Combined association of triglyceride-glucose index and the atherogenic index of plasma with the incidence of cardiovascular diseases among middle-aged and older population Authors: Xueyu Chen, Xuezhen Zhao, Haicheng Fei, Yixin Zhang, Jinhong Pang, Zhang Liu, Wenjuan Wang, Jianhong Qiao, Cuiping Xu Published: 2025 ABSTRACT BackgroundCardiovascular diseases (CVDs) are the leading cause of death worldwide. The atherogenic index of plasma (AIP) reflects atherogenic dyslipidemia and triglyceride-glucose (TyG) index is a surrogate of insulin resistance (IR). Evidence on their combined value for CVDs risk stratification remain limited. In this study, the associations between baseline levels and longitudinal changes of the composite TyG-AIP index and the incidence of CVDs were evaluated among middle-aged and older adults.MethodsData were obtained from the China Health and Retirement Longitudinal Study (CHARLS). A total of 6,986 participants were included in the baseline analysis(2012–2020), and 4,134 participants with repeated biomarker measurements in 2012 and 2015 were included in the longitudinal trajectory analysis. Multivariable cox proportional hazards models and restricted cubic spline(RCS) were applied to evaluate the associations of TyG, AIP, TyG&AIP and TyG-AIP with the incidence of CVDs (stroke and heart disease). K-means clustering characterized longitudinal TyG-AIP patterns. Discrimination was assessed with nomograms and receiver operating characteristic (ROC) curves.ResultsA total of 6,986 participants were included and followed for a median of 8.0 years, during which 1,752 incident CVDs occurred, including 1,343 cases of heart disease and 614 cases of stroke. Across tertiles of the TyG-AIP index, the risk of CVDs increased progressively, with adjusted HRs of 1.16 (95% CI: 1.03–1.31) for T2 and 1.25 (95% CI: 1.10–1.41) for T3 compared with T1. For stroke, the associations were higher in magnitude, with adjusted HRs of 1.40 (95% CI: 1.14–1.74) in T2 and 1.52 (95% CI: 1.23–1.88) in T3. TyG-AIP showed modest but superior discrimination versus TyG or AIP alone (AUC: 0.611 for CVDs; 0.631 for stroke; 0.605 for heart disease). NRI and IDI analyses demonstrated that adding TyG-AIP significantly improved risk reclassification for CVDs (NRI, 0.036–0.054, P < 0.001) and stroke (NRI, 0.096–0.114, P < 0.001). In longitudinal analyses (N = 4,134), participants in the cluster with persistently high and rising TyG-AIP values exhibited the highest risks of CVDs (adjusted HR 1.25, 95% CI: 1.04–1.51), stroke (HR 1.43, 95% CI: 1.05–1.95), and heart disease (HR 1.25, 95% CI: 1.01–1.54).ConclusionBoth baseline and longitudinal changes of TyG-AIP were independently associated with the risk of developing CVDs, especially stroke, in middle-aged and older Chinese adults. Repeated assessment of TyG-AIP captured cardiometabolic deterioration over time and improved identification of individuals at elevated cardiovascular risk. Incorporating long-term monitoring of TyG-AIP into routine health evaluations may enhance population-level CVDs risk prediction and support more effective prevention strategies.Graphical Abstract Supplementary InformationThe online version contains supplementary material available at 10.1186/s12933-025-03038-3. Introduction Cardiovascular diseases (CVDs) remain the leading cause of death worldwide and a major contributor to disability [1]. From 1990 to 2023, deaths attributed to CVDs increased substantially, rising from 13.1 million (95% UI: 12.2–14.0 million) to 19.2 million (95% UI: 17.4–20.4 million) [2]. Additionally, the prevalence of CVDs more than doubled, escalating from 311 million (95% UI: 294–333 million) cases in 1990 to 626 million (95% UI: 591–672 million) in 2023. Such a rapidly increasing global burden highlights the need for more effective early detection and improved risk stratification strategies to better guide prevention and clinical management of CVDs [3]. At the biological level, cardiovascular risk is strongly shaped by disturbances in endothelial function, glucose–lipid metabolism, and chronic inflammation [4–7]. Dysregulation of glucose and lipid homeostasis, together with persistent inflammatory activity, fosters insulin resistance (IR) and atherogenic dyslipidemia, leading to sustained hyperglycemic and hyperlipidemic states that aggravate endothelial injury [8–10]. These processes accelerate macrovascular and microvascular damage affecting the heart and brain [11, 12]. Older adults are particularly vulnerable to such metabolic and vascular deterioration due to age-related declines in vascular integrity and physiological resilience [12]. The atherogenic index of plasma (AIP) has emerged as a novel lipid metabolism marker with strong utility for assessing atherosclerotic risk and IR [13, 14]. Traditional lipid parameters, such as triglycerides (TG) and high-density lipoprotein cholesterol (HDL-C), are widely employed for cardiovascular risk assessment. However, AIP, calculated as the logarithm of the TG/HDL-C ratio, may offer superior characterization of atherogenic dyslipidemia [15]. Current research indicated that elevated AIP correlates closely with increased IR and heightened risk for developing type 2 diabetes mellitus [16]. Such findings highlight AIP could be a potentially valuable tool for identifying individuals at elevated risk for metabolic disorders and CVDs. IR is a central driver of CVDs progression, and growing evidence indicates that the triglyceride-glucose (TyG) index serves as a robust surrogate marker of IR with meaningful prognostic value for CVDs and their related complications [17–19]. Similarly, AIP has been widely recognized as an indicator of atherogenic dyslipidemia and an established predictor of CVDs risk. Although both the TyG and AIP independently contribute to cardiometabolic risk assessment, evidence supporting the combined utility of TyG and AIP is limited, as most previous studies have examined TyG and AIP only in relation to CVDs separately [20, 21]. A recent study evaluated TyG and AIP jointly using binary high-low classifications and reported only modest improvements in CVD prediction, with overall diagnostic performance remaining limited [22]. The earlier binary combination of TyG and AIP provides useful preliminary evidence of their joint relevance. However, this categorical approach reflects only one mode of integrating the two biomarkers. Continuous composite measures, such as TyG-AIP, may yield additional insights, particularly when examined longitudinally. Accordingly, the present cohort study aims to evaluate both the individual associations and the synergistic effect of TyG and AIP indices with CVDs risk among middle-aged and elderly populations. Specifically, the study seeks to determine the relationship between baseline levels of TyG-AIP and changes in these indices during follow-up with the incidence of CVDs, ultimately aiming to identify novel and more effective biomarkers for cardiovascular risk stratification. Methods Study design and population Data for this study were obtained from the China Health and Retirement Longitudinal Survey (CHARLS), a nationwide cohort designed to be highly representative of middle-aged and older adults in China [23]. The CHARLS study recruited participants aged 45 years and above from 450 rural and urban communities spread across 30 provinces in China [24]. The baseline participants for this study were initially recruited between 2011 and 2012, with a total of 17,708 individuals enrolled. In 2013, 2015, 2018, and 2020, face-to-face computer-assisted personal interviews (CAPI) were applied to follow up all study participants. During the follow-up period, standardized questionnaires were administered to collect information on demographics, lifestyle, and health-related data. Physical examinations and blood biochemical data collection were conducted at baseline and during the 2015 survey. The retrospective study analyzed data from the 2012 and 2020 waves of CHARLS. Eligibility criteria for participant selection were as follows: (1) individuals aged at least 45 years old in CHARLS 2012; (2) and having blood sample information in CHARLS 2012 and 2015; (3) and complete questionnaire records. Exclusion criteria were: (1) no informed consent was signed; (2) had major diseases or tumors; (3) incomplete laboratory test data or incomplete CVDs-related questionnaires; (4) individuals with CVDs at baseline. A total of 17,708 participants were initially recruited in 2012. Participants lacking sufficient follow-up visits, missing blood sample information, or incomplete physical examination data were excluded from the analysis, resulting in the exclusion of 10,722 individuals. Consequently, 6,986 participants were included in the correlation analysis examining the relationship between baseline TyG-AIP and the incidence of CVDs from 2012 to 2020. Additionally, accounting the loss to follow-up, 4,134 participants were analyzed to explore the association between longitudinal changes in TyG-AIP and the incidence of new-onset CVDs. A detailed flowchart of the participant selection process is provided in Fig. 1. Fig. 1Flowchart of the study participants selection. Abbreviation: FBG, fasting blood glucose; TG, triglyceride; HDL-C, high density lipoprotein cholesterol; LDL-C, low density lipoprotein cholesterol; TyG-AIP, triglyceride glucose-atherogenic index of plasma; CVDs, cardiovascular diseases. Flowchart of the study participants selection. Abbreviation: FBG, fasting blood glucose; TG, triglyceride; HDL-C, high density lipoprotein cholesterol; LDL-C, low density lipoprotein cholesterol; TyG-AIP, triglyceride glucose-atherogenic index of plasma; CVDs, cardiovascular diseases. Assessment of TyG-AIP The TyG index was a composited value composed of TG and fasting blood glucose levels (FBG) [25]. The TyG was calculated using the formula: TyG = ln [TG (mg/dl) × FBG (mg/dl)/2]. The AIP, defined as the logarithm of the TG to HDL-C ratio, demonstrates superior predictive value compared to traditional lipid metrics [26]. The AIP was calculated with the following formulas: AIP = log10 [TG (mg/dl) / HDL-C (mg/dl)]. Since AIP values can take both positive and negative ranges depending on the balance between TG and HDL-C levels, Min–Max normalization was applied to rescale all AIP values to a 0–1 range. This normalization was performed to ensure the comparability of the composite index, as negative AIP values could otherwise affect the magnitude and interpretation of the integrated indicator. The TyG-AIP index was constructed as the product of TyG and normalized AIP, reflecting the synergistic interaction between glucose- and lipid-related metabolic pathways. Assessment of CVDs The outcome measured was the occurrence of incident CVDs, which included stroke and chronic heart disease. The diagnostic criteria for these conditions were based on self-reported physician diagnosis (including angina, heart attack, congestive heart failure, and other heart problems or stroke) and the current use of medications for treating the disease [23]. Assessment of the change in TyG-AIP during follow-up To characterize longitudinal changes in TyG-AIP between 2012 and 2015, participants were grouped into distinct patterns based on the evolution of TyG-AIP values across the two survey waves. Owing to CHARLS data limitations, TyG-AIP values were available only at these two time points. The grouping was derived using a K-means clustering framework, which classified individuals according to shared patterns of change in TyG-AIP over time. K-means clustering followed by survival analysis to characterize the longitudinal patterns of TyG-AIP changes between 2012 and 2015. Due to limitations of the CHARLS dataset, TyG-AIP values were available only for these two survey waves; therefore, the two observed TyG-AIP measurements were used as the input variables for clustering, and no additional slope variable was included. The optimal number of clusters was determined using the elbow method, which supported a three-cluster solution, and the resulting groups were further evaluated to ensure clinical and biological relevance. Covariates Baseline and follow-up information were obtained through standardized data collection procedures, including interviewer-administered questionnaires, physical examinations, and fasting laboratory assessments. Demographic and lifestyle variables included sex, age, educational level, smoking status, and drinking status. Educational level was categorized as primary school or below, middle school, or high school and above. Smoking status was classified as current smoker or non-smoker, and drinking status as current drinker or non-drinker. physical examinations were obtained following standardized protocols and included height, weight, and body mass index (BMI). Blood pressure was measured using a calibrated digital sphygmomanometer, with systolic (SBP) and diastolic blood pressure (DBP) recorded as the mean of three readings taken after the participant had rested in a seated position for at least five minutes. Information on medical history and medication use was also collected, including physician-diagnosed heart disease, hypertension, diabetes, and dyslipidemia, as well as the use of antihypertensive, lipid-lowering, and antidiabetic medications. Fasting venous blood samples were collected in the morning and analyzed for FBG, total cholesterol (TC), TG, HDL-C, and low-density lipoprotein cholesterol (LDL-C). Participants that met one of the following criteria were considered diabetes: FBG ≥ 126 mg/dL, self-reported doctor-diagnosed diabetes, or currently used anti-diabetic medication. Subjects who met any of the following criteria were classified as having dyslipidemia: TC ≥ 240 mg/dL, HDL-C < 40 mg/dL, LDL-C ≥ 160 mg/dL, or TG ≥ 200 mg/dL [27]. Self-reported doctor-diagnosed dyslipidemia and currently using lipid-lowering medication were also considered to have dyslipidemia. Hypertension was defined as having a SBP ≥ 140 mmHg or a DBP ≥ 90 mmHg [28]. Additionally, participants who had been diagnosed with hypertension by a physician or were currently using antihypertensive medications were also categorized as having hypertension. Statistical analysis The normality of datasets was tested using the Kolmogorov–Smirnov test method. Continuous characteristics were described using mean and standard deviation (mean ± SD), while categorical characteristics were presented as frequencies and percentages (N, %). For characteristics that were not normally distributed, median and interquartile range (IQR) were used. Participants were stratified into three groups based on tertiles of the baseline TyG-AIP index, with T1 representing the lowest group and T3 representing the highest. Additionally, participants were categorized into four classes based on the median values of TyG and AIP at baseline, with Class 1 including individuals with low TyG and low AIP, Class 2 including those with high TyG and low AIP, Class 3 including those with low TyG and high AIP, and Class 4 including those with high TyG and high AIP. Differences among groups were tested using one-way ANOVA for normally distributed continuous variables. For non-normally distributed continuous variables, the Kruskal–Wallis test was utilized, while differences in categorical variables were evaluated using the chi-square (χ²) test. To address missing information, multiple imputation was applied for variables with minimal missingness, whereas individuals with substantial missing data were excluded from the analysis. K-means clustering followed by survival analysis was used to identify longitudinal patterns of TyG-AIP change. The two available TyG-AIP measurements (2012 and 2015) served as clustering inputs, as the CHARLS dataset provides laboratory data only for these two survey waves. K-means is an unsupervised learning algorithm that partitions observations into clusters by minimizing the within-cluster sum of squared Euclidean distances, thereby grouping individuals with similar trajectories of TyG-AIP change. To determine the optimal number of clusters, the elbow method was applied to evaluate reductions in within-cluster variation across increasing values of K [29]. The inspection of the elbow plot indicated that K = 3 yielded the most appropriate clustering structure, as further increases in K provided negligible improvement in model fit. Based on this data-driven assessment, participants were classified into three distinct groups representing different longitudinal patterns of TyG-AIP change, and these clusters were subsequently reviewed to ensure clinical and biological plausibility. Individual follow-up time was calculated from baseline to the earliest occurrence of incident CVDs, death, loss to follow-up, or end of study, whichever occurred first. Participants who died without experiencing a CVDs were censored at the date of death, while those lost to follow-up were censored at the date of their last successful interview. Cox proportional hazards regression models were applied to examine the associations between baseline TyG-AIP groups, the K-means derived clusters, and the incidence of CVDs. The crude model included no covariate adjustments, whereas the adjusted model included age, sex, educational level, smoking status, drinking status, hypertension, diabetes, TC and LDL-C. Prior to variable selection, variance inflation factors (VIFs) were computed to assess multicollinearity, and variables with VIF > 10 (HDL-C and TG) were excluded from the multivariable models. The proportional hazards assumption was assessed using Schoenfeld residuals, and no violations were detected for any model or covariate with all P > 0.05. Hazard ratios (HRs) with corresponding 95% confidence intervals (CIs) were calculated to quantify these associations. The predictive performance of TyG-AIP in forecasting future CVDs events was evaluated using nomogram prediction model and Receiver Operating Characteristic (ROC) curves, with higher area under the curve (AUC) values indicating stronger predictive power. Net Reclassification Improvement (NRI) and Integrated Discrimination Improvement (IDI) were calculated to assess the incremental predictive value of TyG-AIP beyond TyG or AIP. Incremental predictive value was evaluated using NRI and IDI, which provide sensitive statistical assessment of whether the addition of TyG-AIP improves model performance. Restricted cubic spline (RCS) plots were generated to examine the linearity and potential non-linear relationships between changes in TyG-AIP and the risk of CVDs incidence. The RCS models were fitted using four knots placed at the 5th, 35th, 65th, and 95th percentiles of the TyG-AIP distribution within the fully adjusted cox proportional hazards regression models. The threshold of RCS was defined as the TyG‑AIP value at which the HR curve first crossed unity (HR = 1.0) and remained above it, with the 95% CI no longer crossing 1. A two-sided P value < 0.05 was considered statistically significant. All statistical analyses and plotting were performed by the R software (version 4.4.2, package: survival, survminer, survivalROC, rms, splines, survIDINRI, nricens, cluster, NbClust, lcmm, gtsummary, forestploter, and ggplot2. http://www.R-project.org/). Results Baseline characteristics of study participants A total of 6,986 participants (3,695 women and 3,291 men) were included in this study. The mean age was 59.03 years (SD = 9.24). Table 1 presents the baseline characteristics both overall and stratified by the CVDs status. Participants were followed for a median of 8.0 years (IQR: 6.0–8.0), 1,752 individuals developed CVDs, including 1,343 cases of heart disease and 614 cases of stroke. Across tertiles of the TyG-AIP index, statistically significant differences were observed for all baseline characteristics (P < 0.05). Participants in higher TyG-AIP tertiles tended to be younger and more likely to be female, with a lower proportion of current smokers and drinkers. Adverse cardiometabolic profiles were evident in the higher tertiles, including progressively greater BMI, TC, TG, FBG, SBP, and DBP levels, accompanied by markedly lower HDL-C concentrations. Median value of UA levels also increased with rising TyG-AIP. As expected, mean values of TyG increased from 8.07 in T1 to 8.57 in T2 and 9.29 in T3, while median AIP rose from 0.03 (IQR: -0.08–0.11) in T1 to 0.31 (0.25–0.37) in T2 and 0.64 (0.53–0.79) in T3, consistent with the construction of the composite indicator. The prevalence of major cardiometabolic conditions increased across tertiles: hypertension ranged from 30.83% in T1 to 46.46% in T3, diabetes from 7.51% to 22.67%, and overall CVDs from 21.25% to 28.21%. Similar increasing patterns were observed for heart disease (16.96% to 21.04%) and stroke (6.35% to 10.65%). In addition, except for educational level, no statistically significant differences in demographic characteristics were observed between participants included in the final analysis and the overall population prior to exclusions (P > 0.05, Table S1). Table 1Baseline characteristics according to the baseline TyG-AIP tertilesCharacteristicTotalT1T2T3P valuecN = 6986N = 2329N = 2328N = 2329Age (years)59.03 ± 9.2459.56 ± 9.6659.12 ± 9.1658.40 ± 8.85< 0.001SexFemale3695 (52.89)1127 (48.39)1265 (54.34)1303 (55.95)< 0.001Male3291 (47.11)1202 (51.61)1063 (45.66)1026 (44.05)Education level0.001Primary school or below4883 (69.90)1669 (71.66)1646 (70.70)1568 (67.34)Middle school1402 (20.07)458 (19.67)428 (18.38)516 (22.16)High school or above701 (10.03)202 (8.67)254 (10.91)245 (10.52)Current smoking2160 (30.92)789 (33.88)710 (30.50)661 (28.38)< 0.001Current drinking2293 (32.82)939 (40.32)720 (30.93)734 (31.52)< 0.001BMI (kg/m²) a23.38 ± 3.6822.05 ± 3.2523.30 ± 3.6224.78 ± 3.64< 0.001TC (mg/dl) a194.12 ± 37.88188.40 ± 34.69192.43 ± 37.39201.53 ± 40.17< 0.001TG (mg/dl) a125.14 ± 82.2465.34 ± 15.77104.03 ± 20.88206.06 ± 95.04< 0.001HDL-C (mg/dl) a51.78 ± 14.9964.44 ± 13.8350.85 ± 10.0840.06 ± 9.08< 0.001LDL-C (mg/dl) a117.83 ± 34.48113.08 ± 30.40122.66 ± 33.91117.76 ± 38.06< 0.001FBG (mg/dl) a108.42 ± 30.76101.91 ± 20.30105.62 ± 27.58117.74 ± 39.11< 0.001UA (mg/dl) b4.29 (3.56,5.12)4.12 (3.45,4.92)4.21 (3.52,5.01)4.54 (3.76,5.43)< 0.001Systolic (mmHg) a129.90 ± 21.22127.30 ± 20.89129.50 ± 21.26132.91 ± 21.14< 0.001Diastolic (mmHg) a75.58 ± 12.0473.79 ± 11.8375.37 ± 11.9977.58 ± 12.00< 0.001TyG a8.64 ± 0.628.07 ± 0.308.57 ± 0.279.29 ± 0.49< 0.001AIP b0.31 (0.11,0.53)0.03 (-0.08,0.11)0.31 (0.25,0.37)0.64 (0.53,0.79)< 0.001TyG-AIP a3.16 ± 1.391.78 ± 0.472.98 ± 0.324.72 ± 1.06< 0.001Hypertension2663 (38.12)718 (30.83)863 (37.07)1082 (46.46)< 0.001Diabetes964 (13.80)175 (7.51)261 (11.21)528 (22.67)< 0.001CVDs1752 (25.08)495 (21.25)600 (25.77)657 (28.21)< 0.001Heart disease1343 (19.22)395 (16.96)458 (19.67)490 (21.04)0.002Stroke614 (8.79)148 (6.35)218 (9.36)248 (10.65)< 0.001 TyG low3493 (50.00)2243 (96.31)1209 (51.93)41 (1.76)< 0.001high3493 (50.00)86 (3.69)1119 (48.07)2288 (98.24) AIP low3493 (50.00)2329(100)1164 (0.5)0 (0)< 0.001high3493 (50.00)0 (0)1164 (0.5)2329 (100) TyG&AIP Class13020 (43.23)2243 (96.31)777 (33.38)0 (0)< 0.001Class2473 (6.77)86 (3.69)387 (16.62)0 (0)Class3473 (6.77)0 (0)432 (18.56)41 (1.76)Class43020 (43.23)0 (0)732 (31.44)2288 (98.24)a Data are mean ± SD; b Data are median (IQR); c One-way ANOVA test, Kruskal-Wallis rank sum test, Pearson’s Chi-squared test; TyG index, triglyceride–glucose index; AIP, the atherogenic index of plasma; BMI, body mass index; TyG-AIP, TyG × ZAIP; FBG, fasting blood glucose; TC, total cholesterol; TG, triglyceride; HDL-C, high density lipoprotein cholesterol; LDL-C, low density lipoprotein cholesterol; UA, uric acid; SBP, systolic blood pressure; DBP, diastolic blood pressure; CVDs, cardiovascular diseases; Class1, low TyG and low AIP༛Class2, high TyG and low AIP༛Class3, low TyG and high AIP༛Class4, high TyG and high AIP Table 1Baseline characteristics according to the baseline TyG-AIP tertilesCharacteristicTotalT1T2T3P valuecN = 6986N = 2329N = 2328N = 2329Age (years)59.03 ± 9.2459.56 ± 9.6659.12 ± 9.1658.40 ± 8.85< 0.001SexFemale3695 (52.89)1127 (48.39)1265 (54.34)1303 (55.95)< 0.001Male3291 (47.11)1202 (51.61)1063 (45.66)1026 (44.05)Education level0.001Primary school or below4883 (69.90)1669 (71.66)1646 (70.70)1568 (67.34)Middle school1402 (20.07)458 (19.67)428 (18.38)516 (22.16)High school or above701 (10.03)202 (8.67)254 (10.91)245 (10.52)Current smoking2160 (30.92)789 (33.88)710 (30.50)661 (28.38)< 0.001Current drinking2293 (32.82)939 (40.32)720 (30.93)734 (31.52)< 0.001BMI (kg/m²) a23.38 ± 3.6822.05 ± 3.2523.30 ± 3.6224.78 ± 3.64< 0.001TC (mg/dl) a194.12 ± 37.88188.40 ± 34.69192.43 ± 37.39201.53 ± 40.17< 0.001TG (mg/dl) a125.14 ± 82.2465.34 ± 15.77104.03 ± 20.88206.06 ± 95.04< 0.001HDL-C (mg/dl) a51.78 ± 14.9964.44 ± 13.8350.85 ± 10.0840.06 ± 9.08< 0.001LDL-C (mg/dl) a117.83 ± 34.48113.08 ± 30.40122.66 ± 33.91117.76 ± 38.06< 0.001FBG (mg/dl) a108.42 ± 30.76101.91 ± 20.30105.62 ± 27.58117.74 ± 39.11< 0.001UA (mg/dl) b4.29 (3.56,5.12)4.12 (3.45,4.92)4.21 (3.52,5.01)4.54 (3.76,5.43)< 0.001Systolic (mmHg) a129.90 ± 21.22127.30 ± 20.89129.50 ± 21.26132.91 ± 21.14< 0.001Diastolic (mmHg) a75.58 ± 12.0473.79 ± 11.8375.37 ± 11.9977.58 ± 12.00< 0.001TyG a8.64 ± 0.628.07 ± 0.308.57 ± 0.279.29 ± 0.49< 0.001AIP b0.31 (0.11,0.53)0.03 (-0.08,0.11)0.31 (0.25,0.37)0.64 (0.53,0.79)< 0.001TyG-AIP a3.16 ± 1.391.78 ± 0.472.98 ± 0.324.72 ± 1.06< 0.001Hypertension2663 (38.12)718 (30.83)863 (37.07)1082 (46.46)< 0.001Diabetes964 (13.80)175 (7.51)261 (11.21)528 (22.67)< 0.001CVDs1752 (25.08)495 (21.25)600 (25.77)657 (28.21)< 0.001Heart disease1343 (19.22)395 (16.96)458 (19.67)490 (21.04)0.002Stroke614 (8.79)148 (6.35)218 (9.36)248 (10.65)< 0.001 TyG low3493 (50.00)2243 (96.31)1209 (51.93)41 (1.76)< 0.001high3493 (50.00)86 (3.69)1119 (48.07)2288 (98.24) AIP low3493 (50.00)2329(100)1164 (0.5)0 (0)< 0.001high3493 (50.00)0 (0)1164 (0.5)2329 (100) TyG&AIP Class13020 (43.23)2243 (96.31)777 (33.38)0 (0)< 0.001Class2473 (6.77)86 (3.69)387 (16.62)0 (0)Class3473 (6.77)0 (0)432 (18.56)41 (1.76)Class43020 (43.23)0 (0)732 (31.44)2288 (98.24)a Data are mean ± SD; b Data are median (IQR); c One-way ANOVA test, Kruskal-Wallis rank sum test, Pearson’s Chi-squared test; TyG index, triglyceride–glucose index; AIP, the atherogenic index of plasma; BMI, body mass index; TyG-AIP, TyG × ZAIP; FBG, fasting blood glucose; TC, total cholesterol; TG, triglyceride; HDL-C, high density lipoprotein cholesterol; LDL-C, low density lipoprotein cholesterol; UA, uric acid; SBP, systolic blood pressure; DBP, diastolic blood pressure; CVDs, cardiovascular diseases; Class1, low TyG and low AIP༛Class2, high TyG and low AIP༛Class3, low TyG and high AIP༛Class4, high TyG and high AIP Baseline characteristics according to the baseline TyG-AIP tertiles a Data are mean ± SD; b Data are median (IQR); c One-way ANOVA test, Kruskal-Wallis rank sum test, Pearson’s Chi-squared test; TyG index, triglyceride–glucose index; AIP, the atherogenic index of plasma; BMI, body mass index; TyG-AIP, TyG × ZAIP; FBG, fasting blood glucose; TC, total cholesterol; TG, triglyceride; HDL-C, high density lipoprotein cholesterol; LDL-C, low density lipoprotein cholesterol; UA, uric acid; SBP, systolic blood pressure; DBP, diastolic blood pressure; CVDs, cardiovascular diseases; Class1, low TyG and low AIP༛Class2, high TyG and low AIP༛Class3, low TyG and high AIP༛Class4, high TyG and high AIP Association between baseline TyG, AIP, TyG&AIP, and TyG-AIP with the incidence of CVDs, stroke and heart disease Cox regression analysis of TyG and AIP indicated that higher TyG, AIP, and TyG-AIP levels elevated the risks of CVDs, stroke, and heart disease (Table 2). In crude model, HRs for CVDs were 1.32 (95% CI: 1.20–1.45) in high TyG group and 1.34 (95% CI: 1.22–1.48) in high AIP group. As for stroke, HRs were 1.53 (95% CI: 1.30–1.80), 1.44 (95% CI: 1.23–1.69), respectively. For heart disease, HRs were 1.22 (95% CI: 1.10–1.36) and 1.29 (95% CI: 1.16–1.43). After adjusting for all covariates, the associations remained robust. In the adjusted model, participants with high TyG had a higher risk of CVDs (HR = 1.15, 95% CI: 1.04–1.28) and stroke (HR = 1.34, 95% CI: 1.12–1.59) compared with those with low TyG. Relative to individuals with low AIP, those with high AIP demonstrated increased risks of CVDs (HR = 1.24, 95% CI: 1.13–1.37), stroke (HR = 1.32, 95% CI: 1.12–1.56), and heart disease (HR = 1.20, 95% CI: 1.07–1.34). Table 2Association between the baseline TyG, AIP, TyG&AIP and TyG-AIP with the incidence of CVDsCrude modelP valueAdjusted modelP valueHR (95%CI)HR (95%CI)CVDs TyG Low Reference Reference High1.32 (1.20–1.45)< 0.0011.15 (1.04–1.28)0.006 AIP Low Reference Reference High1.34 (1.22–1.48)< 0.0011.24 (1.13–1.37)< 0.001 TyG&AIP Class1 Reference Reference Class21.20 (0.98–1.46)0.0700.97 (0.79–1.20)0.790Class31.29 (1.06–1.55)0.0101.29 (1.07–1.57)0.009Class41.39 (1.26–1.54)< 0.0011.22 (1.10–1.36)< 0.001 TyG-AIP T1 Reference Reference T21.24 (1.10–1.40)< 0.0011.16 (1.03–1.31)0.010T31.39 (1.23–1.56)< 0.0011.25 (1.10–1.41)< 0.001 Stroke TyG Low Reference Reference High1.53 (1.30–1.80)< 0.0011.34 (1.12–1.59)0.001 AIP Low Reference Reference High1.44 (1.23–1.69)< 0.0011.32 (1.12–1.56)0.001 TyG&AIP Class1 Reference Reference Class21.43 (1.04–1.99)0.0301.13 (0.81–1.59)0.460Class31.15 (0.81–1.62)0.4301.14 (0.80–1.61)0.480Class41.58 (1.33–1.88)< 0.0011.39 (1.16–1.67)< 0.001 TyG-AIP T1 Reference Reference T21.48 (1.20–1.83)< 0.0011.40 (1.14–1.74)0.002T31.71 (1.39–2.09)< 0.0011.52 (1.23–1.88)< 0.001 Heart disease TyG Low Reference Reference High1.22 (1.10–1.36)< 0.0011.08 (0.96–1.22)0.180 AIP Low Reference Reference High1.29 (1.16–1.43)< 0.0011.20 (1.07–1.34)0.001 TyG&AIP Class1 Reference Reference Class21.09 (0.86–1.37)0.4800.91 (0.72–1.16)0.440Class31.32 (1.07–1.63)0.0101.33 (1.07–1.65)0.009Class41.30 (1.16–1.46)< 0.0011.15 (1.02–1.31)0.020 TyG-AIP T1 Reference Reference T21.18 (1.03–1.35)0.0201.10 (0.96–1.26)0.170T31.28 (1.12–1.46)< 0.0011.16 (1.01–1.33)0.040Crude model: No covariates were adjusted; Adjusted model: Age, sex, education levels, current smoking, current drinking, hypertension, diabetes, total cholesterol, low density lipoprotein cholesterol were adjusted. Abbreviation: HR, hazard ratio, CI, confidence interval, TyG index, triglyceride-glucose index, AIP, the atherogenic index of plasma, CVDs, cardiovascular diseases; Class1, low TyG and low AIP; Class2, high TyG and low AIP; Class3, low TyG and high AIP; Class4, high TyG and high AIP; T1, the lowest tertile, T2, the middle tertile; T3, the highest tertile Table 2Association between the baseline TyG, AIP, TyG&AIP and TyG-AIP with the incidence of CVDsCrude modelP valueAdjusted modelP valueHR (95%CI)HR (95%CI)CVDs TyG Low Reference Reference High1.32 (1.20–1.45)< 0.0011.15 (1.04–1.28)0.006 AIP Low Reference Reference High1.34 (1.22–1.48)< 0.0011.24 (1.13–1.37)< 0.001 TyG&AIP Class1 Reference Reference Class21.20 (0.98–1.46)0.0700.97 (0.79–1.20)0.790Class31.29 (1.06–1.55)0.0101.29 (1.07–1.57)0.009Class41.39 (1.26–1.54)< 0.0011.22 (1.10–1.36)< 0.001 TyG-AIP T1 Reference Reference T21.24 (1.10–1.40)< 0.0011.16 (1.03–1.31)0.010T31.39 (1.23–1.56)< 0.0011.25 (1.10–1.41)< 0.001 Stroke TyG Low Reference Reference High1.53 (1.30–1.80)< 0.0011.34 (1.12–1.59)0.001 AIP Low Reference Reference High1.44 (1.23–1.69)< 0.0011.32 (1.12–1.56)0.001 TyG&AIP Class1 Reference Reference Class21.43 (1.04–1.99)0.0301.13 (0.81–1.59)0.460Class31.15 (0.81–1.62)0.4301.14 (0.80–1.61)0.480Class41.58 (1.33–1.88)< 0.0011.39 (1.16–1.67)< 0.001 TyG-AIP T1 Reference Reference T21.48 (1.20–1.83)< 0.0011.40 (1.14–1.74)0.002T31.71 (1.39–2.09)< 0.0011.52 (1.23–1.88)< 0.001 Heart disease TyG Low Reference Reference High1.22 (1.10–1.36)< 0.0011.08 (0.96–1.22)0.180 AIP Low Reference Reference High1.29 (1.16–1.43)< 0.0011.20 (1.07–1.34)0.001 TyG&AIP Class1 Reference Reference Class21.09 (0.86–1.37)0.4800.91 (0.72–1.16)0.440Class31.32 (1.07–1.63)0.0101.33 (1.07–1.65)0.009Class41.30 (1.16–1.46)< 0.0011.15 (1.02–1.31)0.020 TyG-AIP T1 Reference Reference T21.18 (1.03–1.35)0.0201.10 (0.96–1.26)0.170T31.28 (1.12–1.46)< 0.0011.16 (1.01–1.33)0.040Crude model: No covariates were adjusted; Adjusted model: Age, sex, education levels, current smoking, current drinking, hypertension, diabetes, total cholesterol, low density lipoprotein cholesterol were adjusted. Abbreviation: HR, hazard ratio, CI, confidence interval, TyG index, triglyceride-glucose index, AIP, the atherogenic index of plasma, CVDs, cardiovascular diseases; Class1, low TyG and low AIP; Class2, high TyG and low AIP; Class3, low TyG and high AIP; Class4, high TyG and high AIP; T1, the lowest tertile, T2, the middle tertile; T3, the highest tertile Association between the baseline TyG, AIP, TyG&AIP and TyG-AIP with the incidence of CVDs Crude model: No covariates were adjusted; Adjusted model: Age, sex, education levels, current smoking, current drinking, hypertension, diabetes, total cholesterol, low density lipoprotein cholesterol were adjusted. Abbreviation: HR, hazard ratio, CI, confidence interval, TyG index, triglyceride-glucose index, AIP, the atherogenic index of plasma, CVDs, cardiovascular diseases; Class1, low TyG and low AIP; Class2, high TyG and low AIP; Class3, low TyG and high AIP; Class4, high TyG and high AIP; T1, the lowest tertile, T2, the middle tertile; T3, the highest tertile Setting the group with both low TyG and low AIP (Class 1) as reference group, the adjusted model indicated an HR of 1.39 (95% CI: 1.16–1.67) for the high TyG & high AIP group (Class 4) in stroke. While in the cases of CVDs and heart disease, the HR values were 1.29 (95% CI: 1.07–1.57) and 1.33 (95% CI: 1.07–1.65) for the low TyG & high AIP group (Class 3), higher than high TyG & high AIP group (CVDs HR: 1.22, 95% CI: 1.10–1.36; heart disease HR: 1.15, 95% CI: 1.02–1.31). Other groups did not show statistical significance in adjusted model. To further examine whether TyG and AIP exert a synergistic effect beyond their joint associations, a multiplicative interaction analysis was performed. After comprehensive adjustment for potential confounders, the multiplicative interaction analysis indicated that the confidence intervals of the interaction term (TyG × AIP) for CVDs, stroke, and heart disease all crossed 1, and no statistically significant multiplicative interaction was observed (P for interaction > 0.05). TyG-AIP was divided into three tertiles, with T1 as the reference group. In the adjusted model, the HR values for CVDs in T2 and T3 were 1.16 (95% CI: 1.03–1.31) and 1.25 (95% CI: 1.10–1.41), respectively. For stroke, the HR values were 1.40 (95% CI: 1.14–1.74) and 1.52 (95% CI: 1.23–1.88), respectively. Additionally, the HR for heart disease in T3 was 1.16 (95% CI: 1.01–1.33). Nomograms based on Cox regression adjusted model were established for TyG-AIP (Fig. 2). The contribution of TyG-AIP to stroke risk prediction was higher compared to CVDs and heart disease (CVDs: 24%; stroke: 34%; heart disease: 16%). Fig. 2Nomograms for predicting 8-year risk of CVDs, stroke, and heart disease based on the TyG-AIP index.The upper axis assigns points according to the value of TyG-AIP. Total points are summed and projected onto the lower scales to obtain the linear predictor and the estimated 8-year survival probability. Abbreviation: CVDs, cardiovascular diseases; TyG-AIP, triglyceride-glucose and atherogenic index of plasma. Nomograms for predicting 8-year risk of CVDs, stroke, and heart disease based on the TyG-AIP index.The upper axis assigns points according to the value of TyG-AIP. Total points are summed and projected onto the lower scales to obtain the linear predictor and the estimated 8-year survival probability. Abbreviation: CVDs, cardiovascular diseases; TyG-AIP, triglyceride-glucose and atherogenic index of plasma. Predictive performance and dose-response relationships of TyG, AIP, and TyG-AIP for CVDs, stroke and heart disease ROC analysis indicated that TyG-AIP demonstrated superior predictive performance for CVDs, stroke, and heart disease compared to TyG and AIP, with respective AUC values of 0.611, 0.631, and 0.605 (Fig. 3). To evaluate whether TyG-AIP provides additional predictive value beyond TyG or AIP, NRI and IDI analyses were conducted. As shown in Table S2, incorporating TyG-AIP significantly improved risk reclassification for CVDs (NRI = 0.054 vs. AIP; 0.036 vs. TyG; both P < 0.001) and stroke (NRI = 0.096 vs. AIP; 0.114 vs. TyG; both P < 0.001). Corresponding IDI values also indicated meaningful improvement in discrimination for CVDs (IDI = 0.004 and 0.002; both P < 0.001) and stroke (IDI = 0.009 and 0.008; both P < 0.001). For heart disease, TyG-AIP yielded modest but statistically significant improvements in both NRI and IDI relative to AIP and TyG (P < 0.05). RCS analysis demonstrated a nonlinear relationship between TyG-AIP and CVDs (P = 0.018) and stroke (P = 0.003), with a linear relationship for heart disease (P = 0.103). A TyG-AIP level above 2.95 was identified as the threshold associated with increased risk for the future incidence of CVDs, stroke, and heart disease (Fig. 4). Fig. 3ROC curves for the predictive performance of TyG, AIP, and TyG-AIP in identifying CVDs. Abbreviation: ROC, receiver operating characteristic; TyG index, triglyceride-glucose index; AIP, the atherogenic index of plasma; TyG-AIP, triglyceride glucose-atherogenic index of plasma; CVDs, cardiovascular diseases ROC curves for the predictive performance of TyG, AIP, and TyG-AIP in identifying CVDs. Abbreviation: ROC, receiver operating characteristic; TyG index, triglyceride-glucose index; AIP, the atherogenic index of plasma; TyG-AIP, triglyceride glucose-atherogenic index of plasma; CVDs, cardiovascular diseases Fig. 4Restricted cubic spline analysis of the relationships among TyG, AIP, TyG-AIP, and the increased risk of CVDs. Model was unadjusted. Abbreviation: TyG index, triglyceride-glucose index; AIP, the atherogenic index of plasma; TyG-AIP, triglyceride glucose-atherogenic index of plasma; HRs, hazard ratios; 95% CIs, 95% confidence intervals; CVDs, cardiovascular diseases Restricted cubic spline analysis of the relationships among TyG, AIP, TyG-AIP, and the increased risk of CVDs. Model was unadjusted. Abbreviation: TyG index, triglyceride-glucose index; AIP, the atherogenic index of plasma; TyG-AIP, triglyceride glucose-atherogenic index of plasma; HRs, hazard ratios; 95% CIs, 95% confidence intervals; CVDs, cardiovascular diseases Incidence of CVDs, stroke and heart disease according to TyG-AIP variations The K-means algorithm was used to classify TyG-AIP variations into three categories based on the values of TyG-AIP in 2012 and 2015 (Fig. 5B). Overall, 4,134 individuals had complete TyG-AIP data across both survey waves (2012 and 2015) and were therefore included in the longitudinal analysis of TyG-AIP changes. In Cluster 1 (N = 1,706), the values of TyG-AIP increased slightly from 1.95 to 2.16, representing a persistently low level with a mild upward trend. In Cluster 2 (N = 1,770), TyG-AIP increased from 3.48 in 2012 to 3.84 in 2015, indicating a moderate TyG-AIP and increasing trend. While Cluster 3 (N = 658) maintained the highest TyG-AIP with the fastest increase, progressing from 5.53 to 5.97 (Table S3). The incidence of CVDs, stroke, and heart disease exhibited progressively across the three clusters (Fig. 5E). CVDs incidence rose from 23.51% in Cluster 1 to 29.83% in Cluster 2 and 31.31% in Cluster 3. Similarly, stroke incidence rose from at 7.62% to 10.00% and 11.85% across Clusters 1, 2, and 3. Heart disease displayed a comparable gradient, with incidences of 18.41%, 23.67%, and 24.16% incidence in Cluster 1, 2, and 3. Fig. 5Clustering of the variations in the TyG-AIP from 2012 to 2015. A, Three classes were found using the K-means method with Euclidean distance: the x- and y-axes are principal components of the change in the TyG-AIP; B, The change of TyG-AIP from 2012 to 2015: Cluster1, the TyG-AIP changed from 1.95 ± 0.53 to 2.16 ± 0.59; Cluster2, the TyG-AIP changed from 3.48 ± 0.51 to 3.84 ± 0.54; Cluster3, the TyG-AIP changed from 5.53 ± 0.98 to 5.97 ± 0.99; C, The distribution of TyG-AIP in 2012; D, The distribution of TyG-AIP in 2015; E, Incidence rates of CVDs categorized by classes of the TyG-AIP. PC, principal component. Abbreviation: TyG-AIP, triglyceride glucose-atherogenic index of plasma Clustering of the variations in the TyG-AIP from 2012 to 2015. A, Three classes were found using the K-means method with Euclidean distance: the x- and y-axes are principal components of the change in the TyG-AIP; B, The change of TyG-AIP from 2012 to 2015: Cluster1, the TyG-AIP changed from 1.95 ± 0.53 to 2.16 ± 0.59; Cluster2, the TyG-AIP changed from 3.48 ± 0.51 to 3.84 ± 0.54; Cluster3, the TyG-AIP changed from 5.53 ± 0.98 to 5.97 ± 0.99; C, The distribution of TyG-AIP in 2012; D, The distribution of TyG-AIP in 2015; E, Incidence rates of CVDs categorized by classes of the TyG-AIP. PC, principal component. Abbreviation: TyG-AIP, triglyceride glucose-atherogenic index of plasma Association between TyG-AIP variability with the incidence of CVDs, stroke and heart disease Longitudinal variations in TyG-AIP were examined in relation to the incidence of CVDs, stroke, and heart disease among participants with repeated measurements (Table 3). In the crude model, compared with Cluster 1, Cluster 3 showed a significantly higher risk of developing CVDs, with an HR of 1.40 (95% CI: 1.19–1.66). For stroke, Cluster 3 exhibited an HR of 1.58 (95% CI: 1.20–2.10), while the HR for heart disease was 1.36 (95% CI: 1.13–1.65). After full adjustment, the HRs for Cluster 3 relative to Cluster 1 were 1.25 (95% CI: 1.04–1.51) for CVDs, 1.43 (95% CI: 1.05–1.95) for stroke, and 1.25 (95% CI: 1.01–1.54) for heart disease. Cluster3 continued to exhibit the highest risk of CVDs, stroke, and heart disease. Table 3Association between the variation of TyG-AIP and the incidence of CVDsCrude modelP valueAdjusted modelP valueHR (95%CI)HR (95%CI)K-means Group of CVDsCluster1 Reference Reference Cluster21.32 (1.16–1.50)< 0.0011.21 (1.06–1.38)0.005Cluster31.40 (1.19–1.66)< 0.0011.25 (1.04–1.51)0.020 K-means Group of Stroke Cluster1 Reference Reference Cluster21.32 (1.05–1.66)< 0.0011.23 (0.98–1.55)0.080Cluster31.58 (1.20–2.10)< 0.0011.43 (1.05–1.95)0.020 K-means Group of Heart disease Cluster1 Reference Reference Cluster21.33 (1.15–1.54)< 0.0011.21 (1.04–1.41)0.010Cluster31.36 (1.13–1.65)< 0.0011.25 (1.01–1.54)0.030Crude model: No covariates were adjusted; Adjusted model: age, sex, education levels, current smoking, current drinking, hypertension, diabetes, TC, LDL-C. Abbreviation: HR, hazard ratio; 95%CI, 95% confidence interval; TyG index triglyceride-glucose index, AIP the atherogenic index of plasma, TyG-AIP, triglyceride glucose-atherogenic index of plasma; CVDs, cardiovascular diseases Table 3Association between the variation of TyG-AIP and the incidence of CVDsCrude modelP valueAdjusted modelP valueHR (95%CI)HR (95%CI)K-means Group of CVDsCluster1 Reference Reference Cluster21.32 (1.16–1.50)< 0.0011.21 (1.06–1.38)0.005Cluster31.40 (1.19–1.66)< 0.0011.25 (1.04–1.51)0.020 K-means Group of Stroke Cluster1 Reference Reference Cluster21.32 (1.05–1.66)< 0.0011.23 (0.98–1.55)0.080Cluster31.58 (1.20–2.10)< 0.0011.43 (1.05–1.95)0.020 K-means Group of Heart disease Cluster1 Reference Reference Cluster21.33 (1.15–1.54)< 0.0011.21 (1.04–1.41)0.010Cluster31.36 (1.13–1.65)< 0.0011.25 (1.01–1.54)0.030Crude model: No covariates were adjusted; Adjusted model: age, sex, education levels, current smoking, current drinking, hypertension, diabetes, TC, LDL-C. Abbreviation: HR, hazard ratio; 95%CI, 95% confidence interval; TyG index triglyceride-glucose index, AIP the atherogenic index of plasma, TyG-AIP, triglyceride glucose-atherogenic index of plasma; CVDs, cardiovascular diseases Association between the variation of TyG-AIP and the incidence of CVDs Crude model: No covariates were adjusted; Adjusted model: age, sex, education levels, current smoking, current drinking, hypertension, diabetes, TC, LDL-C. Abbreviation: HR, hazard ratio; 95%CI, 95% confidence interval; TyG index triglyceride-glucose index, AIP the atherogenic index of plasma, TyG-AIP, triglyceride glucose-atherogenic index of plasma; CVDs, cardiovascular diseases The cumulative incidence of CVDs, stroke, and heart disease The Kaplan-Meier curves depicting the cumulative incidence of CVDs, stroke, and heart disease among all participants are presented in Fig. 6. Compared with participants with low TyG and low AIP, those with high TyG and high AIP exhibited higher cumulative incidences of cardiovascular outcomes, including 27.88% and 28.26% for CVDs, 10.51% and 10.33% for stroke, and 20.81% and 21.39% for heart disease. Consistently, participants in the highest TyG-AIP tertile (T3) experienced the greatest cumulative risk, reaching 28.21% for CVDs, 10.65% for stroke, and 21.04% for heart disease. Incidences varied across the combined TyG and AIP categories. The high TyG & high AIP group (Class4) exhibited the highest cumulative incidence of CVDs and stroke (28.48% and 10.70%, respectively), whereas the low TyG & high AIP (Class 3) group showed the highest incidence of heart disease (21.78%). When longitudinal TyG-AIP patterns were examined, the cluster characterized by persistently high and rising TyG-AIP values (Cluster 3) showed the steepest risk accumulation. For CVDs, stroke and heart disease, Cluster 3 displayed the highest cumulative incidence (CVDs: 31.31%; stroke: 11.85%, heart disease: 24.16%). Fig. 6Fig.6 Kaplan–Meier curves for cumulative incidence of cardiovascular outcomes across TyG, AIP, TyG-AIP based groups and longitudinal clusters. A-D, Cumulative incidence curves for CVDs; E-H, Cumulative incidence curves for stroke; I-L, Cumulative incidence curves for heart disease.Abbreviation: TyG index, triglyceride-glucose index; AIP, the atherogenic index of plasma; TyG-AIP, triglyceride glucose-atherogenic index of plasma; CVDs, cardiovascular diseases; T1-T3, tertiles of TyG-AIP; Cluster1-3, K-means derived groups reflecting longitudinal TyG-AIP trajectories. Fig.6 Kaplan–Meier curves for cumulative incidence of cardiovascular outcomes across TyG, AIP, TyG-AIP based groups and longitudinal clusters. A-D, Cumulative incidence curves for CVDs; E-H, Cumulative incidence curves for stroke; I-L, Cumulative incidence curves for heart disease.Abbreviation: TyG index, triglyceride-glucose index; AIP, the atherogenic index of plasma; TyG-AIP, triglyceride glucose-atherogenic index of plasma; CVDs, cardiovascular diseases; T1-T3, tertiles of TyG-AIP; Cluster1-3, K-means derived groups reflecting longitudinal TyG-AIP trajectories. Discussion Emerging evidence has highlighted the relevance of metabolic indicators in CVDs. In this cohort study examined the associations between TyG, AIP and the composite index TyG-AIP with incident CVDs in a long-term, community-based Chinese population. The findings demonstrated that individuals with elevated baseline levels of TyG, AIP, and TyG-AIP were significantly associated with an increased prevalence of CVDs. Sustained high levels of TyG-AIP further elevated the risk of CVDs, especially stroke. Compared with TyG or AIP alone, the combined index (TyG-AIP) demonstrated modest but consistently greater discriminatory ability for CVDs risk, both at baseline and when evaluated longitudinally. These findings indicate that routine assessment of TyG-AIP levels, along with their longitudinal variation, may provide additional value for identifying individuals at higher cardiovascular risk. TyG is a parameter derived from FBG and TG levels, which can reflect the IR level of the body [30]. In our study, TyG levels were positively associated with the incidence of CVDs, with higher TyG levels significantly increasing the risk of CVDs and stroke by 15% and 34%, respectively. The positive association between baseline TyG level and CVDs was validated in other studies [31–33]. A cohort analysis of 8,358 CHARLS participants reported that individuals in the high-TyG group exhibited a significantly elevated risk of cardiovascular events. Compared with those in the low-TyG group, the high-TyG category showed a 22% higher risk of CVDs and a 39% higher risk of stroke, findings that are broadly consistent with the associations observed in the present study [34]. AIP reflects the abnormal lipid metabolism and the risk of atherosclerosis [35]. Among the study outcomes, individuals classified in the high-AIP group contributed approximately 20% to 32% of all incident cases across the examined cardiovascular conditions. In a cohort study involving 7,708 participants, the HR for CVDs was 1.22 (95% CI: 1.08–1.39) across the tertiles of the baseline AIP, consistent with our findings [36]. After independently assessing TyG and AIP levels, our joint subgroup analysis indicated a notably elevated risk of CVDs (10%– 36%), especially stroke (16%–67%), among participants with concurrent elevations in both TyG and AIP, suggesting a possible synergistic effect. Similar findings have been reported in national cohort data, where individuals with simultaneously high TyG and AIP levels exhibited an increased risk of CVDs (HR = 1.23), closely matching the effect estimate observed in our study (HR = 1.22) [22]. Additional evidence also indicates that combined elevations of AIP and TyG are strongly associated with stroke risk [21]. However, the trend was not entirely consistent across all subgroups, as some subgroups showed a stronger association with CVDs than the subgroup with universally high levels. Such variability could stem from limited sample sizes or potential underlying interactive effects between TyG and AIP. Previous study typically examined associations between CVDs and four distinct combinations of TyG and AIP [22]. To further clarify the combined effects of these indices, we recalculated the TyG-AIP index, categorizing baseline TyG-AIP values into tertiles. The baseline analysis showed that the high TyG-AIP group had hazard ratios of 1.25 for CVDs, 1.52 for stroke, and 1.20 for heart disease, indicating that elevated TyG-AIP levels were associated with a higher risk of these cardiovascular outcomes. Additionally, the TyG-AIP index demonstrated superior predictive performance for all three outcomes compared with TyG and AIP alone, particularly for stroke. Furthermore, prior studies have shown that long-term elevated trajectories of TyG and AIP increased CVDs risk [37, 38]. Combined metabolic indices such as TyG-BMI and TyG-WHtR similarly demonstrated that dynamic changes in these indices correspond to higher CVDs risk over time [39, 40]. Given that TyG-AIP levels may fluctuate over time, cluster analysis was used to characterize the longitudinal changes in TyG-AIP and identified three distinct clusters: for Cluster1, individuals with low sustained TyG-AIP had the lowest risk of CVDs, for Cluster2, those with moderate sustained levels and moderate increases in TyG-AIP had an increased risk of CVDs, for Cluster3, those with the highest sustained levels and fastest increases in TyG-AIP had the highest CVDs risk. Specifically, long-term elevated TyG-AIP levels conferred a significantly higher CVD risk compared to stable low levels, with HRs of 1.25 for CVDs, 1.43 for stroke, and 1.25 for heart disease in Cluster3, emphasizing that persistently high TyG-AIP is a critical risk factor for CVDs, particularly stroke. These observations are consistent with the baseline findings and provide additional evidence that both baseline levels and longitudinal trajectories of TyG-AIP may be useful for identifying individuals at increased risk of cardiovascular outcomes. NRI and IDI analyses further supported the incremental predictive value of TyG-AIP. Compared with TyG or AIP alone, TyG-AIP significantly improved the reclassification of cardiovascular risk, with the greatest gains observed for stroke. These findings indicate that the combined index more effectively captures the metabolic–atherogenic burden relevant to cerebrovascular injury. Although the improvement for heart disease were more modest, TyG-AIP still provided measurable enhancement in prediction. Overall, these results underscore the added predictive utility of TyG-AIP beyond its individual components. The biological rationale for TyG-AIP as a composite indicator is supported by the close mechanistic interplay between IR and atherogenic dyslipidemia. IR (captured by TyG) weakens insulin-mediated inhibition of lipolysis, resulting in excessive release of free fatty acids (FFA) and increased hepatic glucose output [19, 41, 42]. The simultaneous elevation of circulating glucose and fatty acids promotes oxidative stress, stimulates the generation of reactive oxygen species, and triggers endothelial injury, all of which accelerate vascular inflammation and the early stages of atherosclerotic plaque formation [43–45]. Importantly, IR also operates upstream of AIP-related lipid abnormalities. Impaired insulin signaling increases hormone-sensitive lipase activity and augments hepatic VLDL-TG production, driving TG accumulation while simultaneously reducing HDL-C levels through enhanced cholesteryl ester transfer protein activity [46–48]. Thus, IR not only directly contributes to endothelial dysfunction but simultaneously drives the dyslipidemia pattern represented by elevated AIP. When these two metabolic disturbances coexist, their combined vascular impact may exceed that expected from either disturbance alone. Individuals with concomitantly high TyG and AIP are exposed to a metabolic milieu characterized by glucolipotoxicity, intensified oxidative stress, impaired nitric oxide bioavailability, and a heightened pro-inflammatory state [49–52]. The combined disturbances reflected by the TyG-AIP index may contribute to accelerated atherogenesis, increased plaque vulnerability, and microvascular dysfunction. This compounded metabolic burden could help explain why TyG-AIP demonstrated stronger associations with CVDs, particularly ischemic stroke, than either TyG or AIP alone in our study. Cerebral vessels may be especially susceptible to endothelial injury and inflammation-driven thrombosis, making stroke risk more sensitive to these coexisting abnormalities. Accordingly, the TyG-AIP index captures the integrated effects of two closely related yet distinct cardiometabolic pathways and provides a more comprehensive representation of vascular risk than its individual components. The study demonstrated that both baseline and longitudinal TyG-AIP were more strongly associated with stroke than with heart disease, which may reflect differences in the underlying pathophysiological mechanisms of these two conditions. The basic pathological cause of stroke is intravascular thrombosis [53], and the most common pathological types of stroke is ischemic stroke (80%), which is mainly caused by atherothrombosis [54]. Therefore, IR and dysregulated lipid metabolism likely promote stroke occurrence through pathways involving endothelial dysfunction, inflammation, and atherosclerotic progression [55]. Whereas heart disease is a general term for diseases that occur in the heart [56]. The etiology of heart disease is complex, including smoking, diet, obesity, psychological stress and other factors [57]. Meanwhile, heart disease mainly involves coronary vessels [58], while stroke mainly involves brain vessels [59]. Myocardial tissue around coronary arteries tolerates and responds differently to ischemia than brain tissue [60]. Collectively, these differences in pathophysiology, vascular territory, and tissue vulnerability offer a coherent explanation for the stronger association between TyG-AIP and stroke observed in this study. The study had several strengths include the use of the nationally representative CHARLS dataset and a prospective cohort design to enhance causal inference. The prospective cohort design provides temporality between exposure and outcome, thereby improving the robustness of causal inference. Second, the study is among the few to systematically evaluate both TyG and AIP simultaneously, and more importantly, to introduce the composite TyG-AIP index as a continuous cardiometabolic indicator. By examining this combined marker in relation to CVDs, stroke, and heart disease, the study provides new insights beyond prior research that relied primarily on binary classifications of TyG and AIP. Additionally, the study innovatively examines baseline levels of combined index TyG-AIP in relation to CVDs, stroke, and heart disease while also employing the K-means algorithm to track the temporal trajectories of TyG-AIP and assess the progression of CVDs across different levels. Fourth, the study incorporated multiple analytical strategies to comprehensively assess both associations and predictive improvements. This multimethod evaluation strengthens the credibility of the findings and demonstrates the incremental predictive value of TyG-AIP over its individual components. Finally, rigorous adjustments for demographic, lifestyle, and clinical covariates reduce the likelihood of residual confounding. Despite the strengths of this nationally representative cohort, several limitations should be acknowledged. First, the study population was restricted to middle-aged and older adults from the CHARLS cohort, and the sample size, although adequate for primary analyses, may still limit generalizability and reduce statistical power for detecting subtle associations. Second, only two waves of laboratory measurements (2011 and 2015) were available, which constrained our ability to fully characterize long-term dynamic changes in TyG-AIP and may have resulted in an incomplete depiction of its temporal evolution. Third, CVDs outcomes were based on self-reported physician diagnoses of stroke and heart disease, which introduces the potential for misclassification and recall bias. While this method aligns with prior CHARLS-based studies, the absence of adjudicated clinical events remains an inherent limitation. Furthermore, despite comparing included participants with those excluded because of missing data, some degree of selection bias may still exist and cannot be entirely ruled out. Additionally, although TyG-AIP demonstrated statistically higher discrimination than TyG or AIP alone, its absolute AUC values were modest, a common feature of single metabolic biomarkers predicting complex outcomes such as CVDs. Thus, TyG-AIP should be viewed as a simple, low-cost risk indicator that provides incremental improvement in population-level risk stratification. Finally, fasting insulin measurements were not available in CHARLS, precluding the calculation of HOMA-IR and limiting our ability to directly compare TyG-AIP with established IR indices. Future research incorporating larger cohorts, repeated biomarker assessments, adjudicated cardiovascular outcomes, and more comprehensive metabolic phenotyping is warranted to validate and expand upon these findings. Conclusions The cohort analysis demonstrated that both baseline TyG-AIP levels and longitudinal changes in this composite metabolic indicator were significantly associated with the subsequent development of CVDs, particularly stroke. Serial evaluation of TyG-AIP provides meaningful information on cardiometabolic deterioration and may enhance the early identification of individuals at elevated cardiovascular risk. Incorporating long-term monitoring of TyG-AIP into routine health assessments could therefore support more effective risk prediction and prevention strategies among middle-aged and older adults in China. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3 Supplementary Material 4 Supplementary Material 4