Machine learning approaches to enhance diagnosis and staging of patients with MASLD using routinely available clinical information

AIMS

Metabolic dysfunction Associated Steatotic Liver Disease (MASLD) outcomes such as MASH (metabolic dysfunction associated steatohepatitis), fibrosis and cirrhosis are ordinarily determined by resource-intensive and invasive biopsies. We aim to show that routine clinical tests offer sufficient information to predict these endpoints.

METHODS

Using the LITMUS Metacohort derived from the European NAFLD Registry, the largest MASLD dataset in Europe, we create three combinations of features which vary in degree of procurement including a 19-variable feature set that are attained through a routine clinical appointment or blood test. This data was used to train predictive models using supervised machine learning (ML) algorithm XGBoost, alongside missing imputation technique MICE and class balancing algorithm SMOTE. Shapley Additive exPlanations (SHAP) were added to determine relative importance for each clinical variable.

RESULTS

Analysing nine biopsy-derived MASLD outcomes of cohort size ranging between 5385 and 6673 subjects, we were able to predict individuals at training set AUCs ranging from 0.719-0.994, including classifying individuals who are At-Risk MASH at an AUC = 0.899. Using two further feature combinations of 26-variables and 35-variables, which included composite scores known to be good indicators for MASLD endpoints and advanced specialist tests, we found predictive performance did not sufficiently improve. We are also able to present local and global explanations for each ML model, offering clinicians interpretability without the expense of worsening predictive performance.

CONCLUSIONS

This study developed a series of ML models of accuracy ranging from 71.9-99.4% using only easily extractable and readily available information in predicting MASLD outcomes which are usually determined through highly invasive means.

Health-related Quality of Life in Patients With Nonalcoholic Fatty Liver Disease: A Prospective Multi-center UK Study

BACKGROUND & AIMS

It is unclear whether health-related quality of life (HRQoL) is impaired in patients with nonalcoholic fatty liver disease (NAFLD) without advanced fibrosis and how this compares with the general population. We aimed to assess HRQoL in patients with NAFLD in comparison to the general population and any associations of fibrosis severity and metabolic comorbidities with impairments in HRQoL.

METHODS

We prospectively enrolled 513 consecutive patients with NAFLD who completed the EuroQol 5-dimensional questionnaire (EQ-5D) and Chronic Liver Disease Questionnaires (CLDQ). Demographic and clinical information, liver biopsy results, and/or liver stiffness (LS) by transient elastography were recorded. A general population sub-cohort of the Health Survey for England 2018 was used as a comparator (n = 5483), and a 1:1 propensity-score (PS) matching was performed, according to age, sex, body mass index, and type 2 diabetes mellitus (T2DM).

RESULTS

EQ-5D-5L utility was significantly lower in 466 PS-matched patients with NAFLD compared with PS-matched controls (0.77 ± 0.27 vs 0.84 ± 0.19; P < .001), even in those without advanced fibrosis (F ≤2 or LS <8kPa) (0.80 ± 0.24 vs 0.84 ± 0.19; P = .024). HRQoL measures (EQ-5D-5L, EQ-VAS, CLDQ) did not differ between patients with NAFLD with and without advanced fibrosis. LS was independently associated with lower EQ-5D-5L in all patients with NAFLD but not in those without advanced fibrosis. In the latter, lower EQ-5D-5L was associated with female sex, T2DM, and depression.

CONCLUSIONS

Patients with NAFLD, even those without advanced fibrosis, have worse HRQoL compared with the general population. In patients with NAFLD without advanced fibrosis, HRQoL is independently associated with non-liver comorbidities but not LS. Multi-disciplinary management is therefore required in NAFLD, irrespective of fibrosis severity.

Liver Investigation: Testing Marker Utility in Steatohepatitis (LITMUS): Assessment & validation of imaging modality performance across the NAFLD spectrum in a prospectively recruited cohort study (the LITMUS imaging study): Study protocol

Non-alcoholic fatty liver disease (NAFLD) is the liver manifestation of the metabolic syndrome with global prevalence reaching epidemic levels. Despite the high disease burden in the population only a small proportion of those with NAFLD will develop progressive liver disease, for which there is currently no approved pharmacotherapy. Identifying those who are at risk of progressive NAFLD currently requires a liver biopsy which is problematic. Firstly, liver biopsy is invasive and therefore not appropriate for use in a condition like NAFLD that affects a large proportion of the population. Secondly, biopsy is limited by sampling and observer dependent variability which can lead to misclassification of disease severity. Non-invasive biomarkers are therefore needed to replace liver biopsy in the assessment of NAFLD. Our study addresses this unmet need. The LITMUS Imaging Study is a prospectively recruited multi-centre cohort study evaluating magnetic resonance imaging and elastography, and ultrasound elastography against liver histology as the reference standard. Imaging biomarkers and biopsy are acquired within a 100-day window. The study employs standardised processes for imaging data collection and analysis as well as a real time central monitoring and quality control process for all the data submitted for analysis. It is anticipated that the high-quality data generated from this study will underpin changes in clinical practice for the benefit of people with NAFLD. Study Registration: clinicaltrials.gov: NCT05479721.

Machine learning algorithm improves the detection of NASH (NAS-based) and at-risk NASH: A development and validation study

BACKGROUND AND AIMS

Detecting NASH remains challenging, while at-risk NASH (steatohepatitis and F≥ 2) tends to progress and is of interest for drug development and clinical application. We developed prediction models by supervised machine learning techniques, with clinical data and biomarkers to stage and grade patients with NAFLD.

APPROACH AND RESULTS

Learning data were collected in the Liver Investigation: Testing Marker Utility in Steatohepatitis metacohort (966 biopsy-proven NAFLD adults), staged and graded according to NASH CRN. Conditions of interest were the clinical trial definition of NASH (NAS ≥ 4;53%), at-risk NASH (NASH with F ≥ 2;35%), significant (F ≥ 2;47%), and advanced fibrosis (F ≥ 3;28%). Thirty-five predictors were included. Missing data were handled by multiple imputations. Data were randomly split into training/validation (75/25) sets. A gradient boosting machine was applied to develop 2 models for each condition: clinical versus extended (clinical and biomarkers). Two variants of the NASH and at-risk NASH models were constructed: direct and composite models.Clinical gradient boosting machine models for steatosis/inflammation/ballooning had AUCs of 0.94/0.79/0.72. There were no improvements when biomarkers were included. The direct NASH model produced AUCs (clinical/extended) of 0.61/0.65. The composite NASH model performed significantly better (0.71) for both variants. The composite at-risk NASH model had an AUC of 0.83 (clinical and extended), an improvement over the direct model. Significant fibrosis models had AUCs (clinical/extended) of 0.76/0.78. The extended advanced fibrosis model (0.86) performed significantly better than the clinical version (0.82).

CONCLUSIONS

Detection of NASH and at-risk NASH can be improved by constructing independent machine learning models for each component, using only clinical predictors. Adding biomarkers only improved the accuracy of fibrosis.

Performance of non-invasive tests and histology for the prediction of clinical outcomes in patients with non-alcoholic fatty liver disease: an individual participant data meta-analysis

BACKGROUND

Histologically assessed liver fibrosis stage has prognostic significance in patients with non-alcoholic fatty liver disease (NAFLD) and is accepted as a surrogate endpoint in clinical trials for non-cirrhotic NAFLD. Our aim was to compare the prognostic performance of non-invasive tests with liver histology in patients with NAFLD.

METHODS

This was an individual participant data meta-analysis of the prognostic performance of histologically assessed fibrosis stage (F0-4), liver stiffness measured by vibration-controlled transient elastography (LSM-VCTE), fibrosis-4 index (FIB-4), and NAFLD fibrosis score (NFS) in patients with NAFLD. The literature was searched for a previously published systematic review on the diagnostic accuracy of imaging and simple non-invasive tests and updated to Jan 12, 2022 for this study. Studies were identified through PubMed/MEDLINE, EMBASE, and CENTRAL, and authors were contacted for individual participant data, including outcome data, with a minimum of 12 months of follow-up. The primary outcome was a composite endpoint of all-cause mortality, hepatocellular carcinoma, liver transplantation, or cirrhosis complications (ie, ascites, variceal bleeding, hepatic encephalopathy, or progression to a MELD score ≥15). We calculated aggregated survival curves for trichotomised groups and compared them using stratified log-rank tests (histology: F0-2 vs F3 vs F4; LSM: <10 vs 10 to <20 vs ≥20 kPa; FIB-4: <1·3 vs 1·3 to ≤2·67 vs >2·67; NFS: <-1·455 vs -1·455 to ≤0·676 vs >0·676), calculated areas under the time-dependent receiver operating characteristic curves (tAUC), and performed Cox proportional-hazards regression to adjust for confounding. This study was registered with PROSPERO, CRD42022312226.

FINDINGS

Of 65 eligible studies, we included data on 2518 patients with biopsy-proven NAFLD from 25 studies (1126 [44·7%] were female, median age was 54 years [IQR 44-63), and 1161 [46·1%] had type 2 diabetes). After a median follow-up of 57 months [IQR 33-91], the composite endpoint was observed in 145 (5·8%) patients. Stratified log-rank tests showed significant differences between the trichotomised patient groups (p<0·0001 for all comparisons). The tAUC at 5 years were 0·72 (95% CI 0·62-0·81) for histology, 0·76 (0·70-0·83) for LSM-VCTE, 0·74 (0·64-0·82) for FIB-4, and 0·70 (0·63-0·80) for NFS. All index tests were significant predictors of the primary outcome after adjustment for confounders in the Cox regression.

INTERPRETATION

Simple non-invasive tests performed as well as histologically assessed fibrosis in predicting clinical outcomes in patients with NAFLD and could be considered as alternatives to liver biopsy in some cases.

FUNDING

Innovative Medicines Initiative 2.

The Liverpool alcohol-related liver disease algorithm identifies twice as many emergency admissions compared to standard methods when applied to Hospital Episode Statistics for England

BACKGROUND

Emergency admissions in England for alcohol-related liver disease (ArLD) have increased steadily for decades. Statistics based on administrative data typically focus on the ArLD-specific code as the primary diagnosis and are therefore at risk of excluding ArLD admissions defined by other coding combinations.

AIM

To deploy the Liverpool ArLD Algorithm (LAA), which accounts for alternative coding patterns (e.g., ArLD secondary diagnosis with alcohol/liver-related primary diagnosis), to national and local datasets in the context of studying trends in ArLD admissions before and during the COVID-19 pandemic.

METHODS

We applied the standard approach and LAA to Hospital Episode Statistics for England (2013-21). The algorithm was also deployed at 28 hospitals to discharge coding for emergency admissions during a common 7-day period in 2019 and 2020, in which eligible patient records were reviewed manually to verify the diagnosis and extract data.

RESULTS

Nationally, LAA identified approximately 100% more monthly emergency admissions from 2013 to 2021 than the standard method. The annual number of ArLD-specific admissions increased by 30.4%. Of 39,667 admissions in 2020/21, only 19,949 were identified with standard approach, an estimated admission cost of £70 million in under-recorded cases. Within 28 local hospital datasets, 233 admissions were identified using the standard approach and a further 250 locally verified cases using the LAA (107% uplift). There was an 18% absolute increase in ArLD admissions in the seven-day evaluation period in 2020 versus 2019. There were no differences in disease severity or mortality, or in the proportion of admissions with decompensation of cirrhosis or alcoholic hepatitis.

CONCLUSIONS

The LAA can be applied successfully to local and national datasets. It consistently identifies approximately 100% more cases than the standard coding approach. The algorithm has revealed the true extent of ArLD admissions. The pandemic has compounded a long-term rise in ArLD admissions and mortality.

SP5.1.5 Bariatric surgery is associated with greater survival and metabolic health benefits than conventional medical management in people with NAFLD

Aims

Non-alcoholic fatty liver disease (NAFLD) represents the hepatic manifestation of the metabolic syndrome.There are no licensed pharmacotherapies for NAFLD, and it is rapidly becoming the leading cause of liver failure worldwide.Bariatric surgery (BS) is an effective and durable intervention for weight loss and type 2 diabetes (T2D) but is not recommended in the UK’s NICE guidelines for NAFLD management.We aimed to determine whether BS was associated with NAFLD regression.

Methods

1067 people with NAFLD were included in this observational cohort study.Liver histology/validated non-invasive biomarkers were used to diagnose NAFLD.Markers of liver and metabolic health were recorded longitudinally for participants undergoing hepatology-led (n = 702) or BS (n = 365) management in a UK tertiary centre. 1-year outcomes were compared using mixed models. 5-year mortality was compared using Cox proportional-hazards regression.

Results

T2D prevalence was similar between cohorts (p = 0.06). Differences in other baseline factors were observed (age, gender, and body mass index (BMI), all p &lt; 0001). BS showed a significantly lower adjusted 5-year mortality (p = 0.02). Both interventions were associated with improved biomarkers of liver health e.g. ALT (p = 0.00). Greater improvements in targets of NAFLD management such as HbA1c (p = 0.00) and BMI (p = 0.00) were observed after BS.

Conclusions

BS showed an improved 5-year survival compared to hepatology-led management. This may be explained by greater improvements in key targets for NAFLD regression in the BS cohort. However, residual confounding despite adjusted survival analysis cannot be excluded. A prospective randomised trial is recommended to ensure fair comparison of these interventions and enhance the clinical management of NAFLD.

Prognostic value of multiparametric magnetic resonance imaging, transient elastography and blood-based fibrosis markers in patients with chronic liver disease

BACKGROUND & AIMS

Liver cT₁ , liver T₁ , transient elastography (TE) and blood-based biomarkers have independently been shown to predict clinical outcomes but have not been directly compared in a single cohort of patients. Our aim was to compare these tests' prognostic value in a cohort of patients with compensated chronic liver disease.

METHODS

Patients with unselected compensated liver disease aetiologies had baseline assessments and were followed up for development of clinical outcomes, blinded to the imaging results. The prognostic value of non-invasive liver tests at prespecified thresholds was assessed for a combined clinical endpoint comprising ascites, variceal bleeding, hepatic encephalopathy, hepatocellular carcinoma, liver transplantation and mortality.

RESULTS

One hundred and ninety-seven patients (61% male) with median age of 54 years were followed up for 693 patient-years (median (IQR) 43 (26-58) months). The main diagnoses were NAFLD (41%), viral hepatitis (VH, 25%) and alcohol-related liver disease (ArLD; 14%). During follow-up 14 new clinical events, and 11 deaths occurred. Clinical outcomes were predicted by liver cT₁  > 825ms with HR 9.9 (95% CI: 1.29-76.4, P = .007), TE > 8kPa with HR 7.8 (95% CI: 0.97-62.3, P = .02) and FIB-4 > 1.45 with HR 4.09 (95% CI: 0.90-18.4, P = .05). In analysis taking into account technical failure and unreliability, liver cT₁  > 825 ms could predict clinical outcomes (P = .03), but TE > 8kPa could not (P = .4).

CONCLUSIONS

We provide further evidence that liver cT₁ , TE and serum-based biomarkers can predict clinical outcomes, but when taking into account technical failure/unreliability, TE cut-offs perform worse than those of cT₁ and blood biomarkers.

The European NAFLD Registry: A real-world longitudinal cohort study of nonalcoholic fatty liver disease

Non-Alcoholic Fatty Liver Disease (NAFLD), a progressive liver disease that is closely associated with obesity, type 2 diabetes, hypertension and dyslipidaemia, represents an increasing global public health challenge. There is significant variability in the disease course: the majority exhibit only fat accumulation in the liver but a significant minority develop a necroinflammatory form of the disease (non-alcoholic steatohepatitis, NASH) that may progress to cirrhosis and hepatocellular carcinoma. At present our understanding of pathogenesis, disease natural history and long-term outcomes remain incomplete. There is a need for large, well characterised patient cohorts that may be used to address these knowledge gaps and to support the development of better biomarkers and novel therapies. The European NAFLD Registry is an international, prospectively recruited observational cohort study that aims to establish a large, highly-phenotyped patient cohort and linked bioresource. Here we describe the infrastructure, data management and monitoring plans, and the standard operating procedures implemented to ensure the timely and systematic collection of high-quality data and samples. Already recruiting subjects at secondary/tertiary care centres across Europe, the Registry is supporting the European Union IMI2-funded LITMUS 'Liver Investigation: Testing Marker Utility in Steatohepatitis' consortium, which is a major international effort to robustly validate biomarkers that diagnose, risk stratify and/or monitor NAFLD progression and liver fibrosis stage. The European NAFLD Registry has the demonstrable capacity to support research and biomarker development at scale and pace.

Care standards for non-alcoholic fatty liver disease in the United Kingdom 2016: a cross-sectional survey

OBJECTIVE

Guidelines for the assessment of non-alcoholic fatty liver disease (NAFLD) have been published in 2016 by National Institute for Health and Care Excellence and European Associations for the study of the Liver-European Association for the study of Diabetes-European Association for the study of Obesity. Prior to publication of these guidelines, we performed a cross-sectional survey of gastroenterologists and hepatologists regarding NAFLD diagnosis and management.

DESIGN

An online survey was circulated to members of British Association for the Study of the Liver and British Society of Gastroenterology between February 2016 and May 2016.

RESULTS

175 gastroenterologists/hepatologists responded, 116 completing the survey, representing 84 UK centres. 22% had local NAFLD guidelines. 45% received >300 referrals per year from primary care for investigation of abnormal liver function tests (LFTs). Clinical assessment tended to be performed in secondary rather than primary care including body mass index (82% vs 26%) and non-invasive liver screen (86% vs 32%) and ultrasound (81% vs 37%). Widely used tools for non-invasive fibrosis risk stratification were aspartate transaminase (AST)/alanine transaminase (ALT) ratio (53%), Fibroscan (50%) and NAFLD fibrosis score (41%). 78% considered liver biopsy in selected cases. 50% recommended 10% weight loss target as first-line treatment. Delivery of lifestyle interventions was mostly handed back to primary care (56%). A minority have direct access to community weight management services (22%). Follow-up was favoured by F3/4 fibrosis (72.9%), and high-risk non-invasive fibrosis tests (51%). Discharge was favoured by simple steatosis at biopsy (30%), and low-risk non-invasive scores (25%).

CONCLUSIONS

The survey highlights areas for improvement of service provision for NAFLD assessment including improved recognition of non-alcoholic steatohepatitis in people with type 2 diabetes, streamlining abnormal LFT referral pathways, defining non-invasive liver fibrosis assessment tools, use of liver biopsy, managing metabolic syndrome features and improved access to lifestyle interventions.

Absolute quantitation of disease protein biomarkers in a single LC-MS acquisition using apolipoprotein F as an example

LC-MS and immunoassay can detect protein biomarkers. Immunoassays are more commonly used but can potentially be outperformed by LC-MS. These techniques have limitations including the necessity to generate separate calibration curves for each biomarker. We present a rapid mass spectrometry-based assay utilising a universal calibration curve. For the first time we analyse clinical samples using the HeavyPeptide IGNIS kit which establishes a 6-point calibration curve and determines the biomarker concentration in a single LC-MS acquisition. IGNIS was tested using apolipoprotein F (APO-F), a potential biomarker for non-alcoholic fatty liver disease (NAFLD). Human serum and IGNIS prime peptides were digested and the IGNIS assay was used to quantify APO-F in clinical samples. Digestion of IGNIS prime peptides was optimised using trypsin and SMART Digest™. IGNIS was 9 times faster than the conventional LC-MS method for determining the concentration of APO-F in serum. APO-F decreased across NAFLD stages. Inter/intra-day variation and stability post sample preparation for one of the peptides was ≤13% coefficient of variation (CV). SMART Digest™ enabled complete digestion in 30 minutes compared to 24 hours using in-solution trypsin digestion. We have optimised the IGNIS kit to quantify APO-F as a NAFLD biomarker in serum using a single LC-MS acquisition.