A generalized LLMs framework to support public health financing through probabilistic predictions and uncertainty...

Area of Science:

• Public Health Policy
• Health Economics
• Artificial Intelligence in Healthcare

Background:

• Effective public health interventions necessitate a holistic, multisectoral policy approach to integrate "Health-for-All" principles.
• Translating multisectoral strategies into actionable policy requires robust mapping of public budgets to health outcomes and their determinants.
• Current manual budget-tagging methods are resource-intensive and costly, hindering efficient policy implementation.

Purpose of the Study:

• To introduce Categorical Perplexity-based Uncertainty Quantification (CPUQ), a novel Large Language Models (LLMs) framework designed for cost-effective budget-to-indicator and indicator-to-indicator mapping.
• To demonstrate CPUQ's capability in generating interpretable mappings and incorporating model uncertainty for enhanced accuracy and safety in public health budget planning.
• To evaluate the effectiveness of CPUQ in supporting policymakers and advancing the "Health-for-All" agenda.

Main Methods:

• Development of a model-agnostic LLMs framework named CPUQ.
• Utilization of categorical-style prompts to generate interpretable Bernoulli and categorical distributions for a Text-attributed Graph.
• Association of the Text-attributed Graph with descriptions of budget items and health indicators.

Main Results:

• CPUQ framework demonstrates effective alignment with expert annotations for budget-to-indicator mapping.
• CPUQ estimates more nuanced indicator-to-indicator relationships compared to alternative LLMs-based methods.
• The proposed prompting strategy effectively incorporates model uncertainty into the final outputs, enhancing accuracy and safety.

Conclusions:

• CPUQ offers a cost-effective and accurate LLMs-based solution for public health budget planning and indicator mapping.
• The framework enhances decision-making by providing interpretable mappings and quantifying model uncertainty.
• Leveraging LLMs like CPUQ can significantly support the implementation of "Health-for-All" initiatives across governmental and institutional levels.