gms | German Medical Science

After successful phase I, various options are available to design subsequent phase II trials. One approach is to perform a single-arm trial, where the response rate in the intervention group is compared to a pre-fixed value for the proportion. Alternatively, a randomised phase II trial comparing the new treatment with placebo or the current standard may be conducted. However, a problem arises in both approaches when the investigated patient population is very heterogeneous regarding prognostic and predictive factors associated with the response, which is frequently the case, e.g. in oncology. Especially for small sample sizes, the observed response rates may substantially differ from the true response rate since the study population might not well reflect the characteristics of the underlying patient population. Additionally, in a usually small-sized randomised phase II trial, an imbalanced distribution of confounders across treatment arms may cause biased treatment effect estimates [1]. An adjustment may be impeded as imbalanced populations may cause instability in statistical models.

Frequently, a substantial dataset of historical controls exists, for example registry data. For this situation, we propose an approach to enhance the classical single-arm design by including matched control patients. This approach overcomes the previously described disadvantages. The success of a trial within the proposed design can either be defined by a significant hypothesis test comparing treatment and control group at a specified significance level α, or by a successful crossing of a pre-defined threshold [2]. We named our method "matched-threshold-crossing design" since the name describes both the methodology, i.e. the use of a matching procedure for the inclusion of historical controls, and the aim of the trial, i.e. the crossing of an efficacy threshold. The proposed two-stage design with adaptive interim analysis allows to deal with uncertainties in the planning stage, e.g. with respect to the matching rate, the number of matched controls per patient in the intervention group and the treatment effect. Moreover, early stopping for futility after the first stage is implemented. Other approaches to deal with the issue of incoporating historical data into phase II trials have been proposed in the Bayesian literature (see, e.g. [3], [4]). However, our frequentist approach is unique and novel in the sense that, on the one hand, it allows to incorporate matched historical controls within a two-stage single-arm trial ensuring perfectly stable statistical models and, on the other hand, to deal with the uncertainty about trial parameters by means of an interim sample size reassessment. Performance characteristics of the proposed two-stage-design are investigated in comprehensive simulation studies, and application is illustrated with a clinical trial example.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.