A Proof-of-Concept for Explainable Disease Diagnosis Using Large Language Models and Answer Set Programming

TL;DR

McCoy addresses the need for interpretable clinical decision support by merging Large Language Models (LLMs) with Answer Set Programming (ASP). The framework uses LLMs to translate medical literature into ASP rules, enriches these rules with patient data, and employs an ASP solver to generate diagnoses with explicit explanations. Key contributions include automatic knowledge-base construction from literature, integration with patient information, and transparent reasoning traces demonstrated on small-disease datasets with accuracy in the $95$–$100\%$ range. This hybrid approach promises scalable, explainable disease prediction with potential for broader clinical deployment after scaling and rigorous guardrails.

Abstract

Accurate disease prediction is vital for timely intervention, effective treatment, and reducing medical complications. While symbolic AI has been applied in healthcare, its adoption remains limited due to the effort required for constructing high-quality knowledge bases. This work introduces McCoy, a framework that combines Large Language Models (LLMs) with Answer Set Programming (ASP) to overcome this barrier. McCoy orchestrates an LLM to translate medical literature into ASP code, combines it with patient data, and processes it using an ASP solver to arrive at the final diagnosis. This integration yields a robust, interpretable prediction framework that leverages the strengths of both paradigms. Preliminary results show McCoy has strong performance on small-scale disease diagnosis tasks.

Figures (5)

• Figure 1: Solving a problem with ASP. A problem is first modeled as a program, then transformed into ground rules by a grounder. A solver computes the stable models, which are interpreted as solutions.
• Figure 2: McCoy overview. The diagnostic process begins by transforming medical literature into an ASP program using a LLM. An ASP solver then executes the program and produces a final diagnosis based on each patient’s data.
• Figure 3: A prompt that does not yield satisfactory results. The LLM output does not decompose the disease diagnosis into discrete rules.
• Figure 4: An effective prompt that produces accurate and well-structured output. The LLM output is structured into multiple rules that aid explainability and can result in partial diagnoses.
• Figure 5: Causal graph of logic program in \ref{['eq:example-cg']}.