Liver-related and Extrahepatic Events in Patients With Non-alcoholic Fatty Liver Disease

A Retrospective Competing Risks Analysis

Grazia Pennisi; Marco Enea; Manuel Romero-Gomez; Mauro Viganò; Elisabetta Bugianesi; Vincent W.-S. Wong; Anna Ludovica Fracanzani; Giada Sebastiani; Jerome Boursier; Annalisa Berzigotti; Mohammed Eslam; Javier Ampuero; Amine Benmassaoud; Claudia La Mantia; Yuly P. Mendoza; Jacob George; Antonio Craxì; Calogero Camma'; Victor de Ledinghen; Salvatore Petta

Disclosures

Aliment Pharmacol Ther. 2022;55(5):604-615. 

In This Article

Methods

Patient Selection

We retrospectively analysed data from two multicentre cohorts of NAFLD patients with histological (training set) or non-invasive (replication set) assessment of liver fibrosis, prospectively recruited at the Gastrointestinal and Liver Unit of the Palermo University Hospital; at Centre d'Investigation de la Fibrose Hépatique of the Bordeaux University Hospital; at Division of Gastroenterology and Hepatology of McGill University Health Centre of Montreal; at Hepatology Unit of Ospedale San Giuseppe University of Milan; at Hospital Universitario Virgen del Rocío de Sevilla; at Department of Medicine and Therapeutics of the Chinese University of Hong Kong; at Hepato-Gastroenterology Department of Angers University Hospital; at the Swiss Liver Center; at Division of Gastroenterology, Department of Medical Sciences of University of Torino; and at Department of Pathophysiology and Transplantation, Ca' Granda IRCCS Foundation of Policlinico Hospital of University of Milan; and at Storr Liver Centre, Westmead Institute for Medical Research, Westmead Hospital and University of Sydney. Patients were included if they had received follow-up for at least 6 months. Other causes of liver disease were ruled out, including alcohol intake (>20 g/d) as evaluated by a questionnaire; viral (hepatitis B surface antigen, anti-hepatitis C virus and anti-human immunodeficiency virus negativity) and autoimmune hepatitis; hereditary hemochromatosis; and alpha-1 antitrypsin deficiency.

In the training cohort, the Kleiner scoring system[21] was used for histological assessment of NAFLD. Fibrosis stages were assigned from 0 to 4. Liver stiffness measurement (LSM) was used to assess fibrosis non-invasively in the replication cohort. LSM was evaluated with transient elastography by FibroScan (Echosens, Paris, France), using the M and the XL probe when appropriate. LSM was assessed after overnight fasting. LSM < 7.9 kPa was defined as indicating a low risk of F3-F4 fibrosis; LSM 7.9–9.6 kPa as an intermediate risk; LSM ≥ 9.6 kPa as a high risk.[22,23] The study was carried out in accordance with the principles of the Helsinki Declaration and with local and national laws. Approval was obtained from the AOUP "Paolo Giaccone" of Palermo.

Patient Evaluation

Clinical and metabolic data were collected at the time of enrolment. Body mass index (BMI) was calculated in kilograms for weight and in meters for height. Obesity was defined as BMI ≥ 30 kg/m2. The diagnosis of T2D was made according to the American Diabetes Association,[24] using a value of fasting blood glucose ≥126 mg/dl. In patients with a previous diagnosis of T2D, current medications and their changes were documented and used to meet a case definition of T2D. Arterial hypertension was defined by systolic blood pressure ≥140 mm Hg and/or diastolic blood pressure ≥90 mm Hg, or use of blood pressure lowering agents.[25] A 12-hour overnight fasting blood sample was drawn to determine serum levels of ALT, GGT, PLT, total cholesterol and triglycerides.

LRE and EHE Assessment

Incident LRE and EHE were recorded during the entire follow-up period in both groups. LRE was defined as LD (occurrence of either ascites, variceal haemorrhage, encephalopathy, jaundice or HCC). EHE were defined as either CVE (stroke, transient ischaemic attack, myocardial infarction and unstable angina) or EHC not including non-melanomatous skin cancers.

Clinical, biochemical and ultrasound examinations were conducted annually in patients with F0-F2 fibrosis and, for stricter surveillance for HCC and LD, every 6 months in patients with F3 fibrosis or cirrhosis, according to international guidelines.[26] In the presence of cirrhosis, oesophageal gastroscopy was performed at baseline and repeated as recommended by clinical guidelines.[27] Patients with progression to medium or large (F2 or F3) oesophageal varices were treated with b-blockers or underwent elastic banding, whereas no prophylaxis was provided to patients with small (F1) varices. Patients developing LRE during follow-up were evaluated for available therapies and/or for liver transplantation, if appropriate. Evidence of both LRE and EHE was provided by inpatient and outpatient medical records. Patients were censored at the last available visit or in case of death; patients who underwent liver transplantation were censored as dead.

Statistics

Observed counts and incidence rates of hepatic and extra-hepatic events were documented. Patient characteristics were reported within strata: mean ± standard deviation for covariates with symmetrical continuous distributions; median and interquartile range for variables with skewed continuous distributions; and percentage of cases for binary variables. Pairwise comparisons among fibrosis strata, that is F0-F1, F2 and F3-F4, were performed by using t-tests for covariates with symmetrical continuous distributions; Mann-Whitney tests for variables with skewed continuous distributions; and chi-square tests to compare percentages. In all cases, a correction factor for multiple tests was applied.[28]

Univariate analysis was conducted by estimating several univariate models as covariates. All variables that we considered were measured at baseline; consequently, no time-dependent covariates were evaluated. The variables that we considered were: hypertension; T2D; gender; age in years, as well as its categorised version using the median as cut point, 50 for F0-F2 patients, 55 for F3-F4; BMI, as well as the binary version "BMI30" (≤30/>30); platelet counts, as well as Plt150 (platelets > 150.000/Platelets ≤ 150.000); ALT and log(ALT); GGT and log(GGT); cholesterol level; triglyceride level and log(triglyceride level). Robust estimates of the standard errors were used to take into account heterogeneity due to the multicentre study design. The multivariate model included all the covariates resulting significant (P < 5%) in the univariate models. These candidate covariates were then included in the multivariate model one at time, according to a forward selection based on p-value and AIC. Moreover, the selected number of covariates in both the final models for LREs and EHEs agree with the principle of parsimony that avoids overfitting.

Both univariate and multivariate analyses were performed by stratified analyses according to the severity of liver fibrosis. Observed CIFs and competing risks models were used to model the risk of occurrence of LRE and EHE considered as competitive. In particular, CSC and subdistribution hazards (FG)[29] models were fitted. For each of the CSC and FG models, two types of models were employed: the first was a competing risks model fitted on the subset with fibrosis < F3, but using fibrosis with levels F0-F1 and F2 as stratification covariate, with the remaining covariates assumed the same for both strata. The model thus assumes different cause-specific (or subdistribution) baseline hazards by fibrosis strata. The second competing risks model was fitted on the fibrosis stratum F3-F4, by employing a different set of covariates, as a result of the selection process based on P-value, AIC (forward selection), and on clinical criteria. To take into account heterogeneity due to inter-hospital variations, a generalised estimating equation (GEE) approach was used, and robust standard errors were obtained. Predicted CIFs were provided for both hepatic and EHE occurrences by each model.

The fitted models were replicated externally by using a replication cohort of NAFLD patients with non-invasive stratification of liver fibrosis, by comparing the predicted CIFs in the training and in the replication sets. All analyses were performed by using R (version 4.0.2) and the "survival" and "mstate" statistical packages.

processing....