Conclusion: All three simple models had excellent predictive accuracy and were able to stratify risk into clinical meaningful categories. Y HUANG,1,2 W BASTIAAN DE BOER,3 LA ADAMS,1,2 G MACQUILLAN,1,2 E ROSSI,4 M BULSARA,5 GP JEFFREY1,2 1School

of Medicine and Pharmacology, University of Western Australia, Perth, Australia, 2Department of Gastroenterology and Hepatology, Sir Charles Gairdner Hospital, Perth, Australia, 3Department of Anatomical Pathology, PathWest, QEII Medical Centre, Perth, Australia, 4Department of Biochemistry, PathWest, QEII Medical Centre, Perth, Australia, 5Institute of Health and Rehabilitation Research, University of Notre Dame, Perth Australia Background: Collagen proportional area (CPA) is a validated quantitative measure of liver biopsy collagen and is measured using digital image analysis. Compared with Metavir learn more stage, CPA values ≥10% and ≥20% more accurately stratified liver related clinical outcomes. This study aimed to develop a serum model to accurately predict CPA values. Methods: Chronic hepatitis

C patients who had a liver biopsy and serum analyte measurements within six months of biopsy from 1997 to 2012 were included and randomised into a training and validation set (2:1 ratio). A CPA value was obtained for each biopsy using image analysis. Hyaluronic acid (HA), bilirubin, GGT, α2-macroglobulin, ALT, AST, platelet count, prothrombin time, INR, ALP, creatinine and albumin were analysed. Results: 213 patients were included: 142 patients in the training set and 71 in the validation

set. CPA ranged from 1.6% buy Ixazomib to 32.7% in the training set and from 2.8% to 21.3% in the validation set. No significant difference in Metavir stage, CPA value and serum markers were present between the two groups. In the training set univariate analysis found that HA, GGT, α2-macroglobulin, platelet count, INR, prothrombin time, AST and age were significantly correlated with CPA value. HA had the best correlation with a correlation coefficient value of 0.62. These variables were included in multivariate analysis and achieved an R square value of 0.511 to predict PtdIns(3,4)P2 CPA. Using the backwards selection method, three serum markers (HA, α2-macroglobulin and platelet count) which remained significant were included in the final model and achieved an R square value of 0.46 to predict CPA. Using this model the predicted CPA was calculated for each patient. The mean predicted CPA was 7.70 (range: 0.98–28.2) and the mean variance between the predicted and measured CPA was 2.78. The final model had an AUROC of 0.86 (95% CI, 0.78–0.95) to predict those patients with a CPA ≥ 10% and a cut point of 8.7 had a sensitivity of 80.8% and specificity of 85.2%. The AUROC of the model to predict patients with a CPA ≥ 20% was 0.96 (95% CI, 0.91–1.00) and a cut point of 10.7 had a sensitivity of 100% and specificity of 89%. A similar predictive ability of the final model was found in the validation set.