This prospective case-control study demonstrates the association of the ELF test score with liver-related outcomes in a general population of postmenopausal women with risk factors for liver disease in the form of high BMI, high alcohol consumption or both. ELF scores were higher in those participants who subsequently experienced a liver-related event earlier. Time to event analysis demonstrated an association between ELF and LRE, with a hazard ratio of 2 compared to women who do not experience LRE. To the best of our knowledge, this is the first study to evaluate the performance of the ELF test to predict a range of LREs in a general population.
The clinical utility models indicate that the ELF test may have a role in diagnosing liver disease in this low prevalence population, using a single threshold. The AUROC in our study was not highly predictive, and further population studies are required.
Strengths and limitations
Strengths of this population-based study include the prospective design and the independence of data capture for outcomes. This study used ICD-10 codes for CLD that have been used in other studies of cirrhosis, however in an attempt to maximise identification of liver disease we also included codes relating to clinical consequences of advanced cirrhosis. Evaluation of numerous possible confounders including self-reported comorbidities and socioeconomic status minimised bias. Although the number of cases of liver related events was only 58 it is important to recognize that this was the total number of incident cases occurring in a cohort of 95,126 women sampled for this study and so the cases represent a comprehensive sample of participants developing LRE.
ELF tests were performed in one central laboratory, ensuring quality control and consistency, using the proprietary ELF assays. The stability of the ELF test, when applied to samples exposed to a range of common storage conditions, has been demonstrated [
26].
Limitations of this study include the reliance of self-reporting of height and weight and co-morbidities. There is evidence, however, of good reliability of self-reporting height and weight from other studies [
27‐
30], for example in a longitudinal study examining agreement between self-reported and measured height, weight and body mass index in older people [
31]. Several studies have demonstrated good accuracy in recalled weight, with some data indicating underestimation in those with higher BMI [
32‐
35].
As previously discussed, the UKCTOCS alcohol categories do not align with the CMO’s threshold for hazardous drinking of 14 units / week and therefore including the UKCTOCS threshold of 11–15 units / week may have over-estimated ‘high alcohol use’; however excluding this category in this study would have risked excluding some women with hazardous alcohol consumption.
It is possible that reliance on ICD-10 codes to define events may lead to errors due to mis-coding. We interrogated three independent sources in an attempt to reduce risk of non-coding. Further, HES data may not capture clinical events in a number of areas of healthcare, including the private sector. There is no clear definition for liver disease and consequently, due to large variation in data definitions, comparing incidence between studies is difficult. The problem facing clinicians is that nearly 50% of individuals with liver disease only receive the diagnosis when they present to hospital with a decompensating event [
36]. In the community setting the focus must be on identifying liver disease not only at the point of a clinical event but before this where intervention may be more effective. By selecting a group of codes for our data definition, we are contributing to the heterogeneity of definitions in this area. Further work is required to find agreement amongst investigators.
Ninety seven percent of the UKCTOCS population was white. This ethnic homogeneity may have implications for general applicability of the data, in addition to the use of ‘normal’ ELF score ranges. Finally, an evaluation of the ELF test in a secondary care population in Australia reported a positive correlation with age, not seen in the original ELF studies [
37]. More work may be required to determine whether age-specific ranges are required in the general population.
Other studies
The incidence of CLD continues to rise [
19,
38‐
40] and the asymptomatic nature of liver fibrosis progression, leading to cirrhosis results in individuals often presenting with life-threatening features of decompensation in the form of ascites, variceal haemorrhage, hepatic encephalopathy, liver failure or hepatocellular carcinoma [
36]. Noninvasive evaluation of liver fibrosis is now established in clinical practice but remains largely confined to secondary and tertiary care settings where it is applied to patients with known or suspected liver disease. There is an urgent need to identify liver disease and the risk of progressive fibrosis in primary care, not least in those with risk factors, where the reliance on measurement of serum liver enzymes may be falsely reassuring [
41‐
43].
Transient elastography (TE) has been the most extensively investigated and independently validated non-invasive test in the general population, with smaller numbers of studies evaluating and validating serum-based markers [
44]. However, there is a lack of consensus on thresholds for levels of liver disease, both between tests and within the same test. This results in a wide variation in prevalence estimates. Furthermore, although elastography is a relatively rapid test which produces an instant result, automated blood tests, like the ELF test, can be included with other routine blood tests in primary care and, unlike TE, require no training (which would be more difficult to provide in primary care) and have a lower failure rate compared to TE [
45].
TE is operator dependent and although good inter-operator variability in performance has been reported [
46,
47], this is reduced at lesser stages of fibrosis and in individuals with hepatic steatosis, high BMI and in particular waist circumference. In its development of guidelines for management of NAFLD, NICE performed an extensive health economic comparison of non-invasive modalities, concluding that the ELF test was the most cost-effective test in this context [
48]. More recently a prospective study in primary care has shown that the ELF test improved the detection of liver disease and in conjunction with the Fibrosis-4 (Fib-4) score, reduced inappropriate referrals of patients with NAFLD to secondary care [
49]. The choice of modality utilised in a community-based setting is likely to be influenced by local expertise and experience and the prevailing capital and processing costs.
A recent meta-analysis collected data on TE values in healthy individuals identifying 26 studies and a total of 16,082 participants [
50]. The mean liver stiffness in non-obese individuals was 4.68 kPa, with increased stiffness measurements in individuals with diabetes, increasing waist circumference, obesity, elevated serum transaminases or hypertension. Dedicated studies within primary care are, however, required.
A community-based study of participants with risk factors for liver disease comprising excess alcohol use, type 2 diabetes or elevated serum transaminases used simple serum marker algorithms with high negative predictive values to rule out significant liver fibrosis. Those with results indicating liver disease were invited to attend for TE. 12.1% had a normal initial test and of those with a valid elastography result, 27% had elevated liver stiffness [
43].
A large study based in primary care examined the natural history of standard ‘liver function tests’ (LFT) measured in over 95,000 patients with no liver disease, followed up for a median of 3.7 years [
41]. 1.14% developed liver disease and at least one abnormality within the LFT panel was predictive of developing liver disease, for example the HRs for mild ALT and severe ALT rises were 4.23 and 12.67, respectively. Health economic analyses indicated that the most cost-effective strategy in those with abnormal LFTs with no obvious liver disease was to re-test in primary care and in those with high risk neither re-testing nor secondary care referral dominated. In a group of individuals from the general population participating in the National Health and Nutrition Examination Survey (NHANES) III survey, serum markers were applied to those with NAFLD and followed up for a median of 14.5 years. Increasing NAFLD fibrosis score, aspartate aminotransferase (AST) to platelet ratio (APRI) and Fib-4 scores were associated with increasing mortality, although the low number of liver related deaths was too small to analyse [
51].
The ELF test has been evaluated in a group of obese patients undergoing bariatric surgery and who had suspected NAFLD with a significantly higher ELF score in those with non-alcoholic steatohepatitis (NASH) and / or fibrosis on biopsy compared to those with normal histology or steatosis [
52]. Using histology as the reference standard, performance of the ELF test, FibroTest, elastography, and other simple serum marker panels (including APRI and Fib-4) were compared in a primary care cohort with a history of excess alcohol use. Using a cut off value of 10.5, the ELF test diagnosed advanced fibrosis with high accuracy (AUROC = 0.89), with similar performance to FibroTest and elastography, but without test failures, and was more accurate than the simple marker panels [
21].
The ELF test has been shown to predict clinical outcomes in mixed liver disease [
9,
37], and in disease-specific populations, for example primary sclerosing cholangitis [
53] and chronic hepatitis C [
54]. A study in the general population evaluated the performance of the ELF test to predict development of hepatocellular carcinoma, demonstrating that an ELF score of ≥9.89 had an odds ratio of 25 for predicting an event [
55].
Several studies have defined the normal ELF score in healthy populations. In a South Korean cohort where heart disease, diabetes, metabolic syndrome, hepatitis B, hepatitis C and liver dysfunction were excluded, the ELF test score in females was between 5.89 and 8.67 [
56]. In a study using serum samples from 400 blood donors, the ELF test score in females was found to be between 6.6 and 9.3 [
57]. In both studies, the average ELF score was higher in males. Our data adds to the understanding of normal ranges of ELF, showing a higher ELF score in a general population with risk factors, and higher still in those experiencing LRE.