TrialDura: Hierarchical Attention Transformer for Interpretable Clinical Trial Duration Prediction

TL;DR

TrialDura tackles the gap in predicting clinical trial duration by integrating multimodal data (phase, drugs, diseases, eligibility) through Bio-BERT embeddings and a hierarchical attention Transformer. The end-to-end model yields superior predictive accuracy, achieving approximately $MAE \approx 1.04$ years and $RMSE \approx 1.39$ years on a large ClinicalTrials.gov dataset, outperforming traditional baselines by about 7–9%. Interpretability is provided via hierarchical attention and SHAP-like analyses to reveal which text features drive predictions. This approach enables more reliable budgeting, scheduling, and risk management in drug development, with potential to streamline clinical trial management and optimize resource allocation.

Abstract

The clinical trial process, a critical phase in drug development, is essential for developing new treatments. The primary goal of interventional clinical trials is to evaluate the safety and efficacy of drug-based treatments for specific diseases. However, these trials are often lengthy, labor-intensive, and expensive. The duration of a clinical trial significantly impacts overall costs, making efficient timeline management crucial for controlling budgets and ensuring the economic feasibility of research. To address this issue, We propose TrialDura, a machine learning-based method that estimates the duration of clinical trials using multimodal data, including disease names, drug molecules, trial phases, and eligibility criteria. Then, we encode them into Bio-BERT embeddings specifically tuned for biomedical contexts to provide a deeper and more relevant semantic understanding of clinical trial data. Finally, the model's hierarchical attention mechanism connects all of the embeddings to capture their interactions and predict clinical trial duration. Our proposed model demonstrated superior performance with a mean absolute error (MAE) of 1.04 years and a root mean square error (RMSE) of 1.39 years compared to the other models, indicating more accurate clinical trial duration prediction. Publicly available code can be found at: https://anonymous.4open.science/r/TrialDura-F196.

Figures (3)

• Figure 1: Overview of our model. Our model takes phase information, disease, drug molecule and eligibility criteria as the input and predicts the trial duration. We design a hierarchical attention mechanism to model semantic features from eligibility criteria. The lower-level attention mechanism aggregates word embeddings while higher-level attention gathers sentence embeddings.
• Figure 2: Model performance comparison on prediction accuracy.
• Figure 3: Visualization of text segments in the TrialDura model's output, illustrating Shapley values derived from Clinical Trials (NCT03553810). Shapley values correspond to attention weights, with darker colors indicating higher weights. The number and color at the beginning of each sentence represent the attention weight for the entire sentence.

Theorems & Definitions (5)

• definition 1: Clinical Trial Phases
• definition 2: Treatment Set
• definition 3: Target Disease Set
• definition 4: Trial Eligibility Criteria
• definition 5: Clinical Trial Duration