Assessment of the global two-stage method to EC50 determination

Preclinical studies of a new drug supply sets of concentration-effect observations. For each experiment, EC50 is estimated through fitting of the Emax model. Naive pooling of these individual estimates, also called the Standard Two-Stage method (STS), is usually performed. A better combination is obtained by the Global Two-Stage method (GTS) which takes into account the variability of the nonlinear regression estimation errors. The performances of STS and GTS are compared on real and simulated data. The results show that GTS performs better than STS in terms of bias or RMSE, especially in case of poorly designed experiments. Degradation of the quality of STS results in simulations appears to be mainly due to some experiments that are usually rejected by experimenters. Such rejections are avoided when using GTS, which hence is particularly suitable for systematic treatment of this kind of data.