Randomized controlled trials are considered the gold standard for therapeutic evaluation. However, empirical evidence showed that the current planning, conduct and reporting of randomized controlled trials is suboptimal. Protocols contain preventable errors that could be easily fixed at the time of conception, and academic papers lack crucial information.
This PhD project will explore how artificial intelligence — and particularly Large Language Models (LLMs) — can automatically detect these issues, and potentially reduce research waste by providing targeted feedback to researchers at crucial stages: before a trial is launched (when the protocol is developed) and before academic papers are published (when first drafts are submitted to journal editors).
When ? starting January 2026
Where ? Hotel Dieu, Paris, France
Supervision: Isabelle BOUTRON (MD, Phd), Thomas STARCK (Phd)
Background sought: MSc in epidemiology, biostatistics or data science; interest in clinical research and AI; English fluency.
To apply: send CV and motivation letter with names of referees to thomas.starck@polytechnique.org
More details :
1. Missions and objectives
The PhD candidate will carry out a research project aiming to develop, evaluate and test artificial intelligence (AI) tools — in particular Large Language Models (LLMs) — to automatically detect and correct frequent methodological flaws and missing information in randomized controlled trials (RCTs).
The overall goal is to reduce research waste by providing automated methodological support to researchers at key stages of the research process:
- at the design stage, when studies are registered on public registries (e.g. ClinicalTrials.gov)
- at the manuscript submission stage, before they are published
The project will focus on three main objectives:
- Assess to what extend AI and LLM are used to improve clinical research planning and reporting, and what are the reported advantages and risks (living scoping review)?
- Automated detection of frequent and easily correctable flaws in the planning of randomized controlled trials (RCTs), for instance on clinical trial registries
- Automatic detection of incomplete reporting of RCTs (using the CONSORT reporting guidelines)
2. Main activities
- Contribute to the design of study protocols and analysis plans
- Build and manage the required datasets
- Develop and test LLM prompts and pipelines for information extraction and detection of methodological and reporting flaws
- Design and conduct evaluation studies (LLM detection performance validation and prospective randomized feedback trials)
- Perform data analyses (diagnostic accuracy, impact analyses, cost–benefit evaluation)
- Draft scientific articles, conference abstracts and oral presentations, and the PhD thesis
- Participate in regular meetings with supervisors and institutional partners
3. Required skills
Knowledge
- Clinical research methodology, especially randomized trials
- Basic concepts and strong interest in Large Language Models
- Performance evaluation methods (diagnostic accuracy, classification metrics, statistical inference)
- Scientific best practices in biomedical research
Practical skills
- Ability to manage and analyze scientific datasets
- Ability to develop and evaluate AI tools (Python or R programming, use of LLM APIs)
- Scientific writing in English
- Project planning, organization and documentation
Personal skills
- Scientific curiosity and ability to learn autonomously
- Rigor, critical thinking and initiative
- Ability to work in a multidisciplinary and international environment
- Clear and concise communication
4. Candidate profile
- MSc in epidemiology, biostatistics, data science, or a related field; interest in clinical research and AI; English fluency.
- Understanding of clinical research and strong motivation to improve research quality
- Interest in AI/LLMs and digital tools applied to science
- Good command of written and spoken English (working language of the project)
- Programming experience (Python or R) is an asset but not mandatory if the candidate is motivated to learn quickly
