Performance of Biochemical Response Criteria for Predicting Hepatic Decompensation in Cirrhotic Patients with Primary Biliary Cholangitis

Background and Aims Biochemical response (BR) criteria following ursodeoxycholic acid (UDCA) therapy are widely used to risk-stratify patients with primary biliary cholangitis (PBC), but their prognostic utility in established cirrhosis is uncertain. We compared four established BR criteria for predicting hepatic decompensation in cirrhotic PBC patients receiving UDCA. Methods In a retrospective cohort of adults with PBC and cirrhosis treated with UDCA for at least one year, BR was assessed at 12 months using the Barcelona, Paris I, Paris II, and Rotterdam criteria. The primary outcome was first or further hepatic decompensation within 12 months following BR assessment. Predictive performance was evaluated using sensitivity, specificity, predictive values, accuracy, logistic regression (with penalized likelihood where separation occurred), and receiver operating characteristic (ROC) analysis with the DeLong test on the common set of evaluable patients. Results Eighty-four patients were included; 13 (15.5%) developed decompensation within 12 months (ascites 9, jaundice 2, variceal bleeding 2, hepatic encephalopathy 1). Eligibility for BR assessment varied across criteria (Barcelona 79.8%, Paris I/II 90.5%, Rotterdam 39.3%). Paris I and Paris II non-response were significantly associated with decompensation (Firth OR 4.45 [95% CI 1.35–15.81], p=0.017; and OR 5.73 [1.53–31.34], p=0.020), whereas Barcelona non-response was not (OR 1.26 [0.34–5.60], p=0.74). Positive predictive value was modest across criteria (17.8%–44.4%); negative predictive value was high (86.4%–100%) but only modestly exceeded the 84.5% baseline compensated proportion. In MELD-3.0–adjusted penalized models, Paris I and Paris II non-response remained independently associated with decompensation and improved discrimination over MELD-3.0 alone (AUC 0.82 and 0.74 vs 0.57; DeLong p=0.013 and 0.043). Among baseline variables, lower platelets and albumin, higher INR, prior decompensation, and higher Fib-4/APRI were associated with decompensation, whereas MELD-3.0 and cholestatic markers were not; in a parsimonious penalized model, thrombocytopenia was independently associated and Paris II non-response showed a borderline-significant independent association. Conclusions In PBC with established cirrhosis, Paris I and Paris II non-response identify patients at higher risk of decompensation and add prognostic information beyond MELD-3.0. Because the positive predictive value of non-response is modest and the incremental rule-out value of response is limited, BR should be integrated with established markers of cirrhosis severity and fibrosis rather than used in isolation. Given only 13 events in a single-center cohort, these findings are hypothesis-generating.