Participants
Study participants were patients who attended the Hepatology Department, Sir Charles Gairdner Hospital (Perth, Australia) from 1992 to 2016 with NAFLD and underwent liver biopsy. Initially 581 patients were identified with further exclusion of patients with a second aetiology of chronic liver disease (hepatitis B and C, drug induced liver disease and alcohol excess; n = 22), had clinical event prior to liver biopsy (n = 8), unusable liver tissue for Sirius Red staining (n = 14) or no available clinical records for follow-up (n = 90). A cohort of 295 participants were included in the final analysis. The detailed exclusion process was described in Supplementary Fig. 1. The study was approved by the Human Research Ethics Committee at Sir Charles Gairdner Hospital (RGS0000001592).
Clinical and biochemical data collection
Clinical and laboratory data were collected in a standardized fashion from electronic hospital records and pathology records. Demographic features included age, gender and ethnicity. BMI was calculated using the formula: weight (kilograms)/height2 (meters). All biochemical assays including liver function tests, fasting lipids, glucose and insulin was conducted by the State referral laboratory (Pathwest, Nedlands, WA). Non-invasive markers including Hepascore, FIB-4 and APRI were calculated at baseline according to published algorithms [4, 13, 14].
Clinical outcomes collection
Clinical outcomes were collected based on medical records and population-based data-linkage. The primary end point was liver-related death (LRD; death from liver failure or a liver related complication, HCC or transplantation). Secondary outcomes included total mortality (including liver transplantation), liver decompensation (defined by ascites, hepatic encephalopathy, variceal bleeding, hepatorenal syndrome, or spontaneous bacterial peritonitis) and development of HCC. HCC was diagnosed according to standard diagnostic criteria guidelines.[15, 16] Combined liver outcome was defined as liver decompensation, HCC or LRD, whichever occurred first. Clinical outcomes were determined by two methods; firstly, a comprehensive review of the patient medical record, laboratory, pathology and radiology records with outcomes confirmed by experienced hepatologists. Secondly, by ICD-10 codes from the Western Australian (WA) Data Linkage Unit which is a validated population-based data-linkage system including registries for cancer, hospital admissions and mortality for the state of WA. Codes for liver related death included K72.1, I98.3, C22.0; codes for liver decompensation included R18, K72.1, I98.3, K76.7, L65, C22.
Liver biopsy assessment
Liver histology was reviewed by an experienced histopathologist in Pathwest laboratory (Nedlands, WA). Fibrosis was staged according to the NASH-CRN scoring system. Briefly, fibrosis was classified on a 5-point scale: stage 0 = no fibrosis, stage 1 = zone 3 perisinusoidal/perivenular fibrosis, stage 2 = zone 3 and periportal fibrosis, stage 3 = septal/bridging fibrosis, stage 4 = cirrhosis.
Collagen Proportionate Area Determination
Liver biopsy samples were cut to 4µm slides from stored formalin fixed paraffin embedded tissue blocks for Sirius Red Staining performed by Pathwest. Briefly, slides were immersed in 0.4mM Picro-sirius red solution for an hour after washed by distilled water. Slides were then sealed with coverslips followed by washing in 0.5% acetic acid for 3 minutes twice. Dried slides were scanned by Aperio ScanScope XT (Leica biosystems, Wetzlar, Germany) under 20× magnification. Computerized image morphometry was performed by Aperio ImageScope v12.3.3 (Leica biosystems, Wetzlar, Germany). The image analysis process is illustrated in Supplementary Fig. 2. Briefly, the pen tool was selected to select the whole tissue are as the valid testing area. Liver capsule, extra-hepatic tissue and highly fragmented tissue were then counter-selected as the excluded areas from further analysis (Supplementary Fig. 2d and e). Analysis parameters were set as followed: hue value = 0.95; hue width = 0.3; colour saturation threshold = 0.04; Iwp (High) = 255; Iwp (low) – Ip (High) = 175; Isp (low) = 0; Inp (High) = -1. Parameter Ip (Low) = Isp (High) were tuned between 110–175 until all
the brown area (strongly positive) represented all the positive collagen staining area, the blue area (negative) represented all the blank area (including macrocellular steatosis and ballooning), and the orange and yellow area (medium and weakly positive) covered the rest. All other parameters were kept the same for all samples except for Ip (Low) = Isp (High). Analysis was then started and CPA was calculated as (Number of strong positive pixels/Number of total pixels) × 100%. The intraobserver variability were tested in 20 random samples and the absolute mean difference was 0.11 ± 0.04%. CPA stratification was determined and adjusted by a previous study {Huang, 2014 #8}. Group 1 was catagorised as CPA < 3%; group 2 CPA 3–5%; group 3 CPA 5–10% and group 4 CPA > 10%.
Statistical analysis
Statistical analyses were performed by IBM® SPSS® Statistics v 26.0 (IBM Corporation, NY, US). Baseline data were presented as mean and standard deviation (SD) or number and percentages as appropriate. Cox regression analysis was used to demonstrate the association between CPA values and risks of long-term clinical outcomes. Confounding factors including age, diabetes, and hypertension were adjusted in the multi-variate Cox regression model. Area under the receiver operating characteristic curve (AUROC) was used to demonstrate the prediction capacity for CPA and NIT markers. A p value less than 0.05 was considered as statistically significant.