We agree that very old publications declared obsolete or false in the year 2000 could cause the duration of survival to be overestimated if they were in fact obsolete or false many years earlier. However, this risk for survival overestimation does not concern the true conclusions that represent 60% of events.

For each obsolete or false conclusion, we estimated the year in which it became obsolete or false. Most often it was 1980 for biochemical liver tests; 1980 for corticosteroid treatment of hepatitis; and 1990 for non-A, non-B hepatitis, -globulin studies, and studies of portal hypertension. A new survival analysis was performed by using this new estimate of survival duration. With this new estimate, there was a small decrease of truth survival at 50 years for the mean percentage (±SE) of studies without false or obsolete conclusions (22% ± 4% vs. 26% ± 4% with the fixed estimate) but no difference in the percentage of studies without false conclusions (52% ± 5% vs. 53% ± 5% with the fixed estimate). The conclusions and the factors associated or not associated with truth survival did not change.

The 20-year mean survival (±SE) of studies without false or obsolete conclusions was lower in meta-analyses than in other studies (57% ± 10% vs. 73% ± 3%, respectively; log-rank = 11.3 [P = 0.008]); with the fixed estimate, these figures were 57% versus 87%. At 50 years, the survival was higher for negative than for positive conclusions (59% ± 10% vs. 15% ± 3%, respectively; log-rank = 19.4 [P < 0.001]). In randomized trials, the survival rate was higher for negative than for positive conclusions (76% ± 7% vs. 4% ± 4%, respectively, log-rank = 19.4 [P < 0.001]). Still, there was no significant difference between high (56% ± 10%) and low (21% ± 4%) methodologic quality (log-rank test = 0.32; P > 0.2).

The history of interferon in the treatment of hepatitis B remains the best example of the hazards of evidence-based medicine in hepatology. In 1980, clinicians perceived a negative conclusion from a single randomized trial as a major evidence-based result. We still need to be critical about our evidence-based "certainties."