On the relative efficiency of model-assisted designs: a conditional approach

Abstract

In phase I dose-finding trials, model-assisted designs are a novel class of designs that combine the simplicity of algorithm-based methods with the superior performance of model-based methods. Examples of model-assisted designs include the modified toxicity probability (mTPI), Bayesian optimal interval (BOIN) and keyboard designs. To achieve simplicity, these model-assisted methods model only "local" data observed at the current dose, typically using a binomial model, to guide dose assignments. This potentially causes efficiency loss, however, by ignoring the data observed in other doses. To investigate this issue, we propose a conditional approach that utilizes the data from both current and adjacent (i.e., next higher or lower) doses to make the doseassignment decisions. Specifically, we propose the conditional optimal interval (COIN) design, as the conditional approach extension of the BOIN design. We investigate the theoretical properties of the COIN design, and conduct extensive numerical studies to examine its performance in comparison with existing model-assisted designs. We also present the conditional approach to the keyboard design. We observe that the conditional approach improves patient allocation, but yields similar maximum tolerated dose (MTD) identification accuracy as the model-assisted designs, suggesting only minor efficiency loss using local data under the model-assisted designs.

An overview of the BOIN design and its current extensions for novel early-phase oncology trials

“… TITE-BOIN uses DLT data only at the current dose for dosing decisions, in contrast to TITE-CRM, which considers data from all dose levels. However, simulations show that the effect of using only the current dose data leads to negligible efficiency loss [ 25 , 38 ]. TITE-BOIN assumes that the time to DLT is distributed uniformly over the DLT assessment window, similar to what TITE-CRM does.…”

Commentary on “Improving the performance of Bayesian logistic regression model with overdose control in oncology dose‐finding studies”

“…This explains why BOIN and keyboard design generally yield comparable performance to CRM/BLRM, which uses a dose‐toxicity model to borrow information across doses. Lin and Yuan 5 showed that incorporating data other than the current dose into BOIN provides virtually no efficacy gain. In addition, at the end of the trial BOIN actually uses all data across doses to estimate and identify the MTD by isotonic regression.…”

Bayesian Optimal Interval Design with Multiple Toxicity Constraints

“…These model‐assisted interval designs can be implemented in a simple way that is similar to the design, but yield performance that is comparable to the more complicated model‐based designs, (Oron et al, ; Liu et al, ). Lin and Yuan () and Zhou et al, () showed that, although the model‐assisted designs only utilize the accumulated local data at the current dose for decision making, there is no significant loss of efficiency due to not borrowing information across the doses. In addition, studies show that the BOIN and keyboard designs have higher accuracy of identifying the MTD and lower risk of overdosing patients than the mTPI design (Yan et al, ).…”