Leveraging Large Language Models and Machine Learning for Success Analysis in Robust Cancer Crowdfunding Predictions:...

Area of Science:

• Computational linguistics
• Health informatics
• Machine learning applications in healthcare

Background:

• Large language models (LLMs) offer advanced capabilities for analyzing complex text data.
• Medical crowdfunding campaigns present unique linguistic and social nuances influencing success.
• Existing methods struggle to capture these deeper psychosocial and clinical factors.

Purpose of the Study:

• To develop an integrated framework using LLMs and machine learning for predicting medical crowdfunding success.
• To automatically extract nuanced linguistic, social, and clinical features from campaign narratives.
• To identify key predictors of campaign success beyond structured data.

Main Methods:

• Utilized GPT-4o to extract linguistic and social determinants of health features from cancer crowdfunding narratives.
• Employed a random forest model with permutation importance for feature ranking.
• Evaluated four machine learning algorithms (random forest, gradient boosting, logistic regression, elastic net) using 10-fold cross-validation.

Main Results:

• Gradient boosting demonstrated superior sensitivity (0.786-0.798) in identifying successful campaigns.
• Key predictors identified include medical condition severity, income loss, chemotherapy, clear communication, cognitive understanding, family involvement, empathy, and social behaviors.
• LLM-derived features significantly improved the prediction of campaign success.

Conclusions:

• LLMs like GPT-4o effectively extract nuanced features from medical crowdfunding narratives, providing deeper insights than traditional methods.
• Combining LLM features with machine learning enhances the identification of critical success predictors.
• Findings support using LLMs to improve predictive modeling for health-related crowdfunding and inform targeted support strategies for cancer patients.