087	−£12,472	£2802

Statement of principal findings

Our results suggest that CYP2C19 testing followed by tailored treatment is likely to be effective and cost-effective in both adult populations modelled (non-minor ischaemic stroke and TIA/minor ischaemic stroke) providing cost savings and increased QALYs compared with not testing. We found this finding was robust to assumptions made in the model. There is uncertainty around the best test to use (laboratory or POCT), and there is no direct evidence for alternative treatment dipyridamole for patients with LOF allele status. There was no evidence in children for any of the objectives.

Evidence identified for Objectives 1–3 suggests that people with LOF alleles are more likely to experience a secondary vascular event and treatments with alternatives to clopidogrel may reduce this risk. Alternative treatments investigated included in the studies were short-term HD clopidogrel (150 mg) followed by long-term aspirin alone, ticagrelor, aspirin and triflusal. Some studies combined these either with an initial (up to 30 day) course of aspirin or with longer-term aspirin. There were no studies of dipyridamole, one of the antiplatelets that is likely to be offered as an alternative to clopidogrel in the UK, if CYP2C19 testing were to be introduced.

We only identified two small studies, both at high RoB, that evaluated a ‘test-and-treat’ strategy (Objective 1). This is the ideal study design to investigate the benefits of introducing genetic testing to identify CYP2C19 LOF alleles. There was a suggestion that testing plus treating based on LOF allele status was associated with a reduced incidence of ischaemic stroke, TIA and a compositive outcome of secondary vascular events, but CIs were wide and overlapped the null.

Objective 2 identified seven studies investigating whether people who have LOF alleles have a reduced risk of secondary vascular occlusive events if treated with alternative antiplatelet therapies compared to treatment with clopidogrel. Four were judged at low RoB, three had concerns regarding the potential for bias due to missing data and lack of information on allocation concealment. There was evidence that ticagrelor was associated with a lower risk of secondary vascular events, including ischaemic stroke, than clopidogrel. One study suggested that ticagrelor was associated with an increased risk of any bleeding event; the other found no difference in the risk of bleeding with ticagrelor compared to clopidogrel. There was no statistical evidence for differences between antiplatelet treatment strategies for any of the other comparisons or bleeding outcomes.

Objective 3 identified 25 studies (20 cohort studies and 5 trials) that compared people with and without LOF alleles, all of whom were treated with clopidogrel (alone or combined with aspirin or other antiplatelet drugs) to see whether the risk of secondary vascular occlusive events differed between groups. Six studies were judged at high RoB due to loss to follow-up that could potentially be related to incidence of vascular events. There was strong evidence that people with LOF alleles treated with clopidogrel (or clopidogrel plus short-term aspirin) have a greater incidence of secondary vascular events (HR 1.72, 95% CI 1.43 to 2.08), stroke (HR 1.46, 95% CI 1.09 to 1.95) and ischaemic stroke (HR 1.99, 95% CI 1.49 to 2.64) than those without LOF alleles. Metaregression suggested that there was evidence of a reduced effect of LOF alleles in patients given a loading dose of clopidogrel compared to those who were not; in patients taking clopidogrel plus long-term aspirin relative to those taking only clopidogrel or clopidogrel plus short-term aspirin; and in studies that included patients with stroke and TIA as primary events compared with only patients with stroke. The stratified analysis based on clopidogrel regimen suggested that there was no evidence of a difference in the risk of secondary vascular events between those taking clopidogrel plus aspirin (HR 0.92; 95% CI 0.50 to 1.74). However, the analyses based on loading dose of clopidogrel and clopidogrel regimen should be considered exploratory as analyses were not pre-specified. There was no difference in the risk of bleeding between those with and without LOF alleles (HR 0.98, 95% CI 0.68 to 1.40).

Objective 4 identified 11 relevant studies evaluating the accuracy of the POCT in scope. All evaluated Spartan versions of the Genomadix Cube – either Spartan Cube, Spartan RX or Spartan FRX, against a laboratory reference standard – there were no studies on the accuracy of Genedrive. All studies were judged at low RoB. None of the studies were conducted in a stroke population. The Genomadix (Spartan) CYP2C19 tests were found to have very high accuracy with summary estimates of sensitivity and specificity both 100% (95% CI 94% to 100% for sensitivity and 99% to 100% for specificity) for the detection of *2 and/or *3 LOF alleles. There were very few disagreements between the Genomadix Cube and laboratory-based reference standards – 8 of the 11 studies reported perfect agreement between the tests. There was no suggestion of a difference across the three different versions of the test evaluated.

Objective 5 identified 17 relevant studies evaluating the technical performance of the two POCTs in scope and conducted a survey of genomic laboratories to gather information on the technical performance of laboratory-based tests. One study evaluated Genedrive and the others all evaluated Genomadix (Spartan) CYP2C19 tests. Only one of the studies was conducted in a stroke population. There was substantial variation in estimates of test failure rate for Genomadix (Spartan) CYP2C19 tests, which ranged from 0.4% to 19% across studies – the true test failure rate is therefore unclear but has the potential to be a large proportion of samples. Some studies provided data on the prevalence of the different variant forms of CYP2C19; however, these were relatively small samples with little information on ethnicity which is a major factor determining the prevalence of LOF alleles. Studies reporting time to results for Genomadix (Spartan) CYP2C19 tests were consistent with the estimate provided by the manufacturer (64 minutes) – most studies reported that time from buccal swab to results was around 1 hour, although two studies reported higher estimates of 90 minutes and 90–120 minutes. One study of Genedrive reported that it gives results in around 40 minutes. Studies generally suggested that Genomadix (Spartan) CYP2C19 tests were simple, user friendly and can be conducted by staff who have received minimal training. Limitations highlighted include storage conditions (samples need to be frozen and stored between −15 and −80 degrees), only one sample can be genotyped at a time, and it only tests for *2, *3 and *17 alleles. The study that evaluated Genedrive, noted that the test is simple, portable, rapid and does not require analytes to be frozen, and tests for *2, *3, *4, *8 and *17 alleles. One study estimated the cost per patient test at 225 Euros for Spartan (Genomadix) RX but did not provide any information on how this estimate was reached. Genedrive and Genomadix provided information on the platform cost, assay cost and cost of external control kits, which were used in our economic model.

The survey of genomic laboratory hubs had an excellent response rate. Of the 10 labs to home the survey was sent, 8 labs, including those in Northern Ireland, Scotland and Wales, completed the survey. All but one of the labs reported that they had at least one form of sequencing technology and all had at least one form of targeted CYP2C19 gene variant detection, most commonly PCR-based SNP genotyping assays using fluorescent reporter systems, such as TaqMan (ThermoFisher). Preferred technologies for performing CYP2C19 testing varied across labs and included: NGS (two labs), MassARRAY (three labs), LAMP (three labs), PCR-based SNP genotyping assays using fluorescent reporter systems, such as TaqMan (ThermoFisher) and QuantStudio 12K Flex Real-Time PCR System or X9 Real-Time PCR System (one lab). Resource requirements varied across labs, making it difficult to estimate the exact resources that would be required for CYP2C19 testing, and this may vary according to the specific lab test that would be used. Costs per test varied from around £15 (MassARRAY, although another lab estimated this as £100) to £250 for next-generation gene sequencing. Most labs reported that their preferred test could be performed by existing staff members with standard training or that the test was fully automated, although one lab stated that their preferred test (QuantStudio 12K Flex Real-Time PCR System or X9 Real-Time PCR System) would be new to their lab and so training would be required. Most labs reported that it was difficult to estimate the proportion of samples that would not return a valid result, but most expected this to be < 1%. Testing capacity and TATs also varied across labs. Current testing capacity ranged from 0 to 200 tests per week, and TAT ranged from 24 to 72 hours up to 1–2 weeks, although five laboratories estimate that this would be between 72 hours and 1 week. Most laboratories reported that additional testing capacity and faster TAT would be possible with additional resources – including staff, laboratory space, increased automation and equipment. Most labs also confirmed that the test could be performed in local laboratories although additional staff training and/or equipment would be needed. The major barriers to implementing CYP2C19 testing were the scale of the predicted activity and current capacity (four labs), with one highlighting that they do not currently perform any tests of this scale in the NHS and so do not have the infrastructure for this. Most labs stated that while it should be possible to implement POCTs within the laboratory workflow, this may not be the most efficient process for the number of samples that would need to be tested.

We developed a decision-analytic model to evaluate the cost-effectiveness of POCT and laboratory tests compared with no testing (Objective 6), in two populations: (1) TIA/minor ischaemic stroke and (2) non-minor ischaemic stroke; to reflect the different treatment pathways and event rates in these populations. Results were also obtained for a mixed ischaemic stroke and TIA population. We modelled patients moving between five health states: no recurrent stroke, minor stroke, major bleed or ICH, moderate stroke and severe stroke, with a mortality rate depending on health state. A decision tree was used to capture short-term (90-day) outcomes, and a Markov model with a 1-year cycle captured outcomes beyond 90 days over a patient’s lifetime.

In our base case for all populations, we found that testing was cost-effective, with laboratory and point-of-care CYP2C19 testing strategies dominated no testing, that is, CYP2C19 testing generated more QALYs and lower costs compared with no testing. This finding was robust to assumptions made in the model. There was uncertainty, however, on the most cost-effective test (laboratory or POCTs), and it should be noted that results for Genedrive were illustrative only, and conclusions should be focused on comparisons between the laboratory test and the Genomadix Cube CYP2C19 test.

The laboratory test and POCTs gave very similar mean QALYs, and so we compare the tests in terms expected net monetary benefit at a willingness to pay of £20,000 per QALY from our probabilistic analysis, preferring tests with the highest expected net monetary benefit. In the non-minor ischaemic stroke population, the expected net monetary benefits were £6230, £6214 and £6138 for Genedrive, the laboratory test and the Genomadix Cube CYP2C19 test, respectively. In the TIA/minor stroke population, the expected net monetary benefits were £2932, £2802 and £2829 for Genedrive, the laboratory test and the Genomadix Cube CYP2C19 test, respectively. Net monetary benefit is greatest in the non-minor ischaemic stroke population due to the higher event rate in the long term, and hence greater benefit of appropriate treatment in this population. In the combined TIA/ischaemic stroke population, the expected net monetary benefits were £5211, £5163 and £5119 for Genedrive, the laboratory test and the Genomadix Cube CYP2C19 test, respectively. In all populations, expected net monetary benefit is similar, suggesting little difference between the tests. Omitting Genedrive, where there are insufficient data, the highest expected net benefit is for the laboratory test in the non-minor ischaemic stroke population, and the Genomadix Cube CYP2C19 test in the TIA/minor stroke population.

It should be noted that due to limited information on Genedrive, we assumed all model inputs were the same as for the Genomadix Cube CYP2C19 tests with the exception of the costs, which is why Genedrive appears the more cost-effective of the POCTs. The results for Genedrive should therefore be considered exploratory only and this finding may change as further evidence on the accuracy and performance of Genedrive becomes available. We conducted a threshold analysis for the diagnostic test accuracy of Genedrive, reducing sensitivity and specificity (but keeping them equal). We found that the Genedrive system was found to be cost-effective at sensitivity and specificity equal and over 99% versus the Genomadix Cube, equal and over 98% versus the lab test.

The model inputs that have the biggest impact on the cost-effectiveness results were the costs of the different stroke states, and the treatment effects for stroke in patients with CYP2C19 LOF, and the HR for major bleed/ICH on aspirin relative to clopidogrel. However, varying these parameters did not change the overall finding that CYP2C19 testing is cost saving and generates more QALYs compared with no testing. Accounting for uncertainty in a probabilistic analysis gave very similar results. Cost-effectiveness acceptability curves show that there is a high probability that one of the testing strategies is the most cost-effective, with Genedrive having the highest probability of being cost-effective. The laboratory test has a low probability of being cost-effective in the TIA/minor stroke population due to the delay in receiving results.

The overall finding that CYP2C19 testing is cost saving and generates more QALYs compared with no testing was robust in all the scenarios that we explored. The scenarios where CYP2C19 testing was most cost-effective were when prevalence of CYP2C19 LOF was high and for younger cohorts of patients. The scenarios where CYP2C19 testing was least cost-effective were when we assumed that only 69.9% of LOF patients actually receive alternative treatment, and when the alternative treatment was ticagrelor. However, CYP2C19 testing was still cost saving but with a smaller increase in QALYs.

Strengths and limitations of the assessment

Our systematic review followed published guidance on the conduct of systematic reviews and is reported according to PRISMA-2020 guidance33 and PRISMA-DTA guidance,34 making our review processes transparent and robust. The protocol was pre-registered on the PROSPERO database (CRD42022357661) and published on the NICE website for this appraisal. 1 Following protocol registration, we increased the scope of studies eligible for inclusion for Objective 4 or 5, from diagnostic cohort studies (only) to include primary studies of any design. This increased the scope of potentially eligible studies, and we consider it a strength of the work.

We conducted extensive literature searches designed to maximise retrieval of relevant studies and did not apply any language or date restrictions to these searches. We included attempts to locate ongoing and unpublished studies through searches of trial registries and screening of the manufacturers’ submissions. We defined clear, unambiguous inclusion criteria and documented reasons for exclusion of studies at full-text review and for all studies included in the manufacturers' submissions to ensure transparency in how we applied our inclusion criteria. We conducted a formal assessment of the RoB of included studies using the RoB 2 tool for RCTs,37 the ROBINS-E tool for observational studies38 and the QUADAS-2 tool for diagnostic test accuracy studies. 39 These tools are the most robust tools available for these study designs with a clear focus on the RoB within each study. We modified QUADAS-2 to exclude the assessment of applicability; the other tools do not include an applicability assessment. Instead of a formal assessment of applicability, we extracted details on information that could result in variation across studies and considered this in our synthesis of results. All stages of the review process involved at least two reviewers to minimise the RoB or error in the review. Our synthesis included a meta-analysis for Objectives 2, 3 and 4. For Objective 2, different studies evaluated different intervention comparisons. We investigated the potential to carry out a NMA, but the networks either did not connect or were a simple chain of evidence, and so a NMA would give the same results as an analysis of each comparison pair separately. Most comparisons were made by a single study except for ticagrelor versus clopidogrel and aspirin versus clopidogrel, where we pooled the results from two studies.

For Objective 3, where all studies compared outcomes in those with and without LOF alleles taking clopidogrel, there was some evidence of heterogeneity which we explored using metaregression. However, we did not pre-specify all variables that we later considered potential sources of heterogeneity; analyses based on these variables should be considered exploratory and interpreted with caution. There were too few studies to investigate differences across studies for Objectives 1 and 2. For Objective 4, all studies reported consistently high estimates of the accuracy of Genomadix (Spartan) CYP2C19 tests, and so there was no evidence of heterogeneity in outcomes across studies. We formally assessed the potential for publication bias/small-study effects for Objective 3 and found no evidence to suggest the presence of publication bias. We did not include a formal assessment of publication bias due to the small number of included studies for Objectives 1 and 2, and for Objective 4 due to the difficulties in assessing publication bias for diagnostic test accuracy studies where there is no clear threshold for ‘significance’.

We assessed the accuracy of the two POCTs in scope in relation to laboratory tests which were considered the reference standard for this appraisal, this assumes that they have 100% accuracy. Although there are a variety of laboratory tests available there is no clear ‘reference standard’ amongst these, so it would not be possible to specify a reference standard for evaluation of laboratory-based tests. Instead, we assumed that they all have 100% accuracy, following advice from our clinical experts. Given that we found the Genomadix (Spartan) CYP2C19 tests to have close to 100% accuracy (Objective 4), this assumption appears reasonable. For Objective 5, we reviewed studies of the technical performance for our two POCTs; however, for laboratory-based tests, we took a different approach conducting a survey of genomic laboratories to obtain information on the technical performance of laboratory-based tests. We considered that data obtained by a survey would be specific to our research question, provide information of direct relevance to the NHS and provide the most up-to-date data, which would be unlikely to be the case had we reviewed the existing literature on these tests. The survey of the genomic laboratories had a very high response rate with 8 of the 10 labs invited to participate completing the survey.

While there have been previous economic evaluations of the cost-effectiveness of CYP2C19 testing for targeted antiplatelet treatment for patients following a TIA or ischaemic stroke,103–106 there has only been one previous model in a UK population which compared Genedrive versus no testing. 103 Our model has a similar structure to that of Wright et al.,103 but we model stroke severity state rather than number of strokes experienced, and have used more recent evidence sources as inputs to our model. Our conclusions are in line with those found by Wright et al. 103 for Genedrive, but our model also includes the Genomadix Cube CYP2C19 test and laboratory testing options.

We made a range of assumptions in our model (Objective 6), but our scenario analyses suggest that the main conclusion that CYP2C19 testing is cost saving and generates additional QALYs was robust to these assumptions. The different testing strategies had similar QALYs under our model assumptions, and choice between them is largely one of cost minimisation and ease of implementation. To estimate per test costs, we needed to assume a lifetime of the devices which was based on estimates from Genedrive, Genomadix, and an assumption for the lab test. Some costs were excluded because they become negligible when evaluated per test, including freezers to store the Genomadix Cube CYP2C19 test, and staff training costs, and changes to processes to ensure results are recorded in patient records. However, these would represent an upfront investment if testing is adopted in the NHS. Our survey suggested different laboratories would use different platforms; we used the Agena MassARRAY iPlex as a typical platform when estimating lab test costs, but this cost is likely to vary across laboratories. We conducted a threshold analysis for the cost of the lab test which provides lab costs below which the lab test is cost-effective compared with each POCT and for each population.

In our base case, we assumed that all patients with a LOF test result would receive targeted treatment; however, Swen et al. 159 found that physician adoption of pharmacogenetic recommendations was only 69.9% for a range of genes including CYP2C19. Our scenario analysis suggests that although costs increase and QALYs decrease as uptake falls, CYP2C19 is still cost saving and generates more QALYs compared with no testing.

We assumed that the alternative treatment that CYP2C19 LOF patients would receive is dipyridamole plus aspirin, and that the efficacy of dipyridamole plus aspirin does not depend on CYP2C19 LOF status. This assumption was necessary because we did not identify any studies of dipyridamole plus aspirin in patients with CYP2C19 LOF in our reviews and instead used evidence from the PRoFESS RCT. 158 The CHANCE study52 found that the efficacy of aspirin varied with CYP2C19 LOF status, and so it may be the case that dipyridamole plus aspirin also depends on CYP2C19 LOF status. Further research would be required to assess this.

In the absence of test-and-treat studies, we used RCTs comparing treatments for LOF and NoLOF patients where these were available. For some treatments (aspirin, ticagrelor), relative to clopidogrel, we had information for LOF patients, but no information for NoLOF patients. In these cases, we used the result for LOF patients and then applied a HR for clopidogrel for LOF versus NoLOF estimated from the clinical review (Objective 3). The results were not found to be sensitive to changes in this assumption.

We included treatment discontinuation due to treatment-related adverse events, including major bleeds/ICH. It was assumed that in all cases patients would switch to low-dose aspirin following treatment discontinuation. If patients taking clopidogrel would instead switch to dipyridamole plus aspirin, this may change the cost-effectiveness results, because dipyridamole plus aspirin is more effective at preventing future strokes than aspirin. However, dipyridamole plus aspirin also has a high risk of bleeding events, so it is not clear how this would change the results.

Some patients may not take aspirin either due to drug sensitivity or patient or physician choice, and for these patients who have a CYP2C19 LOF variant the alternative treatment is likely to be dipyridamole monotherapy. We did not explicitly model these patients, but the benefit of testing is likely to be reduced for these patients due to lower efficacy of dipyridamole monotherapy relative to dipyridamole plus aspirin. 129

We incorporated the impact of stroke on carers by applying a utility decrement for carers of patients who had a moderate or major stroke (one carer per patient). We assumed that patients do not require a carer following a minor stroke or TIA. Applying the carer utility decrement to patients in the major stroke state led to negative contributions to QALYs, which we felt was reasonable given the very low quality of life for patients who have had a major stroke.

For all objectives, we took a pragmatic approach where the presence of LOF alleles was considered to be the presence of at least one *2 or *3 LOF allele – these are the two most common LOF alleles, although the prevalence of *3 (0.3%) is much lower than the prevalence of *2 (16%). Genomadix Cube only detects these LOF alleles; Genedrive also detects *4, *8 and *35, and the prevalence of these is very low (0.2% and 0.1%, respectively). Both tests are also able to identify the *17 allele, which is associated with increased function; we did not investigate how this could have impacted on outcomes (Objectives 1–3) or on the accuracy of the test (Objective 4). Laboratory tests could, in principle, detect all LOF alleles. There has been some suggestion that having one LOF allele and one increased function allele (e.g. *2/*17 or *17/*3) could, in effect, cancel each other out and lead to normal function, but there is little evidence to support this. 15 It would be possible to offer alternative treatments to rapid metabolisers, those with one normal function allele and one increased function allele (e.g. *1/*17) or ultrarapid metabolisers, those with two LOF alleles (*17/*17). There is also the potential that having two LOF alleles (e.g. *2/*2 or *2/*3) could lead to poorer metabolism of clopidogrel than having only one LOF allele (e.g. *2/*1 or *1/*2). While some studies included in the review did classify participants into normal (no LOF alleles), intermediate (one LOF allele) and poor (two LOF alleles), in this situation, we combined data across categories to dichotomise into normal and poor (intermediate and poor combined) metabolisers. For example, one of the studies included for Objective 1 applied different treatments for those carrying no LOF alleles, one LOF allele and two LOF alleles. There is potential for considerable complexity in how treatment is tailored based on CYP2C19 testing. In practice, a similar pragmatic approach to the approach that we have taken is most likely to be adopted where individuals are dichotomised into low and normal metabolisers. For example, although the CPIC guidance15 classifies people into five metaboliser categories, treatment is dichotomised so that standard treatment with clopidogrel is recommended for ultrarapid (two increased function alleles), rapid (one increased function alleles and one normal function allele) and normal metabolisers (two normal function alleles); alternative treatment is recommended for poor (two LOF alleles) or IMs (one LOF allele, one normal function allele). 15 A related limitation is that some studies only evaluated the *2 LOF allele and did not look for the presence of *3 mutations. Given the very low prevalence of *3 in most populations, this is unlikely to have had a substantial impact on results.

Our review and model focused on evaluation of the accuracy of POCTs against a laboratory reference standard. It could also be of interest to compare the accuracy of both POCT and laboratory CYP2C19 tests (genotype tests) to P2Y12-pathway specific platelet function tests (phenotype test) as a reference standard, or to consider the potential utility of these tests as an alternative to genetic testing. Platelet function testing of blood samples can assess the response to antiplatelet medication and give information on whether the phenotype (in this case, clopidogrel resistance) is encoded by the gene is expressed. These tests must be performed while a patient is taking clopidogrel. This was considered as part of the scoping process for this appraisal, but it was determined that this was not in scope. Such testing is not currently in use in the NHS, and results can be difficult to interpret and are affected by factors other than the genotype alone, for example, a person’s size. 174 While such tests would potentially have a benefit in that they can predict response to clopidogrel based on both genetic and environmental factors, the prognostic value and clinical utility of platelet reactivity testing are still unclear. 175 Further, the need for patients to have started clopidogrel testing for these tests to be used limits their potential usefulness within the setting of this appraisal.

There were a number of limitations in the evidence base. The ideal study to evaluate the potential impact of a new diagnostic test is a RCT where participants are randomised to either be tested and treated accordingly or to not test and receive standard care. We only identified two ‘test-and-treat’ studies; however, neither of these were randomised and both appeared underpowered to detect differences in secondary vascular events between groups. We therefore needed to draw on less robust types of evidence to determine whether people with LOF alleles have better outcomes if treated with alternative antiplatelet drugs. There were also very FEW data for Objective 2, which looked at whether people with LOF alleles had better outcomes when treated with alternative antiplatelet drugs compared to treatment with clopidogrel. A major limitation for our review was that none of the studies for Objective 1 or 2 evaluated dipyridamole + aspirin, the most likely alternative treatment to clopidogrel that would be offered in the NHS. However, aspirin alone may potentially be considered as an alternative treatment option in the NHS, and we did identify two studies comparing clopidogrel plus aspirin for the first 21/30 days with aspirin alone. These studies suggested that there was no difference in the risk of secondary adverse events between those taking clopidogrel plus aspirin and long-term aspirin alone, suggesting that aspirin alone may be an appropriate treatment option. Furthermore, our scenario analysis found that test and treat with aspirin as an alternative treatment was cost saving and generated more QALYs compared with no-test, although the cost savings and QALY benefits were not as great as when the alternative treatment was dipyridamole plus aspirin. While there was evidence that ticagrelor may reduce secondary vascular events compared to clopidogrel in those with LOF alleles who have had a stroke, this is not currently licensed for stroke patients in the UK. One study also suggested an increased risk of bleeding with ticagrelor compared to clopidogrel, although a second study reported no difference in bleeding risk. We ran a scenario analysis using ticagrelor as an alternative treatment for LOF patients and found that CYP2C19 testing was still cost saving and gained QALYs, but that these were less than for dipyridamole plus aspirin as the alternative treatment. Triflusal was evaluated in two studies – this is not currently a realistic treatment option as it is unavailable in the UK.

There was the lack of consistency in which outcome results were reported by the included studies. We performed meta-analyses of HRs, and estimated HRs and their standard errors (SEs) from event frequencies and follow-up time, or 2 × 2 tables of event numbers,52 where these were N/R. This relies on an assumption of proportional hazards. Although hazards of recurrent ischaemic events are variable and higher early after the primary event, it is plausible that that the ratio of hazards remains proportional. Some studies reporting HRs reported testing this assumption and found no evidence against it. 52,53 We were unable to explore relaxing this assumption based on the data available in the primary study reports. A related, and potentially more significant, limitation is that some studies did not report a mean follow-up time per arm, and so projected follow-up times had to be used instead. For these studies, real follow-up time and time at risk are likely to be lower, and could potentially differ between groups.

For Objective 3, there was some overlap between studies that treated patients with clopidogrel plus long-term aspirin and that included a clopidogrel loading dose, both characteristics found to be associated with improved outcomes in LOF carriers. There were not enough studies to assess if the effects from these interventions were independent from each other.

Uncertainties

Our systematic reviews and economic modelling have identified several areas where uncertainties remain. As highlighted above, for this appraisal, we dichotomised into poor and normal metabolisers based on the presence of at least one *2 or *3 LOF allele. All studies that evaluated the accuracy of POCT evaluated their concordance with a laboratory-based reference standard that was testing for the presence or absence of the same LOF alleles – *2 and/or *3. It is unclear how accuracy would have differed had the laboratory-based reference standard tested for the presence of any LOF allele. As *2 and *3 are the most common LOF alleles, this may have had limited impact on findings. There were insufficient data to consider the impact of testing for other LOF alleles, or to evaluate whether treatment should take a more complex approach with different categories based on different combinations of LOF, normal function and increased function alleles – there are up to five categories that could be considered for this, but the benefit of treating each of these metaboliser categories separately is unclear, and in particular how those with increased function alleles (e.g. *17) should be treated. It is also unclear exactly which alleles should be tested for, while some alleles such as *2, *3, *4 and *8 have been clearly linked to LOF, for others such as *9 and *10 the evidence based is still evolving and these are currently categories as ‘indeterminate’ or ‘likely LOF’. 15 These alleles are also much rarer than some of the other alleles and so the benefit of testing for alleles beyond those most frequently associated with LOF is unclear. This is of particular importance when considering which test would be most appropriate to implement within the NHS. Genomadix Cube only detects *2, *3 and *17, while Genedrive detects *4, *8 and *35 in addition to these alleles. There is more flexibility in which tests can be targeted by laboratory-based tests. Some, such as Sanger sequencing, will detect all variants, others will target specific LOF alleles, although probes can potentially be developed for all alleles of interest.

There have been a number of other reviews looking at the role of CYP2C19 testing to guide antiplatelet treatment. The most recent of these was the review by McDermott et al. published in 2022. 176 This non-systematic review included studies describing the interaction between CYP2C19 genotype and clinical outcomes following ischaemic stroke or TIA. The authors concluded that there was good evidence that CYP2C19 LOF allele carriers of Han Chinese ancestry have increased risk of further vascular events when treated with clopidogrel. This is in line with our findings, and although the majority of our studies (13/25 for Objective 3) were from China, we found that results from these studies were consistent with results from studies conducted in other settings. Most of the observed heterogeneity across studies for Objective 3 was explained by whether a loading dose of clopidogrel was applied and by whether clopidogrel was given alone or in combination with aspirin with a smaller difference seen between LOF and NoLOF carriers when a loading dose was applied and clopidogrel given in combination with aspirin. The McDermott review did not carry out a formal statistical synthesis of results across studies and so was not able to carry out the formal investigation of differences across studies that we conducted as part of our review. It also did not differentiate between the different objectives of the studies in the same way that we have done, which allowed us to draw conclusions both on whether people with LOF alleles have poorer outcomes when treated with clopidogrel compared to those without LOF alleles, and on whether those with LOF alleles have a reduced risk of secondary ischaemic events when treated with an alternative to clopidogrel.

Although our review showed consistently high estimates of accuracy across the included studies, a number of questions remain regarding the accuracy and technical performance of CYP2C19 POCT, including accuracy of Genomadix Cube compared to Spartan versions of the test; accuracy of Genedrive; accuracy in stroke patients; and test failure rate.

All the evaluations that contributed data on Genomadix Cube were actually Spartan versions of the test. Two of these evaluated Spartan Cube; this appears equivalent to Genomadix Cube and was renamed when Spartan Bioscience’s assets were acquired by Genomadix. Other studies evaluated Spartan RX and one study evaluated FRX. The difference between the RX and FRX versions is unclear, but it appears that FRX is either an earlier version of the RX or the test was renamed – the study of the FRX version was earlier than other evaluations. Genomadix as part of their submission to NICE explained that

the original Spartan RX CYP2C19 System and the subsequent Genomadix Cube CYP2C19 system both test for the same CYP2C19 alleles; however, the Genomadix Cube test is not a newer version of the Spartan RX test. Significant differences include the mechanisms used to heat and cool the samples, the storage, use and stability of the specimens on the swab, the optical system, and the test workflow. A direct comparison of the performance of the two systems is not available.

A related NICE methods innovation briefing on Spartan Cube explained that this was released as a successor to Spartan RX and they differ in that ‘the 3 reaction tubes are integrated into a single test cartridge, the swabs and test cartridges are packaged separately, and the DNA analyser device is smaller’. 177 It is therefore unclear whether the performance of the Genomadix Cube can be considered equivalent to that of the Spartan RX; however, there was no suggestion of a difference in performance between the two studies that evaluated Spartan Cube and those that evaluated Spartan RX. Further studies that directly compare the two tests are required to confirm this.

The only data available on Genedrive were one small unpublished early evaluation of the test that did not report data on accuracy. The manufacturer confirmed that there were no studies currently ongoing, but these were planned to start from the first quarter of 2023; no details were provided on what these studies would evaluate. They also highlighted that the Genedrive system has been reviewed by NICE for an alternative assay – the Genedrive MT-RNR1 ID Kit, under the MIB29021 and this is also currently under review as an Early Value Assessment, currently in progress. 21 Genedrive offers some potential benefits compared with Genomadix Cube: it is slightly quicker (40 vs. 60 minutes), cheaper (we estimated the cost at £104 per test for Genedrive and £197 for Genomadix), tests for a greater number of alleles – *8 and *35 in addition to the *2, *3 and *17 that are detected by Genomadix Cube. This may be of particular value in ethnicities such as those with Asian and Jewish origins where these alleles are found at higher frequencies. A further benefit of this test is that it does not require the frozen storage of test kits that is required for Genomadix Cube. It also allows patient data to be uploaded directly into patient records, where data are only stored locally with Genomadix Cube. In the absence of data on accuracy and performance of Genedrive, we assumed these would be the same as for the Genomadix Cube in our economic model, which meant that Genedrive was slightly cheaper with the same benefits as Genomadix. However, these findings are uncertain and may change when data on the accuracy and performance of Genedrive become available.

None of the studies that evaluated the accuracy of the POCT evaluated these tests in our population of interest – people who have had an ischaemic stroke or TIA. While for many tests, the population in which the test is conducted can lead to substantial variation in estimates of accuracy,178 we consider this less likely to be the case for genetic tests. With genetic markers, the LOF alleles are either present or absent and will not be affected by other factors that could influence test performance such as comorbidities, age, sex, setting and disease prevalence.

There was substantial variation in test failure rate across studies – this ranged from 0.4% to 18.9% for studies of Genomadix (Spartan) CYP2C19 tests. Reasons for this substantial variation were unclear and so it is difficult to determine the true failure rate. There was no information on the failure rate of Genedrive. Most laboratories expected their favoured laboratory test for CYP2C19 testing to have a failure rate of < 1%. Failure rate could be an important factor in deciding which CYP2C19 test to implement and so further, accurate data are required on failure rate for all tests, both POCT and laboratory based. Another factor that could influence which test should be recommended for CYP2C19 testing is how test results are recorded in patient records and communicated across settings. Genedrive allows results to be added directly to patient records; this is not possible with the Genomadix test. It is important that the results of CYP2C19 testing are not just available at the point of testing but are recorded in the patient’s record for future reference in case they may need to be prescribed clopidogrel again in the future. It is also important that test results are not lost as patients are discharged from hospital after the test is requested but before results are available. This would reduce the uptake of appropriate targeted treatment, which we found led to a large reduction in the cost savings and QALY benefits of testing in our scenario analyses.

There are proposals to implement pre-emptive panel testing across the NHS. A recent multinational cluster randomised trial of a 12-gene pharmacogenetic panel found that guided treatment based on this panel significantly reduced the incidence of clinically relevant adverse drug reactions and was feasible across different European healthcare systems organisation and settings, including the UK. 159 Panel testing may initially only be introduced in some areas as a pilot scheme but longer term could become standard – this may mean CYP2C19 status is already available for certain patients. A recent report on ‘Personalised Prescribing’ by the RCP and British Pharmacological Society produced a set of recommendations for embedding pharmacogenomics, including CYPC19 testing, in the NHS. 179 It is currently unclear whether and when this will be commissioned in NHS care, but a pilot project, the PROGRESS project, is currently underway. 180 If such a programme were implemented, then it may reduce the appropriateness of specific CYP2C19 testing. However, a potential limitation in implementing any form of pharmacogenetic testing is the potential unwillingness of clinicians to act upon pharmacogenetic test results. The multinational trial of the 12-gene panel found that up to 30% of clinicians did not accept the recommendation of a treatment change based on the genetic test. 159

We did not identify any studies comparing test-and-treat strategies with dipyridamole + aspirin as the alternative treatment, which is the most likely alternative treatment to clopidogrel that would be offered in the NHS, nor did we identify any studies comparing the efficacy of dipyridamole + aspirin compared with clopidogrel according to CYP2C19 LOF status. In order to include dipyridamole + aspirin as the alternative treatment in our model, we assumed that the efficacy of dipyridamole + aspirin does not depend on CYP2C19 LOF status. There is therefore uncertainty as to the costs and benefits of using dipyridamole + aspirin as the alternative treatment. Note, however, that CYP2C19 testing was still found to be cost saving and have increased QALYs compared to no testing in our scenarios where aspirin or ticagrelor were the alternative treatment. This suggests that a test-and-treat strategy with dipyridamole + aspirin as the alternative treatment is very likely to be cost-effective, regardless of the exact relative treatment effects by CYP2C19 LOF status.

Genetic variability is not the only factor to affect the efficacy of clopidogrel. Other factors that may be associated with clopidogrel efficacy include other drugs, for example, PPIs may inhibit clopidogrel,3 smoking status, weight, diabetes, hypertension14 and the influence of other rare genetic variants.

We did not find any evidence in a paediatric population in our searches and did not have sufficient evidence to include children in our economic model. However, in children, aspirin, rather than clopidogrel, is currently recommended to prevent recurrence of stroke, and so in that case, there would not be any benefit of testing for CYP2C19 LOF status. We ran a scenario analysis for a cohort of younger adults at index TIA/ischaemic stroke and found that the cost-saving and quality of life benefits improved. This suggests that if clopidogrel were to be considered for a child, then CYP2C19 testing is likely to be cost-effective.

Equality, diversity and inclusion

Our research was based on existing literature and so we had no control over the participants enrolled. We were broad in our inclusion criteria such that studies from any country and in any language of publication were eligible. CYP2C19 LOF is more prevalent in some populations than others and is particularly high in Asian populations. We explored ethnicity as a potential source of heterogeneity in our analyses for Objective 3, where sufficient data were available to do so, but found that overall there was no association between ethnicity and how those with and without LOF alleles respond to clopidogrel. Our economic analysis found that testing was cost-effective for the range of prevalence observed across ethnicities, suggesting that this would not introduce inequity.

Our survey received a broad response from laboratories across England, Wales, Scotland and Northern Ireland providing the full UK perspective on the potential for providing CYP2C19 testing.

Our team included researchers with a broad range of experience and expertise. The lead authors are junior researchers within Bristol TAG, who were given the opportunity to lead on the writing of this report to help develop their research skills and portfolio. They were supported by the two senior authors, professors and co-directors of Bristol TAG, who provided advice and mentorship to the junior researchers leading on the reviews and health economic modelling. The team included those with expertise in systematic reviews, health economics and medical statistics. We also included two clinical experts from the South West Genomic Medicine Service Alliance who provided clinical expertise, particularly around genetic testing. Where needed, we were able to draw on the broader expertise of the seven specialist DAC members.

Patient and public involvement

Our team included two patient representatives who have lived experience of stroke. The patients attended a meeting near the beginning of the project with the full team to share their experience of stroke and gave suggestions of important outcomes to be considered in the economic model. They also helped to ensure the findings of the review are comprehensible by giving feedback on the plain language summary.

Implications for service provision

Our results suggest that CYP2C19 testing followed by tailored treatment is likely to be effective and cost-effective in both populations modelled (non-minor ischaemic stroke and TIA/minor ischaemic stroke) providing cost savings and increased QALYs compared with not testing. We found this finding was robust to assumptions made in the model. There is uncertainty around the best test to use (laboratory or POCT), and there is no direct evidence for alternative treatment dipyridamole for patients with LOF allele status. There was very little difference in cost-effectiveness estimates between the three tests. Omitting Genedrive, where there are insufficient data, the highest expected net benefit is for the laboratory test in the non-minor ischaemic stroke population, and the Genomadix Cube CYP2C19 test in the TIA/minor stroke population. The choice of test to be adopted is likely to be based on practical considerations as to which test would be most appropriate and practical within an NHS setting. Our survey of laboratories suggested that, given the high volume of testing required, it may be more appropriate to implement testing through the genomic labs, or in local labs to allow batching of tests; POCTs are only able to run a single test at a time. However, POCTs have the potential advantage of providing results more quickly (< 1 hour compared to around 1 week), which ensures appropriate treatment can be started as soon as possible and reduces the risk of results not being actioned if the patient has been discharged from secondary care. There are logistical difficulties to implementing POCTs in stroke care. The Genomadix Cube would require appropriate storage in a freezer which would require an investment in both freezers and space. Considerations on how results would be integrated into patient’s medical records both for future prescribing where clopidogrel may be considered, but also on the impact of CYP2C19 variants on a wide range of other medicines. Genedrive is able to add results directly to patient records, which is not currently possible for Genomadix Cube. Our survey of laboratories indicated that while facilities are available for CYP2C19 testing, there would be capacity issues to incorporate routine testing into existing workflow. There was also substantial variation in preferred laboratory test for CYP2C19 testing. If laboratory testing is adopted, this would require investment in equipment, staff, laboratory space and automation of processes.

There is some suggestion of reluctance by clinicians to act on pharmacogenetic variations. Implementation of CYP2C19 testing would therefore need to be accompanied by training and education of staff on the benefits of targeted treatment, to ensure that results are acted upon. There is also a question of what should happen to those already on clopidogrel treatment. Consideration needs to be given as to how such patients should be treated – it may be appropriate to test all those currently on clopidogrel treatment and adapt treatment as necessary.

As discussed above, there are proposals for introducing pharmacogenetic panel testing to test for a range of genetic variants associated with adverse treatment reactions. It is currently unclear when and if such programmes would be introduced in the NHS, but a pilot programme is currently in process. If such panel testing is likely to be introduced into the NHS, then the benefits of specific CYP2C19 testing in a TIA/ischaemic stroke population at this point need to be carefully considered.

Suggested research priorities

The section on uncertainties (see Uncertainties) highlights a number of areas where further research is needed. A clear gap in the evidence is on the clinical and cost-effectiveness of alternative antiplatelet strategies, in particular on dipyridamole plus aspirin, in a stroke population with LOF alleles. Further studies are required to determine the true effectiveness of dipyridamole plus aspirin in this population. The ideal study would randomise patients to be tested and treated based on LOF status – those with LOF alleles would receive dipyridamole plus aspirin, those without would receive clopidogrel. Outcomes would then be compared across groups. Such a study could incorporate a full economic analysis and be cluster randomised by hospital.

There is a lack of data on Genedrive. For the health economic model, we assumed that accuracy and test failure rate would be equivalent to that of Genomadix Cube; however, it is unclear how likely this is to be the case. Further test accuracy studies that also provide information on the technical performance of the test (e.g. test failure rate, cost, time to perform the test) are needed; Genedrive highlights that testing will be starting this quarter but details on what studies are proposed are not available. The true test failure rate of Genomadix Cube remains unclear; there was substantial variation in estimates of test failure rate. Further studies are required to determine what the test failure rate would be if the test were to be implemented into routine practice in the NHS.

The value of testing additional alleles beyond *2 and *3 is unclear, as is the clinical significance of the *17 LOF allele. Our review and economic model took a pragmatic approach where we dichotomised action based on testing such that those with at least one LOF allele were considered; however, it remains unclear whether this is the appropriate approach. Future studies should consider how those with *17 LOF allele respond to clopidogrel and whether those with two LOF alleles have worse outcomes than those with one LOF allele. This would inform whether the dichotomy used in our appraisal is the appropriate strategy to implement in practice.

Contributions of authors

Joe Carroll (https://orcid.org/0009-0004-7006-0938) developed and coded the health economic model, produced all model results and drafted the sections of the report describing the model, sensitivity and scenario analyses, and cost-effectiveness results. He also drafted the cost-effectiveness section of the report.

Catalina Lopez Manzano (https://orcid.org/0009-0009-8883-0167) led the review of Objectives 1–3. She also drafted sections of the clinical effectiveness report.

Eve Tomlinson (https://orcid.org/0000-0002-0969-602X) led the review of Objectives 4–5, contributed as second reviewer to Objectives 1–3, and checked the literature searches. She also drafted sections of the clinical effectiveness report.

Ayman Sadek (https://orcid.org/0009-0000-6565-0790) conducted the reviews of previous economic evaluations of CYP2C19 testing for clopidogrel resistance and previous models of secondary prevention of recurrent stroke, reviewed evidence for model inputs, contributed to model validation and drafted the cost-effectiveness review and model inputs sections of the report. He also drafted the cost-effectiveness section of the report.

Chris Cooper (https://orcid.org/0000-0003-0864-5607) designed and undertook the literature searches, acted as second reviewer for Objectives 4 and 5 (including review of company submissions), and worked on the review of cost-effectiveness. He drafted the sections of the report related to searching.

Hayley E Jones (https://orcid.org/0000-0002-4265-2854) provided statistical supervision for Objectives 1–4.

Lorraine Rowsell provided a patient perspective on the project and edited the plain language summary.

John Knight provided a patient perspective on the project and edited the plain language summary.

Andrew Mumford (https://orcid.org/0000-0002-5523-511X) provided clinical advice for the project, particularly on genetic testing.

Rachel Palmer (https://orcid.org/0009-0006-6253-3330) provided clinical advice for the project, particularly on genetic testing.

William Hollingworth (https://orcid.org/0000-0002-0840-6254) contributed to model conceptualisation and protocol development.

Nicky J Welton (https://orcid.org/0000-0003-2198-3205) provided oversight of the cost-effectiveness analysis, contributing to model conceptualisation, protocol development, review of previous models, identification of inputs to the model, model validation, interpretation and discussion of results of the cost-effectiveness analysis. She also drafted the cost-effectiveness section of the report.

Penny Whiting (https://orcid.org/0000-0003-1138-5682) provided oversight of the clinical effectiveness sections of report. She drafted the clinical effectiveness sections of the protocol, led the survey of laboratories and contributed to the reviews of effectiveness (Objectives 1–3) and accuracy (Objectives 4–5). She also drafted sections of the clinical effectiveness report.

All authors were involved in commenting on the final report. Penny Whiting is the senior author and guarantor.

Acknowledgements

We would like to thank the following specialist Diagnostic Appraisal Committee (DAC) members for clinical advice relating to this project: Dr Dagan Lonsdale (St George’s, University Of London), Prof Albert Ferro (Kings College London), Dr Daniel Burrage (Whittington Health NHS Trust), Dr Alexander Doney (University Of Dundee), Dr Dheeraj Kalladka (Imperial College Healthcare NHS Trust), Mr Paresh Parmar (London North West University Healthcare NHS Trust) and Dr Salim Elyas (Royal Devon University Healthcare NHS Foundation Trust).

We would also like to thank:

Professor Katherine Payne and Dr Stuart Wright (University of Manchester) for discussions on their health economic model of CY219Y testing.

Dr Cinzia Del Giovane (UniMore, Università degli Studi di Modena e Reggio Emilia) and Dr Irene Tramacere (Fondazione I.R.C.C.S. Istituto Neurologico Carlo Besta) for providing data from their Network Meta-Analysis of antiplatelets for secondary prevention in stroke/TIA patients.

Those that responded to our survey on behalf of each of the five regional genomic laboratory hubs (Central and South; East; North West; South East; and North East and Yorkshire) and the Scottish (NHS North Tayside), Welsh and Irish services.

Charlene Wisdom-Trew, Bristol TAG, for providing administrative support.

Data-sharing statement

All data extracted for the systematic review and the results of the risk of bias assessments are provided in full in the appendices to this report. The economic model can be obtained from the corresponding author and will be shared upon reasonable request for academic collaboration.

Ethics statement

The majority of the research included in this report is secondary research and as such did not require ethical approval. The survey of genomic laboratory hubs was shared with the Health Sciences Faculty Research Ethics Officers from the University of Bristol Research Ethics Committee to determine whether ethical approval was required. They determined that the survey constituted an audit of practice/service evaluation and so ethical review was not required.

Information governance statement

There were no personal data involved in the production of this report.

Disclosure of interests

Full disclosure of interests: Completed ICMJE forms for all authors, including all related interests, are available in the toolkit on the NIHR Journals Library report publication page at https://doi.org/10.3310/PWCB4016.

Primary conflicts of interest: Joe Carroll reports consulting fees from Wickenstones Ltd. Rachel Palmer reports UKCPA Genomics Committee Membership Lead. Nicky J. Welton reports honoraria from ABPI and payment from Takeda. All other authors have no interests to disclose.

Disclaimers

This article presents independent research funded by the National Institute for Health and Care Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, the HTA programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, the HTA programme or the Department of Health and Social Care.

Effectiveness and accuracy searches

We used two searches: one for Objectives 1–3 and a separate search for Objectives 4 and 5. The searches were not limited by study design, date of publication, or language. This allowed us to use these searches to identify studies for our review of cost-effectiveness.

Supplemental cost-effectiveness searches

Studies included in the review showing primary and secondary reports

Primary reports are the primary publication for the study and are used to refer to that study throughout text and tables.