Answers are grounded in this paper and cite their sources.
Causal inference tools, in particular those of variance decomposition, hierarchical data structures and counterfactuals, are applied to the study of the methodology of dose-finding studies in oncology. A detailed variance decomposition brings into a much sharper focus the relative performance of different designs. We develop and present new results on the role played by the order of patient inclusions into a sequential dose-finding study. These results make it clear why, previously, authors could easily be misled into a conclusion that different designs enjoy similar performances. This is not so and we show how to avoid making that mistake. We highlight our findings via both theoretical and numerical studies. • We reconsider the problem of critical evaluation of competing early phase clinical trial designs from the viewpoint of causal inference. • The causal inference approach allows us much deeper insights into a design’s performance. New evaluation metrics are suggested. • This increased sensitivity enables investigators to observe important differences between designs that had remained less clear-cut via a traditional analysis. • A variance decomposition provides significant insights, in particular into the role played by the order of patient inclusion in the study.
No takes yet. Share an insight, caveat, or question.
Clertant et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: