Development of a biomarker-enhanced arterial age model for young-to-middle-aged adults with type 2 diabetes

Abstract

Direct vascular aging assessment is not always feasible in routine diabetes care. We aimed to derive a control-based arterial age reference using estimated pulse wave velocity, develop a biomarker-enhanced model for predicting arterial age gap, and evaluate its ability to identify accelerated arterial aging in young-to-middle-aged adults with type 2 diabetes mellitus. This study included 300 participants (150 T2DM, 150 age- and sex-matched controls). Arterial age was derived from a control-group ePWV-age regression. Age gap was defined as estimated arterial age minus chronological age. Candidate predictors were evaluated using multivariable linear regression, and the final model was internally validated by 10-fold cross-validation. Compared with controls, participants with T2DM had higher ePWV, older estimated arterial age, larger age gap, lower adropin, and higher oxLDL (all p < 0.001). Accelerated arterial aging was more frequent in T2DM than controls (76.0% vs. 20.0%). The final model integrating HbA1c, adropin, and oxLDL explained 42% of the variance in age gap (adjusted R²=0.418), showed good discrimination for accelerated arterial aging (AUC 0.889; 95% CI 0.828–0.910), and retained acceptable internal calibration (slope 0.943). A biomarker-enhanced arterial age model integrating HbA1c, adropin, and oxLDL provided an interpretable framework for identifying accelerated arterial aging in young-to-middle-aged adults with T2DM. Although promising for translational implementation, external validation and direct pulse wave velocity benchmarking are required before clinical application. The model also enabled a three-level arterial aging classification and web-based implementation for research use, supporting its potential as a practical risk-communication tool prototype.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-55753-w.

Introduction

Type 2 diabetes mellitus (T2DM) is a significant contributor to vascular dysfunction and the acceleration of vascular aging on a global scale. The prevalence of diabetes has been reported as an estimate of 589 million individuals aged between 20 and 79 in 2024. And this number is expected to reach nearly 853 million by 2050, which underscores the risk of cardiometabolic diseases in middle age and later in life¹. T2DM is characterized by chronic metabolic stress, endothelial dysfunction, inflammation, and oxidative injury, all of which contribute to vascular complications^2–5. With this given background, early vascular aging has thus emerged as a clinically significant feature in T2DM. Among several phenotypes, arterial stiffness serves as a key manifestation of vascular aging and a possible intermediate phenotype with prognostic value. In diabetes populations, higher PWV is associated with increased risk of future cardiovascular events and mortality, supporting its usefulness for upstream cardiovascular risk stratification^6,7.

The concept of vascular age is useful in clinical settings because it turns vascular injury into an age-equivalent concept that has been linked to future cardiovascular risk in prospective cohort settings^8,9. Yet, direct tools for arterial stiffness assessment are not widely available in routine care due to the need for specialised devices, operator training, and cross-device standardization. Also, recent validation recommendations and guidelines discussions highlight these practical barriers¹⁰. Estimated PWV (ePWV), calculated from age and blood pressure, provides a more accessible surrogate of arterial stiffness and has been associated with vascular aging indicators, cardiovascular events, and mortality risk^11–14. However, ePWV is an estimate rather than a stiffness measurement, and biomarker-enhanced arterial-age tools that capture diabetes-specific vascular biology are not yet available. Vascular aging roadmaps also note uncertainty about which biomarkers to prioritise for scalable clinical translation⁹.

To move beyond age- and blood pressure-derived estimates and better reflect diabetes-related vascular biology, we focused on three mechanistically complementary axes. Those are chronic glycemic burden (HbA1c), vasoprotective/endothelial signaling (adropin), and oxidative vascular injury (oxLDL). HbA1c was selected as a clinically established marker of chronic glycemic exposure, which is linked to arterial stiffness progression and cardiovascular risk in T2DM^15–18. Meanwhile, lower circulating adropin in T2DM is associated with increased arterial stiffness, and higher adropin levels are associated with lower carotid atherosclerosis burden, supporting its use as an endothelial-vasoprotective signal marker^19–21. Following this, oxLDL, a marker of oxidative stress, has also been linked to vascular dysfunction. Several experimental and clinical studies have reported that oxLDL levels are markedly elevated in individuals with macrovascular complications. Furthermore, it has been proposed that the oxLDL/LDL-C ratio can be used to detect severe coronary atherosclerosis in T2DM^22,23. Collectively, these biomarkers were selected to capture complementary components of vascular aging biology in T2DM within a clinically interpretable framework. Therefore, in this study, we aim to derive a control-based arterial age reference using ePWV, construct a biomarker-enhanced model for predicting arterial age gap, and assess its potential to detect accelerated arterial aging in young-to-middle-aged individuals with type 2 diabetes. We hypothesized that a combined model that includes HbA1c, adropin, and oxLDL will outperform single-domain models and give a clinically interpretable arterial age framework.

Materials and methods

Study design and participant selection

This is an analytical cross-sectional study with a case-control design conducted in SRM Medical College Hospital and Research Centre, which comprises a total of 300 participants, with 150 T2DM participants and 150 age and sex-matched controls. This study is primarily designed to develop, internally validate, and translationally implement a biomarker-enhanced arterial age prediction model in young-to-middle-aged individuals with T2DM. The study protocol was approved by the institutional ethics committee (Approval no. ECR/8856/INST/TN/2013/RR-19) and strictly adhered to both the institution and the Declaration of Helsinki standards. Also, all participants were comprehensively informed about the research methodologies employed in this study, and informed consent was obtained before their recruitment.

Participants were recruited from both the diabetology OPD and the master health checkup during the study period of June 2024 – October 2025.

Inclusion criteria

Participants were eligible for inclusion if they met the following criteria:

1. Age between 25 and 55 years at the time of enrolment.

2. Should undergo a screening process which includes medical history, anthropometric assessment, blood pressure measurement, and relevant laboratory and clinical investigations.

3. For cases, participants with a confirmed diagnosis of type 2 diabetes mellitus with a disease duration of at least 2 years.

4. For the control group, participants were apparently healthy adults without diabetes, defined by HbA1c < 5.7% and fasting plasma glucose < 100 mg/dL.

5. Participants with no history of chronic medical illness, no recent acute illness, no prior cardiovascular disease, and not receiving long-term medication were included as controls.

Exclusion criteria

Participants were excluded if they had a history of cardiovascular disease, autoimmune disease, or chronic kidney disease. Participants were also excluded from the study if they were pregnant, had been hospitalised, had a fever, or had surgery within 45 days before the study began.

Clinical and anthropometric assessment

Demographic data, such as age and sex, were documented using a data collection form. Height and weight were assessed, and BMI was calculated. Supplementary clinical information, including blood pressure, duration of diabetes, current medications, smoking status, and alcohol use, was also recorded at enrollment.

Blood sample collection and biochemical measurements

4 mL of fasting venous blood samples were collected under aseptic conditions. Samples were centrifuged and used for biochemical investigations. Fasting Plasma Glucose and lipid profiles were analyzed using the Beckman Coulter DXC 700 automated analyser. HbA1c was analyzed using BioRad D10. Circulating adropin and oxidised LDL were measured using commercially available ELISA kits from Origin Diagnostics & Research. The manufacturer mentioned analytical sensitivities and detection ranges for adropin 14.2 pg/mL (31.25–2000 pg/mL), and OxLDL ng/mL 0.51 ng/mL (1.57 ng/mL–100 ng/mL). As per kit instructions, all assays exhibited intra-assay CVs < 8% and inter-assay CVs < 10%. The average optical-density measurements from duplicate samples were used to calculate the concentrations. Each test run contained standard and blank wells per the manufacturer’s specifications. ELISA was carefully performed according to the manufacturer’s instructions.

Statistical analysis

All statistical analyses were performed using IBM SPSS 26. Normality of the data was assessed using the Kolmogorov-Smirnov test. Continuous variables were presented either as mean ± standard deviation or median with interquartile range, depending on their distribution. Categorical variables were reported as counts and percentages. The independent-samples t test or Mann-Whitney U test for continuous variables and the chi-square test for categorical variables were used for between-group comparisons. Spearman’s rank correlation coefficient was used for correlation analyses. Both univariate and multivariate linear regression analyses were carried out. Multicollinearity was examined using tolerance and variance inflation factor (VIF), with low VIF values supporting model stability. All tests were two-tailed, and P < 0.05 was considered significant.

Blood pressure measurement and ePWV calculation

As direct carotid-femoral pulse wave velocity was not available in this cohort, we used estimated pulse wave velocity (ePWV) as the surrogate marker for the vascular aging variable. Instead of replacing carotid-femoral pulse wave velocity, ePWV was used as a feasible internal vascular-aging surrogate to set an arterial age framework for the study population. We used a validated automatic digital sphygmomanometer (Omron HEM-7361T, Omron Healthcare Co., Ltd., Japan) with an atrial fibrillation (AFib) indicator to assess blood pressure. After a five-minute rest, all measures were taken while seated. Three continuous measurements were taken at 30-second intervals using the device’s automated three-reading approach. Each participant’s brachial blood pressure was determined by averaging these three values. Following this, mean blood pressure (MBP) was calculated as DBP + 0.4 (SBP-DBP). ePWV was calculated from MBP and age using the following equations.

1. ePWV = 9.58748315543126 − 0.402467539733184*Age + 4.56020798207263 × 10^{− 3} *Age² − 2.6207705511664 × 10^{− 5}*Age²*MBP + 3.1762450559276 × 10^{− 3}*Age*MBP − 1.83215068503821 × 10^{− 2}*MBP.

2. ePWV = 4.62–0.13*age + 0.0018*age² + 0.0006*age*MBP + 0.0284*MBP¹².

Participants with cardiovascular risk factors use Eq. 1, while participants without cardiovascular risk factors use Eq. 2. Here, individuals who were non-smokers without metabolic syndrome components or a history of myocardial infarction or stroke were considered participants without cardiovascular risk factors.

Derivation of the control reference equation

A regression model associating ePWV to chronological age was developed in the control group alone in order to provide an internal age-referenced vascular standard. The reference equation for vascular aging was obtained from this control model. Alternative formulations were tested to observe whether adding sex or a quadratic age factor improved model fit. Candidate models were examined for explanatory performance, residual behaviour, collinearity diagnostics, and interpretability. The simplest model that fit was the final reference equation. The estimated arterial age (EAA) for each participant was then calculated from the observed ePWV value by algebraically inverting the retained control equation.

Definition of accelerated arterial aging

A physiologically anchored threshold for accelerated arterial aging was established using the Age Gap distribution in the control group. The control Age Gap distribution was used to calculate percentile-based cut-offs. Participants were divided into non-accelerated and accelerated arterial aging status based on a primary threshold chosen from this reference distribution. When necessary, additional percentile criteria were kept for sensitivity or exploratory analysis.

Candidate predictors for model development

Because estimated arterial age integrates chronological age by construction and may therefore be disproportionately influenced by age, Age Gap was the key modelling outcome rather than estimated arterial age itself. Demographic, anthropometric, glycemic, lipid, and biomarker variables that were accessible to all participants comprised the candidate predictor set. These parameters encompassed age, sex, BMI, HbA1c, lipid profile parameters, adropin, and oxLDL. As many continuous variables showed skewed distribution, non-parametric correlation analysis was used for preliminary screening of associations with Age Gap. Predictor selection was based on a combination of.

1. strength and direction of association with Age Gap,

2. biological plausibility,

3. non-redundancy among related lipid or biomarker variables, and.

4. translational suitability for implementation in a calculator.

A small set of possible multivariable linear regression models was used to create clinical, biomarker-only, and combination clinical-biomarker models. Explained variance, standard error of estimate, predictor significance, multicollinearity diagnostics, and practical usability were used to compare the models.

Multivariable linear regression with the enter method was used to model arterial age gap as a continuous outcome. Selected clinical, biochemical, lipid, and biomarker variables found through preliminary association analysis and biological relevance were used to build candidate models. Model selection was guided by several factors: explanatory performance, the statistical significance of predictors, multicollinearity diagnostics, biological interpretability, and translational feasibility. The ultimate model was formulated as follows:

where denotes the intercept, X₁to X_n denote the predictors retained in the final model, and β₁to β_n represent their corresponding regression coefficients.

Predicted arterial age was then calculated using the following formula:

Predicted Arterial Age = Chronological Age + Predicted Age Gap.

Discrimination analysis and classification framework

To evaluate the continuous prediction model’s ability to identify accelerated arterial aging, we used receiver operating characteristic curve analysis. The binary arterial aging status, determined by the control threshold, served as the reference outcome in this analysis. The optimal operational cut-off for predicted AgeGap was selected from the ROC coordinate table. To facilitate interpretation and translational deployment, a three-level arterial aging framework was subsequently defined by combining:

1. the biologically derived control-based threshold, and.

2. the ROC-derived operational cut-off.

Based on the predicted Age Gap, participants were classified as normal, moderate, or accelerated arterial aging.

Internal validation

The final prediction model was internally validated using 10-fold cross-validation. The entire dataset was randomly divided into ten roughly equal folds. In order to produce out-of-fold predictions, the model was trained in nine folds and then applied to the corresponding held-out fold in each iteration. The method was repeated until all participants received out-of-fold predicted Age Gap values. The out-of-fold predictions were used to assess:

• mean prediction error,

• mean absolute error (MAE),

• mean squared error (MSE),

• root mean squared error (RMSE),

• cross-validated explained variance, and.

• calibration.

We tested the calibration by plotting the actual Age Gap against the cross-validated predicted Age Gap. Systematic bias and possible overfitting were evaluated using calibration intercept and slope. An intercept around 0 and a slope near 1 indicated good calibration. ROC analysis was conducted using cross-validated projected Age Gap values to determine internally validated discrimination for accelerated arterial aging. The cross-validated predictions were then given a set operational cut-off to find the overall classification accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and positive predictive value.

Web calculator development and deployment

The final regression model was turned into a web-based calculator to make it easier to use. This calculator was designed to take the user’s age and the final predictors from the final model into account. As a result, it provides the predicted Age Gap, the expected arterial age, and the arterial aging category. To maintain model output consistency with laboratory values, the implementation kept the derivation dataset analytical units.

The calculator was developed as a client-side web application using React and deployed publicly using Cloudflare. Unit guidelines for each biomarker input were included in the public interface, which was designated for research usage only. Functional verification was carried out by comparing calculator results to the appropriate regression-based outputs produced by SPSS for representative tests. The complete reporting framework of this study has been presented as a flowchart in Fig. 1.

Fig. 1.

Framework for derivation, validation, and deployment of arterial age prediction model.

Results

Participant characteristics and derived vascular aging measures

A total of 300 participants were included in the final analysis, with an overall mean chronological age of 47.77 ± 5.52 years. Table 1 shows that participants with T2DM had significantly higher fasting plasma glucose (p < 0.001) and HbA1c (p < 0.001) levels than controls. Participants with T2DM had slightly elevated blood pressure indices compared to controls. They showed an adverse lipid profile, characterized by higher total cholesterol, triglycerides, LDL-C, TC/HDL-C ratio, LDL-C/HDL-C ratio, and TG/HDL-C ratio, together with lower HDL-C. ePWV was also observed to be significantly higher in participants with T2DM than in controls (8.02 vs. 7.21 m/s, p < 0.001), indicating increased arterial stiffness in the diabetic group. Adropin levels were significantly lower in T2DM than in controls (245.27 vs. 834.24 pg/mL, p < 0.001), whereas oxLDL concentrations were markedly higher in T2DM (p < 0.001).

Table 1.

Baseline biochemical, biomarker, and vascular characteristics of controls and participants with T2DM.

Parameter	Overall (n = 300)	Control (n = 150)	T2DM (n = 150)	p value
Chronological age, years	47.77 ± 5.52	47.4 ± 5.26	48.08 ± 5.76	0.135
Male, n (%)	152 (50.6)	79 (52.6)	73 (48.6)	0.488
FPG, mg/dL	100 (90,132)	91 (86, 95)	132.5 (110, 176)	< 0.001
SBP, mmHg	120 (115,126)	116.5 (112,120)	126 (120,132)	< 0.001
DBP, mmHg	76 (73,80)	74 (71,76)	80 (76,86)	< 0.001
MBP, mmHg	91 (87,95)	88 (85,90)	95 (91,102)	< 0.001
HbA1c, %	5.8 (5.2,7.4)	5.2 (5.0, 5.4)	7.4 (6.6, 9.3)	< 0.001
Total cholesterol, mg/dL	170 (145,194)	154 (130, 181.5)	180 (161.75, 216)	< 0.001
Triglycerides, mg/dL	110 (83.25, 148.5)	97 (71, 123.75)	120.5 (95, 163.5)	< 0.001
HDL-C, mg/dL	45 (39.25, 51)	46 (41, 52)	43.5 (38, 49)	0.021
LDL-C, mg/dL	118 (100, 137)	112.5 (91, 128.5)	124 (107, 148.5)	< 0.001
TC/HDL-C	3.87 ± 0.95	3.56 ± 0.82	4.19 ± 0.98	< 0.001
LDL-C/HDL-C	2.71 ± 0.80	2.49 ± 0.73	2.93 ± 0.80	< 0.001
TG/HDL-C	2.85 ± 1.60	2.4 ± 1.03	3.29 ± 1.7	< 0.001
ePWV, m/s	7.52 (7.04,8.1)	7.21 (6.86, 7.70)	8.02 (7.39, 8.49)	< 0.001
Adropin, pg/mL	500 (222, 946)	834.24 (653.4, 1884.9)	245.27 (119.85, 528.74)	< 0.001
oxLDL, ng/mL	64.6 (41.1, 102)	41.87 (29.61, 53.1)	102.5 (84, 130.16)	< 0.001
Estimated arterial age, years	51.87 ± 8.84	47.67 ± 5.98	56.09 ± 9.26	< 0.001
Age gap, years	2.5 (-0.35,6.4)	0.30 (-1.99, 2.38)	6.4 (3.08, 13.46)	< 0.001
Accelerated arterial aging, n (%)	144 (48)	30 (20.0)	114 (76.0)	< 0.001

Independent t- test (normally distributed) and Mann-Whitney U test (Not normally distributed) have been performed for continuous variables, and Chi-square for categorical variables. Data presented in Median (IQR = 25th Quartile, 75th Quartile); p value < 0.05 is considered significant.

Derivation of the control ePWV-age reference equation

To establish an internal reference framework for vascular aging, the relationship between chronological age and ePWV was first examined in the control group. In controls, ePWV showed a strong positive correlation with age (r = 0.884, p < 0.001), supporting the use of age-based regression modelling to derive a reference equation for arterial aging.

The primary age-only linear regression model demonstrated strong explanatory performance (R² = 0.781, adjusted R² = 0.780, p < 0.001) and yielded the following reference equation:

This model was retained as the final control reference equation. Alternative control-group models were also explored to determine whether the inclusion of sex or a quadratic age term improved model fit. Adding sex produced only a marginal improvement in explanatory power (R² = 0.785, adjusted R² = 0.783), and sex was not independently associated with ePWV (B = -0.070, p = 0.087). We also examined a centred quadratic model including age and age², which resulted in only a small increase in fit (R² = 0.788, adjusted R² = 0.785). Comparative performance of the alternative control-group models is presented in Supplementary Table S1. The final control-derived equation was then algebraically inverted to estimate arterial age for each participant from the observed ePWV value. The relationship between chronological age and ePWV in the control group is illustrated in Fig. 2.

Fig. 2.

Relationship between chronological age and ePWV in controls.

Distribution of estimated arterial age and arterial age gap and definition of accelerated arterial aging

Using the final control-derived ePWV-age reference equation, estimated arterial age (EAA) was calculated for each participant by algebraic inversion and applied to each participant’s observed ePWV value to estimate arterial age.

Arterial age gap (Age gap) was then calculated as:

In the control group, the age gap was close to zero, indicating good concordance between vascular age and chronological age in the non-diabetic reference population. The median control-group Age gap was 0.3093 years, supporting the biological appropriateness of the derived reference equation. To define abnormal vascular aging within the study cohort, percentile thresholds were derived from the control-group Age gap distribution. The corresponding percentiles for the 25th, 50th, 75th, 80th, and 90th were − 1.9937, 0.3093, 2.3830, 2.8903, and 3.9706 years. From this, the 80th percentile (2.8903 years) was selected as the primary threshold for accelerated arterial aging. Using the equation, the mean estimated arterial age for the entire study population was 51.87 ± 8.84 years, while the arterial age gap ranged from − 0.35 to 6.4 years. When stratified by study group, the mean estimated arterial age was higher in T2DM than in controls (56.09 ± 9.26 vs. 47.67 ± 5.98 years), and the mean Age gap was markedly greater in T2DM (0.30 vs. 6.4 years). Similarly, the frequency of accelerated arterial ageing was higher among T2DM participants than among controls (76.0% vs. 20.0%, p < 0.001). The accelerated arterial aging threshold was observed by 30 out of 150 individuals (20.0%) in the control group and by 114 out of 150 participants (76.0%) in the T2DM patients. In contrast, 120 of 150 controls (80.0%) and 36 of 150 T2DM individuals (24.0%) were normal [Table 1].

Correlation analyses of arterial age gap with predictors

Table 2 shows that the age gap had no significant association with chronological age (ρ = 0.088, p = 0.130) or sex (ρ = -0.060, p = 0.300). But significant positive correlations were observed for BMI (ρ = 0.244, p < 0.001), HbA1c (ρ = 0.671, p < 0.001), total cholesterol (ρ = 0.304, p < 0.001), triglycerides (ρ = 0.284, p < 0.001), and LDL (ρ = 0.222, p < 0.001). Following this, HDL also showed a weak inverse association (ρ = -0.127, p = 0.028). Among the measured vascular biomarkers, adropin showed an inverse correlation with Age gap (ρ = -0.547, p < 0.001), indicating that lower adropin levels were associated with increased arterial age. Conversely, oxLDL (ρ = 0.580, p < 0.001) showed positive correlations with Age gap.

Table 2.

Spearman correlations between arterial age gap and predictors.

Parameter	Spearman ρ	p value
Age	0.088	0.130
Sex	-0.060	0.300
BMI	0.244	< 0.001
HbA1c	0.671	< 0.001
Total cholesterol	0.304	< 0.001
Triglycerides	0.284	< 0.001
HDL-C	-0.127	0.028
LDL-C	0.222	< 0.001
Adropin	-0.547	< 0.001
oxLDL	0.580	< 0.001

Spearman correlation has been done. p-value < 0.05 is considered significant.

Model development and predictor selection

To identify the most informative predictors of the age gap, we evaluated several multivariable linear regression models. We compared all models using overall explanatory performance, adjusted R2, the standard error of the estimate, the statistical significance of individual predictors, and multicollinearity diagnostics.

The most inclusive model incorporated BMI, HbA1c, LDL-C, adropin, and oxLDL. This model explained 45.4% of the variance in Age gap (adjusted R2 = 0.443; SEE = 4.965; F = 40.636, p < 0.001). However, not all included predictors contributed independently to model performance. In particular, BMI (p = 0.064) and LDL-C (p = 0.604) were not statistically significant in the fully adjusted model. Another model including HbA1c, adropin, and oxLDL retained similar explanatory performance (R2 = 0.447, adjusted R2 = 0.440; SEE = 4.979; F = 59.707, p < 0.001), with all four predictors remaining statistically significant. A biomarker-only model containing adropin, and oxLDL was also examined. Although this model remained statistically significant overall (R2 = 0.349, adjusted R2 = 0.343; SEE = 5.393; F = 52.967, p < 0.001), its explanatory performance was lower than that of the combined models.

To enhance clinical usability and facilitate downstream web-based implementation, a three-predictor model incorporating HbA1c, adropin, and oxLDL was then evaluated as a translational model. The model reported 42% of the variance in Age gap (adjusted R2 = 0.418; SEE = 5.075; F = 72.562, p < 0.001), and all predictors remained statistically significant [Table 3]. Furthermore, multicollinearity was minimal, as indicated by VIF values between 1.353 and 1.857. Therefore, this model was chosen for the final prediction equation, to perform discrimination analysis, conduct internal validation, and implement the web calculator. Detailed model coefficient outputs are provided in Supplementary Tables S2–S5.

Table 3.

Sequential development and selection of models for the prediction of arterial age gap.

Model	Variables entered	Significant retained predictors	Adjusted R²	SEE	Model decision
Model 1	HbA1c	HbA1c	0.361	5.319	Baseline clinical model
Model 2	Adropin, oxLDL	Adropin, oxLDL	0.343	5.393	Biomarker-only model
Model 3	BMI, HbA1c, LDL-C, Adropin, oxLDL	HbA1c, Adropin, oxLDL	0.443	4.965	Highest apparent fit; BMI and LDL-C not independently retained
Final model	HbA1c, Adropin, oxLDL	HbA1c, Adropin, oxLDL	0.418	5.075	Selected for final equation, validation, and web implementation

In all models, confounding factors like duration of diabetes, current medications, smoking status, and alcohol intake were adjusted.

The final regression equation for the prediction of the age gap was:

Predicted arterial age was subsequently derived by adding the predicted Age gap to chronological age.

Agreement between observed and predicted vascular aging metrics

Table 4 shows that the predicted arterial age gap (Predicted Age gap) exhibited a significant positive correlation with the observed Age gap (r = 0.651, p < 0.001). Predicted Age gap also correlated significantly with estimated arterial age (r = 0.529, p < 0.001). Following this, the predicted arterial age showed a strong positive association with EAA (r = 0.821, p < 0.001). A moderate but still significant association was observed between predicted arterial age and observed Age gap (r = 0.430, p < 0.001). Taken together, these results indicate that the final HbA1c-adropin-oxLDL model showed good concordance with the internally derived vascular aging framework.

Table 4.

Correlations between observed and predicted vascular aging metrics.

Variables	Correlation coefficient (ρ)	p value
Observed Age Gap vs. Predicted Age Gap	0.651	< 0.001
Estimated arterial age (EAA) vs. Predicted Age gap	0.529	< 0.001
Estimated arterial age (EAA) vs. Predicted arterial age	0.821	< 0.001
Observed Age Gap vs. Predicted Arterial Age	0.430	< 0.001

Spearman correlation has been done. p-value < 0.05 is considered significant.

ROC analysis and classification performance of the final model

The predicted arterial age gap (Predicted Age Gap) resulted in an AUC of 0.89 (95% CI 0.83–0.91, p < 0.001). The optimal operational cut-off for Predicted Age gap, identified from the ROC coordinate table, was 4.469 years. This threshold was subsequently used to generate a binary classification variable using a model for accelerated arterial aging. At this cut-off, 107 of 144 participants classified as accelerated arterial aging by the reference definition were correctly identified by the model, yielding a sensitivity of 74.31%. Among the 156 participants below the arterial aging threshold, 154 were correctly classified as non-accelerated, yielding a specificity of 98.72%. Overall, 261 of 300 participants were correctly classified, giving an accuracy of 87%. The positive predictive value was 98.17%, and the negative predictive value was 80.63% [Table 5, Supplementary Fig. 1].

Table 5.

ROC-derived discrimination and classification performance of the final model.

Parameter	Value
AUC	0.89
95% CI for AUC	0.83–0.91
p value	< 0.001
Optimal cut-off for Predicted Age gap, years	4.47
Sensitivity, %	74.31
Specificity, %	98.72
Accuracy, %	87
Positive predictive value, %	98.17
Negative predictive value, %	80.63

Three-level arterial aging classification

Predicted Age gap was classified into a three-level arterial aging framework utilising the biologically derived control-based threshold and the ROC-derived operational cut-off to enhance interpretability and translational applicability. Participants were classified as having normal arterial aging if the predicted Age gap was < 2.8903, intermediate vascular aging risk if the predicted Age gap was ≥ 2.8903 and < 4.469, and accelerated arterial aging if the predicted Age gap was ≥ 4.469.

Using this approach, 156 of 300 individuals (52.0%) had normal arterial aging, 35 (11.7%) intermediate vascular aging risk, and 109 (36.3%) accelerated arterial aging. In the control group, 139 of 150 people (92.7%) had normal arterial aging, 9 (6.0%) intermediate, and 2 (1.3%) accelerated. In comparison, 17 of 150 (11.3%) T2DM subjects had normal arterial aging, 26 (17.3%) intermediate-risk, and 107 (71.3%) accelerated arterial aging [Figure 3, Supplementary Table S6].

Fig. 3.

Distribution of normal arterial aging, intermediate vascular aging risk, and accelerated arterial aging categories in controls and participants with T2DM.

Internal validation of the final model

The internal stability of the final HbA1c-adropin-oxLDL model was evaluated by 10-fold cross-validation, wherein fold-specific models were constructed using 90% of the dataset and subsequently applied to the corresponding held-out 10% to produce out-of-fold predictions for all individuals. Cross-validated prediction error was low. Table 6 shows that the mean prediction error was 0.0728 years, with an absolute error of 3.8221 years and a squared error of 26.7545. These results suggest that there was little bias in the predictions. Moreover, the model’s cross-validated root mean squared error (RMSE) was 5.17 years, indicating a high level of predictive accuracy after internal validation.

Table 6.

Internal validation metrics of the final HbA1c–adropin–oxLDL model.

Parameter	Value
Mean prediction error	0.073
Mean absolute error (MAE), years	3.822
Mean squared error (MSE)	26.754
RMSE, years	5.17
Calibration intercept	0.304
95% CI for intercept	−0.493 to 1.101
Calibration slope	0.943
95% CI for slope	0.810 to 1.076
Cross-validated R²	0.395
Cross-validated adjusted R²	0.393
Cross-validated SEE	5.183

The model’s calibration remained acceptable. Upon regressing observed Age gap against cross-validated projected Age gap, the calibration intercept was 0.304 (95% CI − 0.493 to 1.101, p = 0.453), demonstrating the absence of any systematic over- or under-prediction. The calibration slope, measured at 0.943 (95% CI 0.810 to 1.076, p < 0.001), suggested a degree of restricted optimism and no indication of overfitting. Furthermore, the cross-validated regression demonstrated consistent model performance during internal validation, as evidenced by an R² of 0.395 (adjusted R² = 0.393) and a standard error of estimate of 5.183 [Table 6, Supplementary Fig. 2].

Web calculator implementation

To enable the practical use of the final model, the HbA1c-adropin-oxLDL equation was developed into a web-based arterial age calculator. The calculator takes chronological age, HbA1c, adropin, and oxLDL as inputs and returns Predicted Age gap, arterial age, and arterial aging category based on preset categorisation thresholds. The online tool implemented the final prediction framework in a format appropriate for real-time application and was designed to maintain the analytical units utilised in the derivation dataset, namely HbA1c in %, adropin in pg/mL, and oxLDL in ng/mL. The classification output was based on the predefined three-level framework. The calculator was publicly deployed at https://vascularage.in and labelled for research use only. Screenshot of the publicly accessible calculator is shown in Fig. 4.

Fig. 4.

Screenshot of the publicly accessible biomarker-enhanced arterial age calculator.

Discussion

In our study, we developed a biomarker-enhanced arterial age framework that was anchored to a control-derived ePWV reference. We found that individuals with T2DM had higher ePWV, older estimated arterial age, a much larger arterial age gap, and a significantly greater incidence of accelerated arterial aging than age and sex matched controls. The final model integrating HbA1c, adropin, and oxLDL explained a significant portion of the arterial age gap variance, distinguished accelerated arterial aging effectively, and supports a clinically interpretable three-level classification of normal arterial aging, intermediate vascular aging risk, and accelerated arterial aging. These findings are consistent with recent evidence that arterial stiffness is increased across the dysglycemic range and that higher PWV predicts adverse cardiovascular outcomes in diabetes^7,24.

The clinical anchor of the final model was HbA1c, which exhibited the most significant positive correlation with the arterial age gap. This association is both biologically plausible and clinically coherent. Chronic hyperglycemia causes endothelial dysfunction and arterial wall remodeling through non-enzymatic glycation of vascular wall proteins, advanced glycation end-product accumulation, oxidative stress, inflammatory activation, and impaired nitric oxide bioavailability^4,5. A recent longitudinal study by Fang et al. reported that chronic glycemic burden is relevant to both prevalent vascular injury and worsening vascular stiffness over time, further supporting our findings¹⁶. Thus, HbA1c appears to be a practical marker of cumulative metabolic injury that anchors the arterial age model to a clinically and mechanistically meaningful diabetes-related exposure.

Following this, both adropin and oxLDL strengthen the model as they capture biological domains that are not adequately represented by glycemia alone. Mechanistically, the inverse relationship between adropin and arterial age gap may be due to endothelial nitric oxide signaling and vascular redox homeostasis deterioration. Adropin is a vasoprotective peptide that enhances endothelial nitric oxide synthase activity and supports endothelium-dependent vasodilation through VEGFR2-, PI3K/Akt-, and ERK1/2-related signaling pathways^25,26. A recent study by Jurrissen et al. further revealed that reduced circulating adropin is accompanied by greater arterial stiffening, supporting the view that hypoadropinemia is linked to impaired vascular compliance¹⁹. More recent mechanistic data also suggest that adropin can attenuate oxidative stress through Nrf2/HO-1 signaling²⁷. In this context, lower adropin in T2DM may therefore promote arterial stiffening by decreasing nitric oxide bioavailability, antioxidant defense, endothelial dysfunction, and vascular wall remodeling. Following this, the positive association between oxLDL and arterial age gap is physiologically consistent with LOX-1-mediated endothelial damage. OxLDL may increase reactive oxygen species formation, suppress endothelial nitric oxide signaling, activate endothelial cells, and alter vascular reactivity, generating a pro-stiffening environment^28,29. Recent research suggests convergence on common pathways rather than a direct adropin–oxLDL axis. Thus, lower adropin reduces vasoprotective and antioxidant signaling, whereas increased oxLDL enhances LOX-1-driven oxidative and inflammatory damage. Our model may capture both endothelial protection loss and oxidative vascular damage within the same arterial-aging process, making their inclusion physiologically consistent.

This framework has significant translational effects. The high specificity at the cut-off shows that the model may be best for identifying those at high risk of accelerated arterial aging rather than as a universal screening tool. In addition, the model is easier to interpret clinically than isolated biomarker readings because it converts the continuous predicted age gap into a three-level arterial-aging categorization and predicted arterial age output. This is important because vascular-age prediction models may assist doctors and patients in clearly expressing risk when direct vascular testing is unavailable. However, the current data do not demonstrate clinical usefulness, superiority over existing cardiovascular risk assessments, or value for regular care decision-making. The model was developed and internally tested within a single cohort, therefore its effectiveness requires validation in multiple populations before stronger translational claims can be made. This study has several strengths, including the use of a control-derived reference framework, the integration of glycemic, vasoprotective, and oxidative biology, the internal comparison of candidate model types, and the intuitive predicted arterial age output that allows for clinically interpretable categorization.

However, there are several limitations in our study. Initially, causal inference is not feasible due to the cross-sectional design. Secondly, the vascular-aging phenotype was defined using ePWV rather than the reference standard for arterial stiffness, as cfPWV was not directly measured. Although participants with known chronic kidney disease were excluded, quantitative renal function parameters such as estimated glomerular filtration rate and urinary albumin-to-creatinine ratio were not available for all participants and therefore could not be incorporated into the model. The generalizability of ethnic and geographic variables is restricted by the single-center design. Lastly, the model’s performance is partially influenced by the prediction of an internally defined phenotype rather than validation against a gold-standard vascular endpoint. Future research should thus emphasize external validation in more heterogeneous cohorts, prospective outcome validation, and direct comparison with instrument-based pulse wave velocity. Following this systematic testing of the web application within actual clinical workflows, and assessment of whether additional biomarkers enhance performance without compromising robustness and usability. Collectively, our data support the translational potential of a biomarker-enhanced arterial age framework for assessing accelerated arterial aging in young-to-middle-aged individuals with type 2 diabetes.

Conclusion

By integrating a control-derived ePWV reference with HbA1c, adropin, and oxLDL, this study offers a practical and physiologically informed arterial-aging model in T2DM. Internally, the model performed well and allowed a simple three-level arterial aging categorization. Despite its potential as a translational research and risk-communication tool, external validation, direct PWV benchmarking, and prospective outcome-based evaluation will determine its utility. (The calculator was publicly deployed at https://vascularage.in.)

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AGEs

Advanced glycation end products

BMI

Body mass index

BP

Blood pressure

cfPWV

Carotid–femoral pulse wave velocity

CV

Coefficient of variation

CVD

Cardiovascular disease

DBP

Diastolic blood pressure

eNOS

Endothelial nitric oxide synthase

ePWV

Estimated pulse wave velocity

ERK1/2

Extracellular signal–regulated kinase 1/2

EAA

Estimated arterial age

FPG

Fasting plasma glucose

HDL

C–High–density lipoprotein cholesterol

HbA1c

Glycated hemoglobin

HO

1–Heme oxygenase–1

IQR

Interquartile range

LDL

C–Low–density lipoprotein cholesterol

LOX

1–Lectin–like oxidized low–density lipoprotein receptor–1

MAE

Mean absolute error

MBP

Mean blood pressure

MSE

Mean squared error

Nrf2

Nuclear factor erythroid 2–related factor 2

NO

Nitric oxide

oxLDL

Oxidized low–density lipoprotein

PI3K

Phosphoinositide 3–kinase

PPV

Positive predictive value

PWV

Pulse wave velocity

ROC

Receiver operating characteristic

RMSE

Root mean squared error

SBP

Systolic blood pressure

SEE

Standard error of estimate

T2DM

Type 2 diabetes mellitus

VEGFR2

Vascular endothelial growth factor receptor 2

VIF

Variance inflation factor

Author contributions

R.S and K.A. conceptualised the study, curated data, and developed methodology. R.S performed data analysis, created visualisations, and wrote the original draft. V.M. and J.S. contributed to formal analysis, supervision, and validation, and reviewed/edited the manuscript.

Funding

Open access funding provided by SRM Institute of Science and Technology for SRMIST – Medical & Health Sciences. This research received funding from the SRM Selective Excellence Research Initiative, under grant number SRMIST/R/AR(A)/SERI2024/174/71–342.

Data availability

All data generated or analysed during this study are included in this published article and its supplementary information files.

Declarations

Ethics approval and consent to participate

This study was approved by the SRM institutional Ethics Committee, under approval number ECR/8856/INST/TN/2013/RR-19. Written informed consent was obtained from all participants before their inclusion in the study.

Consent for publication

Not applicable. This manuscript does not contain any person’s data in any form (including individual details, images, or videos).

Declarations

During the preparation of this work, the author(s) used Quillbot to improve writing language. After using Quillbot, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.