Truace
Subscribe

Conceptualising Health Economic Models Using a Graph Retrieval-Augmented, Agentic AI Framework with Human-in-the-Loop Validation: Proof-of-Concept Study

Objective: Health economic models (HEMs) are crucial for assessing the cost-effectiveness of healthcare interventions, informing policy decisions and guiding reimbursement strategies. This study examines the feasibility of employing generative artificial intelligence (Gen-AI) for the conceptualisation of HEMs. Two proof-of-concept studies were conducted, implementing a first-of-its-kind agentic Gen-AI-augmented approach for model conceptualisation. Methods: An agentic Gen-AI framework was developed, incorporatin…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1418· cite & verify
Conceptualising Health Economic Models Using a Graph Retrieval-Augmented, Agentic AI Framework with Human-in-the-Loop Validation: Proof-of-Concept Study
"Dental Diagram, Copied for Health Department" by Queensland State Archives is marked with Public Domain Mark 1.0. To view the terms, visit https://creativecommons.org/publicdomain/mark/1.0/.

In brief

Researchers built an agentic generative AI framework to help conceptualise health economic models, which are used to judge the cost-effectiveness of healthcare interventions. It combined chain-of-thought reasoning with self-consistency, ran in an external Python environment, and linked a language model to a curated graph-based retrieval database.

The work matters because such models inform policy decisions and reimbursement strategies, so faster conceptualisation could support those processes. The study is a proof of concept with two cases, and the source gives no measures of time saved or accuracy, so wider performance remains uncertain.

Main points

  1. Two case studies were run: breast cancer as a data-rich disease and transthyretin amyloid polyneuropathy as a data-sparse rare disease.
  2. The framework produced a disease concept diagram showing progression and health states.
  3. It also produced a model structure diagram depicting patient transitions between health states.
  4. Outputs were validated against published literature and expert review by health economic modelers and clinicians.
  5. The authors describe it as the first study to combine human-in-the-loop, CoT-SC and graph-based RAG for this purpose.

The gain

An agentic generative AI framework produced disease concept and model structure diagrams for breast cancer and a rare disease that were checked against literature and by modelers and clinicians.

The rundown

The framework used an agentic execution layer that autonomously orchestrated a multi-step retrieval, reasoning and generation task. Prompt engineering delivered chain-of-thought reasoning, and self-consistency was added on top.

Retrieval relied on a graph-based technique connecting the language model to an external database holding user-selected data specific to health economic models. Human-in-the-loop validation was part of the approach.

The authors conclude that the methods can enhance problem definition, disease and treatment effect pathway development, model structure recommendations, and drafting of model specification plans. They also point to wider uses in health economics and outcomes research.

Sources

  1. Peer-reviewedPharmacoEconomics - Open2026-10-12

The debate