Development of a transparent reporting tool for AI-based diagnostic prediction models of disease and syndrome in Chinese medicine (TRAPODS-CM): a Delphi study

In recent years, artificial intelligence (AI) prediction models have gained traction in Chinese medicine (CM) diagnostics. However, existing reporting guidelines, largely based on modern medicine diagnostic frameworks, have limited applicability to CM due to fundamental differences in diagnostic approaches. This highlights the need for a tailored reporting standard. This study develops a transparent reporting guideline for CM diagnostic AI models using the Delphi consensus method. An initial checklist was created by a steering committee and core working group, drawing on a prior systematic review, existing guidelines, and the unique characteristics of CM diagnostics. A multidisciplinary panel of 31 experts evaluated each domain and item on a 1–5 Likert scale across multiple Delphi rounds. In the first round, domain consensus was decent (Kendall’s W = 0.498), but item consensus was low (W = 0.271), prompting substantial revisions, including 5 modifications and 1 addition for domains, and 20 modifications, 13 deletions, and 4 additions for items. In the second round, domain consensus remained strong (W = 0.487), while item consensus improved significantly (W = 0.349), indicating convergence on the checklist. A final consensus meeting led to the refinement and finalization of the Transparent Reporting Tool for AI-Based Diagnostic Prediction Models of Disease and Syndrome in Chinese Medicine (TRAPODS-CM). This guideline aims to provide a reporting framework for CM diagnostic AI prediction models where the unique diagnostic principles of CM are specifically addressed, ultimately fostering more transparent, reliable and reproducible AI-driven clinical applications in this field.