Magnetic Resonance Elastography (MRE) -Based Prediction Model for Hepatic Decompensation in NAFLD
3-Year Risk of Hepatic Decompensation:
5-Year Risk of Hepatic Decompensation:
Retrospective cohort study with data from six centers from the United States, Europe and Asia:
- The University of California San Diego
- Mayo Clinic Rochester
- Cedars Sinai
- Musashino Red Cross Hospital
- Yokohama City University
- Ankara University School of Medicine
Follow-up time started at the time of the MRE. Participants were followed until development of hepatic decompensation, HCC, death, or the last clinical encounter. Follow-up assessment was performed by a retrospective chart review.
Patient characteristics, including demographic, laboratory, imaging and outcome data are reported as median (interquartile range [IQR]) for continuous variables and N (%) for categorical variables. For the primary outcome, the cohort was randomly split in 1:1 ratio into training and validation sets.
Inclusion and Exclusion CriteriaInclusion
- Adults age ≥ 18 years with NAFLD and liver stiffness (LS) measurement by MRE who were assessed for hepatic decompensation, HCC, and death.
- NAFLD was defined as hepatic steatosis on imaging or a historical liver biopsy in the absence of significant alcohol consumption.
- Previous history of hepatic decompensation or HCC before enrollment or within 3 months of enrollment, follow-up duration of < 3 months, and incomplete critical laboratory data (Figure 1).
Population Data - Medians
Training Cohort (n=627)
- Age: 61 Years
- BMI: 29.1 kg/m2
- MRE: 3.5 kPa
- Average Follow-up: 3 Years
- Age: 60 Years
- BMI: 28.8 kg/m2
- MRE: 3.4 kPa
- Average Follow-up: 4 Years
68 (5.4%) met the composite primary outcome of hepatic decompensation including varices needing treatment, ascites, or hepatic encephalopathy.
Analysis & Results
- In univariable analysis in the training set, age, DM, ln (MRE), square (albumin), ln (AST), and platelet count were significantly associated with hepatic decompensation
- In multivariable analysis in the training set, age, ln (MRE), square (albumin), ln (AST), and platelet count met the statistical threshold for inclusion in the multivariable model (p<.10)
|S0(t) at 3 Years||S0(t) at 5 Years||Mean Score||Individual Score|
|0.9681||0.9490||0.059||=0.024(Age) + 0.949ln(MRE) - 0.122(Albumin)2 + 0.734ln(AST) - 0.016(Platelets)|
Final Risk Estimation Calculated as: 1 - S0(t)e(Individual Score - Mean Score)
The equation derived in the validation cohort was evaluated in quartiles and stratified the risk of hepatic decompensation in the validation cohort (p<.0001) (Figure 2).
In Context With Published Literature
- Fibrosis stage on liver biopsy has been adopted as a surrogate marker for future liver-related outcomes in clinical trials
- Recently, studies have demonstrated the direct association of non-invasive tests (NITs) such as FIB-4 index (33, 34), liver stiffness on vibration-controlled transient elastography (35), and MRE (25, 27, 36, 37) on liver-related events.
- Here, we evaluated a candidate set of variables and then formed a multivariable model for hepatic decompensation that retained excellent diagnostic accuracy in a multi-ethnic validation cohort.
- Importantly, the score differentiated the risk of hepatic decompensation in patients with and without cACLD, despite the low risk (1%) for decompensation in the subset of patients without cACLD.
Strengths and Limitations
- Large sample size of > 1,200 patients
- High number of incident hepatic decompensation events (n=68, 5.4%) allowed for adequate power to assess multivariable models in a training and validation cohort
- First, MRE was only assessed at a single time point in this study. This score was developed to assess prognosis at a single time point and future studies will be required to assess the dynamics of changes in this score over time and its impact on a patients risk of hepatic decompensation
- Second, primarily because MRE has become available in the clinical practice more recently compared to other NITs, e.g. VCTE or other blood-based markers, our study has a relatively short median follow-up duration.
- Last, since all patients in the present study belonged to retrospective cohorts at academic medical centers, a subset of clinical events may not have been captured.
Future prospective studies evaluating this MRE-based model with systematic assessment of hepatic decompensation are recommended