This paper introduces a novel methodological framework for the optimal design of longitudinal trials with discrete-time survival endpoints. We address two critical limitations of existing methodologies: (1) the problematic dependence of model complexity on the number of observation periods, and (2) the narrow focus on treatment effect estimation through D- or \(\hbox {D}_s\) -optimality, which often neglects important predictive targets such as cumulative incidence functions. We propose a prediction-focused optimal design method based on the extended partial logistic regression model and present rigorous equivalence theorems to validate the optimality of the design. Additionally, we introduce a general \(\hbox {D}_{\varvec{A}}\) -optimal criterion for comprehensive comparison. Through four carefully constructed case studies encompassing diverse experimental conditions, we systematically examine the characteristics of our proposed designs. The results reveal distinct support patterns and weight allocation preferences between estimation-focused ( \(\hbox {D}_{\varvec{A}}\) -optimal) designs and prediction-focused ( \(\hbox {I}_{\varvec{ B}}\) -optimal) designs. This work provides researchers with a more flexible and practical framework for designing discrete-time survival studies that better align with their analytical needs.
