Developing a transparent reporting tool for AI-based diagnostic prediction models of disease and syndrome in Chinese medicine: a Delphi protocol

Jieyun Li, Wei Song Seetoh, Jiekee Lim, Xin'ang Xiao, Kehu Yang, Si Yong Yeo, Boyun Sun, Jinhua Liu, Zhaoxia Xu, Linda L. D. ZhongFrontiers Digit. Health 2025 · DOI

TL;DR

This paper introduces a protocol for developing a new reporting guideline, TRAPODS-CM, specifically designed for AI models used in diagnosing diseases and syndromes within Traditional Chinese Medicine (TCM). This is crucial because existing guidelines are tailored to Western medicine and don't account for TCM's unique theoretical framework, leading to a need for standardized, transparent reporting in this rapidly growing field.

What Problem Does This Paper Address?

The integration of Artificial Intelligence (AI) into Traditional Chinese Medicine (TCM) for diagnostic prediction is rapidly advancing. However, a significant challenge arises from the fundamental differences between TCM and Western medicine. TCM diagnoses are based on a holistic perspective, considering "Bing" (disease) and "Zheng" (syndrome), which represent imbalances in the body's functional state. This contrasts with Western medicine's focus on specific disease entities. Current reporting guidelines for AI diagnostic models, such as TRIPOD and STARD, are primarily designed for Western medical contexts. They fail to adequately capture the unique theoretical frameworks, terminology, and classification systems of TCM. This lack of specialized guidelines leads to issues with transparency, reproducibility, and clinical relevance in TCM-related AI research, hindering the reliable adoption of these technologies in clinical practice.

Methods

The study employs a structured, multi-phase Delphi protocol to develop the TRAPODS-CM (Transparent Reporting Tool for AI-based Diagnostic Prediction Models of Disease and Syndrome in Chinese Medicine). This method involves gathering expert consensus through iterative rounds of questionnaires and discussions.

The process begins with a core working group conducting a systematic review of existing literature on TCM diagnostic prediction models to create an initial pool of potential reporting items. This pool is then refined into a draft checklist. Subsequently, multidisciplinary experts in TCM, computer science, and evidence-based methodology are recruited. They will receive Delphi questionnaires to evaluate and rate the importance and relevance of each item. Expert authority is calculated based on their familiarity with the topic and their judgment basis (practical experience, theoretical analysis, etc.). The number of Delphi rounds will be determined by the convergence of expert opinions, with final consensus achieved through online meetings. The study also outlines specific criteria for expert scoring and active coefficients to ensure the quality of the consensus.

Datasets

Not discussed in the paper. The paper focuses on the protocol for developing a reporting guideline, not on the development or evaluation of specific AI models using datasets.

Key Insights and Ideas

  • The unique nature of TCM requires specialized reporting: The paper strongly emphasizes that TCM's holistic approach and syndrome-based diagnostics necessitate reporting guidelines distinct from those used in Western medicine.
  • Bridging the gap between AI and TCM: TRAPODS-CM aims to be a crucial bridge, ensuring that AI models developed for TCM are transparent, reproducible, and clinically relevant, fostering trust and adoption.
  • Expert consensus is key for standardization: The Delphi method is chosen as the primary tool to leverage the collective knowledge of diverse experts, ensuring the developed guideline is comprehensive and practical.
  • Transparency is paramount for AI in healthcare: The underlying motivation is to improve the quality of research reporting, which directly impacts the reliability and trustworthiness of AI applications in healthcare.

Key Results

As this paper presents a protocol for developing a reporting tool, it does not yet have quantitative results from the application of the tool itself. The key "results" at this stage are the established methodology and the plan for developing the TRAPODS-CM checklist. The paper outlines the structure of the initial item pool (53 potential items across 10 domains) and the rigorous process for its refinement and validation through expert consensus.

Limitations and Future Work

The authors acknowledge several limitations:

  • Potential expert selection bias: The selection of experts could inadvertently introduce bias, potentially limiting the diversity of insights.
  • Risk of groupthink: Despite anonymized responses, there's a risk of groupthink suppressing minority or innovative opinions during Delphi iterations.
  • Geographical representation: While efforts will be made to include international experts, regional factors might still influence the panel's composition, potentially favoring China-based experts due to the nature of TCM.

Future work will involve the actual development and validation of the TRAPODS-CM checklist. The authors plan to disseminate the finalized checklist widely through various platforms to maximize its academic and clinical impact. They also aim for extensive international cooperation to further validate and optimize the tool, ensuring its global applicability and promoting the deep integration of CM and modern technology.

Summary updated

BibTeX
@article{li2025developing,
  author = {Jieyun Li and Wei Song Seetoh and Jiekee Lim and Xin'ang Xiao and Kehu Yang and Si Yong Yeo and Boyun Sun and Jinhua Liu and Zhaoxia Xu and Linda L. D. Zhong},
  title = {{Developing a transparent reporting tool for AI-based diagnostic prediction models of disease and syndrome in Chinese medicine: a Delphi protocol}},
  journal = {Frontiers Digit. Health},
  year = {2025},
  doi = {10.3389/fdgth.2025.1575320},
}
Figure