Non-invasive risk scores do not reliably identify future cirrhosis or hepatocellular carcinoma in Type 2 diabetes: The Edinburgh type 2 diabetes study

Background: The incidence of cirrhosis and hepatocellular carcinoma (HCC) is increased in Type 2 diabetes, primarily secondary to non-alcoholic fatty liver disease (NAFLD). European guidelines recommend screening for NAFLD in Type 2 diabetes. American guidelines, while not advocating a screening protocol, suggest using noninvasive markers of fibrosis for risk-stratification and guiding onward referral. Aims: To test the ability of individual fibrosis scores and the European screening algorithm to predict 11-year incident cirrhosis/HCC in an asymptomatic community cohort of older people with Type 2 diabetes. Methods: The Edinburgh Type 2 Diabetes Study investigated men and women with Type 2 diabetes (n = 1066, aged 60–75 at baseline). Liver markers were measured at baseline and year 1; steatosis and fibrosis markers were calculated according to independently published calculations. During 11 years of follow-up, cases of cirrhosis and HCC were identified. Results: Forty-three out of 1059 participants with no baseline cirrhosis/HCC developed incident disease. All scores were significantly associated with incident liver disease by odds ratio ( P < .05). The ability of the risk-stratification tools to accurately identify those who developed incident cirrhosis/HCC 2


| INTRODUC TI ON
People with Type 2 diabetes have a higher incidence of cirrhosis and hepatocellular carcinoma (HCC) than the general population. [1][2][3] The commonest cause of liver disease in Type 2 diabetes is non-alcoholic fatty liver disease (NAFLD) with estimates of prevalence from 40% to 70%. [4][5][6][7] It would be valuable to identify those at high risk of developing cirrhosis/HCC because NAFLD (at the pre-cirrhotic stage) is potentially reversible by weight loss, and it would direct screening and early treatment for varices and HCC, while promoting intensive management of increased cardiovascular risk. 8,9 A significant problem in creating appropriate risk assessment tools for NAFLD is that no consistent risk factors for progressive disease have been identified. Cohort studies report variable results and in meta-analyses the only consistent factor predicting progressive disease is histological identification of liver fibrosis. 10,11 However, liver biopsy is an invasive procedure, with a complication rate that is not acceptable for population screening. Several groups have developed non-invasive risk scoring models to identify those with fibrosis (including the Fibrosis 4 Index (FIB-4), the NALFD Fibrosis Score (NFS), AST:ALT ratio, the AST to Platelet Ratio Index (APRI) and the Enhanced Liver Fibrosis test (ELF)). [12][13][14][15][16] These scores have been validated in cohorts with NAFLD. However, subsequent studies have shown variable performance with the strength of association with incident cirrhosis, HCC, the need for liver transplantation and death varying significantly between cohorts. [17][18][19][20][21] Most of these studies have been small and only included people under secondary care hepatology services. In addition, when applied to specific groups, literature based cut-offs result in very variable proportions of populations being classed as 'high risk' with poor agreement between the top 5% of the distribution of risk scores. 22,23 Consensus guidelines on the management of NAFLD, published by the European Association for the Study of the Liver, the European Association for the Study of Diabetes and the European Association for the Study of Obesity (EASL-EASD-EASO) recommend screening for NAFLD as part of routine care in Type 2 diabetes. 8 These guidelines suggest a screening algorithm that advises referral for specialist hepatology assessment if there is evidence of steatosis and non-invasive markers suggest medium or high risk of fibrosis; or if there is a raised alanine aminotransferase (ALT), aspartate aminotransferase (AST) or gamma-glutamyltransferase (γGT) (Figure 1). The American Association for the Study of Liver Diseases (AASLD), while not recommending a specific screening algorithm, states that there should be 'a high index of suspicion for NAFLD and NASH in Type 2 diabetes'. 9 The AASLD suggests the use of existing liver fibrosis risk scores or assessment methodologies (such as the FIB-4, NFS or transient elastography) to assess at-risk patients. 9 One study of the EASL-EASD-EASO referral algorithm reported that around one third of people routinely attending a diabetes clinic would fulfil the criteria for hepatology referral; the incidence of subsequent cirrhosis and HCC in that cohort was not reported. 24 It is possible that the ability of the non-invasive tests to accurately identify incident disease may be affected by low event rates in community populations. Moreover, it has been suggested that current risk scores may be less accurate in people with Type 2 diabetes than in those without. 25 There remains significant uncertainty about the utility of these screening methods in Type 2 diabetes.

| AIMS
We aimed to assess the ability of individual fibrosis scores and of the EASL-EASD-EASO screening algorithm to predict 11-year incident

Funding Information
The Edinburgh Type 2 Diabetes Study was funded by a grant from the U.K. Medical Research Council (Project Grant G0500877) and the Chief Scientist Office of Scotland (Programme Support Grant CZQ/1/38), and the liver substudy was supported by a grant from Pfizer (Unrestricted Investigator Led Grant). The study sponsor was not involved in the design of the study; the collection, analysis and interpretation of data; writing the report; or the decision to submit the report for publication.
Handling Editor: Salvatore Petta Conclusions: In a cohort with a moderately low incidence of cirrhosis/HCC, existing risk scores did not reliably identify participants at high risk. Better prediction models for cirrhosis/HCC in people with Type 2 diabetes are required.

K E Y W O R D S
cirrhosis, hepatocellular carcinoma, non-alcoholic fatty liver disease, risk prediction, screening, type 2 Diabetes

Lay summary
The incidence of end-stage liver disease is increased in people with Type 2 diabetes, primarily because of nonalcoholic fatty liver disease. Screening for liver disease in Type 2 diabetes is recommended. This study showed that, in a cohort of older people with Type 2 diabetes, current recommended screening pathways did not reliably identify those at risk of developing end-stage liver disease. cirrhosis and/or HCC in an asymptomatic community cohort of older people with Type 2 diabetes.

| The Edinburgh type 2 diabetes study
The Edinburgh Type 2 Diabetes Study (ET2DS) is a population-based prospective cohort study, designed to investigate the progression of complications in people with Type 2 diabetes. The full methods have been described previously. 26 In summary, in 2006/07 participants aged 60-74 with Type 2 diabetes were randomly selected (in age and sex bands) from the Lothian Diabetes Register (a database of almost 30 000 patients with diabetes living in Lothian, Scotland, UK, managed in both primary and secondary care). Invitations to participate were sent to 5454 people, of whom 1066 (20%) attended baseline assessment. These people have been shown to be representative of all those invited and thus of the target population. 26 All who attended the baseline clinic were invited to re-attend a clinical and liver assessment at year 1 and 4. A total of 939 attended the year 1 clinic (of the original baseline cohort, deceased n = 15, unable to contact n = 19, unable to attend n = 93) and 831 at year 4 (of the baseline cohort, deceased n = 88, unsuitable for clinical reasons n = 26, unable to contact n = 23, unable to attend n = 98). The characteristics of the cohort who attended the year 1 clinic were similar to the whole cohort at baseline. 6 All 1066 participants were followed up for outcome assessment to death (320 participants throughout the study) or end of follow-up.

| Data collection-baseline biomarker assessment
Research clinics were undertaken at the Wellcome Trust Clinical Research Facility, Western General Hospital, Edinburgh, UK. F I G U R E 1 EASL-EASD-EASO algorithm (with permission of the initial publisher). Diagnostic flow-chart to assess and monitor disease severity in the presence of suspected NAFLD and metabolic risk factors. 1 Steatosis biomarkers: Fatty Liver Index, Steato Test, NAFLD Fat score (see Tables). 2 Liver tests: ALT AST, γGT. 3 Any increase in (FibroTest, FibroMeter, ELF). 5 Low risk: indicative of no/mild fibrosis; Medium/ high risk: indicative of significant fibrosis or cirrhosis Standardized operating procedures were used for every aspect of data collection as previously detailed. 26 ALT, AST, γGT, platelets and triglycerides were measured on fasting venous samples at the baseline research clinic and were analysed using a Vitros Fusion chemistry system (Ortho Clinical Diagnostics, Bucks, UK). The Enhanced Liver Fibrosis test (ELF) was measured on fasting venous blood samples from the year 1 clinic and was analysed using the ADVIA Centaur immunoassay system (Siemens Healthcare Diagnostics Inc., New York, USA) at the iQur laboratory (London, UK). Ultrasound was undertaken at the year 1 clinic following a 4-hour fast (Sonoline Elegra Ultrasound Imaging System (Siemens Medical Systems Inc., Washington, USA)).
Ultrasounds were graded for hepatic steatosis using established criteria (0=normal liver, 1=indeterminate, 2=mild steatosis, 3=severe steatosis) and validated by three different graders and 1 H MRI spectroscopy in a subset, as previously described. 27 This showed a median fat fraction in those with 'severe' steatosis of 19.4% (interquartile range 12.9-27.5), compared to 4.1% (interquartile range 3.1-8.5) in those with 'indeterminate'/ 'mild' steatosis and 4.2% (interquartile range 1. 2-5.7) in those with 'no steatosis'. As a result of this validation which showed significant overlap between grade 0-2 steatosis, only those with grade 3 steatosis on ultrasound assessment were deemed to have 'definite steatosis'. Individuals with an ultrasound grading of 0-2 were considered to have 'no definite steatosis'.
Participants underwent full diagnostic liver screen (including Hepatitis B and C serology, liver autoantibody titres, alpha-foeto protein, ferritin) and history to assess alcohol status, medication use and past medical history. Any participant with routine liver enzyme tests above the laboratory upper limit of normal (ALT >50 U/L, AST >45 U/L, γGT >55 U/L, alkaline phosphatase >125 U/L), AST:ALT ratio >1, hyaluronic acid >100 µ g/L (in the absence of known joint disease), positive liver autoantibodies, ferritin >1000 ng/mL, alpha-foeto protein >6 ng/mL, positive hepatitis B or C serology, spleen diameter >13 cm, platelets <150 × 10 9 /L in the absence of known haematological cause, or suspected cirrhosis on ultrasound was referred for specialist hepatology review.
Steatosis and Fibrosis scores were calculated and cut-off levels used as per published literature.

| Data collection-identification of liver disease
Possible prevalent liver disease was identified through a patient clinical history questionnaire at the baseline clinic. Possible cases were confirmed if a clinician diagnosis was recorded in primary or secondary care medical records.
Incident cirrhosis and HCC cases were identified and corrob- it has no scale, but lower values suggest improved performance.
Because of our mixed population of screen-detected and cliniciandiagnosed outcomes, possibly skewing our time-to-event data as those who were screen-detected were often diagnosed at a pre-

| Ethics
Ethical permission for the study was granted by Lothian Medical Research Ethics Committee (REC reference 16/SS/0098). All participants gave written informed consent.
Seven people had prevalent cirrhosis/HCC.

| Incident cirrhosis/HCC
Out of 1059 people without cirrhosis/HCC at baseline, 43 developed this outcome over 11 years of follow-up (11-year incidence 4.1%) ( Figure 2   The algorithm, regardless of steatosis marker or fibrosis score inserted, performed variably in how the 'advise to refer' outcome associated with incident cirrhosis/HCC (sensitivity 79%-90%, specificity 36%-73%). PPV was low (5-10%) indicating that a 'advise to refer' outcome was not a good predictor of incident cirrhosis/HCC. NPV was high at 99% but may again reflect the relative rarity of the outcome.

| Performance of the EASL-EASD-EASO algorithm in predicting incident cirrhosis/HCC
False-negative rates were lower when using algorithm compared to fibrosis score alone, but were still 10-20%, which would have resulted

| Sensitivity analysis
Two sensitivity analyses were undertaken. The first demonstrates that there is no improvement in test performance when an outcome of 'presence of varices, ascites or encephalopathy in the context of cirrhosis or HCC' was used (  (Figure 1). a ELF and ultrasound measured at year 1 only-so calculated 10 not 11-year incident cirrhosis/HCC. *P < .05, **P < .01, ***P < .001.
to those presented for the whole cohort with mixed aetiology disease above (Supporting Information 2).
Additionally, analysis was re-run using competing risks regression methodology with the competing risk being non-liver death. Results were similar to those obtained from logistic regression methodology with all risk scores showing a significant association with the development of cirrhosis/HCC and APRI providing the best improvement from null model by BIC (Supporting Information 3). scores were designed to identify advanced fibrosis as opposed to cirrhosis/HCC. However, given the time span of follow-up we would have expected those with advanced fibrosis to progress to cirrhosis over 11 years and there thus to be a correlation. In addition, a significant proportion of our population underwent ultrasound at year 1. All abnormal ultrasounds were followed up and those diagnosed with fibrosis at year 1 progressed to cirrhosis over the period of the study.

| D ISCUSS I ON
ET2DS is a study of moderate size that has reviewed long-term liver outcomes in individuals with Type 2 diabetes who were asymptomatic of liver disease at baseline. Almost all other studies have examined outcomes in people recruited from secondary care hepatology clinics, with known NAFLD and a higher likelihood of cirrhosis/HCC. Although the ET2DS studied a cohort at higher risk of cirrhosis/HCC than the general population, the absolute probability of cirrhosis/HCC was moderately low. Therefore, validated risk scores and a European consensus algorithm have been tested in a cohort where the pre-test probability is low; in contrast to previous studies. However, this represents precisely the scenario in which European guidelines recommend screening for liver disease.
Participants in the ET2DS were well characterized at baseline allowing accurate documentation of baseline risk factors, and have been followed longitudinally and extensively using multiple sources of information.
There are limitations to our study. ET2DS is a single centre study, implications. Firstly, as the natural history of NAFLD progression is very prolonged, it is possible that those who were diagnosed following referral from screening had cirrhosis/HCC at baseline and had prevalent rather than incident disease. However, the range of time from year 1 clinic to diagnosis overlaps significantly in the 'screen-detected' and 'clinician-detected' groups. Moreover, several of those who were 'screen-detected' were not identified with cirrhosis/HCC on initial hepatology review, but follow-up was continued because of concern regarding 'high-risk' features and they were diagnosed with cirrhosis/HCC several years later. Therefore, we defined prevalent disease as that which was clinically apparent at baseline. Secondly, the screening process may have led to an earlier diagnosis of cirrhosis/HCC in some people who may have died from other causes before cirrhosis/HCC was clinically apparent, inflating incidence rates. However, 58% of those identified with cirrhosis de-  It remains unclear why the fibrosis risk scores perform better in people without diabetes than those with diabetes. It is possible that there are confounders influencing the biomarkers used in the non-invasive scores that are affected by diabetes. For example it has been described that measurements of AST and ALT in mouse models are affected by hyperglycaemia. 25 Future research is required to identify improved methods of predicting incident cirrhosis/HCC in this high-risk population, possibly through combining existing risk scores, examining whether serial monitoring is a more effective screening strategy or investigating novel or alternative biomarkers.
Type 2 diabetes is associated with an increased rate of cirrhosis/HCC. 2,3 Risk prediction scores and international guidelines have attempted to provide non-invasive methods of assessing risk of incident disease in this high-risk population. This study shows only modest performance of these risk scores and screening algorithm.
Use would lead to significant pressure on hepatology services from high referral rates coupled with increased patient anxiety generated by false-positive results. Furthermore, the risk scores fail to identify a significant proportion of the population that are potentially vulnerable to incident disease. Future work to improve prediction methods in this population is necessary.