Sequential allocation trial design in anesthesia: an introduction to methods, modeling, and clinical applications

Summary Estimation of the dose–response curve for new anesthetic protocols typically focuses on identifying minimum effective doses. The application of a sequential experimental method is appropriate, as it minimizes sample size requirements by updating dose assignments based on information accrued from successive subjects. One approach is the up‐and‐down method for estimating the median effective dose in a patient population ( ED 50 ). Designs better suited for achieving greater than 50% effectiveness, include the biased coin approach, and continual reassessment method. In this review we introduce different sequential design methods, provide examples of their use, and show through simulation how the method employed influences sample size and the accuracy of the estimated dose. Simulation studies are presented to illustrate the effects of dose parameter and stopping rule choice for up‐and‐down method and biased coin approach. For continual reassessment method, the effects of assumed dose–response model, prior guess, and cohort size are simulated. A binary response regression curve was fit to the data in Saidman and Eger's endtidal halothane dose‐finding study to provide a dose–response curve for generating simulations. A range of options exist when designing a study using sequential allocation with biased coin approach or continual reassessment method. Method choice influences the required sample size and confidence in estimated effect. In the halothane example, up‐and‐down method decreases the required sample size by 20–30% when the choice of design parameters is optimal. For both up‐and‐down method and biased coin approach designs, greater sample sizes, arising from adjusted stopping criteria, might be required to achieve reliable estimates. The continual reassessment method is only efficient if a limited range of doses can be chosen a priori . In conclusion the up‐and‐down method can be more efficient than nonsequential designs for the estimation of the median dose/intervention level for a given intervention ( ED 50 ). The biased coin approach or continual reassessment method are preferred for the estimation of higher or lower tail quantiles such as ED 90 or ED 10 . Continual reassessment method may be superior if knowledge of the dose–response relationship is available for the drug of interest.