The model-based dose-finding method for the combination of two agents consists of the following three components; 1) dose-toxicity model, 2) start-up dose allocation rule before model-based dose-finding stage, and 3) restriction on skipping dose levels in the dose-finding algorithm. Although many authors have developed flexible dose-toxicity models as well as the start-up dose allocation rule, the restriction on skipping dose levels during the trial, has not been adequately studied. In this paper, we propose a new restriction that permits the dropping of dose combinations with toxicity probabilities that are expected to be statistically high, during the trial. We also compared the operating characteristics of the proposed strategy with those of conventional restrictions using simulation studies. Based on the results of the simulation studies, we were able to determine the performance of these strategies and provide some recommendations for their uses.
Measuring the relative performance of a new treatment in the post-marketing environment is often challenging due to the lack of suitable control. We explore a case-only method to estimate the relative incidence of adverse events in disseminated intravascular coagulation (DIC) patients for comparison of two treatments. We proposed a heuristic approach that borrows its intuition from the self-controlled case series method with modification to the selection of control period. In DIC patients, the method of taking the post-treatment period for self-control is inappropriate because patients generally begin with a high risk of adverse events, which gradually improve with treatment. Instead, we randomly matched the timelines of DIC patients receiving competing treatments for baseline adjustment. Since only the timelines with at least an adverse event, that is, the “cases,” were required, the required sample sizes were smaller. Estimates were comparable to those obtained from the randomized trial and cohort method, and better than results based on the 1-to-1 matched case-control and binary response cohort design. Although the proposed approach loses the benefit of self-matching, with careful matching of patients and repeated random matching of time periods, the method has a potential to be useful for the post-market monitoring of new treatment for DIC or similar diseases.
It is expected to develop new drug more efficiently by incorporating historical data into the current study data. Borrowing historical data which is sufficiently similar to the current data allows increasing power and improving the accuracy of the estimated treatment effect. On the other hand, if the historical data is not similar to the current data, there is a potential for bias and inflated type I error rate. Power prior and hierarchical model are widely known as the Bayesian approaches with borrowing strength from historical information. They have the advantage of deciding the amount of historical information continuously depending on the similarity between historical data and current data. Our goal is to introduce power prior and hierarchical model while showing some examples, and provide a review of points to keep in mind when these approaches are used in the clinical trials.