Bridging Data Gaps in Oncology: Large Language Models and Collaborative Filtering for Cancer Treatment...

Area of Science:

• Artificial Intelligence in Oncology
• Clinical Trial Data Analysis
• Personalized Medicine

Background:

• Rare cancer patients face limited evidence-based treatment options due to sparse clinical trials.
• Large language models (LLMs) and recommendation algorithms can leverage existing clinical trial data to improve treatment decisions.

Purpose of the Study:

• To develop and validate a novel approach for identifying effective cancer treatments using LLMs and recommendation algorithms.
• To address the challenge of limited treatment options for rare cancers by systematically analyzing clinical trial data.

Main Methods:

• Utilized LLMs to extract and standardize over 100,000 cancer trials from ClinicalTrials.gov.
• Annotated each trial using a custom scoring system for cancer-treatment interactions.
• Implemented three collaborative filtering algorithms to recommend treatments across various cancer types.

Main Results:

• Created a comprehensive database covering 78 cancer types and 5,315 interventions.
• Recommendation models achieved high predictive accuracy (cross-validated RMSE: 0.49-0.62).
• Identified clinically meaningful novel treatments for melanoma, validated by experts.

Conclusions:

• Demonstrated a proof of concept for combining LLMs and recommendation algorithms to identify novel cancer treatments.
• This integrated approach can accelerate the discovery of effective therapies for rare cancers.
• Potential to improve patient outcomes by providing evidence-based treatment recommendations where data is scarce.