Automating clinical trial matching using LLMs 🔍🤖

Inspiration ✨

Clinical trial matching is a significant issue in healthcare. Many patients miss out on potentially life-saving treatments due to the complexities involved in finding suitable trials. By leveraging AI and automation, we can simplify this process, making it more efficient and accessible for both patients and healthcare providers.

What it does 💡

Our solution takes patient clinical notes and uses them to search for matching trials on ClinicalTrials.gov by generating relevant keywords through Large Language Models (LLMs). GPT-4 is then utilized to determine a patient’s eligibility by analyzing the inclusion and exclusion criteria. Our patient-to-trial matching approach automates pre-screening, streamlines the evaluation of eligibility criteria, and provides step-by-step reasoning.

How we built it 🛠️

LLM --> ClinicalTrials.gov API --> GPT4 assessment

• Python Script: We developed a Python script to automate the process of fetching clinical trials from ClinicalTrials.gov using their API.
• Keyword Generation: LLMs were used to identify between 1-3 keywords for each patient, ensuring that the search is precise and relevant.
• Eligibility Assessment: We prompted GPT-4 to assess the patient’s eligibility for each trial step by step. The model assigns a score:
  • Inclusion Criteria: Adds a score of (100 / number of inclusion criteria) for each criterion met.
  • Exclusion Criteria: Deducts a score of (100 / number of exclusion criteria) * 2 for each criterion met.
• Classification: Based on the scores, patients are classified as "eligible," "ineligible," or "irrelevant."

Challenges we ran into 🚧

• Problem Research: Understanding the intricacies of clinical trials and eligibility criteria required extensive research.
• Government Website: Integrating with ClinicalTrials.gov and handling its API presented several technical challenges.
• Safe AI Implementation: Ensuring that our AI models operate safely and ethically, especially in a sensitive domain like healthcare, was a major concern.

Accomplishments that we're proud of 🏆

• Efficiency: Created an AI tool that significantly improves and accelerates the process of matching patients with clinical trials.
• Accuracy: Our system provides accurate and detailed assessments of patient eligibility, reducing the workload on healthcare professionals.
• Innovation: Successfully integrated advanced AI techniques to solve a real-world problem in healthcare.

What we learned 📚

• Clinical Trials Terminologies: Gained a deep understanding of the terms and conditions associated with clinical trials.
• AI Implementation: Learned how to effectively implement AI models in a practical application, ensuring they meet real-world requirements.

What's next for MedMatch 🚀

• EHR Integration: Plan to integrate our solution directly into Electronic Health Record (EHR) systems like Epic and Meditech using their API development platforms. This will allow for seamless integration into existing healthcare workflows, making our tool even more accessible and useful for healthcare providers.

Built With

• api
• clinicaltrials.gov
• gpt-4o
• llm