Data-dependent contrast test for dose-finding clinical trials

We propose a simple and powerful data-dependent contrast test with ordinal-constraint contrast coefficients of the dose response determined from observed responses. The contrast coefficients are easily calculated using a pool-adjacent-violators algorithm and making assumptions for the contrast coefficients. Once the dose response is determined for p < 0.05 in the data-dependent contrast test, the best dose-response model is selected from multiple dose-response models. Using the best model, a recommended dose is identified. We demonstrate the data-dependent contrast test for sample data. In addition, we calculate the ordinal-constraint contrast coefficients and test statistic for an actual study, and we obtain a recommended dose. Finally, we perform a simulation study with 11 scenarios to evaluate the performance of the data-dependent contrast test by comparing multiple comparison procedures with modeling techniques. We confirm the dose response for both the sample data and the actual study. In the simulation study, the data-dependent contrast test is more powerful than the conventional method on the simulation datasets generated using non-dose-response models. In addition, the type-1 error rate of the data-dependent contrast test remains at a significant level when there is no difference between the treatment groups. We conclude that the data-dependent contrast test can be applied unproblematically in a dose-finding clinical trial.