I am no expert in statistics. However, it seems in group sequential analysis model the interim analysis can go either way since we chose same criteria for early exit. Unless the design has flaw by the designer on purpose I see no reason the interim analysis will overestimate the effect of intervention. Adding a post-hoc power analysis will be more convincing.

Thank you Dr Teng

Supposing that the null hypothesis is (classically) that the "intervention" performs no better than the control and that interim analysis has been properly planned (usually in the context of a group-sequential design) it does not create any bias in favour of the intervention. Actually, depending on the accuracy of the hypotheses underlying the power calculation, it can often result in the opposite, that is a larger and more expensive trial than a classical single-analysis design.

This is why the choice of an adaptive design should result not from the false hope of favouring the intervention but from other considerations, for instance, ethical issues, or uncertainties about the magnitude of the effects or the statistical properties of the evaluation criteria.

Thank you Paul Preziosi 

 Early stopping of randomized trials for apparent benefit following interim alalysis is problematic. It is suggested that stopping clinical trials early for apparent benefit may overestimate treatment effects.

see : Randomized Trials Stopped Early for Benefit: A Systematic Review. Montori VM et al JAMA.2005;294:2203-2209

Frédéric Barbot Hi, Barbot,i'm wondering what if the the interim analysis conflict with the final analysis results. It seems that you have some experience, can you discussion something about it?

Jun Zhou , planning one or more interim analyses usually involves a stopping rule if that analysis allows a statistical conclusion. In such case you do not reach the final analysis. When you go on to the final analysis and reach a conclusion, it is usually because you have more data and a higher statistical power, thus that result should be authoritative .

Paul Preziosi I agree with your opinion. I mean Interim analysis (only half of the patients) reaches a significant level at alpha=0.02. In order to collect more safety data, all patients completed the trial. then the sensitivity analysis show no statistical significance at alpha=0.05。 As you say, more data and a higher statistical power , but if it indicates that interim analysis have some shortcomings?

In my opinion there are statistical problems with stopping clinical trial early for an apparent benefit :

Multiplicity. To deal with multiplicity, 4 methods used , O , Brien-Flemming, Peto and Lan-DeMets. and bayesian approach.

But to minimize the overestimation of the treatment effect with early stopping, choose a low p value (0.001)as the threshold for stopping at the different times of interim analysis, and if possible wait for a large number of events and and continue enrollment to confirm the positive trend.

Montori et al, Jama 2005, showed that methodological information following clinical trials with early stopping was poor ( lack of information concerning sample size, number of interim analyses, stopping rules used....)