# AutoTrial: Prompting Language Models for Clinical Trial Design

**Authors:** Zifeng Wang, Cao Xiao, Jimeng Sun  
**arXiv:** [2305.11366](https://arxiv.org/abs/2305.11366)  
**Date:** 2023-05-19  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.11366v2/

## Key points

- Clinical trials are crucial for drug development, and appropriate eligibility criteria are essential for success.
- AutoTrial is a method that uses language models to assist in designing clinical trial protocols.
- AutoTrial incorporates controllable generation, scalable knowledge incorporation, and multi-step reasoning to generate high-quality criteria texts.
- The method aims to reduce the risk of clinical trial failure by ensuring proper trial design and power evaluation of therapies.
- AutoTrial's effectiveness depends on the quality of data used for training the language model.
- Human clinicians should make final decisions regarding trial design and safety considerations despite AutoTrial's capabilities.
- AutoTrial outperforms baseline methods based on GPT-2 in generating criteria texts for different target diseases.
- Regular updates to training data and careful consideration of limitations are necessary for maximizing the utility of AutoTrial in clinical trial protocol design.

## Summary

, , , , 

Clinical trials play a crucial role in the development of new drugs, and designing appropriate eligibility criteria is essential for their success. In this paper, the authors introduce AutoTrial, a method that utilizes language models to assist in the design of clinical trial protocols. AutoTrial incorporates controllable generation under instructions, scalable knowledge incorporation, and multi-step reasoning to generate high-quality criteria texts that are fluent, coherent, and clinically accurate. The method aims to reduce the risk of clinical trial failure by ensuring proper trial design and adequate power to evaluate proposed therapies. <break>

<break>

While AutoTrial shows promising results in generating criteria texts, it is important to acknowledge its limitations. The method's effectiveness relies on the quality of data used to train the language model, highlighting the importance of using high-quality and up-to-date clinical trial databases for training. Additionally, AutoTrial may not account for unexpected or rare side effects that could impact the safety and efficacy of treatments during trials. Therefore, it is crucial for human clinicians to make final decisions regarding trial design and safety considerations. <break>

In-depth analysis of AutoTrial's performance across different target diseases reveals superior results compared to baseline methods based on GPT-2. The method demonstrates higher median performance and more stable score distribution across various metrics. Furthermore, qualitative analysis showcases consistent wins over baselines when targeting common diseases/conditions. <break>

Overall, AutoTrial presents a valuable tool for designing clinical trials by leveraging language models. While it offers significant support in generating eligibility criteria texts, human expertise remains essential in evaluating overall trial design and safety aspects. Regular updates to training data and cautious consideration of potential limitations are necessary for maximizing the utility of AutoTrial in clinical trial protocol design.