Incremental Value of Traditional Chinese Medicine Features in Two-Stage Pre-Endoscopic Prediction

Colorectal Neoplasms, Crohn Disease (CD), Inflammatory Bowel Diseases, Ulcerative Colitis (Disorder)

Traditional Chinese Medicine, TCM-integrated model, Western medicine model, Machine learning, Decision curve analysis

This observational study evaluates whether adding Traditional Chinese Medicine (TCM)-related clinical features improves the performance of machine-learning models that estimate disease risk before colonoscopy. We compare two modeling strategies that use the same sequential (two-stage) framework: Western-only models, based on demographic, clinical, and laboratory information commonly used in Western medicine. TCM-integrated models, which use the same types of Western information plus structured TCM features. Stage 1 aims to distinguish people more likely to have functional bowel problems from those more likely to have inflammatory or neoplastic colorectal disease. Stage 2 focuses on people in the higher-risk pathway and aims to distinguish ulcerative colitis from other inflammatory or neoplastic conditions. Models are developed in one dataset and tested in independent external data from other centers. We assess discrimination (including AUC), clinical usefulness with decision-curve analysis, and which TCM features contribute most in the integrated models. This study does not assign treatments. It analyzes existing clinical information to support future pre-endoscopic triage research. Results will not replace colonoscopy or clinician judgment.

Background and rationale. Colonoscopy remains essential for diagnosing ulcerative colitis (UC) and related colorectal inflammatory or neoplastic conditions, but pre-endoscopic risk stratification may help prioritize evaluation. Machine-learning models based on Western clinical and laboratory variables have been developed for sequential triage. Whether structured TCM features provide incremental value beyond a Western-only strategy is less clear and is the focus of this record (TCM-PREDICT). Study design. Multicenter, observational, retrospective diagnostic modeling study with model development and external validation. No investigational drug, device, or care pathway is assigned by the protocol. Modeling framework. Two parallel strategies share the same two-stage prediction structure and are compared using matched algorithm indices where applicable: Western-only strategy: candidate predictors limited to Western demographic, clinical, and laboratory variables. TCM-integrated strategy: Western variables plus rule-structured TCM features collected in routine integrative practice. Stage 1 addresses separation of a lower-risk / functional-predominant pathway from an inflammatory-neoplastic pathway. Stage 2 addresses separation of UC from other inflammatory or neoplastic conditions within the higher-risk pathway. Analysis overview (high level). Primary emphasis is strategy-level incremental value (TCM-integrated versus Western-only), summarized across internal validation, external validation, and decision-curve analysis. Feature-level interpretation for the TCM-integrated models (e.g., SHAP-based importance of TCM variables) is secondary and explanatory. Subgroup summaries (e.g., by center or age band) may be reported descriptively. Data and participants. De-identified clinical records from participating centers are used under institutional approvals. Eligibility, sample size, outcomes, and calendar dates are specified in the corresponding registry fields and should not be repeated here at length. Patient care implications. Findings are intended for research on pre-endoscopic prediction and triage. Models are not a substitute for endoscopy or clinical decision-making.

No related papers found yet.

None listed