Defining the Minimum Acceptable Diagnostic Accuracy of Noninvasive Fibrosis Testing in Cirrhosis: A Decision Analytic Modeling Study

Abstract

No studies explore the clinical consequences of using noninvasive tests (NITs) compared to liver biopsy (LB) in diagnosing cirrhosis. Our aim was to combine two decision analytic models to determine the minimum diagnostic accuracy criteria for NITs to diagnose cirrhosis with equivalence to LB in terms of mortality. We further evaluated selected existing NITs used alone and sequentially. A decision tree was constructed with associated 2-year mortality incorporating an LB or NIT strategy to diagnose cirrhosis in a hypothetical cohort of 1,000 asymptomatic patients. Cirrhosis prevalence was modeled at 5%, 20%, and 50%. Decision curve analyses were performed, expressing the net benefit of tests over a range of threshold probabilities (P_t ). The NIT deriving from the two models that could diagnose cirrhosis with at least equal mortality to LB was termed "mNIT." Existing NITs were then compared using both decision models. The combined mNIT minimum sensitivity and specificity to diagnose cirrhosis with equivalence to LB at 5%, 20% and 50% cirrhosis prevalence were; 89% and 88%, 94% and 85%, and 94% and 87%, respectively at P_t = 0.20. Sequential NITs performed better than single NITs at any prevalence. Combining both decision models, FibroTest plus vibration-controlled transient elastography (VCTE) and VCTE alone were the only existing NITs that were better than or equal to LB at diagnosing cirrhosis at 5% prevalence. At 20% and 50% prevalence, only FibroTest high specificity cutoff plus VCTE was equivalent to or better than LB. Conclusion: Decision analytic models were used to determine the minimum acceptable diagnostic accuracy of NITs for diagnosing cirrhosis; we recommend that such models should be used as the standard in evaluating the diagnostic performance of NITs.