Distribution of Traditional Chinese Medicine Pattern Elements in Patients with Gastric Cancer: A Cross-Sectional Study

Summary

This study establishes a reproducible protocol for identifying and quantifying Traditional Chinese Medicine (TCM) syndrome elements in patients with gastric cancer, enabling systematic analysis through cluster, correlation, and principal component analysis methods.

Abstract

Gastric cancer remains a major cause of morbidity and mortality worldwide, with a particularly high burden in China. Traditional Chinese Medicine (TCM) emphasizes holistic regulation and syndrome differentiation, yet reproducible methods for quantifying syndrome elements in gastric cancer are limited. We conducted a cross-sectional study of 60 patients at Wuxi Second Traditional Chinese Medicine Hospital (December 2022-March 2024) and present a standardized, reproducible protocol for syndrome-element evaluation. Demographics, medical history, and TCM diagnostic information were captured with a standardized case-report form; syndrome elements were identified using a weighted scoring system with dual-expert adjudication. Analysis followed a prespecified multivariate pipeline-two-stage H-K clustering (hierarchical Ward.D2 to suggest k followed by k-means refinement), correlation-network mapping, and principal component analysis. The most frequent elements were qi stagnation, qi deficiency, blood stasis, spleen deficiency, and dampness; strong associations were observed between qi stagnation and blood stasis and between qi deficiency and blood stasis, with additional links involving dampness/phlegm. Principal component analysis indicated that a small set of elements accounted for the largest share of between-patient variation, supporting clinically meaningful stratification. This protocol operationalizes TCM syndrome-element assessment into measurable steps with shareable forms and code, enhancing objectivity and reproducibility and providing a methodological basis for clinical decision-making and multicenter validation. The framework is adaptable to other malignancies, facilitating broader integration of TCM within evidence-based oncology.

Introduction

Gastric cancer remains one of the most prevalent and lethal malignancies worldwide, with China bearing a disproportionately high incidence and mortality burden^1,2. Despite advances in surgical resection, chemotherapy, and targeted therapies, the overall prognosis remains unsatisfactory, underscoring the need for complementary approaches that address both disease biology and patient quality of life³. Traditional Chinese Medicine (TCM), emphasizing holistic regulation and syndrome differentiation, is widely used in oncology; however, the lack of standardized, replicable methods for syndrome identification has limited systematic integration of TCM into evidence-based cancer care^4,5.

Recent developments in syndrome element differentiation provide a structured framework to quantify the key pathological factors underlying TCM syndromes, enabling more precise patient classification and tailored interventions⁶. Applying this quantitative framework to gastric cancer offers a way to bridge traditional theory with modern analytics, yet reproducible, protocolized workflows remain scarce⁷. To our knowledge, this is the first protocol to operationalize syndrome-element assessment in gastric cancer with explicit forms, dual-expert adjudication, and a prespecified multivariate pipeline (two-stage H-K clustering, correlation network, and PCA), packaged for replication and multicenter adoption. By translating syndrome differentiation into measurable variables and prespecifying analytic parameters, the present study provides a replicable foundation for clinical practice and multicenter research, clarifies the distribution of key elements in gastric cancer populations, and offers a concrete path toward more precise TCM-based oncology strategies.

Protocol

Ethics and Oversight

This study was approved by the Ethics Committee of Wuxi Second Traditional Chinese Medicine Hospital (20220904). All participants provided written informed consent in accordance with the Declaration of Helsinki; procedures were consistent with ICH-GCP principles. Any identifiable media used for demonstration was separately consented. To reduce assessment bias, physicians who evaluated TCM syndrome elements were blinded to the American Joint Committee on Cancer (AJCC) stage and treatment status throughout scoring and adjudication; blinding was maintained by a data steward who removed those fields from assessor-facing sheets.

Patient recruitment and case sources

Consecutive patients with gastric cancer were enrolled between December 1, 2022, and March 31, 2024, from the outpatient and inpatient departments of Wuxi Second TCM Hospital. Screening was logged prospectively (order approached, eligibility decision, reason for exclusion, consent outcome). A de-identified screening checklist and the blank case-report form (CRF) are provided as a table in the public repository under Data Availability. Baseline demographics and staging summaries for the enrolled cohort are presented in Table 1 to contextualize subsequent analyses.

Diagnostic and staging criteria

Diagnosis followed the norms for the diagnosis and treatment of common malignant tumors issued by the Medical Administration Department of the Ministry of Health of the PRC, with pathological confirmation required in every case. Tumor stage was assigned according to the AJCC Cancer Staging Manual, 8th edition, and verified against pathology and imaging source documents before enrollment⁸.

Eligibility criteria and withdrawal rules

Adults (≥18 years) with pathologically confirmed gastric cancer and Karnofsky Performance Score (KPS) ≥ 60 were eligible, provided they could complete the standardized TCM assessment. Exclusions included pregnancy, unstable severe comorbidities compromising consent/safety, or incomplete core baseline data. Patients receiving or having received chemotherapy, radiotherapy, or targeted/immunotherapy were eligible; treatment status was recorded (Never/Ongoing/Past) and used in stratified and sensitivity analyses. Participants could be withdrawn if data were insufficient for syndrome differentiation, if rapid clinical deterioration occurred, or upon personal request, without impact on subsequent care.

Data collection and management

Trained researchers completed a standardized CRF titled Collection Form of Syndrome Element Information for Gastric Cancer Patients. Data were double-entered using spreadsheet software (version ≥15.30), independently cross-checked, and reconciled before database lock. A data dictionary, coding sheet, and audit log were maintained from the first entry to analysis. Analytical datasets were de-identified and stored with role-based access control; a frozen, time-stamped copy was archived to support replication and auditing.

Standardized TCM assessment

Clinical evaluation adhered to the four classical examinations-inspection, listening/smelling, inquiry, palpation-in a fixed sequence under consistent lighting (~5,000-6,500 K; CRI ≥90). Patients were seated facing the light source; tongue and pulse features were recorded using predefined descriptors. Assessors underwent calibration with pilot cases and exemplar vignettes before formal scoring. All observations were recorded immediately on the CRF to minimize recall error. The cohort-level distribution of composite TCM syndrome patterns derived from this assessment is summarized in Table 2 (footnotes specify whether a single primary pattern or multiple patterns were allowed).

Syndrome-element scoring and adjudication

Items were scored according to the Syndrome Element Scale (full item list and codes in Supplementary Table 1). Inter-rater reliability (present/absent) was quantified using Cohen's κ with 95% CIs and summarized in Supplementary Table 2. Severity weights were: mild = 0.7, moderate = 1.0, severe = 1.5. Weighted items were summed within each element to yield the element score. Presence of an element was defined at threshold = 14; robustness at 12 and 16 is reported in Supplementary Table 3. Two TCM oncology physicians independently scored each case; discordances were resolved in a consensus meeting with a written rationale. Element-level prevalence across patients is reported in Table 3, and a worked example of item-to-element scoring is shown in Supplementary Figure 1.

Statistical analysis

Software and reproducibility

All analyses were performed in R (v4.3). A runnable, version-locked script and the full sessionInfo() output are archived with the dataset (https://doi.org/10.6084/m9.figshare.30405196.v2). A fixed random seed was set before all stochastic procedures. Core functions/packages were: base stats (prcomp, hclust, kmeans), cluster (silhouette), and igraph (network visualization)⁹. The element-level distribution is summarized in Supplementary Table 1, and the blank case-report form (CRF), data dictionary, and tidy input matrices are available in the public repository referenced in the Data Availability section.

Two-stage H - K clustering

A two-stage procedure was used to derive clinically interpretable groups.

Hierarchical stage: Patient-by-element weighted scores were z-standardized, Euclidean distances were computed, and agglomeration used Ward.D2 linkage. The dendrogram was inspected to propose candidate k using joint consideration of the elbow in within-cluster dispersion, branch stability, and preliminary silhouette patterns¹⁰.

k-means refinement. For each candidate k, k-means was initialized from the hierarchical partitions and run with nstart ≥ 50; final membership minimized within-cluster sum of squares. Average silhouette width and per-cluster profiles were exported for traceability. The integral heat map summarizing standardized scores is shown in Figure 1; the corresponding dendrogram is shown in Figure 2.

Principal component analysis and correlation network

Element-score matrices were centered and scaled, and PCA was performed with prcomp(center = TRUE, scale = TRUE). The node size reflects the prevalence of elements; labels were manually checked to ensure no overlap and readability. The resulting correlation network is shown in Figure 3. The PCA biplot is shown in Figure 4. Pairwise Pearson correlations across elements were computed with pairwise-complete observations; edges were retained when |r| ≥ 0.30, with edge width proportional to |r|. A force-directed algorithm (Fruchterman-Reingold) was used for layout to improve readability¹¹.

Missing data and sensitivity

Per-variable completeness (counts, %) is summarized in Supplementary Table 4; because missingness was low, complete-case analysis was used for all primary outputs. Robustness was assessed by recomputing element-presence classification under alternative score cutoffs (12 and 16) relative to the prespecified threshold of 14 (Supplementary Table 3), and-where relevant-within treatment-status strata (Never/Ongoing/Past; Supplementary Table 5). Inter-rater reliability for element adjudication by two independent TCM experts was high (overall agreement 95%), as detailed in Supplementary Table 2. A worked example illustrating item-to-element scoring and thresholding is provided in Supplementary Figure 1.

Quality control, timing, and troubleshooting

Quality assurance included weekly 10% source-data verification, periodic assessor recalibration, and automated range/logic checks before database lock. Typical timelines were ~10-15 min for screening/consent, 20-30 min for standardized TCM assessment and scoring, and 10-15 min for data entry/verification per case. The critical steps here include maintaining stable lighting and a fixed examination sequence; preserving assessor blinding to stage and treatment; and documenting reasons for any adjudication. When summarizing distributions, denominators were explicitly stated as N (patients) or non-exclusive element counts, and the same denominator was used consistently across text, tables, and figure captions.

Results

Participant characteristics

Sixty patients with gastric cancer were enrolled between December 2022 and March 2024. Age distribution was: 41-50 years, 9/60 (15.0%); 51-60 years, 18/60 (30.0%); 61-70 years, 15/60 (25.0%); 71-80 years, 13/60 (21.7%); and 81-90 years, 5/60 (8.3%). Overall sex distribution was male 37/60 (61.7%) and female 23/60 (38.3%). Counts by sex within each age band are reported in Tabl...

Discussion

This study developed and applied a standardized, protocol-centric workflow to identify and quantify TCM syndrome elements in gastric cancer. By combining a weighted scoring system, dual-expert adjudication, and a prespecified multivariate pipeline-two-stage H-K clustering, correlation network analysis, and principal component analysis-the method consistently highlighted qi stagnation, qi deficiency, blood stasis, and dampness as the most frequent and clinically relevant elements in this cohort, in line with prior observa...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Case Report Form (CRF)	Wuxi Second TCM Hospital	CRF-001	Standardized form for collecting patient data
cluster R Package	CRAN	R-cluster-008	R package for clustering and silhouette analysis
Examination Light (5,000–6,500 K)	Philips	EX-LIGHT-003	Light used for TCM examination
igraph R Package	CRAN	R-igraph-007	R package for network visualization
Pulse Measuring Device	Omron	PMD-005	Device for measuring pulse during TCM evaluation
R Software	The R Project	R-001	Open-source software for statistical analysis
stats R Package	CRAN	R-Stats-006	R package for statistical analysis (prcomp, hclust, kmeans)
Syndrome Element Scale	Wuxi Second TCM Hospital	SE-Scale-002	Scoring tool for evaluating syndrome elements
Tongue Imaging Device	Canon	TI-IMG-004	Device used to capture tongue images