Intervention-analog association between traditional Chinese medicine syndrome load and lipid/C-reactive protein biomarkers: a single-center retrospective cohort study using target trial emulation.

内容摘要

Traditional Chinese Medicine (TCM) syndrome differentiation is central to individualized diagnosis, but its relationship with objective biochemical markers remains insufficiently characterized within explicit frameworks for intervention-analog causal inference. This study aimed to construct a syndrome load index using confirmatory factor analysis (CFA) and item response theory (IRT), assess measurement invariance, and estimate adjusted intervention-analog associations with lipid and C-reactive protein (CRP) biomarkers. This single-center retrospective cohort study included 1,472 adults with structured TCM assessments and repeated biochemical measurements over 90&#x202f;days. Syndrome load was modeled as a continuous latent variable using CFA/IRT. A Gaussian generalized propensity score with stabilized inverse probability of treatment weighting, combined with double machine learning cross-fitting, estimated associations with low-density lipoprotein-cholesterol (LDL-C), triglycerides (TG), and CRP z-scores, as well as corresponding binary abnormality outcomes. Quartiles were used only for descriptive diagnostics. Negative controls, baseline laboratory sensitivity analyses, weight diagnostics, and E-value analyses assessed robustness. CFA/IRT models demonstrated good fit (CFI&#x202f;=&#x202f;0.96, RMSEA&#x202f;=&#x202f;0.042) and supported metric/scalar invariance across both sex and age groups. The emulated contrast represented a 1-standard deviation (SD) difference in baseline syndrome load as an analytic intervention analog, rather than a literal randomization of a latent construct. After weighting, all absolute standardized mean differences (SMDs) were <0.10, the effective sample size was 1,289, and no material positivity violations were observed. A 1-SD higher syndrome load was associated with higher TG z-scores (+0.18; 95%CI: 0.12-0.24; approximately +0.20&#x202f;mmol/L), LDL-C z-scores (+0.11; 95%CI: 0.05-0.18; approximately +0.10&#x202f;mmol/L), and CRP z-scores (+0.16; 95%CI: 0.09-0.23; approximately +0.31&#x202f;mg/L), corresponding to 3-5% absolute risk increases in biomarker abnormalities. Baseline laboratory sensitivity analyses yielded attenuated but directionally consistent estimates (TG&#x202f;+&#x202f;0.16, LDL-C&#x202f;+&#x202f;0.09, CRP&#x202f;+&#x202f;0.13 to +0.14 z-score units). Associations were stronger among participants aged &#x2265;65&#x202f;years, those with diabetes, or those with body mass index (BMI)&#x202f;&#x2265;&#x202f;28. Negative controls were null, and E-values ranged from 1.54 to 1.92. Higher syndrome load was consistently associated with dyslipidemia and CRP elevation within an explicit intervention-analog framework. These findings support syndrome objectification and personalized cardiometabolic risk stratification; however, prospective intervention studies are required before concluding that the latent syndrome construct itself is directly modifiable.