This paper addresses the challenge of uncertainty quantification and interpretability in clinical trial approval predictions. The authors propose a selective classification (SC) approach integrated with the Hierarchical Interaction Network (HINT), a state-of-the-art model for clinical trial prediction. SC allows the model to abstain from making predictions for ambiguous or low-confidence samples, enhancing both accuracy and interpretability. The method is evaluated on a comprehensive dataset, showing significant improvements in precision-recall metrics across all phases of clinical trials. The results demonstrate that the proposed approach not only enhances the model's performance but also provides valuable insights into the decision-making process, making it a valuable tool for clinical trial management and resource allocation.This paper addresses the challenge of uncertainty quantification and interpretability in clinical trial approval predictions. The authors propose a selective classification (SC) approach integrated with the Hierarchical Interaction Network (HINT), a state-of-the-art model for clinical trial prediction. SC allows the model to abstain from making predictions for ambiguous or low-confidence samples, enhancing both accuracy and interpretability. The method is evaluated on a comprehensive dataset, showing significant improvements in precision-recall metrics across all phases of clinical trials. The results demonstrate that the proposed approach not only enhances the model's performance but also provides valuable insights into the decision-making process, making it a valuable tool for clinical trial management and resource allocation.