Revealing Patient Dissatisfaction With Health Care Resource Allocation in Multiple Dimensions Using Large Language...

Area of Science:

• Health Services Research
• Computational Linguistics
• Public Health Informatics

Background:

• Global challenge in accurately measuring healthcare needs for diverse patient populations.
• Need for advanced computational methods to optimize healthcare resource allocation and reduce waste.

Purpose of the Study:

• To assess patient dissatisfaction with healthcare resource allocation across different diseases.
• To demonstrate the effectiveness and practicality of large language models (LLMs) in healthcare resource assessment.
• To support Sustainable Development Goal 3 (Good Health and Well-being) through optimized resource distribution.

Main Methods:

• Utilized aspect-based sentiment analysis (ABSA) on patient reviews.
• Employed ChatGPT with chain-of-thought (CoT) prompting for ABSA across patient experience, physician skills, and infrastructure.
• Integrated International Classification of Diseases 11th Revision (ICD-11) API for disease-specific sentiment classification.

Main Results:

• ChatGPT 3.5 demonstrated superior performance considering stability, cost, and runtime.
• Achieved a weighted total precision of 0.907 and an average accuracy of 0.893.
• Identified highest dissatisfaction for sex-related diseases and lowest for circulatory diseases; highlighted infrastructure needs for blood-related diseases in China.

Conclusions:

• LLM-driven analysis of patient reviews, combined with ICD-11 classification, effectively assesses disease-specific healthcare needs.
• The proposed method provides a rational approach to healthcare resource allocation.
• This technology can significantly aid public health management and resource optimization.