Background: The distinction between definitive nonalcoholic steatohepatitis (NASH) and other histological patterns of nonalcoholic fatty liver disease (NAFLD) continues to rely on liver biopsy, which has risks and limitations.
Objectives: We explored the usefulness of a Bayesian approach to establish the likelihood ratios (LRs) of different noninvasive diagnostic modalities (circulating cytokeratin-18 fragment levels, hydrogen breath test, transient elastography) for distinguishing definitive NASH from other forms of NAFLD.
Patients and methods: From a series of 235 consecutive patients with biopsy-proven NAFLD enrolled at a tertiary referral hepatology clinic, we identified 135 patients with definitive NASH and 100 with other forms of NAFLD. Bayesian probabilities in the form of LRs were estimated for predicting definitive NASH.
Results: In conditional logistic regression models, the probability of having definitive NASH was associated with diabetes mellitus [odds ratio (OR) 2.89; 95% confidence interval (CI) 1.71-4.85], positive transient elastography results (OR 2.11; 95% CI 1.08-4.09), and apoptosis marker M30 (OR 2.57; 95% CI 1.78-3.93). We then estimated the LRs for variables independently related to definitive NASH, which were as follows: diabetes(+) LR=2.42; M30(+) LR=2.12; and transient elastography(+) LR=1.77. On the basis of the LR form of Bayes' theorem, the prediction model that took into account diabetes, M30 measurements, and the results of transient elastography at a tertiary hospital determined an 81% probability of distinguishing definitive NASH from other forms of NAFLD.
Conclusion: A Bayesian approach combining clinical, laboratory, and imaging data may be useful for identifying patients with the highest probability of having definitive NASH.