Lee, J, F. Thall, P, Msaouel, P. A phase I‐II design based on periodic and continuous monitoring of disease status and the times to toxicity and death. Statistics in Medicine. 2020; 39: 2035– 2050. https://doi.org/10.1002/sim.8528

Patient response to treatment often is very complex, and may include several efficacy and adverse variables, with each observed once or repeatedly over time. In most clinical trial designs, patient outcome is simplified to facilitate application of adaptive rules for safety monitoring, futility or superiority monitoring, dose optimization, treatment selection, or hypothesis testing. The properties of such designs typically are evaluated using the reduced outcomes, which may lead to the putative conclusion that a given design has good properties. Such simplifications often come with a heavy price, since they typically discard important information. When viewed in terms of the full, unreduced outcomes, such designs may have poor performance and lead to flawed inferences. This paper presents a model and phase I-II design to optimize the dose of a new anti-cancer agent in the common clinical setting where, during a pre-specified follow-up period, the times to disease progression, toxicity, and death are monitored continuously, and an ordinal disease status variable, including progressive disease as one level, is evaluated repeatedly by scheduled imaging. A simulation study shows that the proposed full-outcome dose optimization design has greatly superior performance compared to a simpler design using two efficacy and toxicity outcomes obtained by combining the four variables described above. In a wide variety of cases, the full-outcome design has substantially larger probabilities of correctly choosing truly optimal doses and excluding truly unsafe doses. The full-data approach used by the proposed phase I-II design is utility based, and it can be applied much more generally to other clinical trial design settings.