BACKGROUND: Mixed treatment comparisons (MTC) is useful in comparing treatments when not all treatment-pairs have available head-to-head data. A previous MTC analysis (Dakin&James, EASL 2008) showed results that are inconsistent with observed data from clinical trials. We conducted analyses to understand how changing assumptions/implementations affect the MTC results. OBJECTIVES: To evaluate the performance of different MTC models in assessing the relative efficacy of available nucleoside/tides and combinations in antiviral-naïve patients with HBeAg(+) CHB. METHODS: Proportions of HBeAg(+) CHB patients with undetectable HBV DNA at Year 1 were from published trials referenced in D&J. Bayesian MTC analyses were conducted using models and assumptions proposed in Higgins and Whitehead (H&W, 1996) and Lu and Ades (L&A-Unconstrained and L&A-Constrained, 2004). Analyses were implemented in WinBUGS v1.4. Model performance was evaluated by how well it fit observed proportions. RESULTS: The dataset was small relative to the number of comparisons evaluated. Only 10 randomized controlled trials satisfied D&J inclusion/exclusion criteria, yielding data on only 12 of 28 possible head-to-head comparisons among the 8 treatments considered (LAM-lamivudine, ADV-adefovir, ETV-entecavir, TEL-telbivudine, TDF-tenofovir, ADV+LAM, TEL+LAM, and PLB-placebo). The H&W estimates were very similar to D&J results. The estimated proportions from L&A-C are most consistent with observed data (see Table). Results could not distinguish among the four most efficacious treatments at Year 1 (ETV, TDF, TEL, TEL+LAM). CONCLUSIONS: Compared with observed data, the L&A-C model is better able to predict observed results than either the D&J, H&W or L&A-U. With limited data, MTC results can vary across models and model performance should be evaluated against observed data. Proportion of patients achieving undetectable HBV DNA at Year 1: The treatments (No. Trials) for TDF(1): 74%, 75.6% (55.9%, 91.1%), 93.7% (80.0%, 99.3%); ETV(3): 70.1% (58%-76%), 67.9% (54.8%, 78.4%), 73.1% (57.6%, 87.6%); TEL(3): 60.1% (60%-61%), 59.1% (44.7%,71.9%), 62.9% (44.8%,81.7%); TEL+LAM(1): 49%, 48.3% (25.6%,72.3%), 53.3% (21.9%,84.3%); ADV(4): 21.0% (13%-40%), 23.7% (14.7%,37.1%), 48.8% (25.8%,77.5%); LAM(6): 38.9% (32%-43%), 37.3% (26.7%,46.9%), 38.4% (33.9%,42.8%); ADV+LAM(1): 39%, 37.9% (17.6%,61.8%), 37.5% (12.5%,68.7%); PLB(2): 3.7% (0%-17%), 4.7% (1.6%,9.9%), 7.1% (1.5%,18.5%) for Weighted Average1 (Min-Max) Observed, L&A-C Implementation2, D&J EASL20082.  1-By sample size; 2-Estimates (95% Bayesian Credible Interval).