Deep learning methods for clinical trial design, execution, and analysis

Clinical trials are essential for developing new treatments, where drugs are tested on carefully selected cohorts in controlled settings and must pass rigorous safety and efficacy standards across all trial phases before receiving approval and becoming available to patients. Despite their importance, the current process of conducting clinical trials is often lengthy, costly, and burdensome for both patients and researchers. On average, it takes 10~15 years and costs $2.87 billion for a drug to complete all trial phases, with most drug candidates failing the trials. The vast and diverse data generated throughout this process, including clinical documents, medical images, and patient records, presents an opportunity for deep learning methods to leverage historical patterns and streamline clinical trial tasks. This thesis focuses on developing deep learning methods to enhance clinical trials across their entire lifecycle, encompassing design, execution, and analysis. First, we explore AI-driven approaches to streamline the review of clinical studies in the literature, facilitating the generation of clinical evidence and novel hypotheses. Additionally, we investigate patient and trial outcome prediction models, which serve as a foundation for trial simulation and design optimization. Next, we introduce methods that improve clinical trial document drafting through enhanced retrieval and generation, alongside techniques for assessing participant eligibility to streamline recruitment. Finally, we explore AI-driven code generation methods for analyzing clinical trial data, enabling more efficient result reporting and evidence generation. Through these contributions, this thesis advances the application of deep learning in medical science discovery, with a particular emphasis on accelerating and optimizing clinical trials.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois Urbana-Champaign

Year dc:date

2025

Author dc:creator

• Wang, Zifeng

Contributors dc:contributor

• Sun, Jimeng
• Han, Jiawei
• Tong, Hanghang
• Lu, Zhiyong
• Wang, Sheng

• deep learning
• clinical trial
• large language model

Statement dc:rights

• Copyright 2025 Zifeng Wang

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/129374

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F129374