This paper addresses the importance of uncertainty quantification in clinical trial outcome prediction, particularly in medical diagnosis and drug discovery. The authors propose incorporating selective classification into the Hierarchical Interaction Network (HINT) to enhance the model's ability to discern nuanced differences and improve overall performance. Selective classification allows the model to withhold decisions for ambiguous or low-confidence samples, thereby enhancing prediction accuracy. Comprehensive experiments demonstrate significant improvements in key metrics such as PR-AUC, F1, ROC-AUC, and overall accuracy. Specifically, the proposed method achieved relative improvements of 32.37%, 21.43%, and 13.27% in PR-AUC over the base model (HINT) in Phase I, II, and III trial outcome prediction, respectively. The findings highlight the robustness and potential utility of this strategy in clinical trial predictions, setting a new benchmark in the field. The paper also discusses the limitations and future directions of the approach, emphasizing the need for continued development and refinement of models like HINT to improve precision and care in predictive analytics within clinical research.This paper addresses the importance of uncertainty quantification in clinical trial outcome prediction, particularly in medical diagnosis and drug discovery. The authors propose incorporating selective classification into the Hierarchical Interaction Network (HINT) to enhance the model's ability to discern nuanced differences and improve overall performance. Selective classification allows the model to withhold decisions for ambiguous or low-confidence samples, thereby enhancing prediction accuracy. Comprehensive experiments demonstrate significant improvements in key metrics such as PR-AUC, F1, ROC-AUC, and overall accuracy. Specifically, the proposed method achieved relative improvements of 32.37%, 21.43%, and 13.27% in PR-AUC over the base model (HINT) in Phase I, II, and III trial outcome prediction, respectively. The findings highlight the robustness and potential utility of this strategy in clinical trial predictions, setting a new benchmark in the field. The paper also discusses the limitations and future directions of the approach, emphasizing the need for continued development and refinement of models like HINT to improve precision and care in predictive analytics within clinical research.