Answers are grounded in this paper and cite their sources.
Integrating clinical and omics data has the potential to improve survival prediction by combining complementary prognostic information. However, these data types differ substantially in dimensionality, measurement error, and information density, making their effective integration challenging. When directly combined, the much larger number of omics variables may dominate the modeling process, potentially overshadowing clinically informative predictors. To address this challenge, we developed a stacking-based framework, a two-layer ensemble learning approach, for integrative survival prediction using clinical and omics data. Within the stacking-based framework, we proposed two strategies for integrative prediction: (1) component-wise base-learner stacking; (2) component-wise penalty factor base-learner stacking. The latter was further extended to subset stacking by partitioning omics data into multiple subsets. To adapt stacking to survival prediction, we further investigated three meta-learners based on the integrated Brier score, negative log-likelihood, and partial negative log-likelihood loss functions. The proposed stacking-based strategies were compared with existing competitors using real-world and simulated datasets. The proposed stacking strategies generally achieved more accurate and consistent survival predictions across multiple real-world and simulated datasets than existing integration methods. Another advantage of the stacking-based framework is its flexibility in constructing customized base learners (evidenced by subset stacking). For survival stacking, the negative log-likelihood meta-learner achieved comparable accuracy to the integrated Brier score meta-learner but showed greater stability. The proposed stacking framework showed competitive and relatively stable prediction performance across the datasets evaluated in this study, while allowing flexible construction of base learners. These findings support further investigation of stacking as a framework for integrating heterogeneous clinical and omics data in survival prediction.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: