gms | German Medical Science

In early phase crossover trials sample sizes are often not based on formal power calculations but on a given precision for the resulting confidence interval. In many cases such sample size calculations are performed by using software packages that are only able to deal with parallel and paired designs. Thus, if 2×2×2 or other higher-order crossover studies are going to be performed, often the actual sample size needed is only approximated by using a paired design for the calculations. Hence all within-subject comparison studies have to rely on the sample size of a paired design, thereby neglecting the actual trial design. We extended the set of possible study designs allowing for different crossover design options (similar as done within the R package PowerTOST) thereby enabling us to incorporate the actual study design into the calculations. We compared resulting sample sizes and precisions for various crossover designs with the corresponding paired design. Our results show that although the calculations for the actual crossover design is more accurate, calculations using the paired design in general yield a conservative sample size. Additionally a quanitification is provided for the error made when using the paired design for the sample size calculation.