Diagnostic accuracy and clinical impact of MRI-based technologies for patients with non-alcoholic fatty liver disease: systematic review and economic evaluation

Patients who have indeterminate results from fibrosis testing

Results from TE, ARFI and ELF tests may indicate that some level of fibrosis is present but may not be able to confirm the presence of advanced fibrosis (F3) or cirrhosis (F4). Where results show that some level of fibrosis is present, but the level of fibrosis cannot be confirmed, these results are referred to as indeterminate results. The range of values used to define indeterminate results and the language used to describe indeterminate results varies across guidelines and clinical studies (e.g. ‘grey zone’,29 ‘intermediate risk’22 and ‘inconclusive results’30).

In the BSG guidelines,22 it is recommended that clinicians should consider liver biopsy for patients with a TE score between 7.9 kPa and 9.6 kPa (intermediate risk of advanced fibrosis), and for patients with a TE score > 9.6 kPa (high risk of advanced fibrosis). In the European Association for the Study of the Liver (EASL) guidelines,18 it is recommended that a TE score < 8 kPa rules out advanced fibrosis and that a TE score ≥ 8 kPa represents an intermediate to high risk of advanced fibrosis. Clinical advice to NICE24 is that indeterminate results are also possible from ARFI, although the exact values for an indeterminate ARFI result depend on the device manufacturer.

Clinical advice to NICE24 is that indeterminate results are possible from the ELF test. ELF test scores between 7.8 and 10.523 or 7.7 and 9.7 are considered to be indeterminate results. 31 In the EASL guidelines,18 it is recommended that an ELF score < 9.8 rules out advanced fibrosis for patients with NAFLD.

In current NHS practice, a biopsy may be considered for patients with indeterminate results from fibrosis testing. MRI-based testing could therefore be used as an additional, non-invasive, diagnostic test to help clinicians assess the need for a liver biopsy. However, the EAG notes that the range of values used to define an indeterminate result can vary across guidelines for the same test and the terms ‘indeterminate’ and ‘intermediate’ are used interchangeably. It is therefore unclear which range of values from non-invasive tests should indicate an indeterminate result and signal that patients should be referred for MRI-based testing.

Patients for whom transient elastography or acoustic radiation force impulse is unsuitable

TE and ARFI may not be suitable tests for people with a very high BMI or those with significant ascites because excessive amounts of fat and fluid overlying the liver can prevent the propagation of shear waves necessary to assess liver stiffness. 24 The tests may fail, or the clinicians may decide not to refer patients for these tests because they are likely to fail.

Liver biopsy may be considered for this subgroup of patients to determine the stage of fibrosis. MRI-based testing could be used as an additional, non-invasive, diagnostic test to help assess the need for a liver biopsy.

Patients who have discordant results from fibrosis testing

Patients with NAFLD may undergo multiple tests to confirm the presence of advanced fibrosis. If the results from these tests are discordant, then liver biopsy should be considered. For example, in the EASL guidelines18 it is recommended that patients with discordant results, that is, patients for whom one non-invasive test indicates low risk of advanced fibrosis (e.g. TE < 8 kPa or ELF < 9.8) but another indicates intermediate to high risk of advanced fibrosis (e.g. TE ≥ 8 kPa or ELF ≥ 9.8), should be considered for liver biopsy.

Clinical advice to the EAG is that patients who have indeterminate results, patients for whom TE or ARFI is unsuitable, and patients who have discordant results should be considered for a liver biopsy. MRI-based testing could be used as an additional, non-invasive, diagnostic test to help assess the need for a liver biopsy.

LiverMultiScan

LiverMultiScan (Perspectum Ltd) is a non-invasive multiparametric MRI-based imaging software application that provides quantitative analysis of liver fat content, liver iron concentration and fibro-inflammation from non-contrast MRI images. The topic selection oversight panel identified LiverMultiScan software as potentially suitable for evaluation by the Diagnostics Assessment Programme (DAP) based on a MedTech Innovation Briefing32 published by NICE and further information provided by the manufacturer. 24

LiverMultiScan software enables assessment of liver fat content from PDFF, liver iron concentration from T2* mappings and fibro-inflammation from T1 mappings. The T1 analyses for fibro-inflammation are adjusted for iron level to remove artefacts and increase accuracy. 33 This output is referred to as the cT1 score. PDFF is an estimate of the percentage of fat within the liver tissue and is calculated from the ratio of fat versus fat and water in MRI images. PDFF can be computed using the IDEAL (Iterative Decomposition of water and fat with Echo Asymmetry and Least squares estimation) or three-point Dixon method.

LiverMultiScan protocols can be integrated into existing abdominal MRI protocols on Siemens, Philips or GE Healthcare scanners and do not require any contrast agent or additional hardware in addition to the MRI scanner. 24 A 15 minute scan acquisition time is typically required to obtain the MR images for analysis by LiverMultiScan software. 24 Training on how to use the LiverMultiScan protocol takes approximately 3 hours. 24 Technical support from imaging application specialists at Perspectum Ltd is provided by the manufacturer as part of the licence. 34 The imaging data from the MRI scan are sent to Perspectum Ltd via an Amazon-hosted cloud service and are analysed by Perspectum Ltd trained operators. 35 The quantitative analysis is returned to clinicians electronically in report format as a PDF document. 35

Perspectum Ltd suggested to NICE24 that the normal reference range for MRI PDFF is less than 5.6% liver fat content and that the diagnosis indicated by the cT1 output and the clinical recommendations are as follows:

• <800 ms: fatty liver

  • no inflammation present

  • reassess with MRI in 3 years

• 800–875 ms: NASH

  • recommend lifestyle modification

  • manage type 2 diabetes and cardiovascular disease

  • monitor disease status with MRI after 6 months

• >875 ms: high-risk NASH

  • reassess with MRI every 6 months

  • consider liver biopsy if cirrhosis is suspected

  • cancer surveillance

  • consider inclusion in NASH therapeutic trials.

Perspectum Ltd does not propose that LiverMultiScan is suitable for staging fibrosis but considers that LiverMultiScan can stage NAFLD and distinguish between patients with NASH and high-risk NASH. 24 However, in the EASL guidelines18 liver biopsy is recommended as the reference standard for the diagnosis of NASH for patients with NAFLD.

Magnetic resonance elastography

MRE is a non-invasive MRI-based technique that uses a mechanical driver to generate shear waves across the liver during an MRI scan. 36 An MRI sequence with motion-encoding gradients measures the propagation of the shear waves across the liver to produce an image (elastogram) showing the distribution of liver stiffness. 36 MRE requires additional hardware to an MRI scanner, including an active acoustic driver, a passive pneumatic driver and a connector. 37 MRE can be used alongside standardised MRI PDFF and iron-assessment packages offered by scanner manufacturers, such as Siemens, Philips or GE Healthcare scanners, to assess fat and iron. 38

The MRE acquisition is performed during breath-holding and takes 12–15 seconds, and is typically repeated four times. 24 The total acquisition time can last approximately 1 minute. 24 Inadequate breath-holding can produce image artefacts which can affect diagnostic accuracy. 37

NICE guidelines (NG499 and NG5039) do not consider the routine use of MRE for diagnosing NAFLD or liver fibrosis or cirrhosis. However, MRE is used in some NHS centres where it is available, when other diagnostic tests have returned indeterminate results.

The commercially available Resoundant, Inc. MRE platform measures the magnitude of the complex shear modulus of propagating waves to provide liver stiffness outputs (kPa). 40 The complex shear modulus is composed of two components, the storage modulus, which describes tissue elasticity, and the loss modulus, which describes tissue viscosity and the ability to absorb energy. 41 The company, Resoundant, Inc., has suggested to NICE24 that MRE liver stiffness outputs (kPa) can be used to stage liver fibrosis as follows:

• >2.9 kPa: any fibrosis

• >3.3 kPa: significant fibrosis

• >3.9 kPa: advanced fibrosis

• >4.8 kPa: cirrhosis.

Liver biopsy

Liver biopsy, an invasive procedure, is considered the gold standard for staging liver fibrosis, inflammation and steatosis, and for diagnosing NASH. 9 During liver biopsy, a small sample of tissue is percutaneously or transvenously removed from the liver using a needle. 42 However, liver biopsies are associated with inter- and intra-observer variability and sampling error. 43,44 Liver biopsies are expensive because patients require outpatient care, specialists (a gastroenterologist, hepatologist or radiologist) are needed to carry out the biopsy, pathologists are needed to examine and report the biopsy results and clinicians are required to interpret biopsy results and recommend clinical management for patients. 9 Liver biopsies can be painful and are associated with a high risk of complications, including bleeding from the biopsy site (0.3–10.9%) and major intraperitoneal bleeding (0.1–4.6%). 42

In NG50,39 it is recommended that clinicians should consider a liver biopsy to diagnose cirrhosis in patients for whom TE is not suitable. In NG49,9 it is stated that a liver biopsy should not be used to diagnose NAFLD or for monitoring disease progression, and that biopsies should be avoided in children and young people unless there is an unclear diagnosis or concern about rapid disease progression.

Clinical advice to NICE24 is that in some NHS centres, liver biopsy is carried out in a large proportion of patients with suspected significant or advanced fibrosis to either confirm the suspected diagnosis or to obtain a diagnosis to allow entry into clinical trials. Clinical advice to the ERG is that liver biopsy results provide information that can be used to inform treatment decisions and clinical management.

Clinical advice to the EAG is that, even after an MRI assessment, patients would be referred for biopsy if the following diagnoses were suspected:

• advanced fibrosis (≥F3)

• steatosis with Brunt grade ≥ 2

• advanced NASH (NAS ≥ 4 and ≥F3)

• high risk of progressive disease (NASH or >F1).

Clinicians do not always refer patients for liver biopsy if they suspect the patient has cirrhosis. Reasons for not referring a patient for a liver biopsy include old age, significant co-morbidities, and being contraindicated for biopsy (e.g. patients with extrahepatic biliary obstruction or bacterial cholangitis). 42 Clinical advice to the EAG is that some patients (5–10%) do not wish to proceed with liver biopsy, or are treated at centres without access to liver biopsy.

Additional searches (clinical impact review)

Where clinical impact outcome data relating specifically to MRI-based technologies were not identified by the initial search strategy, broader searches were carried out to consider studies of NAFLD populations irrespective of whether MRI-based technologies had been used. MEDLINE and Epub Ahead of Print, In-Process & Other Non-Indexed Citations (via Ovid) were searched, and details of the additional searches are provided in Appendix 2.

Statistical analysis and data synthesis

Subgroup analyses and sensitivity analyses

No subgroup analyses or sensitivity analyses were performed by the EAG (see Appendix 3 for further details).

Studies identified by the manufacturers

The test manufacturers’ evidence submissions included details of studies that were potentially relevant, and should be considered, for inclusion in the EAG review. All the studies suggested by the manufacturers had already been identified by the EAG searches. The studies identified by the manufacturers that were not included in the EAG review are listed in Supplementary material 1 with reasons for exclusion.

Quality assessment

The included studies that provided DTA30,53–64 data were assessed for risk of bias using the QUADAS-2 tool. 49 A summary of the results of the assessment using the QUADAS-2 tool is presented in Table 3. The EAG’s full assessment is presented in Supplementary material 2.

Characteristics of the included studies

The characteristics of the 13 studies30,53–64 included in the DTA review are presented in Table 4.

In line with the final scope24 issued by NICE, all the studies30,53–64 included patients with NAFLD for whom advanced fibrosis or cirrhosis had not yet been diagnosed. However, only the Eddowes study30 recruited patients who were scheduled for non-targeted liver biopsy to (i) stage fibrosis after inconclusive non-invasive assessment of fibrosis or (ii) make a diagnosis after a range of non-invasive tests had not confirmed a diagnosis. The EAG considers that the Eddowes study30 population provides evidence for the population of patients who have indeterminate or discordant results from fibrosis testing. However, it is unclear whether the term ‘inconclusive’ means indeterminate and/or discordant. The EAG notes that the patients in the study30 were scheduled for a biopsy and therefore may not represent all patients with indeterminate and/or discordant results from previous fibrosis testing; clinical advice to the EAG is that not all patients with indeterminate and/or discordant results will have a biopsy.

Two studies30,59 assessed the DTA of LiverMultiScan, ten studies53–55,57,58,60–64 assessed the DTA of MRE and one study56 assessed the DTA of LiverMultiScan and MRE. The two studies30,59 that assessed the DTA of LiverMultiScan were based in the UK, whereas the ten studies53–55,57,58,60–64 that assessed the DTA of MRE were based in Holland,62 Japan,60,61 South Korea,58 Sweden54 and the USA. 53,55,57,63,64 The study56 that assessed the DTA of LiverMultiScan and MRE was based in Japan. Four of the studies53,57–59 reported that they were conducted in tertiary care. The EAG notes that all of the included studies were conducted in hospitals and therefore considers it likely that all studies were conducted in either secondary or tertiary care settings.

According to the corresponding author, the Pavlides 201759 study population included the Banerjee 201465 study population and therefore the EAG does not regard the studies as two independent data sets (Michael Pavlides, University of Oxford, 26 November 2021, personal communication).

Six of the included studies54,55,60,61,63,64 considered patients with liver disease aetiologies other than NAFLD and did not report or provide data upon request specifically for the subpopulation of patients with NAFLD for whom advanced fibrosis or cirrhosis had not been diagnosed. Three of the included studies30,53,57 exclusively considered patients with NAFLD for whom advanced fibrosis or cirrhosis had not been diagnosed. However, one of the studies53 did not report any outcomes of interest and did not provide additional data upon request. For the remaining studies,56,58,59,62 the EAG obtained data for patients with NAFLD for whom advanced fibrosis or cirrhosis had not been diagnosed (Table 5). As a result, the EAG quantitative synthesis includes data from only six of the identified studies. 30,56–59,62

Diagnostic test accuracy results

The absolute numbers of true positive (TP), false positive (FP), true negative (TN) and false negative (FN) LiverMultiScan or MRE test results compared to the reference standard of liver biopsy (i.e. 2 × 2 data) were not presented in any of the included studies. We contacted the authors of all included studies to request these data.

Perspectum Ltd provided 2 × 2 data in response to the EAG request for information for the three LiverMultiScan studies30,56,59 included in the DTA review. The authors of the Troelstra 202162 study of MRE provided 2 × 2 data in response to the EAG request. Data from the Kim 202058 study were obtained from a systematic review, and 2 × 2 data from the Kim 201357 study were calculated using the number of patients with and without the diagnosis of interest, and the estimates of sensitivity and specificity reported in the published paper. The full set of data sources is provided in Table 5.

The EAG’s quantitative synthesis therefore included data from six30,56–59,62 (out of 13) identified studies for which 2 × 2 data were available.

Where studies reported 2 × 2 data (i.e. the number of TP, FP, TN and FN test results), data from individual studies were summarised in forest plots (Figures 3–6) alongside estimates of sensitivity and specificity. The individual study results were grouped by diagnosis of interest, and the cut-off value used to indicate a positive result from the index test was also provided.

FIGURE 3.

Forest plot displaying 2 × 2 data, sensitivity and specificity for LiverMultiScan PDFF from the included studies. Source: see Table 5.

FIGURE 4.

Forest plot displaying 2 × 2 data, sensitivity and specificity for LiverMultiScan cT1 from the included studies. ^*Data for NASH was available from the Imajo 202156 study for two cut-off values, 800 ms and 875 ms. All other studies reported data for the 875 ms cut-off value only. cT1 = iron corrected longitudinal relaxation time. Source: see Table 5.

FIGURE 5.

Forest plot displaying 2 × 2 data, sensitivity and specificity for LiverMultiScan PDFF and cT1 combined from the included studies. cT1 = iron corrected longitudinal relaxation time. Source: see Table 5.

FIGURE 6.

Forest plot displaying 2 × 2 data, sensitivity and specificity for MRE from the included studies. ^*NASH was defined in the Imajo 202156 study as NAS ≥4 with ≥1 hepatocyte ballooning and ≥1 lobular inflammation, and in the Troelstra 202162 study as ≥1 steatosis, ≥1 hepatocyte ballooning and ≥1 lobular inflammation. Source: see Table 5.

Where studies reported area under the receiver operating characteristic (AUROC) curve results, these results are summarised in Appendix 4 (Tables 21 and 22).

Results from External Assessment Group meta-analyses: magnetic resonance elastography

For MRE, there was only one diagnosis (fibrosis ≥F3) where at least three studies56–58 (224 participants) provided DTA data. For this diagnosis, data were available from the Troelstra 202162 study (MRE G’ shear modulus and MRE G’ loss modulus), the Kim 201357 study (complex shear modulus) and the Kim 202058 study (complex shear modulus). The EAG considered it appropriate to include data from the Troelstra 202162 study for the MRE G’ shear modulus rather than for the MRE G’ loss modulus in the meta-analysis; clinical advice to the EAG was that the MRE G’ shear modulus results were directly comparable with the MRE complex shear modulus results. It would not have been possible to include data for both moduli from the Troelstra 202162 study in a meta-analysis as both data sets represented the same group of patients.

As cut-off values varied between the three studies56–58 that reported data for this diagnosis, a summary ROC curve was estimated (Figure 7).

FIGURE 7.

Summary ROC plot for fibrosis (≥F3) data from the MRE test. The solid line is the summary ROC curve. The dashed line indicates sensitivity = 1 − specificity (i.e. an uninformative test). The circles represent individual study results. The EAG notes that the Troelstra 202162 study used an investigational MRE design and not the Resoundant, Inc. MRE platform that is commercially available and was used in the Kim 201357 and Kim 202058 studies.

The summary ROC curve demonstrates how sensitivity and specificity values change as cut-off values vary between the three included studies. 57,58,62 The closer the summary ROC curve is to the top left-hand corner in ROC space (where sensitivity and specificity both equal 100%), the greater the discriminatory power of the test. The summary ROC curve for an uninformative test would be the upward diagonal of the summary ROC plot (the dashed line). The summary ROC curve in Figure 7 therefore indicates high DTA. It is also important to note that the observed study results do not all lie close to the summary ROC curve; this may be due to the fact that small studies are likely to estimate values for test accuracy that are further away from the true test accuracy values than larger studies (i.e. statistical error). Two of the included studies had small sample sizes (n = 35 in the Troelstra 202162 study and n = 47 in the Kim 202058 study). Clinical and/or methodological heterogeneity between the included studies57,58,62 may also explain the fact that observed study results do not all lie close to the summary ROC curve. For example, the EAG notes that the Troelstra 202162 study used an investigational MRE design and not the Resoundant, Inc. MRE platform that is commercially available and was used in the Kim 201357 and Kim 202058 studies. Furthermore, the studies were conducted in different countries (Kim 2013,57 USA; Kim 2020,58 South Korea; Troelstra 2021,62 Holland). These differences may have introduced heterogeneity to the analysis.

Quality assessment

Seven30,53,54,57,59,62,64 of the 1330,53–64 DTA studies were also included in the clinical impact review. The EAG reassessed the methodological quality of the seven DTA studies30,53,54,57,59,62,64 using the NIH study quality-assessment tool. 51 Of the remaining four studies included in the clinical impact review, two were cohort studies,66,67 one was an RCT described in two publications68,74 and one was a qualitative study. 69 Full assessments using the NIH study quality-assessment tool51 for the seven DTA studies30,53,54,57,59,62,64 and the two cohort studies,66,67 the full assessment and summary of the risk of bias assessment for the included RCT68,74 and the full assessment using the CASP qualitative-studies checklist52 for the included qualitative study69 are presented in Supplementary material 2.

Characteristics of the included studies

Only one study30 provided clinical impact results for a population of patients with NAFLD who had indeterminate or discordant results from fibrosis testing. Seven studies30,53,54,57,59,62,64 that were included in the DTA review also provided evidence describing the clinical impact of MRI-based technologies for the assessment of patients with NAFLD. The characteristics of the original seven studies30,53,54,57,59,62,64 are presented in Table 4. In addition to these seven studies,30,53,54,57,59,62,64 the EAG identified four new studies. 66–69 Three studies described LiverMultiScan66,67,69 and one study described MRE. 68 These comprised one prospective cohort study66 based in the UK, one retrospective cohort study67 based in the USA, one RCT68,74 based in Germany, Netherlands, Portugal and the UK and one qualitative study69 based in the UK. The RCT68,74 (RADIcAL trial) was a phase IV, multicentre, international study that evaluated the impact of using LiverMultiScan in the diagnostic pathway compared to standard of care (SoC) for patients with suspected NAFLD and was sponsored by Perspectum Ltd. Information about the RADIcAL trial68,74 is presented in Table 7. The characteristics of the four new studies66–68,69 are presented in Table 8.

Intermediate outcomes

Test failure rate

Three studies30,59,66 reported test failure rate for LiverMultiScan and six studies53,54,57,62,64,67 reported test failure rate for MRE. However, two of the studies59,66 that assessed LiverMultiScan and three of the studies54,64,67 that assessed MRE included patients with other liver disease aetiologies in addition to NAFLD and did not provide data specific to patients with NAFLD.

The test failure rate of LiverMultiScan for patients with all liver aetiologies ranged from 5.3%59 to 7.6%66 and the test failure rate of LiverMultiScan for patients with NAFLD only was 5.6%. 30 The reasons for LiverMultiScan test failure specific to patients with NAFLD were technical failure (n = 1/3), MRI scan cancelled (n = 1/3) and patient unable to tolerate MRI scan (n = 1/3). 30

The MRE test failure rate for patients with all liver aetiologies ranged from 0.0%54 to 7.6%53 and the MRE test failure rate for patients with NAFLD only ranged from 3.9%57 to 7.6%. 53 The EAG performed a fixed-effects meta-analysis to obtain a pooled estimate of test failure rate for patients with NAFLD (test failure rate = 4.2%, 95% CI 2.5% to 6.2%); a forest plot displaying this analysis is provided in Appendix 6 (Figure 13). Minimal statistical heterogeneity was observed between the included studies (I² = 18.9%). The reasons for MRE test failure specific to patients with NAFLD were technical failures (n = 11/24),57,62 patients refusing the test (n = 9/24),53,57 claustrophobia (n = 3/24)53 and the patient being unable to fit in the scanner (n = 1/24). 53

FIGURE 8.

PRISMA flow diagram for the cost-effectiveness review.

FIGURE 9.

Current NHS diagnostic test pathway for patients with inconclusive results from previous fibrosis testing.

FIGURE 10.

Proposed LiverMultiScan plus biopsy diagnostic test pathway for patients with inconclusive tests from fibrosis testing. LMS, LiverMultiScan

External Assessment Group diagnostic test accuracy and clinical impact review

The EAG DTA review identified 13 studies30,53–64 reported in 15 publications. 30,31,53–65 The EAG clinical impact review identified 11 studies30,53,54,57,59,62,64,66–69 reported in 14 publications. 30,31,33,53,54,57,59,62,64–69 However, the EAG was only confident that one study (the Eddowes 201830 study) was carried out in the population described in the final scope24 issued by NICE, namely patients with NAFLD for whom advanced fibrosis or cirrhosis had not been diagnosed:

• patients who have indeterminate results from fibrosis testing

• patients for whom TE or ARFI is unsuitable

• patients who have discordant results from fibrosis testing.

The clinical impact review only identified one RCT: the RADIcAL trial,68 which was carried out by Perspectum Ltd. Results from this study68 showed that, compared with patients in the standard-care arm, fewer patients with non-NAFLD, fewer with NAFLD and fewer patients with no-mild fibrosis (F0 to F1) underwent unnecessary biopsies in the LiverMultiScan arm. Feedback from Perspectum Ltd71 and the McKay study69 was that patients’ and carers’ experiences of using LiverMultiScan were positive.

External Assessment Group quantitative analysis

The only relevant study30 (n = 50) identified by the DTA review focused on the potential of LiverMultiScan to deliver cost savings compared to biopsy and included clinical results (for example, cT1 and PDFF scores). The Eddowes study30 categorised patients according to low or high risk of progressive liver disease. However, it was also possible to interpret the DTA data71 generated by LiverMultiScan as follows: any fibrosis (≥F1), significant fibrosis (≥F2), Brunt grade ≥1, Brunt grade ≥2, NASH and advanced NASH. In response to a request from the EAG, Perspectum Ltd71 also provided data for patients with advanced fibrosis (≥F3).

No DTA data were submitted to NICE by the manufacturer of MRE (Resoundant, Inc.). Eleven studies53–58,60–64 evaluated the DTA of MRE, but none of the studies explicitly included patients with indeterminate or discordant results from previous fibrosis testing.

The EAG carried out a quantitative analysis using data from six studies. 30,56–59,62 Where patients were diagnosed consistently across studies (fibrosis, steatosis and NASH), the EAG carried out meta-analyses using cT1 and PDFF outputs for LiverMultiScan and for MRE. Results from the EAG meta-analyses suggested that the LiverMultiScan cT1 output is more sensitive and specific than the LiverMultiScan PDFF output, and that for the diagnosis of fibrosis (≥F3), MRE has high DTA. However, the meta-analyses were populated with data from small numbers of studies and only one30 of the studies included the population that is the focus of this assessment. This should be considered when interpreting the results from the EAG meta-analyses.

Search strategy

The search strategies used to identify diagnostic and clinical impact evidence for inclusion in the clinical effectiveness review can be found in Appendix 1. To identify published economic evaluations, the EAG appended an economic evaluation-specific search filter to the clinical search strategies (Appendix 7). In addition, two databases of economic publications [EconLit (EBSCO) and the CEA registry] were searched, using the search strategies presented in Appendix 7, from inception until 4 October 2021. The results of the searches were entered into an Endnote X9 library and de-duplicated (MM) before being exported into Covidence.

Study selection and inclusion criteria

The review inclusion and exclusion criteria (Table 9) reflected the decision problem outlined in the final scope24 issued by NICE.

The identified publications were assessed for inclusion in the review using a two-stage process. First, two reviewers (DB and RH) independently screened all the titles and abstracts identified by the electronic searches to find potentially relevant records. Second, full-text copies of these records were obtained and assessed independently by two reviewers (DB and RH) using the inclusion criteria presented in Table 9. Disagreements were resolved through discussion at each stage and, in all cases, a consensus was reached.

Data extraction

A data-extraction form was designed in Microsoft Excel. Extracted data included bibliographic information (author[s] and year of publication), type of economic evaluation, country, perspective, population, intervention and comparators, model structure, model outcomes, and sensitivity analyses undertaken. Data extraction was carried out independently by two reviewers (DB and RH) and the two reviewers agreed the final version of the completed data extraction form.

Quality of cost-effectiveness evidence

The EAG assessed the quality of the included economic evaluations using the Drummond checklist76 for assessing economic evaluations and the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) checklist. 77 Quality assessment was performed by one reviewer (DB) and checked for accuracy by a second reviewer (RH). All disagreements were resolved through discussion. There were no unresolved issues and, therefore, it was not necessary to consult with a third reviewer.

Results of the systematic review of existing cost-effectiveness evidence

The searches resulted in the identification of 253 publications. Once duplicates (n = 49) had been removed, 204 publications remained. Following first-stage screening (titles and abstracts), 31 publications were retrieved for full-text review. After assessing applying inclusion criteria, one publication30 was identified as being relevant. The PRISMA flow diagram48 provides an illustration of the screening and selection process (Figure 8). A list of the studies excluded at the full-text stage, along with reasons for exclusion, is provided in Supplementary material 1.

Quality of the included evidence

The quality of the included study30 was assessed using the Drummond checklist76 (Table 10) and the CHEERS checklist77 (see Supplementary material 2).

The population (n = 50) described in the published paper is patients with inconclusive results from fibrosis testing. The EAG has assumed that inconclusive is an umbrella term for a group of patients with indeterminate and/or discordant results from previous fibrosis testing. The EAG notes that all patients considered in this analysis were scheduled for a biopsy. This means that the study sample does not represent all patients with indeterminate and/or discordant results from previous fibrosis testing; clinical advice to the EAG is that not all patients with indeterminate and/or discordant results will have a biopsy.

Eddowes 201830 repeated the analyses carried out by Blake 201678 using DTA results from their study. Blake 201678 constructed a simple decision tree to compare the costs for three NAFLD diagnostic pathways that use non-invasive techniques. The patients modelled by Blake 201678 did not have inconclusive results from previous fibrosis testing and therefore the Eddowes 201830 cost-saving results are not relevant to this appraisal.

The information provided in the published paper30 is limited and, therefore, it is unclear whether all important costs and consequences were included in the analysis, or whether the included costs and consequences were valued credibly. An incremental analysis was not performed and there is no evidence that any sensitivity or scenario analyses were performed. The authors did not describe the limitations of the CEA, nor the generalisability of results.

Characteristics of the included study

The characteristics of the included study30 are summarised in Table 4. This study30 was also included in the EAG DTA and clinical impact review.

The included study, Eddowes 2018,30 reported results from a cost-utility analysis. The population was adult patients with NAFLD for whom advanced fibrosis or cirrhosis had not yet been diagnosed and who were scheduled for non-targeted liver biopsy to stage fibrosis after inconclusive non-invasive assessment of fibrosis or to make a diagnosis after a range of non-invasive tests had not confirmed a diagnosis. Three diagnostic tools were considered: LiverMultiScan (two cut-offs: 822 ms and 875 ms), TE (two cut-offs: 5.8 kPa and 7.0 kPa), ELF (two cut-offs: 7.7 and 9.8); LiverMultiScan plus TE (four combinations of cut-offs) was also considered. The perspective of the analysis was the UK NHS, and the time horizon was 2 weeks (i.e. LiverMultiScan and TE were performed within 2 weeks of biopsy).

Results were generated by a decision-tree model. The model was populated with clinical effectiveness evidence from a cross-sectional study undertaken at the Queen Elizabeth Hospital Birmingham and the Royal Infirmary of Edinburgh (ISRCTN39463479). Costs were sourced from the NHS tariff and the cost year was 2016. The short time horizon of the model meant that it was not necessary to discount costs and benefits.

Study results and conclusions

External Assessment Group cost-effectiveness review conclusions

The EAG searches for published economic evaluations that assessed the cost-effectiveness of LiverMultiScan and MRE only identified one study. 30 The included study30 assessed the comparative cost savings versus biopsy of LiverMultiScan, TE, ELF and LiverMultiScan plus TE. The authors provided limited data describing the study methods and results and, therefore, study quality and the generalisability of results are unclear.

In the Eddowes 201830 study, clinical effectiveness evidence was collected from a population with inconclusive results from previous fibrosis testing. To generate cost-effectiveness results, Eddowes 201830 study clinical effectiveness data were used to populate the Blake 201678 model. However, the focus of the Blake 201678 model was not to explore cost-effectiveness for patients with inconclusive results from previous fibrosis testing. Therefore, Eddowes 201830 study cost-effectiveness results are not relevant to this appraisal.

Introduction

The EAG cost-effectiveness review did not identify any published economic evaluations that were relevant to this appraisal; the EAG has therefore developed a de novo economic model.

Perspectum Ltd suggest24 that LiverMultiScan results can be used by clinicians to help diagnose patients with fatty liver, NASH and high-risk NASH. Perspectum Ltd71 also provided LiverMultiScan DTA results for a range of other diagnoses, including advanced fibrosis (≥F3). Whilst LiverMultiScan results are unlikely to inform patient treatment plans, they can potentially be used to help identify patients for whom a biopsy may not be appropriate. In contrast, biopsy results provide an accurate diagnosis and data that can be used to inform patient treatment plans, for example, identification of co-factors for liver injury (such as alcohol, iron, or auto-immune hepatitis). However, biopsy is an expensive invasive procedure that is not without risks. If LiverMultiScan results could be used to help identify patients who do not require a biopsy, this would benefit patients by reducing the number of unnecessary biopsies and would save NHS resources. The primary clinical outcome from the EAG model is therefore the number of biopsies avoided if LiverMultiScan were introduced into the diagnostic pathway.

The EAG cost-effectiveness results will be driven by the proportion of patients who, if they had a biopsy, would test positive: that is, population prevalence. This estimate is independent of LiverMultiScan test accuracy (or the accuracy of any other test introduced into the diagnostic pathway). Population prevalence estimates vary depending on two factors, the diagnosis and the population investigated. Published evidence56,58 shows that population prevalence varies by population investigated; it is essential that the prevalence data used to populate the EAG model relate to the population described in the final scope24 issued by NICE and are generalisable to patients treated in NHS clinical practice.

The EAG only identified one study (Eddowes 201830) that provided LiverMultiScan DTA and population prevalence data that were focused on patients who were scheduled for, and received, a biopsy, and who had inconclusive results from previous fibrosis testing.

As DTA data are only available for patients with inconclusive results who received a biopsy, the Eddowes 201830 study population represents a subset of the population described in the final scope24 issued by NICE. Clinical advice to the EAG is that not all patients with inconclusive results from previous fibrosis testing would be referred for a biopsy; reasons for not referring a patient for a biopsy include presence of co-morbidities, personal choice, old age and medical contraindications. The utility of positive LiverMultiScan results for patients who would not be referred for biopsy is unclear and is not considered in the EAG model. No DTA or population prevalence data are available for the full population described in the final scope24 issued by NICE.

Further, LiverMultiScan data are not available for patients for whom TE or ARFI was unsuitable. In addition, no DTA or population prevalence data were available for any of the population described in the final scope24 issued by NICE for patients who had had an MRE.

The EAG cautions that the data presented in the Eddowes 201830 study relate to 50 patients and the data presented by Perspectum Ltd71 relate to 46 patients; however, both sets of data appear to be from the same group of patients, that is, as described in the Eddowes 201830 publication, and are referred to as Eddowes 2018/Perspectum Ltd. 30,71

The EAG model has been developed based on the assumption that the LiverMultiScan DTA results are robust and will be used to stop clinicians from sending patients with a negative result for a biopsy. However, if this assumption does not hold then results from the EAG model should not be used to inform decision-making.

Model structure

The EAG built a decision tree in Microsoft Excel® to estimate the costs and quality-adjusted life years (QALYs) associated with two diagnostic pathways, LiverMultiScan plus biopsy and liver biopsy only. Eight different diagnostic test strategies described in the literature or by Perspectum Ltd71 were investigated. Eddowes 201830 chose to categorise patients according to low or high risk of progressive liver disease; however, Perspectum Ltd71 has provided data for seven other ways of interpreting the DTA data generated by LiverMultiScan (from the same study). The eight different diagnostic test strategies considered by the EAG were:

• T1: Any fibrosis (≥F1)

• T2: Significant fibrosis (≥F2)

• T3: Advanced fibrosis (≥F3)

• T4: Brunt grade ≥1

• T5: Brunt grade ≥2

• T6: NASH (NAS ≥4, ≥1 for lobular inflammation and hepatocyte ballooning)

• T7: Advanced NASH (NAS ≥4 plus ≥F2)

• T8: High risk of progressive disease (NASH or >F1).

In the EAG model, for each of the eight diagnostic test strategies (T1 to T8), if a patient’s LiverMultiScan result exceeds the specific cT1 or PDFF thresholds associated with the test strategy, then the patient is defined as having a positive result and will have a biopsy. The EAG asked the Specialist Committee members to consider the eight diagnostic test strategies and identify any strategies for which a positive LiverMultiScan result would not change their decision to send a patient for a biopsy. The advice from the Specialist Committee was that, for patients with LiverMultiScan test results suggesting a diagnosis of T3, T5, T7 and T8, the decision whether to send the patient for a biopsy would not change: that is, patients who had a positive LiverMultiScan test result would proceed to biopsy. The EAG has presented results for all strategies but considers that the findings from the strategies in bold are the most important.

In the model, LiverMultiScan cT1 or PDFF results lead to the following consequences:

• true positive (TP); LiverMultiScan result and biopsy result are both positive – correctly identified by LiverMultiScan results and patient is appropriately sent for a biopsy

• false positive (FP); LiverMultiScan result positive and biopsy result negative – incorrectly identified by LiverMultiScan results and patient is inappropriately sent for a biopsy

• true negative (TN); LiverMultiScan result negative and biopsy, if performed, would have been negative – correctly identified by LiverMultiScan results and the patient was appropriately not sent for a biopsy, LiverMultiScan repeated at 6 months, result is negative and no biopsy required

• false negative (FN); LiverMultiScan result negative but biopsy, if performed, would have been positive – incorrectly identified by LiverMultiScan results and patient inappropriately not sent for a biopsy, LiverMultiScan repeated at 6 months, biopsy following repeat LiverMultiScan (assumed always to be positive)

• test failure – patients go to straight to biopsy.

The accuracy of liver biopsy does not influence EAG cost-effectiveness results; the model is driven by the congruence of the LiverMultiScan and biopsy results and not by the diagnoses reached following a biopsy.

The assumption that all patients with a negative LiverMultiScan test result will go on to have a repeat LiverMultiScan at 6 months and will then be correctly diagnosed is optimistic and favours the LiverMultiScan plus biopsy pathway for two reasons. First, it seems implausible that the accuracy of a second LiverMultiScan test will be 100%; some patients are likely to have a second FN result and some patients with an initial TN result will have a FP result and will go straight to (an unnecessary) biopsy. Second, the EAG has assumed that patients whose second LiverMultiScan test results are negative will have no further tests as this result is assumed to be a TN.

The population prevalence can be estimated by adding together the number of patients with TP and FN results.

Perspectum Ltd71 has suggested that patients will receive a second LiverMultiScan if their cT1 score is between 800 and 875 ms; however, the EAG has assumed that patients with cT1 scores less than 800 ms will also receive a second LiverMultiScan. The EAG considers that this assumption is appropriate as all tests for this cohort have low specificity (i.e. high rates of FNs).

As all patients are assumed to be correctly diagnosed by 6 months, the LiverMultiScan plus biopsy pathway benefits arise from identifying people who are TNs and removing the costs and lost QALYs arising from these patients having unnecessary biopsies. These benefits are balanced against the LiverMultiScan plus biopsy pathway costs and the QALY loss associated with FNs.

Currently, NHS patients with inconclusive results from previous fibrosis testing may be sent for a biopsy or receive no further diagnostic tests (Figure 9); the proposed LiverMultiScan plus biopsy pathway is shown in Figure 10.

Population

The modelled population is patients with inconclusive results from fibrosis testing who, without access to LiverMultiScan, would be scheduled for and would receive a biopsy. Patient characteristics are based on the population described in the Eddowes 201830 study. All patients (n = 50) had a histologically confirmed diagnosis of NAFLD without secondary causes and without history of alcohol excess; 32 patients had an inconclusive non-invasive assessment of fibrosis and 18 patients had undergone a range of non-invasive tests without a firm diagnosis being made. Over half of the patients were male (56%), their average age was 54 years, 86% were Caucasian, 58% were non-smokers and 10% of patients in the study were post-transplant.

Intervention

For patients with inconclusive results from fibrosis testing who were scheduled for, and received, a biopsy, DTA data and population prevalence data were only available from a population of patients who had received a LiverMultiScan. 30 No MRE DTA data and population prevalence data were available for the population described in the final scope24 issued by NICE.

Cut-off values have been proposed by Perspectum Ltd71 for the staging of fibro-inflammation, associated diagnoses, and clinical management options. The normal reference range for PDFF is ≤5.6% liver fat content. The proposed cT1 cut-off values are:

• Less than 800 ms: ‘Fatty liver’

  • Reassure as no inflammation present

  • Reassess with MRI in 3 years

• 800–875 ms: ‘Non-alcoholic steatohepatitis (NASH)’

  • Lifestyle modification

  • Management of type 2 diabetes and cardiovascular disease

  • Monitor disease status with MRI after 6 months

• More than 875 ms: ‘High-risk NASH’

  • Reassess with MRI every 6 months

  • Consider liver biopsy if cirrhosis is suspected

  • Cancer surveillance

  • Consider inclusion in NASH therapeutic trials.

When compared with the PDFF values from the same cohort of patients and using the same diagnostic test strategies, the cT1 scores always generated the same or higher sensitivity and specificity values. The EAG CEA is, therefore, populated with LiverMultiScan cT1 scores. For completeness, the cost-effectiveness results generated using PDFF values are presented in Appendix 8 (Tables 23–27).

Comparator

The comparator is liver biopsy only which represents current standard of care.

Time horizon, discounting and perspective

The model has a maximum time horizon of 6 months and ends when a patient has a biopsy or has been accurately diagnosed following a repeat LiverMultiScan test. The short model time horizon means that discounting of costs and benefits is not relevant. The cost perspective of the model is the NHS. For patients in the LiverMultiScan plus biopsy pathway, only the costs and outcomes associated with the LiverMultiScan test and biopsy are considered. For patients in the biopsy only pathway, only the costs and outcomes associated with biopsy are considered.

External Assessment Group model parameters

Intervention and comparator costs

Unless otherwise stated, the intervention costs are presented in 2019/20 GBP. The costs prior to receiving a LiverMultiScan or biopsy, whichever test comes first in the pathway, are not included in the EAG analysis. Intervention costs are displayed in Table 13.

Biopsy complications

The Stevenson study80 estimated the average costs (per biopsy) of treating complications associated with a percutaneous biopsy and a transjugular biopsy to be £7 and £13 respectively. An EAG targeted literature search failed to identify more robust estimates. The EAG weighted the Stevenson study80 costs by the proportions of patients (NHS Reference Costs 2019/2079) who had percutaneous and transjugular biopsies (£7.30) and inflated the weighted cost to 2019/20 prices (£8.54) using the NHS Cost Inflation Index (pay and prices index).

Utility values

The only utility values required in the EAG model are the disutilities associated with having a biopsy. The EAG carried out a targeted search of the literature; however, the EAG did not identify any primary studies that reported disutility values specifically associated with liver biopsy for patients with inconclusive results from fibrosis testing. There is no information in NG49,9 the NICE guideline for the assessment and management of NAFLD, about the disutility associated with having a biopsy. However, the Stevenson study80 identified that a loss of utility due to biopsy can be caused by direct pain and anxiety, serious adverse events and death. The EAG also considers that loss of utility can arise from failure to treat patients with advanced liver disease (i.e. LiverMultiScan test FN results).

Summary of base-case assumptions

Parameter assumptions and sources used in the base-case model are summarised in Table 14.

Uncertainty

Uncertainty around parameter values and the impact this could have on cost-effectiveness results has been explored by the EAG by running threshold and scenario analyses.

The EAG undertook three threshold analyses:

• LiverMultiScan test results were assumed to be 100% accurate. For each of the diagnostic test strategies, the proportion of patients who would test positive using the reference standard (biopsy) was varied until the LiverMultiScan plus biopsy pathway versus biopsy pathway only was cost-effective at a threshold of £20,000 (£30,000) per QALY gained.

• For each of the eight diagnostic test strategies, the QALY loss associated with liver biopsy threshold analysis was varied until the LiverMultiScan plus biopsy pathway versus biopsy pathway only was cost-effective at a threshold of £20,000 (£30,000) per QALY gained.

• For each of the eight diagnostic test strategies, the cost at which LiverMultiScan was cost-effective at a threshold of £20,000 (£30,000) per QALY gained was estimated.

The EAG also carried out scenario analyses, for all eight diagnostic test strategies, in which the effects of LiverMultiScan failure rates of 0% and 10% were explored.

External Assessment Group base-case cost-effectiveness analysis results

The EAG has generated base-case analysis cost-effectiveness results for a hypothetical cohort of 1000 patients with inconclusive results from fibrosis testing. Eight diagnostic test strategies were investigated in the EAG base-case analysis:

• T1: Any fibrosis (≥F1)

• T2: Significant fibrosis (≥F2)

• T3: Advanced fibrosis (≥F3)

• T4: Brunt grade ≥1

• T5: Brunt grade ≥2

• T6: NASH (NAS ≥4, ≥1 for lobular inflammation and hepatocyte ballooning)

• T7: Advanced NASH (NAS ≥4 plus ≥F2)

• T8 High risk (NASH or ≥F1).

The EAG base-case CEA results show that there is wide variation between the eight diagnostic test strategies in terms of the number of biopsies that could be avoided if the LiverMultiScan test were introduced into the current diagnostic pathway [minimum: Brunt grade ≥1 (n = 0); maximum: Brunt grade ≥2 (n = 328.9)].

For all eight diagnostic test strategies, the inclusion of the LiverMultiScan test in the pathway increases costs per patient; range: £244 (Brunt grade ≥2) to £412 (Brunt grade ≥1).

For seven of the diagnostic test strategies [any fibrosis (≥F1), significant fibrosis (≥F2), advanced fibrosis (≥F3), Brunt grade ≥1, NASH (NAS ≥4, ≥1 for lobular inflammation and hepatocyte ballooning), advanced NASH (NAS ≥4 plus ≥F2), and high risk (NASH or >F1)], QALY losses were greater for the LiverMultiScan plus biopsy pathway than for the biopsy only pathway. For the remaining diagnostic test strategy (Brunt grade ≥2), the QALY loss was greater for the biopsy only pathway.

For seven of the diagnostic test strategies [any fibrosis (≥F1), significant fibrosis (≥F2), advanced fibrosis (≥F3), Brunt grade ≥1, NASH (NAS ≥4, ≥1 for lobular inflammation and hepatocyte ballooning), advanced NASH (NAS ≥4 plus ≥F2) and high risk (NASH or >F1)], the base-case ICERs per QALY gained show that the LiverMultiScan plus biopsy pathway is dominated by the biopsy only pathway, that is, the biopsy only pathway is less expensive and leads to fewer QALY losses than the LiverMultiScan plus biopsy pathway.

The most cost-effective diagnostic test strategy is Brunt grade ≥2. The incremental cost-effectiveness ratio (ICER), for this strategy, for the comparison of the LiverMultiScan plus biopsy pathway versus the biopsy only pathway, is £1,266,511 per QALY gained. Clinicians suggested that, when considering this strategy, a positive result from a LiverMultiScan test would indicate that a patient should be referred for a biopsy. EAG base-case cost-effectiveness results are provided in Tables 15–19.

Threshold analyses

External Assessment Group scenario analyses

External Assessment Group analyses of uncertainty considered and rejected

Diagnostic test accuracy

In line with the final scope24 issued by NICE, the 13 studies30,53–64 included in the DTA review considered patients with NAFLD for whom advanced fibrosis or cirrhosis had not yet been diagnosed. However, no studies were identified that provided evidence for the DTA of MRI-based technologies for patients with NAFLD for whom TE or ARFI was unsuitable. Of the 13 studies30,53–64 that were included in the DTA review, the EAG was confident that only one study30 provided evidence for the DTA of MRI-based technologies for patients with NAFLD for whom advanced fibrosis or cirrhosis had not yet been diagnosed and who had indeterminate or discordant results from fibrosis testing; the Eddowes 201830 study evaluated LiverMultiScan and reported both PDFF and cT1 outputs. When assessing study quality, for most of the risk of bias and applicability concerns domains, the EAG considered that most studies had low risk of bias. For diagnosis of fibrosis, sensitivity ranged from 50% to 88% and specificity ranged from 42% to 75%. Sensitivity and specificity values for fibrosis testing were consistently higher when using LiverMultiScan cT1 data than when using LiverMultiScan PDFF.

Data from three studies were included in the meta-analyses for LiverMultiScan. For fibrosis (≥F2 and ≥F3), the pooled sensitivity and specificity values were higher for LiverMultiScan cT1 than for LiverMultiScan PDFF. For steatosis (Brunt grade ≥1), the meta-analysis results suggested that LiverMultiScan cT1 had greater sensitivity than specificity. The steatosis (Brunt grade ≥2) results for LiverMultiScan PDFF were fairly consistent with those for LiverMultiScan cT1. For NASH and advanced NASH, meta-analysis results were broadly similar between the LiverMultiScan cT1 and LiverMultiScan PDFF outputs, with the exception of sensitivity for detecting advanced NASH (LiverMultiScan cT1: 66.0%; LiverMultiScan PDFF: 49.4%). All other estimates of sensitivity and specificity ranged from 58.0% to 73.7%.

The sensitivity (fibrosis ≥F2) and specificity (fibrosis ≥F1 and ≥F2) reported for MRE in the four individual studies56–58,62 identified by the EAG were consistently greater when compared to those observed with LiverMultiScan. For fibrosis (≥F2) the sensitivity of MRE ranged from 82% to 95% and specificity ranged from 85% to 100%. For fibrosis (≥F3) the sensitivity of MRE ranged from 71% to 100% and specificity ranged from 79% to 93%. Data from three studies56–58 were used to estimate a summary ROC curve for MRE for advanced fibrosis (≥F3). The summary ROC indicated high DTA but not all observed study results lay close to the curve. The sensitivity and specificity observed in the two studies57,58 that used the Resoundant, Inc. MRE platform that is commercially available ranged from 85% to 100% and from 92% to 93%, respectively. The EAG notes that the DTA results for MRE are for patients with NAFLD for whom advanced fibrosis or cirrhosis had not yet been diagnosed. However, the studies did not specify whether these were patients who had indeterminate results from fibrosis testing, for whom TE or ARFI was unsuitable or who had discordant results from fibrosis testing.

Clinical impact

Eleven studies30,53,54,57,59,62,64,66–69 evaluated the clinical impact of MRI-based technologies for patients with NAFLD for whom advanced fibrosis or cirrhosis had not yet been diagnosed. As in the DTA review, no studies were identified that provided evidence for the clinical impact of MRI-based technologies for patients with NAFLD for whom TE or ARFI was unsuitable. Only one study30 provided evidence for the clinical impact of MRI-based technologies for patients with NAFLD for whom advanced fibrosis or cirrhosis had not yet been diagnosed and who had indeterminate or discordant results from fibrosis testing.

The two studies66,67 that evaluated the prognostic ability of MRI-based technologies included patients with NAFLD for whom advanced fibrosis or cirrhosis had not yet been diagnosed. However, the studies66,67 also included patients with other liver disease aetiologies and did not present results specifically for patients with NAFLD.

One study68 reported that LiverMultiScan could reduce the number of unnecessary biopsies for patients with non-NAFLD, NAFLD and no to mild fibrosis (F0 to F1) when compared to standard care.

Test failure rate in a population of patients with NAFLD was reported in four studies. 30,53,57,62 The test failure rate of the index tests for patients with NAFLD was 5.6%30 for LiverMultiScan and ranged from 3.9%57 to 7.6%53 for MRE. The test failure rate of MRE for patients with NAFLD was estimated by the EAG meta-analysis to be 4.2% (95% CI 2.5% to 6.2%).

Acceptability of LiverMultiScan from patient feedback was generally positive. 69 Patients considered the MRI scan was a painless and comfortable procedure and many highlighted that the ‘non-invasive’ element of the procedure was important. 69

No studies were identified that evaluated the remaining clinical impact outcomes specified in the final scope24 issued by NICE (see Table 2).

Cost-effectiveness

Eddowes 201830 study clinical effectiveness data were collected from a population with inconclusive results from previous fibrosis testing and used to populate the Blake 201678 model. However, the Blake 201678 model was not designed to explore cost-effectiveness for patients with inconclusive results from previous fibrosis testing. Therefore, the Eddowes 201830 study cost-savings estimates are not relevant to this appraisal.

The EAG developed a de novo economic model that enabled a comprehensive assessment (eight different diagnostic test strategies) of the cost-effectiveness of two different diagnostic pathways: LiverMultiScan plus biopsy versus biopsy only. The base-case ICER per QALY gained results for seven diagnostic pathways showed that LiverMultiScan plus biopsy was dominated by biopsy only and for Brunt grade ≥2 the ICER per QALY gained was £1,266,511. The results from the EAG threshold and scenario analyses demonstrated that these results were robust to plausible variations in the magnitude of key parameters.

The EAG also carried out MRE analyses using sensitivity and specificity data from a population that differed from the population described in the final scope24 issued by NICE and, therefore, results should only be considered as illustrative.

Strengths of the assessment

This assessment is the first to evaluate the DTA, clinical impact and cost-effectiveness of MRI-based technologies for three groups of patients with NAFLD for whom advanced fibrosis or cirrhosis has not yet been diagnosed, namely (i) patients with indeterminate results from fibrosis testing, (ii) patients who are unsuitable for testing with TE or ARFI and (iii) patients with discordant results from fibrosis testing. The clinical and cost-effectiveness systematic review processes included extensive literature searches and followed best-practice recommendations. 45–48

Perspectum Ltd71 has provided DTA data that were not previously available from published sources. These DTA data could allow LiverMultiScan outputs to be used to inform treatment decisions for patients with NAFLD (eight different diagnostic test strategies). The EAG used these data, as well published data, to carry out quantitative analyses.

A key strength of the EAG economic evaluation is that the de novo model provides a simple, flexible framework that allows the comparison of eight different diagnostic strategies. It is based on the best available DTA and population prevalence evidence (identified through the systematic review and provided by Perspectum Ltd) and captures the trade-off between high upfront costs of diagnostic tests and the reduction in subsequent biopsies that they may offer. The model design captures all of the main factors that are relevant to the decision problem. It is user-friendly and calculations are transparent. Furthermore, the model can easily be updated to incorporate new DTA and population prevalence evidence if they become available.

Limitations of the assessment

The DTA and population prevalence data available from Eddowes 2018/Perspectum Ltd30,71 are from patients with inconclusive results from previous fibrosis testing. The EAG has assumed that inconclusive is an umbrella term that includes the three subgroups of patients described in the final scope24 issued by NICE; however, the EAG is not confident that the term inconclusive includes patients for whom TE and ARFI are unsuitable.

The EAG quantitative synthesis only included data from six studies. 30,56–59,62 Furthermore, the meta-analyses were populated with data from small numbers of studies and only one30 of the studies included the population that is the subject of this assessment. This should be considered when interpreting results from the EAG meta-analyses.

Data on the clinical impact of MRI-based technologies were scarce for some outcomes (prognostic ability, number of liver biopsies and test failure rate). No data were available for the remaining clinical outcomes listed in the in the final scope24 issued by NICE.

Eddowes 2018/Perspectum Ltd30,71 provided LiverMultiScan DTA data for the relevant population. These data were included in the EAG DTA review and were used to inform the EAG economic model. However, Resoundant, Inc. did not provide any MRE DTA evidence for the relevant population and therefore MRE could not be considered as a comparator in the EAG economic model, although the cost-effectiveness of MRE can be inferred from the model results, that is, MRE is unlikely to be cost-effective in the population described in the final scope24 issued by NICE (using data from Eddowes 2018/Perspectum Ltd30,71) even if test accuracy was 100%.

In the EAG model, LiverMultiScan is positioned as a triage test, that is, LiverMultiScan would be added to the current NHS diagnostic pathway to avoid a more invasive downstream test (biopsy). The LiverMultiScan test is not 100% sensitive or specific for any of the eight diagnostic test strategies considered; the levels of sensitivity and specificity required to provide clinicians with sufficient confidence to use LiverMultiScan test results for patients described in the final scope24 issued by NICE are not known.

Potentially, different proportions of patients with advanced disease will receive a LiverMultiScan test FN result depending on the diagnostic test strategy used. If this did occur, the average impact of a FN result (costs and, notably, QALY losses) would vary depending on diagnostic test strategy used. The inability to resolve this issue is unlikely to be a major limitation of the EAG analyses as results from an EAG scenario analysis that removed the QALY loss associated with a LiverMultiScan test FN result showed that the conclusions that can be drawn from the EAG base-case cost-effectiveness analyses results did not change.

Clinical effectiveness

MRI-based technologies may be useful to identify patients who may benefit from additional testing in the form of liver biopsy and those for whom this additional testing may not be necessary. However, there is a paucity of DTA and clinical impact data for a population that may benefit from implementation of this technology, namely patients with indeterminate or discordant results from previous fibrosis testing or patients for whom TE and ARFI are not suitable.

Cost-effectiveness

Only one small LiverMultiScan study29 provided DTA and population prevalence data for patients described in the final scope24 issued by NICE. It is unclear whether sensitivity and specificity estimates reported by this small study29 will give clinicians sufficient confidence to use LiverMultiScan test results to triage patients with inconclusive results from previous fibrosis testing to biopsy. Cost-effectiveness results from the EAG model are only informative if clinicians have confidence in LiverMultiScan DTA data. Using the available DTA and population prevalence data, EAG cost-effectiveness results showed that LiverMultiScan is unlikely to be cost-effective at current prices when used to triage patients with inconclusive results from previous fibrosis testing to biopsy.

LiverMultiScan data are not available for patients for whom TE or ARFI was unsuitable. Further, no MRE DTA data were available for the population described in the final scope24 issued by NICE. The EAG considers that even if MRE was 100% accurate, due to high population prevalence estimates it is unlikely that MRE would be cost-effective at current prices.

Implications for service provision

If LiverMultiScan were to be recommended by NICE, the implications for NHS service provision would be significant due to the increased staffing levels and changes in infrastructure that would be required to accommodate the high demand for MRI scans for patients with NAFLD.

Suggested research priorities

Only Eddowes 2018/Perspectum Ltd30,71 provided data for a relevant population. Other published studies may also have included these patients; however, this information was not available from the published studies. If, in future, information about results from previous fibrosis testing could be recorded at the time of study enrolment, study DTA results from individual patients or subgroups could be used to inform treatment decisions.

Qualitative studies are required to investigate the impact of non-invasive technology test results on clinical decision-making, their potential to influence the uptake and maintenance of lifestyle modifications and the acceptability of the technologies to patients.

Intermediate outcomes

1. exp Non-alcoholic Fatty Liver Disease/

2. non-alcoholic fatty liver disease.tw,kw.

3. NAFLD.tw,kw.

4. non-alcoholic steatohepatitis.tw,kw.

5. NASH.tw,kw.

6. metabolic dysfunction associated fatty liver disease.tw,kw.

7. MAFLD.tw,kw.

8. or/1–7

9. exp Magnetic Resonance Imaging/

10. MRI.tw,kw.

11. magnetic resonance imag*.tw,kw.

12. LiverMultiScan.tw,kw.

13. Magnetic resonance elastograph*.tw,kw.

14. MRE.tw,kw.

15. or/9–14

16. 8 and 15

17. exp animals/

18. human/

19. 17 not 18

20. 16 not 19

21. limit 20 to english language

22. Clinical Decision-Making/

23. ‘clinical decision making’.tw,kw.

24. 22 or 23

25. 20 and 24

26. 8 and 24

27. exp ‘Predictive Value of Tests’/

28. ((predict* or prognos*) adj (value or ability)).tw,kw.

29. (predict* adj2 (progression or regression)).tw,kw.

30. 27 or 28 or 29

31. 20 and 30

32. exp *’Predictive Value of Tests’/

33. ((predict* or prognos*) adj (value or ability)).ti,kw.

34. (predict* adj2 (progression or regression)).ti,kw.

35. 33 or 34 or 35

36. 8 and 36

37. *Biopsy/ and Liver/

38. ‘number of liver biops*’.tw,kw.

39. (‘number of biops*’ adj3 liver).tw,kw.

40. 38 or 39 or 40

41. 20 and 41

42. 8 and 41

43. (lifestyle adj modif*).tw,kw.

44. 20 and 44

45. (lifestyle adj modif*).ti,kw.

46. 8 and 46

47. (time adj3 result*).tw,kw.

48. 20 and 48

49. 8 and 48

50. (time adj5 diagnos*).tw,kw.

51. Delayed Diagnosis/

52. Early Diagnosis/

53. 51 or 52 or 53

54. 20 and 54

55. ‘time to diagnosis’.tw,kw.

56. 8 and 56

57. (fail* adj3 (rate* or detect* or diagnos*)).tw,kw.

58. 20 and 58

59. 8 and 58

60. ((reduc* or remission) adj5 (fibrosis or inflammation)).tw,kw.

61. 20 and 61

62. ((reduc* or remission) adj3 (liver fibrosis or fibro inflammat* or fibro-inflammat*)).tw,kw.

63. 8 and 63

64. ((reduc* or remission) adj3 (liver adj fat*)).tw,kw.

65. 20 and 72

66. 8 and 72

Clinical outcomes and patient-reported outcomes

1. exp Non-alcoholic Fatty Liver Disease/

2. non-alcoholic fatty liver disease.tw,kw.

3. NAFLD.tw,kw.

4. non-alcoholic steatohepatitis.tw,kw.

5. NASH.tw,kw.

6. metabolic dysfunction associated fatty liver disease.tw,kw.

7. MAFLD.tw,kw.

8. 1 or 2 or 3 or 4 or 5 or 6 or 7

9. exp Magnetic Resonance Imaging/

10. MRI.tw,kw.

11. magnetic resonance imag*.tw,kw.

12. LiverMultiScan.tw,kw.

13. Magnetic resonance elastograph*.tw,kw.

14. MRE.tw,kw.

15. 9 or 10 or 11 or 12 or 13 or 14

16. 8 and 15

17. exp animals/

18. human/

19. 17 not 18

20. 16 not 19

21. limit 20 to english language

22. exp Mortality/

23. (mortalit* or death* or died).tw,kw.

24. 22 or 23

25. 20 and 24

26. (mortalit* or death* or died).ti,kw.

27. 22 or 26

28. 8 and 27

29. 28 not 25

30. exp Morbidity/

31. morbidit*.tw,kw.

32. contraindicat*.tw,kw.

33. complication*.tw,kw.

34. 30 or 31 or 32 or 33

35. 8 and 34

36. (morbidit* or complication* or contraindicat*).ti,kw.

37. exp *Morbidity/

38. 36 or 37

39. 8 and 38

40. 20 and 34

41. 39 not 40

42. exp ‘Quality of Life’/

43. ‘quality of life’.tw,kw.

44. ‘Chronic Liver Disease Questionnaire ‘.tw,kw.

45. CLDQ.tw,kw.

46. 42 or 43 or 44 or 45

47. 8 and 46

48. 20 and 46

49. 47 not 48

50. exp ‘Patient Acceptance of Health Care’/ or exp Patient Satisfaction/

51. acceptab*.tw,kw.

52. (patient* adj3 satisf*).tw,kw.

53. ‘perceived effectiveness’.tw,kw.

54. claustrophobi*.tw,kw.

55. 50 or 51 or 52 or 53 or 54

56. 8 and 55

57. 20 and 55

58. 56 not 57

Clinical impact studies

No studies provided data for the clinical impact outcomes of interest, and limited data were available for intermediate outcomes. There were only sufficient data to perform a meta-analysis for MRE test failure rate. It was not necessary or useful to plot or tabulate the data reported for other outcomes; these data were therefore reported narratively.

If the EAG had tabulated or plotted other clinical and/or intermediate outcome data, binary and categorical data would have been presented as frequencies and proportions, and continuous data would have been presented as means and standard deviations, or medians and interquartile ranges, according to the distribution of the data. If it had been possible to perform meta-analyses for continuous outcomes, the EAG would have expressed continuous data as means and standard deviations or standard errors (calculated from standard deviations or CIs where appropriate), and pooled these data in an inverse-variance meta-analysis using the metan command in Stata version 14.

Very little heterogeneity was observed in the conducted meta-analyses, and therefore it was not necessary to perform subgroup analyses. The EAG also did not perform sensitivity analyses, as there were no studies that the EAG considered to be important to exclude in sensitivity analyses (to investigate the impact of the inclusion of these studies on the overall pooled estimate).

EconLit (via EBSCO)

1. TI ‘non-alcoholic fatty liver disease’ OR AB ‘non-alcoholic fatty liver disease’ OR SU ‘non-alcoholic fatty liver disease’)

2. TI NAFLD OR AB NAFLD OR SU NAFLD

3. TI ‘non-alcoholic steatohepatitis’ OR AB ‘non-alcoholic steatohepatitis’ OR SU ‘non-alcoholic steatohepatitis’

4. TI NASH OR AB NASH OR SU NASH

5. TI ‘metabolic dysfunction associated fatty liver disease’ OR AB ‘metabolic dysfunction associated fatty liver disease’ OR SU ‘metabolic dysfunction associated fatty liver disease’

6. TI MAFLD OR AB MAFLD OR SU MAFLD

7. TI MRI OR AB MRI OR SU MRI

8. TI ‘magnetic resonance imag*’ OR AB ‘magnetic resonance imag*’ OR SU ‘magnetic resonance imag*’

9. TI LiverMultiScan OR AB LiverMultiScan OR SU LiverMultiScan

10. TI ‘Magnetic resonance elastograph* OR AB ‘Magnetic resonance elastograph* OR SU ‘Magnetic resonance elastograph*

11. TI MRE OR AB MRE OR SU MRE

12. S1 OR S2 OR S3 OR S4 OR S5 OR S6

13. S7 OR S8 OR S9 OR S10 OR S11

14. S12 AND S13