Methotrexate and monitoring for hepatotoxicity

National Clinical Guideline Centre (UK)

See 2019 Partial Update

Update information September 2017: The guideline has been revised throughout to link to MHRA advice and NICE technology appraisals that have been completed since original publication. Minor updates since publication August 2019: Links to the MHRA safety advice on the risk of using retinoids in pregnancy have been updated to the June 2019 version.

12Methotrexate and monitoring for hepatotoxicity

Publication Details

The risk of liver fibrosis is an accepted but unknown risk associated with methotrexate. Histological evaluation of a liver biopsy specimen is currently the gold standard for diagnosing, staging and monitoring liver fibrosis due to any cause but the procedure of liver biopsy carries significant morbidity and mortality, and is disliked by patients. The need for liver biopsy is commonly cited as a reason for dissatisfaction with treatment by patients, or for discontinuing therapy when biopsy is felt to be necessary³⁴⁷. In addition, the technique is subject to sample errors, since the samples collected are very small and pathological change may not be evenly distributed, and interpretation varies amongst histologists depending on level of experience, size of biopsy and use of staging/scoring system. Given the limitations of liver biopsy, significant effort has been invested in identifying clinical useful, non invasive markers of liver fibrosis that allow identification and quantification of liver fibrosis²⁴. Fibroelastography (achieved using the FibroScan^®) gives a measure of liver of elasticity (and therefore fibrosis) by measuring reflected ultrasound echoes before and during compression of the liver. The degree of displacement is related to the tissue elasticity stiffness. This method has been used to evaluate and track fibrosis in chronic liver disease⁴⁰⁴, and, as indicated in recent systematic review and economic analysis by the NHS Centre for Evidence-based purchasing⁶⁰, may have clinical utility for the detection and monitoring of fibrosis due to other causes. Serum biomarkers of liver fibrosis focus on indirect markers of liver function or direct markers of extracellular matrix components or the enzymes involved in their turnover. Indirect markers of liver function include aspartate aminotransferase (AST), alanine aminotransferase (ALT), c-glutamyl transpeptidase(c-GT), hyaluronic acid, apolipoprotein A1, bilirubin, a2-macroglobulin, haptoglobin, cholesterol, homeostasis model assessment of insulin resistance, platelets and prothrombin time. Direct markers of liver function include collagen IV, collagen VI, tissue inhibitor of metalloproteases-1 (TIMP-1), laminin, human cartilage glycoprotein-39 (YKL-40), tenascin, undulin, matrix metalloproteinase-2 (MMP-2) and pro-collagen III propeptide (PIIINP)²⁵³. Some of these biomarkers have been combined to improve clinical utility (for example, the European Enhanced Liver Fibrosis ELF panel which combines hyaluronic acid, TIMP-1 and PIIINP measurements).

For the last 5 – 10 years, serial measurement of PIIINP has become standard practice³⁴ for monitoring for liver fibrosis in patients on methotrexate, with elevated levels indicating the need for treatment cessation and/or consideration of liver biopsy. Given the high level of concern amongst clinicians and patients about methotrexate-associated liver dysfunction and the plethora of new indirect markers of liver disease, the GDG agreed it important to review the evidence for the clinical utility and validity of these markers of liver fibrosis in the context of psoriasis and treatment with methotrexate in order to optimise the safe use of this drug, and minimise the need for liver biopsy.

The GDG agreed to pose the following question: in people with psoriasis (all types) who are being treated with methotrexate or who are about to being treatment with methotrexate, what is the optimum non-invasive method of monitoring hepatotoxicity (fibrosis or cirrhosis) compared with liver biopsy?

12.1. Methodological introduction

12.1.1. Review methods

A literature search was conducted for diagnostic cohorts or case control studies that assessed the accuracy of non-invasive diagnostic tools to detect liver fibrosis or cirrhosis in people with psoriasis being treated or considered for treatment with methotrexate, compared with diagnosis by the reference standard of liver biopsy.

No time limit was placed on the literature search and there were no limitations on sample size or duration of follow-up. Indirect populations were excluded.

The relevant population for these diagnostic tools will be those with psoriasis who are at risk of developing liver damage as a result of exposure or planned exposure to methotrexate. The intended role of the index test would be for use by dermatologists to identify those suspected of having clinically significant liver damage in order to refer only these people on for expert assessment and, therefore, reduce the need for the invasive procedure of liver biopsy. Consequently, it is most important that the test is able to accurately rule-out a diagnosis, so that very few people with liver damage are missed for referral, although a reasonable accuracy for ruling-in a diagnosis would also be desirable to avoid referring too many people inappropriately.

The outcomes considered were:

Sensitivity
Specificity
Positive predictive value (PPV)
Negative predictive value (NPV)
Likelihood ratios (LRs)

The comparisons considered were any of the following diagnostic tests compared with liver biopsy:

imaging techniques: liver ultrasound, liver scintigraphy, ultrasound elastography (achieved using the FibroScan)
serum markers: serial pro-collagen III (PIIINP), the enhanced liver fibrosis (ELF) panel (tissue inhibitor of matrix metalloproteinase 1 (TIMP 1), hyaluronic acid (HA) and pro-collagen III), and FibroTest
AST to platelet ratio index (APRI)
Standard liver function tests (e.g., alanine aminotransferase (ALT), alkaline phosphatase (AP), aspartate aminotransferase (AST), total bilirubin, albumin, total protein, lactate dehydrogenase (LDH), gamma-glutamyl transferase (GGT) and prothrombin time (PT))

It was recognised that there was great variability in the literature regarding definitions of abnormal results on both liver biopsy and non-invasive tests. For the liver biopsy findings, any definition of fibrosis or cirrhosis, regardless of the classification scale, was accepted as indicating clinically significant liver damage. However, studies that limited the definition to at least marked fibrosis were excluded as they may overestimate the sensitivity by removing the potentially more difficult to diagnose milder end of the fibrosis spectrum. Additionally, fibrosis and cirrhosis were considered together as there were few cases of cirrhosis reported and many studies did not give the number with fibrosis and cirrhosis separately, although it is accepted that cirrhosis represents a greater clinical burden. The experience of the pathologist assessing the biopsy sample and the adequacy of sampling of the histological specimen are probably more important in terms of accurate diagnosis than the classification system used, but these were rarely stated in the studies. For the non-invasive tests, the definition of abnormal liver function provided in the study was accepted for use in the analysis, because, for example, there are no universally accepted reference ranges for liver function tests and the ranges may differ according to the population being studied (anything above the upper limit of normal was accepted as an abnormal reading in this review).

It was not possible to analyse the data using diagnostic meta-analysis (because there were no cases with at least 5 studies addressing the same reference standard and index tests, population and outcomes) or the standard version of GRADE. Therefore, a modified version of GRADE has been used and a narrative summary provided. The statistics used for this diagnostic review differ from those used in intervention reviews, and a definition for each of them is provided below (Table 155). Although no meta-analysis has been performed, forest plots are provided presenting the sensitivity and specificity of the tools compared with biopsy findings as reported in the studies individually (Appendix J). There are no forest plots for one study²⁷⁸, as insufficient raw data were available.

Table 155

Definitions of summary statistics for diagnostic accuracy studies.

Positive and negative predicative values are dependent on disease prevalence (pre-test probability) and so need to be interpreted together with prevalence, in the context of how test results modify the probability of disease (post-test probabilities). Consider that the lower the prevalence of disease the more certain we can be that a negative test indicates no disease, and the less certain that a positive result truly indicates the presence of disease. A note on how to interpret post-test probabilities/predictive values in the light of the disease prevalence is provided in Appendix Q. Fifteen diagnostic studies²⁸^,³⁴^,⁶⁴^,¹¹⁷^,¹⁴⁹^,²¹⁷^,²³⁹^,²⁴¹^,²⁵⁴^,²⁷⁸^,²⁸⁸^,³⁰⁵^,³³⁵^,⁴³²^,⁴³³ were found that addressed the question and were included in the review. No studies were available that from an exclusively paediatric population.

These studies differed in terms of:

Mean age (range 46 to 55 years)
Gender: % male (range 52 to 71.4%)
Sample size (range N=15 to N=168)
Prevalence of fibrosis and cirrhosis (6.9–69.5%)
Unit of analysis
- 8 studies used only one index test and one reference standard per person³⁴^,¹¹⁷^,¹⁴⁹^,²⁵⁴^,³⁰⁵^,³³⁵^,⁴³²^,⁴³³
- 3 studies included multiple paired index and reference tests per person⁶⁴^,²¹⁷^,²⁴¹
- 1 study included only single pre-MTX tests but multiple paired tests post-MTX²⁸⁸
- In 2 studies it was unclear whether the results were based upon single tests or multiple paired tests per person²⁸^,²⁷⁸
- 1 study included more than one index and reference test per patient, and also more than one index test per reference standard (i.e. the biopsy was paired with more than one index test)²³⁹

A summary of the methodological quality of the included studies according to QUADAS II criteria is provided in Table 156.

Table 156

Summary of study quality.

12.1.2. Study details – methods and results

The study methods are graded in the evidence profile (Table 157) and a summary of the study results is provided in Table 158. In the narrative below, methodological flaws according to the QUADAS II criteria are noted as points to suggest caution when interpreting results.

12.1.2.1. Liver function tests

Methods

Five studies were found that investigated the diagnostic accuracy of liver function tests in people with psoriasis eligible to receive methotrexate. The reference standard biopsy classification varied between the studies; two studies²⁷⁸^,²⁸⁸ used the Roenigk classification system, 2 studies used a system similar to Robinson grading¹⁴⁹^,³⁰⁵ and in one paper the classification system was unclear²¹⁷.

Two of the studies limited the population to those with known²¹⁷ or suspected¹⁴⁹ fibrosis. Two of the studies¹⁴⁹^,²⁷⁸ had an unclear method for determining the index test threshold, which could have meant that a cut-off was chosen in a post-hoc manner to optimise the apparent sensitivity of the test. Three of the studies¹⁴⁹^,²¹⁷^,³⁰⁵ had an unclear period of time between the index test and reference standard.

Results

Sensitivity: of patients with fibrosis or cirrhosis on biopsy, the proportion expected to test positive

Albumin: 19–29%
ALT: 5–40%
AP: 38–57%
AST: 20–43%
Bilirubin: 0–20%
Galactose: 14%
GGT: 33%
Prothrombin time: 1%

Specificity: of patients without fibrosis or cirrhosis on biopsy, the proportion expected to test negative

Albumin: 76–100%
ALT: 85–92%
AP: 71–76%
AST: 86–100%
Bilirubin: 86–96%
Galactose: 94%
GGT: 63%
Prothrombin time: 99%

Positive predictive value (figure in brackets is value-added PPV; the improvement in ability to determine a positive diagnosis over and above the known prevalence): if the liver function test was positive the probability of having liver fibrosis or cirrhosis (PPV) was:

AP: 15–60% (5 to 16%)
ALT: 22–67% (22–39%)
Albumin: 33–100%
Bilirubin: 0–41% (−47 to 23%)
Prothrombin time: 25% (NA)
AST: 29–100% (19–53%)
GGT: 40% (−2.9%)
Galactose: 83% (13.8%)

Negative predictive value: if the liver function test was negative the probability of not having liver fibrosis or cirrhosis (NPV) was:

Albumin: 61–62% (38–39% chance of having liver fibrosis or cirrhosis despite having a negative test)
ALT: 52–80% (20–48% chance of having liver fibrosis or cirrhosis despite having a negative test)
AP: 60–92% (8–40% chance of having liver fibrosis or cirrhosis despite having a negative test)
AST: 62–93% (7–38% chance of having liver fibrosis or cirrhosis despite having a negative test)
Bilirubin: 50–91% (9–50% chance of having liver fibrosis or cirrhosis despite having a negative test)
Galactose: 32% (68% chance of having liver fibrosis or cirrhosis despite having a negative test)
GGT: 56% (44% chance of having liver fibrosis or cirrhosis despite having a negative test)
Prothrombin time: 66% (34% chance of having liver fibrosis or cirrhosis despite having a negative test)

Positive likelihood ratio: in a person with compared to a person without liver fibrosis or cirrhosis, the number of times more likely a positive test result is:

Albumin: infinity
AP: 1.71–2.03
ALT: 2.6–5.2
AST: 3.13-infintiy
Bilirubin: 1.57–4.7
Galactose: 2.19
GGT: 0.89

Negative likelihood ratio: in a person without compared to a person with liver fibrosis or cirrhosis, the number of times more likely a negative test result is:

Albumin: 1.4
AP: 1.3–1.7
ALT: 1.4–1.5
AST: 1.4–1.5
Bilirubin: 0.88–1.2
Galactose: 1.1
GGT: 0.93

Additional information

One study²⁸⁸ assessed subgroups before and during methotrexate treatment and showed no consistent trend among the different liver function tests for differing accuracy before and after treatment was commenced
One study²⁸⁸ assessed the statistical association between abnormal liver function tests and biopsy grade III or IV, adjusted for age and history of cholecystitis. This study found that there was a significant association between grade III or IV biopsy findings and abnormal AST, but not ALP or bilirubin, levels
In one study¹⁴⁹, the one case of cirrhosis was not detected by abnormal liver function tests

12.1.2.2. Liver scintigraphy

Methods

Three studies¹¹⁷^,²⁴¹^,²⁵⁴ were found that investigated the diagnostic accuracy of liver scintigraphy in people with psoriasis eligible to receive methotrexate. The reference standard biopsy classification varied between the studies; one study¹¹⁷ used the Roenigk classification system, one study²⁴¹ graded fibrosis as none, very mild, mild, moderate or severe based on the method of Warin et al (abnormal was defined as at least moderate fibrosis, which maps on to the fibrosis assessed on the Roenigk scale) and the final study²⁵⁴ graded the biopsy according to steatosis, inflammation, fibrosis (graded mild, moderate or severe) and cirrhosis. The definition of abnormal on the liver scan also varied between the studies: one study¹¹⁷ counted the presence of any one from heterogeneous uptake, hepatomegaly, extra hepatic uptake and focal defects; another²⁵⁴ assessed the size of the liver and spleen, the pattern of uptake in these organs and the degree of extrahepatic uptake; and the third²⁴¹ classified abnormal as a portal contribution of <50% of total hepatic uptake of colloid at 30s. None of the studies specified whether the assessors were blinded to the results of the first test.

Results

Sensitivity and specificity: The findings for the sensitivity and specificity of liver scans varied between the studies. The sensitivity ranged from 50.0 to 83.3% and specificity from 64.7 to 81.5%. Sensitivity and specificity were highest in the study that defined abnormal results on the scan as <50% portal contribution, which also had by far the lowest prevalence of liver fibrosis or cirrhosis and used the definition of at least moderate fibrosis.

Positive predictive value/negative predictive value: If the scan was positive the probability of having liver fibrosis or cirrhosis (PPV or proportion of patients with a positive test who are correctly diagnosed) ranged from 25 to 40% and if the scan was negative the probability of not having liver fibrosis or cirrhosis (NPV or proportion of patients with a negative test who are correctly diagnosed) ranged from 78.6 to 98.5% (1.5 to 21.4% chance of having fibrosis or cirrhosis despite having a negative test).

Given that the pre-test probabilities of having fibrosis/cirrhosis were 29.2, 6.9 and 24.5% in the three populations, this means that the liver scan improves the ability to determine a positive diagnosis (over and above the known prevalence) by 10.8 to 18.8% and a negative diagnosis by 5.3 to 7.8%.

Likelihood ratio: A positive test result ranged from 1.62 to 4.50 times more likely in a person with compared to a person without fibrosis/cirrhosis, and a negative test result ranges from 1.5 to 5.0 times more likely in a person without compared to a person with fibrosis/cirrhosis. Both the positive and negative likelihood ratios were much more favourable in the study that defined abnormal results on the scan as <50% portal contribution, which also had by far the lowest prevalence of liver fibrosis or cirrhosis and used the definition of at least moderate fibrosis²⁴¹.

Additional information

One study²⁴¹ noted that the one false negative result had a portal contribution of 51% so a slight alteration in the threshold would have resulted in all patients with portal fibrosis to be detected by the scan.

In one study¹¹⁷, the two cases of cirrhosis were correctly identified.

12.1.2.3. Liver ultrasound

Methods

Two studies⁶⁴^,²⁵⁴ were found that investigated the diagnostic accuracy of liver ultrasound in people with psoriasis eligible to receive methotrexate. The reference standard biopsy classification varied between the studies; one study²⁵⁴ graded the biopsy according to steatosis, inflammation, fibrosis (graded mild, moderate or severe) and cirrhosis and the other study⁶⁴ graded the biopsy by subjective microscopic assessment based on the method of Warin et al of fat, inflammation, fibrosis (each graded 0, 0.5, 1, 2, or 3) and cirrhosis (not graded). The definition of abnormal on the ultrasound scan also varied between the studies: one study counted the presence of abnormalities in any one from liver size, shape, echo pattern and information about the biliary and vascular system according to a standard proforma while the other assessed fatty change and fibrosis (only those showing fibrosis were counted as positive tests).

One study²⁵⁴ did not specify whether the assessors were blinded to the results of the first test.

Results

Sensitivity and specificity: The findings for the sensitivity and specificity of ultrasound scans varied between the studies. The sensitivity ranged from 0 to 19% and specificity from 86 to 100% for detecting any degree of fibrosis and were 25% and 100%, respectively, for detecting portal fibrosis (in accordance with Roenigk criteria).

Positive predictive value/negative predictive value: If the ultrasound scan was positive the probability of having liver fibrosis or cirrhosis (PPV or proportion of patients with a positive test who are correctly diagnosed) ranged from 0 to 100% and if the scan was negative the probability of not having liver fibrosis or cirrhosis (NPV or proportion of patients with a negative test who are correctly diagnosed) ranged from 57 to 73% (27 to 43% chance of having fibrosis or cirrhosis despite having a negative test).

Given that the pre-test probabilities of having fibrosis/cirrhosis were 24.5, 48.2 and 37.0% in the three populations, this means that the liver scan improves the ability to determine a positive diagnosis (over and above the known prevalence) by −24.5 to 63.0% and a negative diagnosis by −2.5 to 6.0%.

Likelihood ratio: A positive test was infinitely more likely in a person with compared to a person without fibrosis/cirrhosis in two studies but equally likely in another study, and a negative test result ranged from 0.86 to 1.2 times more likely in a person without compared to a person with fibrosis/cirrhosis.

The difference in accuracy for detecting any compared with portal fibrosis was less pronounced than with scintigraphy

Additional information

In one study²⁵⁴ ultrasound failed to detect any of the three cases of fibrosis or cirrhosis.

12.1.2.4. PIIINP

Methods

Four studies³⁴^,²³⁹^,³³⁵^,⁴³²^,⁴³³ were found that investigated the diagnostic accuracy of PIIINP assays in people with psoriasis eligible to receive methotrexate. The reference standard biopsy classification varied between the studies; one study²³⁹ used the Roenigk classification system, one study³⁴ graded the biopsy according to steatosis, inflammation, fibrosis and cirrhosis and the other two studies did not define the classification systems used⁴³²^,⁴³³. All studies conducted more than one assessment of PIIINP per person and the threshold for an abnormal PIIINP assay was >4.2 μg/l (based on the reference range in Finnish blood donors); however, the manufacturer’s information leaflet states that the reference range is 2.3–6.4 μg/l based on PIIINP values of apparently healthy adults (19–65 years), although variations in population demographics may mean that slightly different reference limits apply across populations.

Although all studies performed more than one PIIINP assay per person, for the analysis of diagnostic accuracy not all of the test results were always included:

One study³⁴ serially assessed PIIINP and used only the PIIINP assay taken at the time of first biopsy
One study³³⁵^,⁴³³ had serial PIIINP assays in 11 out of 74 participants and used the PIIINP assay taken at the time closest to biopsy
One study²³⁹ included multiple PIIINP assays from serial assessments and multiple biopsies per patient in the analysis (with some biopsies counted more than once as they were paired with more than one PIIINP assay), and only included biopsies with PIIINP tests within 6 months before and 6 months after biopsy
The final study⁴³² serially assessed PIIINP but classed participants as positive on biopsy or PIIINP if at least one of their tests was abnormal (but it is unclear how many abnormal test results they may also have had).

Two studies⁴³²^,⁴³³ had an unclear period of time between the measurement of the index test and the reference standard, which may have meant that the clinical condition of the individual had changed in the time that elapsed between the assessments.

One study⁴³² performed serial analyses of PIIINP and multiple biopsies per patient but did not include all of the PIIINP or biopsy results in the analysis; therefore, those who tested positive (based on at least one abnormal result) could also have had several negative tests. This study was still considered eligible for inclusion as those classed as negative would not have had even a single elevated PIIINP or abnormal biopsy result among the multiple test results, which is informative as we are interested in a screening test most able to accurately determine those who do not have liver abnormalities.

Results

Note that PIIINP elevation can be due to an increase in fibrosis (and so cleaving of pro-collagen) anywhere in the body. Therefore, in those with psoriasis and arthritis it is possible that any elevation in PIIINP is due to the arthritis rather than the liver. In the available studies the proportion with PsA ranged from 22–46%, but was unclear in two studies³⁴^,³³⁵.

In one study³⁴ the range of PIIINP values in a control group of 11 people with PsA and no MTX exposure was 2.2–4.6 ng/ml.

In the study²³⁹ with 22% PsA, 4 of 6 grade II biopsies from 4 patients with inflammatory arthritis had elevated PIIINP in all associated readings and the other two biopsies had some abnormal PIIINP readings.

In one pilot study³³⁵ one out of 11 participants with PsA gave a false positive result, and this participant had steatosis on biopsy. This was the only false positive in the study. Note that in a sub-group analysis of 10 people with PsA and 13 people with psoriasis but no arthritic component the accuracy for ruling out was actually higher in the group with PsA (sensitivity 100% vs 33% and NPV 100% vs 40%); however, the sample sizes in the subgroups were very small.

In the final study⁴³² 38.6% had PsA and one of the two false positives was a participant with PsA.

Sensitivity and specificity: The findings for the sensitivity and specificity of PIIINP varied between the studies. The sensitivity ranged from 62.5 to 100% and specificity from 63.6 to 97.9%. Note that the sensitivity and specificity were high in the study with the highest risk of bias and the lowest prevalence⁴³², which did not include all of the PIIINP assay results in the analysis.

Positive predictive value/negative predictive value: If the PIIINP assay was positive the probability of having liver fibrosis or cirrhosis (PPV or proportion of patients with a positive test who are correctly diagnosed) ranged from 23.4 to 95.0% and if the scan was negative the probability of not having liver fibrosis or cirrhosis (NPV or proportion of patients with a negative test who are correctly diagnosed) ranged from 88.5 to 100% (0 to 11.5% chance of having fibrosis or cirrhosis despite having a negative test).

Given that the pre-test probabilities of having liver fibrosis or cirrhosis were 24.1, 5.8, 13.7 and 34.7% in the four populations, this means that the PIIINP assay improves the ability to determine a positive diagnosis (over and above the known prevalence) by 9.7 to 60.3% and a negative diagnosis by 5.6 to 23.2%. Note that the value-added PPV was markedly higher in the two Zachariae studies⁴³²^,⁴³³.

Likelihood ratio: A positive test result ranged from 1.93 to 36 times more likely in a person with compared to a person without fibrosis/cirrhosis, and a negative test result ranged from 1.79-times to infinitely more likely in a person without compared to a person with fibrosis/cirrhosis.

The two Zachariae studies⁴³²^,⁴³³ demonstrated markedly higher values for sensitivity and PPV than the other two studies.

Additional information

One study²³⁹ noted that three liver biopsies in two morbidly obese patients who also had maturity-onset diabetes were graded II on Roenigk classification but showed signs of NASH (rather than portal fibrosis, which is more often associated with MTX use).
In one study³⁴ the three cases of cirrhosis were all correctly identified and the sensitivity and specificity for detecting fibrosis alone were 81% and 62%, respectively, based on one biopsy per patient.

12.1.2.5. Fibrotest and fibroscan

Methods

One study²⁸ was found that investigated the diagnostic accuracy of Fibrotest and Fibroscan in people with psoriasis eligible to receive methotrexate. The reference standard biopsy classification was based on the Metavir system and the definition of abnormal was Metavir >F2. The definition of abnormal on the Fibrotest was defined by a cut-off of 0.31 and on Fibroscan by a cut-off of 7.1kPa based on the literature.

This study did not state whether the population was based on a consecutive sample and there could have been up to 18 months between the index test and reference standard being undertaken, which could be long enough for the liver to develop fibrosis or cirrhosis. Additionally, for Fibroscan there was some discrepancy between the details in the text and the reported diagnostic accuracy statistics.

Results

Sensitivity and specificity: The sensitivity was 83% for Fibrotest and 50% for Fibroscan, while the specificities were 61% and 88%, respectively

Positive predictive value/negative predictive value: If the Fibrotest was positive the probability of having liver fibrosis or cirrhosis (PPV or proportion of patients with a positive test who are correctly diagnosed) was 42% and if the test was negative the probability of not having liver fibrosis or cirrhosis (NPV or proportion of patients with a negative test who are correctly diagnosed) was 92% (8% chance of having fibrosis or cirrhosis despite having a negative test). The PPV for Fibroscan was 33% while the NPV was 86% (14% chance of having fibrosis or cirrhosis despite having a negative test).

Given that the pre-test probability of having fibrosis/cirrhosis was 25% for the Fibrotest population, this means that the liver scan improves the ability to determine a positive diagnosis (over and above the known prevalence) by 16.7% and a negative diagnosis by 16.7%. It was not possible to calculate the valued-added predictive values for Fibroscan as the population sample used for the calculation of PPV and NPV was unclear.

Likelihood ratio: For Fibrotest, a positive test was 2.14-times more likely in a person with compared to a person without fibrosis/cirrhosis, and a negative test was 3.7-times more likely in a person without compared to a person with fibrosis/cirrhosis. Again, it was not possible to calculate this statistic for Fibroscan as the 2×2 table could not be verified.

Additional information

In nine patients, Fibroscan and Fibrotest resulted in different Metavir scores with a discordance of two stages. In four of them, the total Fibroscan procedure failed because of the presence of obesity. In the remaining five, biopsy length was significantly shorter than the biopsy length of the remaining patients.

12.2. Non-invasive liver tests vs. liver biopsy

12.2.1. Evidence profile

Table 157

Modified GRADE profile for the diagnostic accuracy of tools to detect liver fibrosis or cirrhosis.

12.2.2. Evidence summary

Table 158

Summary statistics for diagnostic accuracy of tools for fibrosis and cirrhosis.

12.2.3. Evidence statements

The following statements are organised by outcome and ordered to list the tests in approximate order from the best to the worst diagnostic accuracy according to that measure.

Sensitivity: of patients with fibrosis or cirrhosis on biopsy, the proportion expected to test positive

PIIINP: 62.5 to 100% [4 studies; 264 participants; moderate to very low quality evidence]³⁴^,²³⁹^,⁴³²^,⁴³³
Scintigraphy (portal contribution): 83% [1 study; 63 participants; very low quality evidence]
Fibrotest: 83% [1 study; 24 participants; very low quality evidence]²⁸
Scintigraphy (abnormalities): 50–57% [2 studies; 73 participants; very low quality evidence]¹¹⁷^,²⁵⁴
AP: 38–57% [3 studies; 200 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
Fibroscan: 50% [1 study; 24 participants; very low quality evidence]²⁸
AST: 20–43% [3 studies; 235 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
Gamma-glutamyl transferase: 33% [1 study; 15 participants; very low quality evidence]³⁰⁵
ALT: 5–40% [2 studies; 186 participants; very low quality evidence]¹⁴⁹^,²⁷⁸
Ultrasound (portal fibrosis): 25% [1 study; 28 participants; very low quality evidence]⁶⁴
Albumin: 19–29% [2 studies; 183 participants; low to very low quality evidence]²⁷⁸^,³⁰⁵
Bilirubin: 0–20% [3 studies; 200 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
Ultrasound (any fibrosis): 0 to 19% [2 studies; 77 participants; low to very low quality evidence]⁶⁴^,²⁵⁴
Galactose: 14% [1 study; 45 participants; very low quality evidence]²¹⁷
Prothrombin time: 1% [1 study; 168 participants; low quality evidence]²⁷⁸

Specificity: of patients without fibrosis or cirrhosis on biopsy, the proportion expected to test negative

Ultrasound (portal fibrosis): 100% [1 study; 28 participants; very low quality evidence]⁶⁴
Prothrombin time: 99% [1 study; 168 participants; low quality evidence]²⁷⁸
Ultrasound (any fibrosis): 86 to 100% [2 studies; 77 participants; low to very low quality evidence]⁶⁴^,²⁵⁴
AST: 86–100% [3 studies; 235 participants; low to very low quality evidence] ²⁷⁸^,²⁸⁸^,³⁰⁵
Bilirubin: 86–96% [3 studies; 200 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
Galactose: 94% [1 study; 45 participants; very low quality evidence]²¹⁷
ALT: 85–92% [2 studies; 186 participants; very low quality evidence]¹⁴⁹^,²⁷⁸
Albumin: 76–100% [2 studies; 183 participants; low to very low quality evidence]²⁷⁸^,³⁰⁵
Fibroscan: 88% [1 study; 24 participants; very low quality evidence]²⁸
Scintigraphy (portal contribution): 82% [1 study; 63 participants; very low quality evidence] ²⁴¹
PIIINP: 63.6 to 97.9% [4 studies; 264 participants; moderate to very low quality evidence]³⁴^,²³⁹^,⁴³²^,⁴³³
Alkaline phosphatase: 71–77% [3 studies; 200 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
Scintigraphy (abnormalities): 65–73% [2 studies; 73 participants; very low quality evidence]¹¹⁷^,²⁵⁴
Gamma-glutamyl transferase: 63% [1 study; 15 participants; very low quality evidence]³⁰⁵
Fibrotest: 61.1% [1 study; 24 participants; very low quality evidence]²⁸

Galactose: 83% (13.8%) [1 study; 45 participants; very low quality evidence]²¹⁷
Albumin: 33–100% (53%) [2 studies; 183 participants; low to very low quality evidence]²⁷⁸^,³⁰⁵
AST: 29–100% (19–53%) [3 studies; 235 participants; low to very low quality evidence] ²⁷⁸^,²⁸⁸^,³⁰⁵
PIIINP: 23.4 to 95.0% (9.7 to 60.3%) [4 studies; 264 participants; moderate to very low quality evidence]³⁴^,²³⁹^,⁴³²^,⁴³³
ALT: 22–67% (22–39%) [2 studies; 186 participants; low to very low quality evidence]¹⁴⁹^,²⁷⁸
AP: 15–60% (5.4 to 16%) [3 studies; 200 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
Fibrotest: 42% (16.7%) [1 study; 24 participants; very low quality evidence]²⁸
GGT: 40% (−2.9%) [1 study; 15 participants; very low quality evidence]³⁰⁵
Scintigraphy (abnormalities): 37.5–40.0% (10.8 to 13.0%) [2 studies; 73 participants; very low quality evidence]¹¹⁷^,²⁵⁴
Bilirubin: 0–41% (−47 to 23%) [3 studies; 200 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
Fibroscan: 33% (NA) [1 study; 24 participants; very low quality evidence]²⁸
Scintigraphy (portal contribution): 25% (18.8 %) [1 study; 63 participants; very low quality evidence]²⁴¹
Prothrombin time: 25% (NA) [1 study; 168 participants; low quality evidence]²⁷⁸
Ultrasound: 0 to 100% (−24.5 to 63.0%) [2 studies; 77 participants; low to very low quality evidence]⁶⁴^,²⁵⁴

Negative predictive value (figure in brackets is value-added NPV; the improvement in ability to determine a negative diagnosis over and above the known prevalence): if the liver function test was negative the probability of not having liver fibrosis or cirrhosis (NPV) was:

PIIINP: 88.5 to 100% (5.6 to 23.2%) [4 studies; 264 participants; moderate to very low quality evidence]³⁴^,²³⁹^,⁴³²^,⁴³³
Scintigraphy (portal contribution): 98.5% (5.4%) [1 study; 63 participants; very low quality evidence]²⁴¹
Fibrotest: 92% (16.7%) [1 study; 24 participants; very low quality evidence]²⁸
Fibroscan: 86% (NA) [1 study; 24 participants; very low quality evidence]²⁸
Scintigraphy (abnormalities): 78.6 to 81.8% (5.3 to 7.8%) [2 studies; 73 participants; very low quality evidence]¹¹⁷^,²⁵⁴
AST: 62–93% (2.6 to 8.7%) [3 studies; 235 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
AP: 60–92% (1.6 to 8.2%) [3 studies; 200 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
Bilirubin: 50–91% (−3.3 to 0.6%) [3 studies; 200 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
ALT: 52–80% (6.4–7.8%) [2 studies; 186 participants; very low quality evidence]¹⁴⁹^,²⁷⁸
Ultrasound: 57 to 73% (−2.5 to 6.0%) [2 studies; 77 participants; low to very low quality evidence]⁶⁴^,²⁵⁴
Prothrombin time: 66% (NA) [1 study; 168 participants; low quality evidence]²⁷⁸
Albumin: 61–62% (8.7%) [2 studies; 183 participants; low to very low quality evidence]²⁷⁸^,³⁰⁵
Gamma-glutamyl transferase: 56% (−1.5%) [1 study; 15 participants; very low quality evidence]³⁰⁵
Galactose: 32% (1.8%) [1 study; 45 participants; very low quality evidence]²¹⁷

Positive likelihood ratio: in a person with compared to a person without liver fibrosis or cirrhosis, the number of times more likely a positive test result is:

Albumin: infinity [2 studies; 183 participants; low to very low quality evidence]²⁷⁸^,³⁰⁵
AST: 3.13-infintiy [3 studies; 235 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
PIIINP: 1.93 to 36 [4 studies; 264 participants; moderate to very low quality evidence]³⁴^,²³⁹^,⁴³²^,⁴³³
Scintigraphy (portal contribution): 4.50 [1 study; 63 participants; very low quality evidence]²⁴¹
Ultrasound: zero to infinite [2 studies; 77 participants; low to very low quality evidence]⁶⁴^,²⁵⁴
ALT: 2.6–5.2 [2 studies; 186 participants; very low quality evidence]¹⁴⁹^,²⁷⁸
Bilirubin: 1.57–4.7 [3 studies; 200 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
Galactose: 2.19 [1 study; 45 participants; very low quality evidence]²¹⁷
Fibrotest: 2.14 [1 study; 24 participants; very low quality evidence]²⁸
Alkaline phosphatase: 1.71–2.03 [3 studies; 200 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
Scintigraphy (abnormalities): 1.62 to 1.85 [2 studies; 73 participants; very low quality evidence]¹¹⁷^,²⁵⁴
Gamma-glutamyl transferase: 0.89 [1 study; 15 participants; very low quality evidence]³⁰⁵

Negative likelihood ratio: in a person without compared to a person with liver fibrosis or cirrhosis, the number of times more likely a negative test result is:

Scintigraphy (portal contribution): 5.0 [1 study; 63 participants; very low quality evidence]
PIIINP: 1.79-times to infinitely [4 studies; 264 participants; moderate to very low quality evidence]³⁴^,²³⁹^,⁴³²^,⁴³³
Fibrotest: 3.7 [1 study; 24 participants; very low quality evidence]²⁸
Alkaline phosphatase: 1.3–1.7 [3 studies; 200 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
AST: 1.4–1.5 [3 studies; 235 participants; low to very low quality evidence] ²⁷⁸^,²⁸⁸^,³⁰⁵
ALT: 1.4–1.5 [2 studies; 186 participants; very low quality evidence]¹⁴⁹^,²⁷⁸
Scintigraphy (abnormalities): 1.4 to 1.5 [2 studies; 73 participants; very low quality evidence]¹¹⁷^,²⁵⁴
Albumin: 1.4 [2 studies; 183 participants; low to very low quality evidence]²⁷⁸^,³⁰⁵
Galactose: 1.1 [1 study; 45 participants; very low quality evidence]²¹⁷
Bilirubin: 0.88–1.2 [3 studies; 200 participants; low to very low quality evidence]²⁷⁸^,²⁸⁸^,³⁰⁵
Gamma-glutamyl transferase: 0.93 [1 study; 15 participants; very low quality evidence]³⁰⁵
Ultrasound: 0.86 to 1.2 [2 studies; 77 participants; low to very low quality evidence]⁶⁴^,²⁵⁴

Conclusions

The available studies mainly have small samples, which, combined with the relatively low prevalence of fibrosis and cirrhosis, mean that the estimates of diagnostic accuracy are imprecise, leading to uncertainty (particularly around the sensitivity of the tests)
All of the tests generally perform better in terms of specificity compared with sensitivity, meaning that they are of greater value for confidently ruling in a diagnosis of clinically significant liver damage if the non-invasive test is positive, but there is less certainty that those who test negative actually do not have fibrosis or cirrhosis
Ruling in a diagnosis:
- The specificity was consistently over 75% for the majority of the tests (ultrasound, prothrombin time, AST, bilirubin, galactose, ALT, albumin and scintigraphy when abnormality was assessed using the % portal contribution to total hepatic uptake of colloid and Fibroscan)
- However, there was great variability in the PPV for each test, with no test showing values consistently above 50% across the different studies (except the galactose tolerance test which was only assessed in one study²¹⁷)
- The positive likelihood ratio was best for AST, albumin, ultrasound and PIIINP
Ruling out a diagnosis:
- Accepting the uncertainty, the tests that may give a useful level of sensitivity are PIIINP, scintigraphy for detecting portal fibrosis and Fibrotest
- Similarly, the NPV was only consistently over 75% for PIIINP, scintigraphy, Fibrotest and Fibroscan
- The negative likelihood ratio was best for PIIINP, scintigraphy for detecting portal fibrosis and Fibrotest.

12.3. Economic evidence

One study⁵⁴ was included that evaluated different methods of monitoring for hepatotoxicity in people with psoriasis being treated with methotrexate. The monitoring strategies evaluated Chalmers and colleagues were defined as follows:

3.: serial PIIINP testing with selective liver biopsy, and

4.: Routine liver biopsy.

This study is summarised in the economic evidence profile below (Table 159 and Table 160). See also the full study evidence tables in Appendix I.

Table 159

Serial PIIINP versus routine liver biopsy – Economic study characteristics.

Table 160

Serial PIIINP versus routine liver biopsy – Economic summary of findings.

No relevant economic evaluations comparing other non-invasive liver monitoring methods were identified. No studies were excluded.

The monitoring strategies evaluated by Chalmers and colleagues were defined as follows:

Serial PIIINP testing with selective liver biopsy:
- Where possible serum should be collected for PIIINP measurement prior to starting methotrexate. It should subsequently be measured every 2–3 months during continued treatment. Indications for considering liver biopsy:
  - Elevation of pre-treatment PIIINP above 8.0 μg L⁻¹
  - Elevation of PIIINP above the normal range (1.7 to 4.2 μg L⁻¹) in at least three samples over a 12 month period
  - Elevation of PIIINP above 8.0 μg L⁻¹ in two consecutive samples
- Indications for considering withdrawal of methotrexate:
  - Elevation of PIIINP above 10.0 μg L⁻¹ in at least three samples over a 12 months period
- The decision whether to perform liver biopsy, withdraw treatment or continue treatment despite raised PIIINP levels must also take into account other factors such as disease severity, patient age and the ease with which alternative therapies may be used in place of methotrexate.
Routine liver biopsy:
- In patients without risk factors for liver damage, perform first liver biopsy after cumulative dose of 1.0 to 1.5 g methotrexate
- Provided no significant abnormalities are found, repeat liver biopsy after each additional 1.5 g methotrexate
- When cumulate dose >4.0 g, perform biopsy after each additional 1.0 g methotrexate
- In patient with risk factors for liver damage, perform liver biopsy within 2–4 months of starting methotrexate and after each additional 0.5 to 1.0 g thereafter.

Based on the findings of the study and if PIIINP measurement cost £22.50:

Monitoring with serial PIIINP and selective liver biopsy is likely to be cost-saving if liver biopsy costs more than £375
Monitoring with serial PIIINP and selective liver biopsy may be more costly if liver biopsy costs less than £375.

None of these cost estimates take into account the additional costs of managing potential complications of liver biopsy. With the risk of developing significant hepatic injury from liver biopsy being approximately 1–2% and the risk of mortality being around 0.01–0.1%, these costs (and impact on health-related quality of life) could be significant. If these costs were included, it is likely that cost of liver biopsy at which monitoring with serial PIIINP becomes cost-saving would be much lower. Table 161 below shows that the current cost of liver biopsy (excluding cost of potential complications) is between £553 for a day case and £816 for patients requiring an overnight stay in hospital.

Table 161

Unit costs of monitoring tests – exclusive of labour costs.

In the event that a monitoring strategy of serial PIIINP measurement with selective liver biopsy is more costly than routine liver biopsy, the additional costs could be justified by improved health outcomes in terms of mortality and morbidity avoided. These would have to be weighed against the risk that some patients with significant liver abnormalities may be missed.

The authors investigated whether changing the threshold value upon which PIIINP was counted as predictive of liver fibrosis would increase the specificity of the test. They found that altering the threshold from 4.2 to 4.9 μg L⁻¹ would have reduced the number of false positives (e.g. those undergoing a liver biopsy who turn out to have normal result or minor abnormalities) by more than half, but at the risk of failing to identify patients with significant liver damage (e.g. false negatives).

The study does not indicate whether any significant abnormalities were missed in the serial PIIINP strategy and what the consequences for these patients might be. The authors assert that the risk of serious harm from liver biopsy outweighs the risk of missing significant liver damage in patients monitored using serial PIIINP.

12.3.1. Unit costs

In the absence of recent UK cost-effectiveness analysis, relevant unit costs are provided below to aid consideration of cost effectiveness.

12.3.2. Evidence statements

One partially applicable cost-consequence analysis with very serious limitations found that for patients with psoriasis undergoing treatment with methotrexate, a strategy of monitoring hepatotoxicity with serial PIIINP and selective liver biopsy was likely to be cost saving compared to routine liver biopsy if the unit cost of liver biopsy was greater than £375.

12.4. Recommendations and link to evidence

Table

Before and during methotrexate treatment, offer the person with any type of psoriasis an evaluation for potential risk of hepatotoxicity. Use standard liver function tests and serial serum procollagen III levels to monitor for abnormalities during treatment (more...)

Publication Details

Copyright

Apart from any fair dealing for the purposes of research or private study, criticism or review, as permitted under the Copyright, Designs and Patents Act, 1988, no part of this publication may be reproduced, stored or transmitted in any form or by any means, without the prior written permission of the publisher or, in the case of reprographic reproduction, in accordance with the terms of licences issued by the Copyright Licensing Agency in the UK. Enquiries concerning reproduction outside the terms stated here should be sent to the publisher at the UK address printed on this page.

The use of registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant laws and regulations and therefore for general use.

The rights of National Clinical Guideline Centre to be identified as Author of this work have been asserted by them in accordance with the Copyright, Designs and Patents Act, 1988.

Publisher

Royal College of Physicians (UK), London

NLM Citation

National Clinical Guideline Centre (UK). Psoriasis: Assessment and Management of Psoriasis. London: Royal College of Physicians (UK); 2012 Oct. (NICE Clinical Guidelines, No. 153.) 12, Methotrexate and monitoring for hepatotoxicity.

Table 155Definitions of summary statistics for diagnostic accuracy studies

Measure	Definition
True positives (TP)	Correct positive test result - number with fibrosis or cirrhosis with a positive index test result
True negatives (TN)	Correct negative test results - number without fibrosis or cirrhosis with a negative index test result
False positives (FP)	Incorrect positive test result - number without fibrosis or cirrhosis with a positive index test result
False negatives (FN)	Incorrect negative test result - number with fibrosis or cirrhosis with a negative index test result
Sensitivity	Proportion of those with the disease (based on reference standard) who are positive on the index test
Specificity	Proportion of those without the disease (based on reference standard) who are negative on the index test
Positive predicative value (PPV)	Probability of having the disease in a patient with a positive index test result
Negative predicative value (NPV)	Probability of not having the disease in a patient with a negative index test result
Positive likelihood ratio (LR+)	The number of times more likely a positive test result is in a person with compared to a person without the disease (therefore LR+ is >1)
Negative likelihood ratio (LR−)	The number of times more likely a negative test result is in a person with compared to a person without the disease (therefore LR− is <1)

Table 156Summary of study quality

Study	N	Index test(s)	Selection criteria	Reporting bias	Verification bias^(a)	Time between tests^(b)	Index test threshold selection	Blinding of assessors	Experienced assessor	Adequate biopsy sample
Liver function tests
Ho 1986 (prospective)	18	LFT: ALT	Consecutive sample, all receiving MTX (Singapore)	Yes	Yes – only included those with high ALT or high total MTX dose	Unclear	Unclear	Yes	Unclear	Unclear
Lenler-Peterson 1982 (retrospective)	45 (151 concurrent tests)	LFT: galactose tolerance test	Consecutive sample, all receiving MTX and having developed fibrosis	Yes	Yes – only included those known to have developed fibrosis	Unclear^(c)	Pre-defined	Unclear	Unclear	Unclear
Newman 1989 (retrospective)	168 (364 biopsies paired with LFTs, 85 before treatment)	LFTs: ALT, AST, bili, AP, PT, alb	Consecutive sample, before and during MTX	Unclear if all analysed	No	3 days	Unclear	Yes	Unclear (but IRR of 3 assessors checked)	Unclear
O’Connor 1989 (retrospective)	78 (147 biopsies paired with LFTs; 52 before and 95 after treatment)	LFTs: AST, bili, AP	Unclear sampling, all had used MTX (normal pre-Tx biopsy)	No	No	Maximum 1 week	Pre-defined normal ranges	Yes	Unclear	Unclear
Paramsothy 1988 (prospective)	15	LFTs: AST, bili, AP, alb, GGT	Unclear sampling, all had used MTX	Yes	No	Unclear	Pre-defined normal ranges	Yes	Unclear	Unclear
Liver scintigraphy and ultrasound scans
Geronemus 1982 (retrospective)	24	Liver scintigraphy: Tc 99m sulphur colloid scan	Unclear sampling, all had long-term MTX use	Yes	No	Maximum 2 months	Pre-specified	Unclear	Unclear	Unclear
McHenry 1992 (retrospective)	63 (87 paired results)	Liver scintigraphy: Tc 99m sulphur colloid scan	Consecutive sample, before and during MTX	No	No	Maximum 4 weeks	Pre-specified	Unclear	Unclear	Unclear
Mitchell 1987 (prospective)	49	Liver scintigraphy: Tc 99m sulphur colloid scan Ultrasound	Unclear sampling, all had long-term MTX use	No	No	1 day	Pre-specified	Unclear	Yes	Unclear
Coulson 1987 (prospective)	28 (54 paired tests)	Ultrasound	Unclear sampling, before and during MTX	No	No	Maximum 1 month	Pre-specified	Yes	Yes for ultrasound; unclear for biopsy	5 μm sections
PIIINP
Boffa 1996 (prospective)	87 (147 paired tests)	PIIINP	Unclear sampling, all had long-term MTX use Note: unclear proportion with PsA	No	No	<1 day	Pre-specified ^(d)	Yes	Unclear	Unclear
Maurice 2005 (retrospective)	34 (46 biopsies with 2–6 assays per biopsy)	PIIINP	Consecutive sample, all receiving MTX Note: 22% had inflamm. arthritis	No	No	Maximum 6 months	Pre-specified ^(d)	Yes	Unclear	18 gauge needle
Zachariae 1989 & Risteli 1987 (prospective)	73	PIIINP	Consecutive sample, all receiving MTX (≥6 months) Note: 45.8% of pilot group had PsA	Yes	No	Unclear	Pre-specified ^(d)	Yes	Unclear	Unclear
Zachariae 2001 (retrospective)	70 (189 biopsies and 329 assays)	PIIINP	Unclear sampling, all had MTX use and normal initial biopsy and PIIINP Note: 38.6% had PsA	Yes	No	69/70 had ≥3 analyses within a year around the time of biopsy	Pre-specified ^(d)	Unclear	Unclear	Unclear
Fibrotest and Fibroscan
Berends 2007 (retrospective)	24	Fibrotest Fibroscan - used median value of successful readings on the same day	Unclear sampling	Yes	No	≤18 months	Pre-specified	Yes	Yes	Variable (only one had <10 portal tracts)

: Alb: albumin; ALT: alanine aminotransferase; AP: alkaline phosphatase; AST: aspartate aminotransferase; bili: bilirubin; GGT: gamma-glutamyl transferase; IRR: inter-rater reliability; LFT: liver function tests; MTX: methotrexate; PIIINP: aminoterminal peptide of type III procollagen; PsA: psoriatic arthritis; PT: prothrombin time; Tc 99m: Technetium-99m isotope; Tx: treatment

a: Verification bias = did all patients in the studies received the same comparison tests, regardless of initial results

b: Clearly, if the time between the index test and the reference standard is too long it is possible that any discrepancy in findings is not accounted for by inaccuracy in the index test but rather y the clinical status of the participant having changed in the intervening period. However, the time for progression to fibrosis is unclear and any cut-off for a maximum time between tests would be arbitrary; therefore, all studies were included regardless of time between tests, although this will be considered as a risk of bias

c: Study methods state that participants were admitted at 1-year intervals for biopsy and galactose test, which implies they were performed on the same day

d: The threshold for an abnormal PIIINP assay was >4.2 μg/l (based on the reference range in Finnish blood donors); however, the manufacturer’s information leaflet states that the reference range is 2.3–6.4 μg/l based on PIIINP values of apparently healthy adults (19–65 years) although variations in population demographics may mean that slightly different reference limits apply across populations.

Table 157Modified GRADE profile for the diagnostic accuracy of tools to detect liver fibrosis or cirrhosis

Study characteristics			Quality Assessment					Summary of findings
No. of studies	Design	N	Limitation	Inconsistency	Indirectness	Imprecision^*	Other consideration	Pre-test probability	Sensitivity	Specificity	Post-test probability positive (if positive result)	Post-test probability negative (if negative result)	Quality
LFTs vs biopsy
AST
Newman 1989	Retrospective	168	VS^a	N^b	N	N		Unclear for full group	20 (13–30)%	90 (84–93)%	49 (33–65)%	70 (62–76)%	⊕⊕○○ LOW
O’Connor 1989	Retrospective	50 tests	S^c	N^b	N	VS	Pre-treatment	9.6%	40 (5–85)%	89 (76–96)%	29 (4–71)%	93 (81–99)%	⊕○○○ VERY LOW
O’Connor 1989	Retrospective	47 (86 tests)	S^c	N^b	N	VS	Post-treatment	24.2%	43 (22–66)%	86 (75–93)%	50 (26–74)%	82 (71–91)%	⊕○○○ VERY LOW
Paramsothy 1988	Prospective	15	VS^d	N	N	VS		46.7%	29 (4–71)%	100 (63–100)%	100 (21–100)%	62%	⊕○○○ VERY LOW
ALT
Newman 1989	Retrospective	168	VS^a	N^b	N	S		Unclear for full group	5 (0.6–17)%	85 (72–94)%	22 (3–48)%	52 (40–63)%	⊕○○○ VERY LOW
Ho 1986	Prospective	18	VS^e	N	S^f	VS	TH >32 U/l	27.8%	40 (7.9–71.3)%	84.6 (72.3–96.7)%	50 (9.8–89.2)%	78.6 (67.1–89.8)%	⊕○○○ VERY LOW
Ho 1986	Prospective	18	VS⁵	N	S^f	VS	TH >40 U/l	27.8%	40 (8.0–58.9)%	92.3 (80.0–99.6)%	66.7 (13.4–98.2)%	80.0 (69.3–86.3)%	⊕○○○ VERY LOW
Bilirubin
Newman 1989	Retrospective	168	VS^a	N^b^,^g	N	N	TH ≥2 μmol/l	Unclear for full group	19 (12–29)%	86 (80–90)%	41 (26–57)%	60 (63–75)%	⊕⊕○○ LOW
O’Connor 1989	Retrospective	50 tests	S^c	N^b^,^g	N	VS	Pre-treatment	9.6%	20 (7–72)%	96 (85–99)%	33 (1–91)%	91 (80–98)%	⊕○○○ VERY LOW
O’Connor 1989	Retrospective	47 (86 tests)	S^c	N^b^,^g	N	S	Post-treatment	24.2%	10 (2–30)%	95 (87–99)%	40 (5–85)%	76 (65–85)%	⊕⊕○○ LOW
Paramsothy 1988	Prospective	15	VS^d	N^g	N	VS	TH ≥18 μmol/l	46.7%	0 (0–41)%	88 (47–100)%	0 (0–87)%	50 (41–58)%	⊕○○○ VERY LOW
Alkaline phosphatase
Newman 1989	Retrospective	168	VS^a	N^b	N	N		Unclear for full group	38 (28–49)%	71 (63–77)%	39 (28–49) %	70 (63–77) %	⊕⊕○○ LOW
O’Connor 1989	Retrospective	50 tests	S^c	N^b	N	VS	Pre-treatment	9.6%	40 (5–85)%	77 (60–87)%	15 (2–45)%	92 (83–97)%	⊕○○○ VERY LOW
O’Connor 1989	Retrospective	47 (86 tests)	S^c	N^b	N	S	Post-treatment	24.2%	57 (34–78)%	72 (60–83)%	40 (23–59)%	84 (72–92)%	⊕⊕○○ LOW
Paramsothy 1988	Prospective	15	VS^d	N	N	VS		46.7%	42.9 (14.1–65.6)%	75.0 (49.9–94.9)%	60.0 (19.8–91.9)%	60.0 (39.9–75.9)%	⊕○○○ VERY LOW
Prothrombin time
Newman 1989	Retrospective	168	VS^a	N^b	N	N		Unclear for full group	1 (0–5) %	99 (94–99) %	25 (6–80) %	66 (61–72) %	⊕⊕○○ LOW
Albumin
Newman 1989	Retrospective	168	VS^a	N^b	N	N	TH ≥35 g/l	Unclear for full group	19 (11–29)%	76 (68–83)%	33 (19–48) %	61 (52–68) %	⊕⊕○○ LOW
Paramsothy 1988		15	VS^d	N	N	VS	TH ≥150 u/l	46.7%	29 (4–71)%	100 (63–100)%	100 (21–100)%	62 %	⊕○○○ VERY LOW
Gamma-glutamyl transferase
Paramsothy 1988	Prospective	15	VS^d	N	N	VS		42.9%	33.3 (6.7–65.8)%	62.5 (42.5–86.8)%	40 (8.0–79.0)%	55.6 (37.8–77.2)%	⊕○○○ VERY LOW
Galactose tolerance test
Lenler-Peterson 1982 1989	Retrospective	45	VS^h	N	Sⁱ	N		69.5%	14.3 (10.2–16.4)%	93.5 (84.1–98.3)%	83.3 (59.5–95.5)%	32.3 (29.1–34.0)%	⊕○○○ VERY LOW
Scintigraphy vs biopsy
Geronemus 1982	Retrospective	24	VS^j	N^b^,^k	N	VS		29.2%	57.1 (22.7–86.7)%	64.7 (50.5–76.9)%	40.0 (15.9–60.7)%	78.6 (61.3–93.3)%	⊕○○○ VERY LOW
McHenry 1992	Retrospective	63	VS^l	N^k^,^m	N	S		6.9%	83.3 (38.0–99.1)%	81.5 (78.1–82.6)%	25.0 (11.4–29.7)%	98.5 (94.4–99.9)%	⊕○○○ VERY LOW
Mitchell 1987	Prospective	49	VSⁿ	N^k	N	S		24.5%	50.0 (24.2–74.9)%	73.0 (64.6–81.1)%	37.5 (18.2–56.2)%	81.8 (72.4–90.9)%	⊕○○○ VERY LOW
Ultrasound vs biopsy
Mitchell 1987	Prospective	49	VSⁿ	N^k	N	VS		24.5%	0%	86%	0%	73%	⊕○○○ VERY LOW
Coulson 1987	Prospective	28	S^o	N^k	N	S	Any fibrosis	48.2%	19.0 (7–39)%	100 (88–100)%	100 (39–100)%	57%	⊕⊕○○ LOW
Coulson 1987	Prospective	28	S^o	N^k^,^m	N	VS	Portal fibrosis	37.0%	25.0 (9–49)%	100.0 (90–100)%	100% (39–100)%	69%	⊕○○○ VERY LOW
PIIINP vs biopsy
Boffa 1996	Prospective	87	S^o	N	N	N	Paired tests	24.1%	81.0 (60.3–93.5)%	63.6 (57.1–67.6)%	41.5 (30.9–47.9)%	91.3 (81.9–97.0)%	⊕⊕⊕○ MODERATE
Zachariae 2001	Retrospective	70	VS^p	N	S^q	VS	Serial PIIINP assays	5.8%	100 (40–100)%	97 (89–100)%	66 (30–84)%	100%	⊕○○○ VERY LOW
Maurice 2005	Retrospective	34	S^r	N^b	N^s	N	Serial PIIINP assays	13.7%	62.5 (42.1–79.8)%	67.5 (64.3–70.3)%	23.4 (15.8–29.9)%	91.9 (87.5–95.6)%	⊕⊕⊕○ MODERATE
Zachariae 1989 and Risteli 1988	Prospective	73	VS^t	N	N	N	Paired tests	34.7%	76.0 (61.8–79.8)%	97.9 (90.3–99.9)%	95.0 (77.2–99.7)%	88.5 (81.6–90.3)%	⊕⊕○○ LOW
		13	VS^u	N	N	VS	No-PsA	69.2%	33.0 (7.0–70)%	100 (40–100)%	100 (33–100)%	40%	⊕○○○ VERY LOW
		10	VS^u	N	N	VS	PsA	40%	100 (40–100)%	83 (36–100)%	80 (40–92)%	100%	⊕○○○ VERY LOW
Fibrotest
Berends 2007	Retrospective	24	S^v	N	N^w	VS		25%	83.3 (40.8–99.1)%	61.1 (46.9–66.4)%	41.7 (20.4–49.6)%	91.7 (70.4–99.6)%	⊕○○○ VERY LOW
Fibroscan
Berends 2007	Retrospective	24	VS^x	N	N^w	VS^y		25%	50 (0.07–0.93)%	88 (0.62–0.98)%	33%	86%	⊕○○○ VERY LOW

*: Imprecision is assessed based on the sensitivity, specificity PPV and NPV of the tests; if there was no majority in the assessment of imprecision across these statistics higher weighting was given to sensitivity and NPV as these are most important for the intended role of the test.

a: Unclear threshold selection; unclear if all patients included in the analysis or received both tests; experience of pathologist and adequacy of biopsy specimen unclear

b: Note that biopsy grading was according to Roenigk (threshold does not include fibrous expansion of portal tracts without extension to the parenchyma and fibrosis not associated with the portal tracts is not scored at all; therefore, NAFLD which may be associated with MTX use will not be detected on this score)

c: Unclear sampling and unclear baseline characteristics; not all patients were included in the analysis due to incomplete data sets/not receiving both tests; experience of pathologist and adequacy of biopsy specimen unclear

d: Unclear sampling; unclear time between tests; experience of pathologist and adequacy of biopsy specimen unclear

e: Unclear patient selection method; unclear time between tests; experience of pathologist and adequacy of biopsy specimen unclear

f: Only included those with an indication of liver damage (either by cumulative dose of methotrexate or raised ALT levels)

g: Thresholds for abnormal enzyme test varied between studies

h: Unclear baseline characteristics; time between tests unclear; unclear if biopsy assessed blinded to clinical and laboratory data; experience of pathologist and adequacy of biopsy specimen unclear

i: Population limited to those known to have developed fibrosis or cirrhosis

j: Unclear if selection was based on a consecutive sample; unclear if tests were interpreted by blinded assessors and unclear who made the assessments; adequacy of biopsy specimen unclear

k: Definition of abnormal result on scan varies between studies

l: Adequacy of biopsy specimen unclear

m: Note that the threshold biopsy grading for abnormal reference test result was at least moderate fibrosis, which corresponded to portal fibrosis consistent with Roenigk grade III

n: Unclear if selection was based on a consecutive sample; unclear if tests were interpreted by blinded assessors and experience of pathologist assessing biopsy unclear; adequacy of biopsy specimen unclear

o: Unclear if selection was based on a consecutive sample; experience of biopsy assessor and adequacy of biopsy specimen unclear

p: Unclear if selection was based on a consecutive sample; experience of biopsy assessor and adequacy of biopsy specimen unclear; unclear blinding of biopsy assessor and unclear order of tests

q: Serial analyses of PIIINP were performed; therefore not a 1:1 relationship with biopsies. Those who tested positive on either test could also have had several negative tests

r: Experience of biopsy assessor and adequacy of biopsy specimen unclear

s: Serial analyses of PIIINP were performed; therefore not a 1:1 relationship with biopsies but data on all assays included (so some biopsies were counted more than once as paired with multiple PIIINP assay results)

t: Unclear if selection was based on a consecutive sample; experience of biopsy assessor and adequacy of biopsy specimen unclear; unclear order and timing between tests

u: Subgroup analysis of pilot group only; unclear if selection was based on a consecutive sample; experience of biopsy assessor and adequacy of biopsy specimen unclear; unclear order and timing between tests

v: Unclear if selection was based on a consecutive sample; maximum time between tests was 18 months

w: Biopsy grading was classed as abnormal if it was Metavir grade F2 or greater (threshold does not include fibrous expansion of portal tracts without septa and fibrosis not associated with the portal tracts is not scored at all, similar to the Roenigk score)

x: Unclear if selection was based on a consecutive sample; maximum time between tests was 18 months; uncertainty in how the diagnostic test accuracy statistics were calculated (unable to reconcile with 2×2 table)

y: No estimate of imprecision available from the paper

Table 158Summary statistics for diagnostic accuracy of tools for fibrosis and cirrhosis

Study	N	Index test threshold	Reference test threshold	Pre-test probability	Sensitivity	Specificity	PPV Value-added PPV	NPV Value-added NPV	Post-test probability of PsA despite test −ve	Positive likelihood ratio (LR+)	Negative likelihood ratio (LR−)
LFTs vs biopsy
AST
Newman 1989	168	≥40 U/L	Roenigk grade 3–4	Unclear for full group	20 (13–30)%	90 (84–93)%	49 (33–65)%	70 (62–76)%	30%	NA	NA
O’Connor 1989 – pre-treatment	50	Unclear (based on ‘normal ranges’)	Roenigk grade 3–4	9.6%	40 (5–85)%	89 (76–96)%	29 (4–71)% 19.4%	93 (81–99)% 2.6%	7%	3.76 (0.97–15)	0.67 (0.33–1.38)
O’Connor 1989 – post-treatment	47 (86 tests)	Unclear (based on ‘normal ranges’)	Roenigk grade 3–4	24.2%	43 (22–66)%	86 (75–93)%	50 (26–74)% 25.8%	82 (71–91)% 6.2%	18%	3.13 (1.49–6.56)	0.66 (0.45–0.95)
Paramsothy 1988	15	≥40 U/L	Fibrosis (any severity)	46.7%	29%	100%	100 (21–100)% 53.3%	62% 8.7%	38%	Infinity (0.31–101)	0.71 (0.44–1.19)
ALT
Newman 1989	168	≥40 U/L	Roenigk grade 3–4	Unclear for full group	5 (0.6–17)%	85 (72–94)%	22 (3–48)%	52 (40–63)%	48%	NA	NA
Ho 1986	18	>32 U/L As defined in study	Fibrosis (septum formation)	27.8%	40 (7.9–71.3)%	84.6 (72.3–96.7)%	50 (9.8–89.2)% 22.2%	78.6 (67.1–89.8)% 6.4%	21.4%	2.60 (0.49–14)	0.71 (0.33–1.50)
Ho 1986	18	>40 U/L consistent with other studies	Fibrosis (septum formation)	27.8%	40 (8.0–58.9)%	92.3 (80.0–99.6)%	66.7 (13.4–98.2)% 38.9%	80.0 (69.3–86.3)% 7.8%	20.0%	5.20 (0.60–45)	0.65 (0.31–1.35)
Bilirubin
Newman 1989	168	≥2 μmol/l	Roenigk grade 3–4	Unclear for full group	19 (12–29)%	86 (80–90)%	41 (26–57)%	60 (63–75)%	40%	NA	NA
O’Connor 1989 – pre-treatment	50	Unclear (based on ‘normal ranges’)	Roenigk grade 3–4	9.6%	20 (7–72)%	96 (85–99)%	33 (1–91)% 23.4%	91 (80–98)% 0.6%	9%	4.7 (0.51–43)	0.84 (0.54–1.30)
O’Connor 1989 – post-treatment	47 (86 tests)	Unclear (based on ‘normal ranges’)	Roenigk grade 3–4	24.2%	10 (2–30)%	95 (87–99)%	40 (5–85)% 15.8%	76 (65–85)% 0.2%	24%	1.57 (0.31–8.00)	0.97 (0.84–1.11)
Paramsothy 1988	15	≥18 μmol/l	Fibrosis (any severity)	46.7%	0%	88%	0 (0–87)% −46.7%	50 (41–58)% −3.3%	50%	0	1.14 (0.80–1.58)
Alkaline phosphatase
Newman 1989	168	≥100 U/L	Roenigk grade 3–4	Unclear for full group	38 (28–49)%	71 (63–77)%	39 (28–49) %	70 (63–77) %	30%	NA	NA
O’Connor 1989 – pre-treatment	50	Unclear (based on ‘normal ranges’)	Roenigk grade 3–4	9.6%	40 (5–85)%	77 (60–87)%	15 (2–45)% 5.4%	92 (83–97)% 1.6%	8%	1.71 (0.52–5.63)	0.78 (0.38–1.63)
O’Connor 1989 – post-treatment	47 (86 tests)	Unclear (based on ‘normal ranges’)	Roenigk grade 3–4	24.2%	57 (34–78)%	72 (60–83)%	40 (23–59)% 15.8%	84 (72–92)% 8.2%	16%	2.03 (1.21–3.41)	0.6 (0.37–0.98)
Paramsothy 1988	15	≥121 u/l	Fibrosis (any severity)	46.7%	42.9 (14.1–65.6)%	75.0 (49.9–94.9)%	60.0 (19.8–91.9)% 13.3%	60.0 (39.9–75.9)% 6.7%	40.0%	1.71 (0.39–7.48)	0.76 (0.36–1.62)
Prothrombin time
Newman 1989	168	≥14.5 s	Roenigk grade 3–4	Unclear for full group	1 (0–5) %	99 (94–99) %	25 (6–80) %	66 (61–72) %	34%	NA	NA
Albumin
Newman 1989	168	≥35 g/l	Roenigk grade 3–4	Unclear for full group	19 (11–29)%	76 (68–83)%	33 (19–48) %	61 (52–68) %	39%	NA	NA
Paramsothy 1988	15	≥150 u/l	Fibrosis (any severity)	46.7%	29%	100%	100 (21–100)% 53.3%	62 % 8.7%	38%	Infinity (0.31–101)	0.71 (0.44–1.19)
Gamma-glutamyl transferase
Paramsothy 1988	15	≥36 u/l	Fibrosis (any severity)	42.9%	33.3 (6.7–65.8)%	62.5 (42.5–86.8)%	40 (8.0–79.0)% −2.9%	55.6 (37.8–77.2)% −1.5%	44.4%	0.89 (0.21–3.76)	1.07 (0.49–2.33)
Galactose tolerance test
Lenler-Peterson 1982 1989	45 (151 concurrent test)	≥3 g/l	Fibrosis (unclear classification)	69.5%	14.3 (10.2–16.4)%	93.5 (84.1–98.3)%	83.3 (59.5–95.5)% 13.8%	32.3 (29.1–34.0)% 1.8%	67.7%	2.19 (0.67–7.20)	0.92 (0.82–1.02)
Scintigraphy vs biopsy
Geronemus 1982	24	Presence of abnormalities^(a)	Roenigk grade 3–4	29.2%	57.1 (22.7–86.7)%	64.7 (50.5–76.9)%	40.0 (15.9–60.7)% 10.8%	78.6 (61.3–93.3)% 7.8%	21.4%	1.62 (0.65–4.02)	0.66 (0.26–1.67)
McHenry 1992	63 (87 paired results)	Portal contribution <50%	Portal fibrosis	6.9%	83.3 (38.0–99.1)%	81.5 (78.1–82.6)%	25.0 (11.4–29.7)% 18.8%	98.5 (94.4–99.9)% 5.4%	1.5%	4.50 (2.52–8.04)	0.20 (0.03–1.23)
Mitchell 1987	49	Presence of abnormalities^(b)	Fibrosis (any severity)	24.5%	50.0 (24.2–74.9)%	73.0 (64.6–81.1)%	37.5 (18.2–56.2)% 13.0%	81.8 (72.4–90.9)% 5.3%	19.2%	1.85 (0.85–4.02)	0.69 (0.38–1.25)
Ultrasound vs biopsy
Mitchell 1987	49	Presence of abnormalities^(c)	Fibrosis (any severity)	24.5%	0%	86%	0% −24.5%	73% −2.5%	27%	0	1.16 (0.95–1.33)
Coulson 1987	28 (58 paired observations)	Presence of abnormalities^(d)	Fibrosis (any severity)	48.2%	19.0%	100%	100 (39–100)% 51.8%	57% 5.2%	43%	Infinity (0.69–204)	0.81 (0.67–0.99)
Coulson 1987	28 (58 paired observations)	Presence of abnormalities^(d)	Fibrosis (at least moderate – portal fibrosis^(e))	37.0%	25.0%	100.0%	100% (39–100)% 63.0%	69% 6.0%	31%	Infinity (1.07–315)	0.75 (0.58–0.97)
PIIINP vs biopsy
Boffa 1996	87 (147 paired tests)	>4.2 ng/ml	Fibrosis	24.1%	81.0 (60.3–93.5)%	63.6 (57.1–67.6)%	41.5 (30.9–47.9)% 17.4%	91.3 (81.9–97.0)% 15.4%	8.7%	2.23 (1.52–3.26)	0.30 (0.12–0.74)
Zachariae 2001	70 (189 biopsies and 329 assays)	>4.2 ng/ml	Fibrosis (any severity)	5.8%	100%	97%	66 (30–84)% 60.2%	100% 5.8%	0%	32 (6.80–83)	0 (0.01–1.44)
Maurice 2005	34 (70 biopsies and 306 assays)	>4.2 ng/ml	Roenigk grade 3–4	13.7%	62.5 (42.1–79.8)%	67.5 (64.3–70.3)%	23.4 (15.8–29.9)% 9.7%	91.9 (87.5–95.6)% 5.6%	8.1%	1.93 (1.31–2.83)	0.56 (0.33–0.94)
Zachariae 1989 and Risteli 1982	73	>4.2 ng/ml	Fibrosis (any severity)	34.7%	76.0 (61.8–79.8)%	97.9 (90.3–99.9)%	95.0 (77.2–99.7)% 60.3%	88.5 (81.6–90.3)% 23.2%	11.5%	36 (5.07– 251)	0.25 (0.12–0.49)
Risteli 1982 – no PsA subgroup	13	>4.2 ng/ml	Fibrosis (any severity)	69.2%	33.0 (7.0–70)%	100 (40–100)%	100 (33–100)% 30.8%	40% 9.2%	60%	Infinity	0.67
Risteli 1982 – PsA subgroup	10	>4.2 ng/ml	Fibrosis (any severity)	40%	100 (40–100)%	83 (36–100)%	80 (40–92)% 40%	100% 40%	0%	6.00 (0.99,18 )	0.00 [0.01,1.82)
Fibrotest
Berends 2007	24	>0.31	≥F2 on Metavir system	25%	83.3 (40.8–99.1)%	61.1 (46.9–66.4)%	41.7 (20.4–49.6)% 16.7%	91.7 (70.4–99.6)% 16.7%	8.3%	2.14 (1.08,4.23)	0.27 (0.04,1.69)
Fibroscan
Berends 2007	24	>7.1 kPa	≥F2 on Metavir system	25% (20% of those assessable )	50%	88%	33% ?	86% ?	14%	-	-

: NA: Not available

: NPV: Negative predictive value

: PPV: Positive predictive value

a: The abnormalities assessed were heterogeneous uptake, hepatomegaly, extra hepatic uptake and focal defects

b: The abnormalities assessed were size of liver and spleen, pattern of uptake in these organs and degree of extrahepatic uptake

c: The abnormalities assessed were liver size, shape, echo pattern and information about the biliary and vascular system according to a standard proforma

d: The abnormalities assessed were fatty change and fibrosis; only those showing fibrosis were counted as positive tests

e: This is in accordance with Roenigk criteria

Table 159Serial PIIINP versus routine liver biopsy – Economic study characteristics

Study	Limitations	Applicability	Other comments
Chalmers 2005	Very serious limitations ^(g)	Partially applicable ^(h)	Cost analysis conducted alongside a multicentre prospective audit in UK and Ireland Costs included biopsy, overnight hospital stay, histology, PIIINP analysis

g: Given that treatment with methotrexate may continue for more than 2 years, time horizon may be insufficient. Does not report incidence of adverse events/complications associated with liver biopsy and any effect on costs. Within trial analysis and so does not incorporate all available evidence on differences between monitoring methods but results appear consistent with results of clinical review.

h: QALYs not used (cost consequence analysis).

Table 160Serial PIIINP versus routine liver biopsy – Economic summary of findings

Study	Incremental cost	Incremental effects	ICER	Uncertainty
Chalmers 2005	£25 ^(a)	Fewer liver biopsies per patient per year; fewer normal biopsies	PIIINP is more costly, but reduces number of biopsies (normal and abnormal) performed	Whether serial PIIINP with selective liver biopsy was more or less costly than routine liver biopsy was dependent on the unit cost of liver biopsy
Chalmers 2005	−£49 ^(b)		PIIINP is less costly and reduces number of biopsies (normal and abnormal) performed

a: Where PIIINP measurement costs £22.50 and liver biopsy costs £270.00 (Essex)

b: Where PIIINP measurement costs £22.50 and liver biopsy costs £577.00 (Manchester)

Table 161Unit costs of monitoring tests – exclusive of labour costs

Item	Unit Cost	Notes
Liver function tests	£4.12 per batch	Shepherd and colleagues 2006³⁶⁷
Liver scintigraphy	£180?	HRG RA36Z (Nuclear medicine – category 2) Other categories range from £170 to £700
Liver ultrasound	£53	HRG RA23Z; average of outpatient, direct access and other categories of care
PIIINP	£21.64	Woolacott and colleagues 2006⁴²⁷
Liver biopsy	Elective inpatient: £816 Day case: £553	HRG GB04Z; NHS Reference Costs

: Source: NHS Reference Costs 2009–10⁷⁴

Recommendations on methotrexate and monitoring for hepatotoxicity	Methotrexate and monitoring for hepatotoxicity 93. Before and during methotrexate treatment, offer the person with any type of psoriasis an evaluation for potential risk of hepatotoxicity. Use standard liver function tests and serial serum procollagen III levels to monitor for abnormalities during treatment with methotrexate, taking into account pre-existing risk factors (for example obesity, diabetes and alcohol use), baseline results and trends over time. 94. When using serum procollagen III levels to exclude liver fibrosis or cirrhosis, be aware that the: test cannot be used in children and young people results may be unreliable in people with psoriatic arthritis estimated positive predictive value is 23–95% and the estimated negative predictive value is 89–100%. 95. Provide advice on modifiable risk factors for liver disease prior to and during therapy, including alcohol intake and weight reduction if appropriate in line with ‘Alcohol-use disorders: preventing harmful drinking’ (NICE public health guidance 24), and ‘Obesity’ (NICE clinical guideline 43). For further advice on how to support attitude and behavioural change see ‘Behaviour change’ (NICE public health guidance 6). 96. Seek timely specialist advice and consider referral to a clinician with expertise in liver disease if the results of liver tests are abnormal.
Future research recommendations	25. What is the clinical utility and validity of non-invasive markers of liver fibrosis (for example, FibroScan, FibroTest and ultrasound) in people with psoriasis receiving methotrexate or other treatment interventions?
Relative values of different outcomes	Standard accuracy outcomes for diagnostic tests were looked for: Sensitivity and specificity Positive predictive value (PPV) Negative predictive value (NPV) Likelihood ratios. The GDG felt the most important characteristics of a test for dermatology use (i.e. for use as a screening test) are: Very good accuracy to rule out those who do not have liver damage and refer all who may have the disease for specialist hepatology assessment, so that no true cases are missed (high sensitivity, NPV and LR−). Reasonable accuracy for ruling in a diagnosis, to avoid wasting resources by making inappropriate referrals to hepatology (specificity, PPV and LR+). The test should also be practical for use in a dermatology setting.
Trade off between clinical benefits and harms	The GDG agreed not to recommend scintigraphy on the grounds that it is impractical and involves a radioactive isotope.
Economic considerations	Limited evidence was available to inform the GDG about the cost- effectiveness of alternative methods for monitoring hepatotoxicity associated with methotrexate treatment. One costing study showed that serial testing with procollagen III N-terminal propeptide (PIIINP) and selective liver biopsy was likely to be cost saving compared to routine liver biopsy if the cost of liver biopsy was less than £375. NHS reference costs from 2009–10 indicate that liver biopsy as a day case procedure costs £553; therefore, the GDG concluded that it is highly likely that PIIINP with selective liver biopsy is likely to be the optimal monitoring strategy for patients taking methotrexate. The GDG also considered that the addition of liver function tests to PIIINP is unlikely to add significant costs and may improve the identification of patients needing further investigation. In addition to these cost considerations, the GDG considered the risk of serious harm associated with liver biopsy (1–2% risk of injury; 0.01–0.1% risk of mortality). They considered that the potential risk of missing significant liver damage in patients monitored with serial PIIINP to be outweighed by the risks, costs and inconvenience of performing routine liver biopsy on all patients. The GDG discussed the importance of getting these monitoring methods right, as the other treatments available to patients with moderate to severe psoriasis are increasingly toxic and/or costly. Reducing the number of people who are being successfully managed by methotrexate, a very cost-effective treatment, who have a false positive test result and thus move on to more toxic and/or costly strategies could result in a more efficient use of NHS resources.
Quality of evidence	The GDG noted important variables between the studies: There were differences in whether all participants recruited had a known or suspected diagnosis of liver disease. Prevalence of fibrosis and cirrhosis varied from 6.9% to 69%. Unit of analysis: there was variation in whether the study reported one set of paired tests or more than one of each of the tests per patient: Eight studies used only one index test and one reference standard per person Three studies included multiple paired index and reference tests per person (Coulson, Lenler, McHenry) One study included only single paired tests before methotrexate but multiple paired tests after methotrexate (O’Connor) In two studies it was unclear whether the results were based upon single tests or multiple paired tests per person (Berends, Newman) One study included more than one index and reference test per patient, and the biopsy was paired with more than one index test (Maurice) Different scales were used to assess severity of fibrosis on a liver biopsy and it is not possible to map them to a common scale. Study limitations: Multiple tests per patient could introduce bias by weighting towards those with multiple biopsies (which could be because there was an indication of abnormal liver function or because they had been receiving methotrexate for longer; but it could also be that those who develop abnormal liver function are taken off methotrexate which would bias the results in the other direction). However, there was no clear/consistent impact of different unit of analysis on results. The GDG noted that multiple PIIINP tests are standard, as three or four per year are advised for the purposes of sensitivity. The GDG also noted that although liver biopsy is used as the gold standard: It is associated with sampling error and different results can be seen depending on where the sample is taken from in the liver; so if sampling was inadequate the result may misrepresent the true state of the liver Some classification schemes may not detect non-alcoholic fatty liver disease (NAFLD) Possible inadequate grading and diverse classification schemes. Experience of the person assessing the sample is important and this was not reported in the majority of the studies. For index tests assessments, there was unclear reporting of methods and different definitions of abnormal were used. Most studies were retrospective and population sampling methods were unclear (i.e. they may not have used consecutive or random sampling, and difficult to diagnose cases could have been excluded thus introducing bias). The time between tests was also unclear in a number of studies. The GRADE rating for most results was low/very low quality evidence. Findings for fibrotest and fibroscan were based on a small study and so were insufficient to base a recommendation upon; this is an area for future research. Ultrasound is very specific but there are only two studies which are very old and may not reflect current ultrasound technology. Further research into ultrasound is desired and the GDG agreed to make a future research recommendation for ultrasound.
Other considerations	The GDG discussed the psoriatic arthritis (PsA) population and whether PIIINP is useful for this group. Serum procollagen III is cleaved off from collagen when fibrotic tissue is broken down. The PIIINP assay is not liver specific. Therefore the test may be less useful in people with arthritis, as the result could be elevated due to arthritis not liver damage. Fibroscan is currently a research tool and its use is not widespread in dermatology practice (although it is used in hepatology departments). For people with a BMI >30, a special probe is needed for fibroscan, which costs an additional £30K. The data on LFTs is counter-intuitive, and the GDG discussed whether some of the tests should not be used. The GDG only looked at the endpoints of fibrosis and cirrhosis. LFTs detect other issues including idiosyncratic hepatotoxic reaction to methotrexate, non- alcoholic fatty liver disease and excessive alcohol consumption. Therefore the GDG agreed not to make a ‘do not use’ recommendation for any of the LFTs. The GDG wished to capture people in the recommendations who have serial abnormal test results over time due to development of fibrosis. Acute hepatotoxic reaction to methotrexate would be detected by an acute rapidly rising abnormality of LFTs. The GDG agreed to recommend that people with psoriasis taking methotrexate should be assessed for fibrosis or cirrhosis using PIIINP and LFTs. People with serial abnormal results should be referred for specialist opinion to assess risks and benefits of continuing methotrexate. Where specialist opinion is required it could be sought from a hepatologist or gastroenterologist in view of the potential lack of availability of hepatologists in some areas. The GDG debated whether methotrexate should be stopped while waiting for a referral. Stopping treatment for three months in someone with severe disease and/or with arthritis could have devastating consequences. There is not an urgent need for expedient referrals in this group; the group in which urgent referral would be needed is people with bone marrow failure. Therefore no recommendation was made about stopping methotrexate while waiting for specialist appointment. People with psoriasis have comorbidities that may predispose them to abnormal liver function, such as obesity and diabetes. This does not preclude the use of methotrexate, but it is extremely important to test liver function prior to therapy and monitor during therapy in case fibrosis develops in this high risk population. Methotrexate is known to be a hepatotoxic drug in the short term (at least) and certain factors especially prevalent in people with severe psoriasis (diabetes, obesity, alcohol-related morbidity) are also associated with liver dysfunction. The GDG therefore agreed that the recommendations needed to highlight that when using methotrexate, any clinical factors that might impact on liver function should be taken into account, and that abnormalities that develop need to be considered in this context and advice given on safe alcohol intake.