In their cost-effectiveness analysis of interferon treatment for chronic hepatitis B infection, Wong and colleagues [1] used probabilities of seroconversion to hepatitis B e antigen (HBeAg) negative status that were substantially higher than those of other studies [2, 3]. It is troubling that this rate was derived from a briefly described, previously unpublished metaanalysis. Rates used in pooling were omitted, and trials with doses of less than 480 MU and a crossover trial, both study exclusion criteria, were apparently included.

Using the same meta-analytic method [4], we could not reproduce the reported rate difference (36.5%), nor could we identify the reported 552 patients meeting all inclusion criteria. In our closest approximation (n = 560), the rate difference between the patients receiving the highest doses and the controls was 32.8% (95% CI, 21.0% to 44.6%). Unlike Wong and associates, we rejected the hypothesis of homogeneity (P = 0.001), indicating that pooling may be inappropriate. If patients treated with less than 480 MU are excluded, the rate difference is 30.5% (CI, 18.6% to 42.4%; P = 0.014 for homogeneity). If the crossover trial [5] is also excluded, the rate difference decreases to 25.2% (CI, 16.1% to 34.3%), and the hypothesis of homogeneity is supported (P = 0.223).

Our recalculations show that small changes in selection criteria substantially affect the outcome, particularly when heterogeneity is present. Interpretation of model-dependent analyses must include evaluation of assumptions and input data. Our failure to replicate results for one end point may not invalidate the entire cost-effectiveness analysis but does indicate a need for closer scrutiny. Even the most meticulous reader, however, would not find adequate information in the report to evaluate other major assumptions. The primarily univariate sensitivity analysis as presented could not compensate for systematic error in the rates and their variances.

The inconsistencies we noted are particularly distressing because they occur in a study funded by the pharmaceutical company that manufacturers the product being evaluated. Such studies should be held to an especially high standard of peer review.