Concurrent Prediction of Multiple Survival Outcomes with a Refined Stacking Algorithm
该研究改进了多任务预测算法，使其能处理多个生存结局的并发预测，通过加速失效时间模型和弹性网模型，在模拟和实际数据中表现优于单任务模型，适用于共病疾病建模。
Xing et al. (2019) developed prediction algorithms, termed multi-task prediction algorithms using revised stacking (MTPS), to enable us to conduct concurrent prediction for multiple outcome variables with high-dimensional predictors integrated into the prediction process. Their algorithms employed the strategy of the stacking algorithm to construct a multi-task learner through a two-step procedure, where separate single learners are constructed in Step 1, and mutually carried information among those learners is then facilitated in Step 2. While their methods handle both continuous and binary outcomes, as well as a mix of them, they are not applicable to the context of survival data, which arises commonly in applications.Expanding their work to handle the prediction of multiple survival outcomes, we develop a new concurrent prediction algorithm by utilizing the revised residual stacking framework, where the parametric accelerated failure time (AFT) model and Elastic Net AFT model are employed. Through simulation studies and a real-data application, we demonstrate that the novel enhancement of MTPS for survival outcomes surpasses the performance of their single learners. Consequently, this newly refined MTPS is recommended for modelling comorbidity diseases. This research offers a new dimension to MTPS, allowing a diverse array of applications spanning various domains.