1. Introduction
In October 2023, the American Heart Association released a Presidential Advisory defining cardiovascular-kidney-metabolic (CKM) syndrome as a systemic disorder driven by pathological interactions among metabolic risk factors, chronic kidney disease (CKD), and the cardiovascular system.[1] These interactions contribute to multi-organ dysfunction and substantially elevate the risk of adverse cardiovascular outcomes.[1]
Chronic low-grade inflammation constitutes the core shared pathological basis linking CKM spectrum disorders and arthritis. The advisory highlighted that several factors, including chronic inflammatory diseases such as rheumatoid arthritis, increase the likelihood of CKM stage progression and the risk of cardiovascular and renal endpoints. However, arthritis prevention itself was not emphasized.[1] Accumulating evidence indicates that both the onset and progression of CKM are closely linked to osteoarthritis and rheumatoid arthritis.[2,3] As a prevalent and clinically detectable condition, osteoarthritis may serve as an early indicator of individuals at high risk for multi-organ metabolic decline, supporting the integration of arthritis status into CKM risk stratification frameworks.
The triglyceride-glucose body mass index (TyG-BMI) is widely employed as a surrogate marker of insulin resistance.[4,5] It can be easily calculated from routine fasting blood glucose, triglycerides and body mass index (BMI) without the detection of insulin, making it suitable for large population cohorts such as CHARLS. By integrating the triglyceride-glucose index (TyG) with BMI, triglyceride-glucose body mass index (TyG-BMI) comprehensively evaluates insulin resistance combined with adiposity and has shown superior predictive ability for arthritis compared to the TyG alone, which has been validated in multiple epidemiological studies.[6–10] Nevertheless, the relationship between TyG-BMI and arthritis in individuals with CKM syndrome remains unclear. Given the established role of CKM in arthritis pathogenesis, investigating the association between TyG-BMI and incident arthritis across CKM stages 0–3 is warranted.
Such research could enhance the understanding of progressive arthritis and inform early, comprehensive intervention strategies to alleviate the associated disease burden. Therefore, the primary objective of this study was to examine the epidemiological association between baseline TyG-BMI and subsequent incident self-reported physician-diagnosed arthritis or rheumatism among adults across CKM stages 0 to 3 within a nationally representative Chinese cohort. This study focuses on association analysis and does not conduct formal causal inference or develop clinical risk prediction models.
Subgroup and mediation analyses were predefined as secondary exploratory analyses.
2. Method
2.1. Data source and study population
Data for this study were derived from the China Health and Retirement Longitudinal Study (CHARLS). The CHARLS dataset can be accessed through the official website of the CHARLS (http://charls.pku.edu.cn/). All researchers are required to submit a formal application and sign a data use agreement, and data access is restricted by the official data sharing policies and ethical requirements of CHARLS.
The original CHARLS survey was approved by the Institutional Review Board of Peking University. The household survey involving anthropometric measurements received ethical approval (IRB00001052-11015), and biomarker collection was separately approved under IRB00001052-11014. All procedures were conducted in accordance with the ethical principles of the Declaration of Helsinki, and written informed consent was obtained from every participant prior to data collection.
The present secondary analysis was conducted using de-identified publicly available data in full compliance with the CHARLS data use agreement. The CHARLS is a nationally representative longitudinal cohort exclusively designed for community-dwelling Chinese adults aged 45 years and older.[11] Both CKM impairment and arthritis are age-associated chronic conditions with markedly elevated prevalence after middle age; therefore, this population is appropriate to explore the epidemiological association between TyG-BMI and incident arthritis. The study protocol was reviewed and approved by the Institutional Review Board of Peking University.
The CHARLS employed a multistage stratified probability proportional to size sampling design to ensure national representativeness, covering 28 provinces, 150 county-level units, and 450 villages or urban communities. The baseline survey was conducted between June 2011 and March 2012, with follow-up surveys performed in 2013, 2015, 2018, and 2020. The CHARLS collected comprehensive data through face-to-face interviews, including detailed information on demographic characteristics, family structure, and health status and functioning, lifestyle and health behaviors, and activities of daily living, mental health, and cognitive function.
In specific survey waves, physical measurements were taken, encompassing blood pressure, height, weight, waist circumference, and grip strength. Fasting venous blood samples were collected during the 2011 (baseline) and 2015 waves for biomarker analysis.
For the present analysis, laboratory data from the baseline (2011) fasting blood samples were utilized. Participants were instructed to fast for at least 8 to 12 hours overnight prior to phlebotomy. Collected blood samples were immediately processed, transported under strict cold-chain conditions, and stored at −70°C until laboratory assay.
A standardized protocol was used to measure biomarkers including fasting plasma glucose, glycated hemoglobin, a full lipid profile (total cholesterol, triglycerides, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol), and renal function parameters (e.g., blood urea nitrogen (BUN), serum creatinine, uric acid). The detailed laboratory methodology has been described in prior CHARLS publications.[12] From the initial CHARLS baseline cohort of 17,708 individuals, we applied the following sequential exclusion criteria to arrive at the analytical sample: age <45 years; classification into CKM stage 4 or missing CKM stage data; missing values for key TyG-BMI components (triglycerides, fasting glucose, or BMI); prevalent arthritis at baseline; missing arthritis status during the follow-up period. Loss to follow-up before 2020, potentially including death, migration, and refusal of reinterview, contributed to missing outcome data.
After applying these criteria, a total of 5018 participants with CKM stages 0 to 3 and complete data for analysis were included. A flowchart detailing the participant selection process is provided in Figure 1.
2.2. Calculation of triglyceride-glucose BMI
The TyG-BMI was derived by multiplying the BMI by the TyG. The TyG index was calculated using the formula: Ln [fasting triglycerides (mg/dL) × fasting plasma glucose (mg/dL)/ 2]. BMI was computed as weight in kilograms divided by height in meters squared (kg/m2).[9]
2.3. Definition of arthritis
The primary outcome was incident self-reported doctor-diagnosed arthritis or rheumatism ascertained at the 2020 CHARLS follow-up. Participants with prevalent arthritis/rheumatism reported at the 2011 baseline were excluded. Case status was determined via the chronic disease questionnaire item: “Have you been diagnosed with arthritis or rheumatism by a doctor?[11]” Affirmative responses defined incident cases.
Notably, the dataset lacked information to differentiate osteoarthritis, rheumatoid arthritis, gout, or other specific rheumatic disorders; accordingly, the outcome represents a composite measure rather than a single subtype of arthritis confirmed by medical records.
2.4. Definition of CKM stages 0 to 3
CKM stages were defined following the American Heart Association Presidential Advisory.[1] Stage 0 indicated no CKM risk factors. Stage 1 represented excess adiposity or adipose dysfunction. Stage 2 comprised metabolic risk factors and CKD.
Stage 3 was defined by subclinical cardiovascular disease (CVD) proxies: estimated glomerular filtration rate <30 mL/minutes/1.73m2 or Framingham 10-year ASCVD risk score ≥20%.[13] Estimated glomerular filtration rate was calculated via the Chinese modified MDRD equation[14]; CKD definition complied with KDIGO guidelines.[15] Full operational definitions, CHARLS variable mapping and diagnostic thresholds for CKM staging are summarized in Supplementary Table S1, Supplemental Digital Content 1.
2.5. Covariates
Covariates were selected based on available CHARLS data. Sociodemographic variables included age, sex, marital status, education level, and place of residence. Medical history included hypertension, diabetes, liver diseases, cancer, coronary heart disease, stroke, lung diseases, kidney diseases, digestive diseases, dyslipidemia, and CVD.
Lifestyle factors included smoking status, drinking status, and night sleep duration measured in hours.
2.6. Statistical analysis
We clarified the study timeline: all statistical analyses used 2011 baseline variables and incident arthritis outcomes ascertained in 2020 following sample eligibility screening. Complete-case analysis was adopted for primary analyses. Participants with missing TyG-BMI components, missing CKM stage information, or missing 2020 arthritis status were excluded; those with prevalent baseline arthritis were also excluded.
Covariates were selected based on prior epidemiological literature on arthritis risk factors. Complete-case analysis may introduce selection bias under nonrandom missingness. Multiple sensitivity analyses including multiple imputation, models additionally adjusted for BMI, and comparative analyses of TyG, BMI and TyG-BMI were conducted to test the robustness of our findings.
Physical activity was not included in primary models owing to missingness and comparability limitations across survey waves. Medication information was utilized to define cardiometabolic conditions where relevant CHARLS variables were accessible, while incomplete medication data represents a study limitation.
Participants were stratified based on arthritis status at the 2020 follow-up and by quartiles of TyG-BMI. Normally distributed continuous variables are presented as mean ± standard deviation (SD), whereas non-normally distributed variables are summarized as median (interquartile range). Between-group differences were compared using independent samples t tests or Mann–Whitney U tests as appropriate.
Categorical variables were reported as number (percentage).
Three sequential multivariable logistic regression models with incremental covariate adjustment were constructed to evaluate the association between TyG-BMI and incident arthritis. Model 1 was the unadjusted crude model without any covariates. Model 2 was partially adjusted for age, sex, marital status, education level, residential area, smoking and drinking status.
Model 3 was the fully adjusted model, which further controlled for multiple chronic disease histories. TyG-BMI was entered into the models in 2 distinct ways: as a continuous variable per SD increment, and as categorical quartile variables.
Restricted cubic spline (RCS) models with 4 knots were applied to examine nonlinear dose-response relationships, and formal nonlinearity test results were reported. Subgroup and interaction analyses were conducted to evaluate effect modification. Stratification by CKM stage was prespecified in advance, while subgroup analyses stratified by sleep duration and other factors were post hoc exploratory analyses.
Causal mediation analysis was applied to decompose the total effect into direct and indirect effects, and the key identification assumptions required for mediation analysis are presented in the manuscript. Owing to the inherent limitations of observational research, definite causal pathways cannot be confirmed, hence all mediation results are interpreted with caution.
Finally, we assessed whether TyG-BMI provided incremental discriminatory information beyond BMI. A base logistic regression model was constructed using age, sex, marital status, educational level, place of residence, smoking status, drinking status, nighttime sleep duration, and histories of hypertension, diabetes, liver disease, cancer, heart disease, stroke, lung disease, kidney disease, digestive disease, dyslipidemia, and CVD. BMI, TyG, and TyG-BMI were then added separately to the base model.
We further added TyG or TyG-BMI to the model already containing BMI to evaluate their incremental value beyond adiposity. Model discrimination was assessed using the area under the receiver operating characteristic curve (AUC). Differences in AUC were evaluated using paired DeLong tests, while continuous net reclassification improvement and integrated discrimination improvement (IDI) were calculated.
The corresponding 95% confidence intervals (CI) were estimated using 5000 nonparametric bootstrap resamples.
The core statistical analyses were performed using IBM SPSS Statistics Version 26.0 (IBM Corp., Armonk) and Zstats online statistical platform. A two-tailed α level of 0.05 was set as the threshold for statistical significance, with P < .05 considered statistically significant.
3. Results
3.1. Baseline characteristics of the study population
Table 1 presents the baseline demographic and clinical characteristics of the participants. A total of 5018 individuals with CKM stages 0 to 3 were included in this study. Among them, 2307 participants (45.97%) developed incident arthritis during the follow-up period.
Participants in the arthritis group were older than those in the non-arthritis group, with a mean age of 61.47 ± 8.25 years compared to 58.92 ± 8.90 years. The median age was also higher in the arthritis group.
The arthritis group contained a higher proportion of women, individuals with lower education levels, and those residing in rural areas. The proportion of married participants was lower in the arthritis group. In terms of lifestyle factors, a history of smoking and drinking was more common among participants with arthritis, and the median night sleep duration was shorter.
Regarding laboratory indicators, median levels of TyG-BMI, total cholesterol, triglycerides, low-density lipoprotein cholesterol, and HDL-C were higher in the arthritis group. In contrast, median levels of creatinine and hemoglobin were lower. Furthermore, participants with arthritis had a higher prevalence of hypertension, heart disease, stroke, lung diseases, liver diseases, kidney diseases, digestive diseases, and CVD.
3.2. Baseline characteristics according to TyG-BMI quartiles
Table 2 presents the baseline characteristics of participants stratified by TyG-BMI quartiles. With increasing TyG-BMI quartiles, the proportion of participants classified into CKM stage 2 and stage 3 showed a progressive increase. Specifically, the proportion of stage 2 increased from 46.77% in Q1 to 69.72% in Q4, while the proportion of stage 3 increased from 14.10% in Q1 to 26.69% in Q4.
Conversely, the proportions of participants in CKM stage 0 and stage 1 decreased across the quartiles. These differences were statistically significant.
The prevalence of arthritis also increased across TyG-BMI quartiles. In the 2020 follow-up, arthritis prevalence rose from 43.51% in the lowest quartile (Q1) to 48.21% in the highest quartile (Q4).
3.3. Association between TyG-BMI and incident arthritis in CKM stages 0 to 3
Table 3 summarizes the association between TyG-BMI and incident arthritis among participants with CKM stages 0 to 3. In the unadjusted model, each one-SD increase in TyG-BMI was associated with a 7.0% higher risk of arthritis. After adjusting for age, sex, marital status, education level, place of residence, smoking status, and drinking status, each one-SD increase in TyG-BMI was associated with a 9.0% elevated risk.
In the fully adjusted model, the positive association remained statistically significant, with an 11.0% increased risk of arthritis per one-SD increment in TyG-BMI.
Figure 2 illustrates the dose-response association between TyG-BMI and risk of incident self-reported arthritis or rheumatism. RCS regression identified a statistically significant overall association (P for overall association = .001), while the test for nonlinearity did not reach statistical significance (P for nonlinearity = .095). Accordingly, the exposure–outcome relationship was approximated as linear.
An apparent inflection point at a TyG-BMI value of 197.89 was observed in the fitted curve. This exploratory turning point is merely a descriptive feature of the curve within the present sample. At TyG-BMI ≥ 197.89, the odds ratio was >1 and increased gradually alongside rising TyG-BMI.
For TyG-BMI values below this level, the odds ratio remained below unity without a clear directional trend.
3.4. Subgroup analysis
We examined multiplicative interactions between TyG-BMI and 2 factors: CKM stage and sleep duration. Stratified analyses were carried out for significant interactions, with subgroup results shown in Figure 3. CKM stratification was predefined for our primary aim, whereas sleep duration and other subgroup stratifications were post hoc exploratory tests.
Both CKM stage and sleep duration acted as effect modifiers for the studied association. Among individuals with sleep duration of <7 hours, the exposure was associated with a 19% higher risk of the outcome (odds ratio [OR] = 1.190, 95% CI: 1.080–1.310). In contrast, no statistically significant association was observed in those with sleep duration of 7 hours or longer (OR = 1.010, 95% CI: 0.920–1.120).
As a post hoc exploratory finding, this sleep duration subgroup result is susceptible to multiple testing bias and residual confounding and requires independent external confirmation. For CKM stage, the magnitude of association appeared strongest in participants with stage 1, where the exposure was associated with a 57% higher risk of the outcome (OR = 1.570, 95% CI: 1.200–2.050). No statistically significant association was detected in individuals with CKM stages 0, 2, or 3.
Statistically significant associations between the exposure and elevated outcome risk were also observed in several additional exploratory subgroups: male sex, rural residence, age <60 years, married status, absence of hypertension, diabetes, heart disease, stroke, lung disease or kidney disease, presence of digestive disease, and never drinking alcohol. In contrast, no statistically significant association was found in subgroups including female sex, urban residence, presence of chronic conditions such as hypertension and diabetes, sleep duration ≥7 hours, and unmarried status. These exploratory subgroup differences suggest the observed risk association may be less evident in these populations or could be attributable to residual confounding or chance.
3.5. Mediation analysis
Results of the mediation analysis are presented in Table 4. HDL-C showed a statistically significant negative mediating effect in the association between TyG-BMI and arthritis. The average causal mediation effect was negative, whereas the average direct effect remained positive.
The proportion mediated was 22.87%. This observational pattern suggests that the elevated arthritis risk related to higher TyG-BMI may be partially explained by lower HDL-C concentrations; however, we cannot distinguish whether HDL-C functions as an intermediate along a causal pathway or merely serves as a correlated marker of underlying metabolic dysfunction. In contrast, BUN exhibited only a marginally statistically significant negative mediating effect, with a proportion mediated of 1.36%, pointing to a negligible contribution within this analytical pathway.
All mediation findings should be interpreted cautiously as exploratory and require independent confirmation in future investigations.
3.6. Incremental discrimination and reclassification analyses
The base covariate model yielded an AUC of 0.602 (95% CI, 0.597–0.640). After the separate addition of BMI, TyG, and TyG-BMI, the AUCs were 0.606 (95% CI, 0.599–0.644), 0.604 (95% CI, 0.598–0.642), and 0.606 (95% CI, 0.599–0.645), respectively. When TyG-BMI was added to the model already containing BMI, the change in AUC was 0.00004 (95% CI, −0.00066 to 0.00444), the continuous net reclassification improvement was 2.68% (95% CI, −4.96% to 11.31%), and the IDI was −0.0002% (95% CI, −0.0025%–0.1559%)(Table 5).
4. Discussion
Despite accumulating evidence confirming independent correlations between individual metabolic risk factors and arthritis, prominent gaps persist in the current body of literature. First, most studies exploring the link between cardiometabolic disorders and joint diseases fail to adopt the standardized AHA cardiovascular-kidney-metabolic (CKM) staging framework proposed by the American Heart Association.[1] Few longitudinal cohorts have stratified participants across the entire CKM continuum (stages 0–3) to assess graded arthritis risk; moreover, existing CKM-related research overwhelmingly focuses on cardiovascular and renal adverse outcomes, with only a small number of domestic studies preliminarily exploring its connection with joint disorders via cross-sectional or short-term cohort designs.[2,3] Second, most prior analyses exploring the association between TyG-derived insulin resistance indicators and arthritis adopted cross-sectional designs and were mainly performed in Western and Korean populations.[6–10] Even national epidemiological surveys conducted among Chinese adults only reported cross-sectional correlations between combined TyG-obesity indices and prevalent arthritis, without prospective follow-up to observe incident arthritic outcomes.[7] Accordingly, longitudinal evidence regarding the predictive value of TyG-BMI for new-onset arthritis in Chinese middle-aged and older adults remains lacking. Furthermore, no previous research has combined CKM staging with mediation analyses to clarify lipid or renal intermediate pathways mediating the TyG-BMI–arthritis association.
Mechanistically, chronic low-grade inflammation, aberrant adipokine secretion and impaired high-density lipoprotein function act as shared pathogenic pathways connecting insulin resistance, progressive CKM multimorbidity and irreversible articular cartilage damage.[16,17] Against this backdrop, the present national prospective cohort study was conducted to delineate the prospective association between TyG-BMI and incident arthritis across CKM stages 0 to 3, as well as quantify the mediating roles of lipid and renal biomarkers, so as to fill the abovementioned gaps in epidemiological evidence.
Using baseline and follow-up data from the CHARLS, we performed this analysis among 5018 individuals with CKM stages 0 to 3. To our knowledge, this is the first investigation specifically focused on this population and the first to explore potential mediating mechanisms underlying this relationship.
Our fully adjusted primary analysis revealed that each one-standard-deviation increase in TyG-BMI independently raised the odds of incident arthritis by 11.0% (OR = 1.11, 95%CI 1.04–1.19, P = .001), alongside a clear dose-response trend: participants in the highest TyG-BMI quartile had a 36% greater arthritis risk relative to the lowest quartile. This modest yet consistent risk elevation has practical clinical value. As a convenient marker of adiposity and insulin resistance, TyG-BMI can be used for early arthritis risk screening in adults with cardiometabolic abnormalities, and metabolic optimization targeting reduced TyG-BMI may help prevent arthritis onset in this group.
A RCS analysis confirmed a positive linear dose-response association between TyG-BMI and incident arthritis, which strengthened notably above the TyG-BMI cutoff of 197.89. Consistent with our quartile analyses, this linear trend shows arthritis risk rises gradually as adiposity and insulin resistance worsen, stressing the importance of sustained early metabolic intervention. This threshold needs external validation and cannot be directly used for clinical practice now.
Existing epidemiological studies have documented an association between the TyG index and arthritis.[6,8,9] The combination of the TyG index with obesity indicators is often considered relevant to the occurrence or progression of arthritis. Research has shown that TyG-BMI and TyG–WHtR are positively correlated with the prevalence of arthritis in Chinese and American populations, demonstrating stronger diagnostic relevance compared to TyG alone.[7,10] Moreover, the association between TyG–WHtR and osteoarthritis (OA) appears more pronounced in younger populations.[10] This study further suggests that TyG-BMI may represent an independent risk marker for arthritis in individuals at CKM stages 0 to 3. The underlying mechanism is likely related to the integrated characteristics of insulin resistance, obesity, and metabolic disorders represented by TyG-BMI, which jointly disrupt joint homeostasis.
First, insulin resistance may alter the secretion of adipokines and inflammatory cytokines, thereby affecting joint structure.[18,19] Second, obesity not only increases mechanical load on joints but also induces chronic low-grade inflammation through the endocrine function of adipose tissue[20–22] Furthermore, in this study, CKM stages showed a gradient progression with increasing TyG-BMI quartiles. The proportions of stage 2 and stage 3 increased from 46.77% and 14.10% in Q1 to 69.72% and 26.69% in Q4, while the proportions of stage 1 and stage 4 declined, potentially due to the progression of stage 1 patients and the relative increase in stages 2 and 3. This observation further supports the compounding effect of metabolic abnormalities and joint injury, consistent with findings that chronic kidney and metabolic disorders exacerbate systemic inflammation.[23] Collectively, these mechanisms significantly elevate the risk of arthritis development.
Existing CKM research primarily concentrates on cardiovascular and renal outcomes.[1,23] Limited cross-sectional data from China has tentatively linked CKM disorders to arthritis.[2,3] Our longitudinal analyses further extend the spectrum of CKM-related adverse outcomes to incident arthritis via TyG-BMI assessment. Still, causal pathways underlying this association cannot be verified in the current observational design.
Subgroup analysis identified effect modification by sleep duration and baseline CKM stage. CKM stratification was prespecified as the core subgroup analysis, with the strongest TyG-BMI–arthritis association observed in participants with CKM stage 1. The sleep duration interaction was a post hoc exploratory endpoint.
The amplified arthritis risk among short-sleep participants may stem from sleep loss-induced metabolic disturbance and elevated pro-inflammatory cytokines (TNF-α, IL-6),[24–26] forming a synergistic “double-hit” effect with high TyG-BMI. Sufficient sleep alleviates such metabolic damage by maintaining metabolic homeostasis and promoting chondrocyte repair.[27,28] However, this sleep-related interaction carries risks of multiple testing bias and requires external validation, so it should not be overemphasized. Our core predefined finding highlights the need for targeted arthritis risk surveillance among individuals with CKM stage 1 and elevated TyG-BMI.
Mediation analysis revealed a statistically significant negative mediating association of HDL-C between TyG-BMI and incident arthritis, accounting for 22.87% of the total observed association. Higher TyG-BMI was correlated with lower HDL-C concentrations, which may attenuate HDL’s anti-inflammatory and lipid-clearing protective effects on joint tissue and coincide with elevated arthritis risk. This observation is in line with prior epidemiological findings linking impaired HDL metabolism to joint damage in individuals with metabolic disorders.[17,29] In contrast, BUN showed merely a marginal, minimal mediating association (P = .08, mediation proportion = 1.36%).
The enrolled participants only spanned CKM stages 0 to 3 with mild renal involvement at most, which explains the negligible contribution of urea-related chondrocyte deterioration across the cohort.[30,31] Importantly, all mediation results are exploratory statistical observations derived from observational data. HDL-C and BUN should be interpreted as correlated metabolic and renal biomarkers, rather than validated causal mediators underlying this relationship.
Notably, TyG-BMI incorporates anthropometric information from BMI. Unlike BMI, which merely reflects adiposity, TyG-BMI further integrates triglyceride and fasting glucose levels to capture insulin resistance. As a composite metric, it captures both adiposity burden and metabolic dysfunction, and may yield predictive information beyond BMI when assessing arthritis risk in metabolically impaired populations.
These findings not only propose TyG-BMI as a novel composite biomarker for evaluating arthritis risk among middle-aged and older adults aged ≥45 years with CKM abnormalities, but also provide tentative strategies for arthritis prevention via comprehensive metabolic regulation, including ameliorating insulin resistance, optimizing blood lipids, preserving HDL function, and regulating sleep habits. Such preliminary clinical relevance is most prominent for individuals with elevated TyG-BMI, insufficient sleep, and CKM stage 1. However, given the inherent limitations of observational research, definitive TyG-BMI cutoff values suitable for routine clinical application cannot be established merely based on the present dataset.
Clinically, TyG-BMI may serve as an inexpensive, readily available auxiliary marker to preliminarily screen individuals at high risk of arthritis during routine cardio metabolic follow-up visits. Mediation analyses further revealed that sustained stable HDL-C levels might confer joint-protective effects in populations with metabolic dysfunction. Nevertheless, causal associations cannot be confirmed by this longitudinal observational study.
Additional interventional trials are warranted to verify whether targeted reduction of TyG-BMI can effectively decrease the incidence of arthritis.
From a public-health perspective, the findings underscore the importance of holistic integrated management covering metabolic, renal, cardiovascular and musculoskeletal health among middle-aged and older adults. Accumulation of further prospective and interventional evidence is required prior to the widespread adoption of TyG-BMI in population-level arthritis screening and preventive initiatives.
5. Strengths and limitations
This study has several notable strengths. First, it utilized data from a nationally representative longitudinal cohort (CHARLS) with a substantial sample size (n = 5018) and a long follow-up duration of 9 years. The research strictly adhered to the STROBE reporting guidelines, and the data, including laboratory-measured blood indices, underwent professional assessment and rigorous quality control, thereby helping to minimize selection and information biases.[11,32] Second, the participant selection was highly targeted.
By focusing for the first time on individuals at CKM stages 0 to 3: a population at high risk for arthritis: the study design mitigated interference from “disease heterogeneity” commonly found in general populations. This approach allowed for a more precise elucidation of the risk value of TyG-BMI within a cohort characterized by cardio metabolic-kidney-cardiovascular multimorbidity.
However, several limitations must be acknowledged. First, arthritis cases were identified based on self-reported physician-diagnosed outcomes rather than official medical records. Even with strict quality control procedures in CHARLS, misclassification bias could not be eliminated.
Non-differential misclassification generally attenuates the true association, while unequal medical accessibility may cause differential misclassification and bias our results. Second, slight disparities existed in baseline characteristics between included and excluded individuals; thus, selection bias could not be entirely excluded, which should be taken into account during result interpretation (Supplementary Table S2, Supplemental Digital Content 2). Third, residual confounding could not be fully excluded despite extensive covariate adjustment.
Smoking status, alcohol consumption and sleep duration were adjusted for in our statistical models. However, detailed quantitative data on dietary composition and daily physical activity were not available in the database; incomplete capture of these lifestyle indicators, together with unmeasured occupational loading, inevitably brought about residual confounding. We used educational attainment and residential area to approximate socioeconomic status, whereas detailed indicators including household income, total assets and medical insurance coverage were not included.
We also failed to perform stratified analyses by socioeconomic strata, which further retained socioeconomic-related confounding. Meanwhile, recall bias, interviewer bias, inconsistent measurements across survey waves and laboratory testing errors were unavoidable, potentially leading to inaccurate TyG-BMI calculations. Moreover, the CHARLS dataset does not collect genetic susceptibility information, so genetic confounding could not be accounted for, and the genetic roles underlying the association between TyG-BMI and incident arthritis could not be assessed in the present study.
Fourth, limitations inherent to logistic regression should be noted. Given the moderate incidence of arthritis in our elderly population, odds ratios inevitably overestimate the actual relative risk. Although our outcome event count satisfied the standard event-per-covariate requirement, adjusting for numerous comorbidities still raised risks of collinearity and mild model instability.
For these statistical reasons, all adjusted effect sizes were interpreted cautiously. Fifth, potential reverse causation serves as another critical limitation. Subclinical and early undiagnosed arthritis can lead to joint discomfort, decreased physical activity, aggravated insulin resistance and lipid metabolic disorders, which may increase TyG-BMI values accordingly.
This reverse causal pathway could partially explain the positive association we found. Given the inherent characteristics of observational epidemiological research, reverse causality cannot be fully excluded in the present study. Sixth, the follow-up duration from 2011 to 2020 was finite.
This relatively limited observation window may miss later incident arthritis cases, restricting our ability to evaluate long-term associations between TyG-BMI and arthritis. Finally, our analytical sample was restricted to Chinese adults aged ≥45 years with CKM stages 0 to 3 via strict eligibility criteria, which greatly limits the external generalizability of our findings. The results cannot be generalized to younger populations, individuals without metabolic disorders, or ethnic groups from other areas.
Besides, all analyses were completed within the single CHARLS cohort without external validation in an independent population. Future studies integrating genetic data and multi-center repeated validations are required to consolidate our conclusions.
6. Conclusions
In summary, this study clarified the relationship between TyG-BMI and incident arthritis among individuals with CKM stages 0 to 3. Elevated TyG-BMI served as an independent risk factor for new-onset arthritis. HDL-C showed a statistically significant inverse correlative relationship with the TyG-BMI–arthritis association.
Nevertheless, we cannot confirm HDL-C acts as a true causal mediator given the inherent limitations of observational research. Effect modification by sleep duration and CKM stage was observed, with the strongest correlation detected among participants with sleep duration <7 hours or those classified as CKM stage 1.
These exploratory findings imply that regular TyG-BMI monitoring, improved sleep habits and HDL-C regulation may aid arthritis risk management for individuals in CKM stage 1. The TyG-BMI inflection point of 197.89 observed in our cohort is merely a sample-specific descriptive feature and is not validated for clinical cutoff application.
Future studies should evaluate longitudinal TyG-BMI trajectories, confirm the findings in independent populations, distinguish arthritis subtypes, and test whether improving metabolic health, lipid profiles, and sleep patterns can reduce arthritis or rheumatism incidence.
Acknowledgments
The authors acknowledge the China Health and Retirement Longitudinal Study (CHARLS) team for making the data available for this research. We are also deeply grateful to all the study participants and field staff for their invaluable contributions to the study.
Author contributions
Data curation: Haokun Yan.
Formal analysis: Yaojia Yang.
Methodology: Yaojia Yang, Baixiao Zhao.
Conceptualization: Baixiao Zhao.
Funding acquisition: Baixiao Zhao.
Writing – original draft: Haokun Yan.