Selection of Quantitative Decision-Making Criteria Using Weighted Decision Error Rates.

Abstract

Quantitative decision-making frameworks provide objective criteria for advancing a drug development program. Within this process, two types of decision error can occur: proceeding to the next phase and failing (incorrect go) or not proceeding when success would have been achieved (incorrect no-go). This paper discusses strategies to reduce these errors in the context of a phase 2/3 development programme by selecting an optimal go threshold, which minimises the combined risk of an incorrect go or incorrect no-go decision. If the importance of these two errors differs, we propose the use of a weighted decision error rate (WDER) to select the optimal go threshold. We explore the influence of the choice of weighting and the prior beliefs about the drug efficacy on the optimal decision rule and demonstrate how the WDER can guide phase 2 sample size determination, often resulting in a larger sample size paired with a more stringent go threshold for efficiency. We advocate for a new definition of the probability of success in phase 2, emphasizing that it should reflect the probability of making the correct decision about proceeding, rather than just the probability of advancing to phase 3. This approach ensures that decision-making in drug development is both robust and tailored, enhancing the risk-benefit assessment for each phase transition.