Endothelial Function Is Associated With Incident Diabetes Mellitus: The ELSA‐Brasil Cohort Study Authors: Karina P. M. P. Martins, Sandhi M. Barreto, Lidyane V. Camelo, Sara T. Menezes, Naomi M. Hamburg, Maria de Fatima H. S. Diniz, Antonio L. P. Ribeiro, Luisa C. C. Brant Published: 2026 ABSTRACT ABSTRACTAimsDiabetes mellitus is associated with endothelial dysfunction, but the bidirectional nature of this relationship remains debated. We aimed to evaluate the association between microvascular endothelial function and incident diabetes in Brazilian adults and to assess potential effect modification by obesity status.Material and MethodsWe analysed participants from the ELSA‐Brasil cohort without diabetes at baseline and with valid endothelial function measurements. Microvascular endothelial function was assessed by peripheral arterial tonometry (PAT), using mean basal pulse amplitude (BPA), reflecting basal vascular tone and the PAT ratio, indicating vasodilatory capacity. Higher BPA and lower PAT ratio represent worse endothelial function. Sex‐specific quartiles were analysed. Incident diabetes was assessed after a mean follow‐up of 8.8 years (SD = 2.1). Cox proportional hazards models were adjusted for sex, age, education, physical activity, smoking, alcohol use and body mass index (BMI) or waist circumference.ResultsAmong 1268 participants (mean age 51 ± 8 years; 45% women; mean BMI 26.4 ± 4.4 kg/m2), 159 developed diabetes. Incidence rates increased across BPA quartiles and were higher in lower PAT ratio quartiles. After multivariable adjustment, impaired endothelial function was associated with increased diabetes risk, particularly for BPA (HR 2.13; 95% CI 1.24–3.66). The association for PAT ratio was weaker (HR 1.54; 95% CI 0.90–2.60). No effect modification by obesity status was observed.ConclusionImpaired endothelial function at baseline was independently associated with incident diabetes in Brazilian adults, regardless of obesity. These findings highlight the role of vascular dysfunction in cardiometabolic disease development. Introduction Diabetes mellitus is a metabolic disorder associated with an increased risk of all‐cause and cardiovascular mortality [1]. In 2022, approximately 828 million adults (95% CI 757–908 million) were living with diabetes worldwide, representing an increase of 630 million cases compared with 1990 [2]. By 2050, more than 1.31 billion (1.22–1.39) people are projected to have diabetes, with expected age‐standardised total diabetes prevalence rates greater than 10% in two super‐regions: 16.8% (16.1–17.6) in north Africa and the Middle East and 11.3% (10.8–11.9) in Latin America and Caribbean [3]. Atherosclerotic cardiovascular disease (CVD) is the major cause of morbimortality in individuals living with diabetes. Brazil is among the five countries with a greater number of individuals living with diabetes, with 15.7 million in 2021, and an estimated 23.2 million by 2045, as the prevalence is rising along with the increasing prevalence of obesity [3, 4, 5]. In this context understanding the pathophysiologic mechanisms and the determinants of diabetes are fundamental to advance prevention and reduce adverse events. Endothelial dysfunction is an early expression of vascular disease, related to cardiovascular risk factors, and can predict cardiovascular events [6]. We and others [7, 8, 9] have previously demonstrated that more impaired microvascular vasodilatory function was cross‐sectionally associated with metabolic risk factors, including higher body mass index (BMI), total cholesterol/HDL and triglycerides, and with obesity, even in the absence of metabolic abnormalities [10], but more longitudinal studies are needed to understand the direction of this association. Studies show that endothelial dysfunction predicts diabetes [11, 12, 13, 14, 15]. This finding may be related to the effects of hyperglycemia and insulin resistance, which may result in reduced bioavailability of vascular nitric oxide (NO) and inappropriate production of oxygen‐free radicals, culminating in endothelial dysfunction [16]. However, these studies used diverse methodologies to evaluate endothelial function, some of which are complex and/or invasive, and only two studies evaluated microvascular function using non‐invasive methods in individuals from Europe [13, 15]. Although the relationship between endothelial dysfunction and diabetes is likely bidirectional, growing evidence suggests that endothelial dysfunction may occur early in the course of metabolic disease and contribute to impaired insulin delivery, altered glucose uptake, and microvascular dysfunction [15]. However, longitudinal evidence evaluating whether endothelial dysfunction precedes the development of diabetes remains limited and inconsistent. As such, we aim to investigate the association between endothelial function and incident diabetes in the diverse Brazilian context, and evaluate whether there was effect modification of this relation according to obesity status, as a secondary analysis. We hypothesised that endothelial dysfunction would predict diabetes in Brazilian adults, with greater effect size in those individuals living with obesity. Methods Study Design, Population and Setting The study sample derived from the ELSA‐Brasil study, a multicenter cohort of 15 105 active or retired civil servants from six universities and research institutes in six Brazilian capitals, aged 35–74 years, who volunteered to participate. We collected data during four visits (2008–2010, 2012–2014, 2017–2019 and 2022–2024) and through annual telephone surveillance [17]. Other details of concept, design and cohort profile were published elsewhere [18]. The current study sample derived from the participants enrolled at the Minas Gerais Investigation Center, where peripheral arterial tonometry (PAT) data acquisition began partway through the baseline visit (2008–2010). A detailed description of patients who performed endothelial function at baseline has been previously published [9]. From the 1659 participants who were eligible for PAT exams, 124 participants were excluded due to: Raynaud's disease (n = 1), radical mastectomy (n = 3), refusal (n = 2), hand abnormality (n = 5), acute illness 12 h before the PAT (n = 58), incomplete data (n = 2), technically inadequate studies (n = 50). Hence, there were 1535 participants who had valid PAT at baseline. For the current study, all ELSA‐MG participants who attended baseline, underwent PAT exam, and were free of diabetes were eligible. Demographic and clinical variables as age, sex, smoking status, and history of a previous medical diagnosis of diabetes and anti‐diabetic medication use were collected through interviews and followed the ELSA‐Brasil's study protocol [17, 19]. In each visit, after an overnight fasting > 8 h, we measured weight, height, and waist circumference (WC) following internationally standardised protocols. Plasma glucose was measured using the hexokinase method (Cobas c501R, Roche Diagnostics, Rotkreuz, Switzerland) and glycated haemoglobin (HbA1c) by high‐pressure chromatography (HPLC—Bio‐Rad Laboratories, Hercules, CA, USA) [20]. We followed the STROBE guidelines for cohort studies to report this analysis. Explanatory Variables: Microvascular Endothelial Function Endothelial function was assessed by PAT. PAT exam was performed through Endo‐PAT2000 (Itamar Medical Ltd., Caesarea, Israel) by two certified examiners [18, 20]. Briefly, the cuff was placed on the participant's nondominant arm, 2 cm above the cubital fossa, and PAT probes were placed on each index finger. Baseline pulse amplitude was measured for 5 min. Arterial flow was interrupted on one side for 5 min by inflating the cuff at suprasystolic pressure. After 5 min, the cuff was deflated to induce reactive hyperemia, and the PAT signal was recorded for another 5 min. The contralateral finger was used to control for systemic changes. Two variables from PAT were used: the first is mean basal pulse amplitude (BPA) that reflects the basal vascular tone and is calculated by logarithmically transforming the mean values of the BPAs of both arms, and for which a higher value reflects worse endothelial function. The second is PAT ratio, which reflects the response to reactive hyperemia for which a lower value reflects a more impaired response. The reproducibility of PAT in the ELSA‐Brasil study was previously reported [21]. In this study, BPA and PAT ratio were categorised into sex‐specific quartiles, as women were observed to have better endothelial function, characterised by lower BPA and higher PAT ratio (Figure S1). Endothelial function measures were analysed in sex‐specific quartiles to allow for non‐linear associations with incident diabetes. We used the first quartile as the reference category for BPA and the fourth quartile as the reference category for PAT ratio, for both sexes. Outcome Variable: Type 2 Diabetes Mellitus The response variable was the time until the first diagnosis of diabetes occurring after study entry up to 31 December 2019. Participants who died of unrelated causes or who were lost to follow‐up before an event was observed were censored at the time of last observation. Diabetes was dichotomously defined in participants who: (i) reported a medical diagnosis of diabetes or current use of medication for diabetes or (ii) had a laboratory measurement reaching the thresholds for fasting plasma glucose (FPG) (≥ 7.0 mmol/L; 126 mg/dL), 2 h post‐load glucose (PG) (≥ 11.1 mmol/L; 200 mg/dL), or HbA1c (≥ 48 mmol/mol; 6.5%) [22, 23]. We ascertained incident diabetes at follow‐up visits based on the same criteria. We additionally included incident cases of diabetes if the participant reported a diagnosis of diabetes on at least two annual telephone interviews after the last clinical visit [19, 24, 25]. Covariables Educational level was categorised as incomplete elementary school, complete elementary school, complete secondary school, or university degree. Physical activity was assessed by the long version of the International Physical Activity Questionnaire and participants were classified as sedentary (< 600) or active (> 600 metabolic equivalent‐min/week) [26]. Smoking status was self‐reported and stratified as smoker or not smoker. The participant was classified as having excessive alcohol consumption when a consumption ≥ 210 g alcohol/week for men and ≥ 140 g alcohol/week for women was reported. BMI was calculated as weight in kilogrammes divided by height in square metres. WC was not included in the same models as BMI due to high collinearity, as confirmed by the estat vce command from Stata, which identified a correlation of 0.92 between the two variables. However, as a sensitivity analysis, we replaced BMI by WC in the fully adjusted models and in the evaluation of multiplicative interaction. We chose to test for the presence of interaction because BMI/WC may represent either confounders, as obesity can be related to diabetes and endothelial dysfunction, or effect modifiers [27]. Statistical Analysis Characteristics of the study population at baseline visit were described as proportions for categorical variables, and as means and standard deviations or medians and interquartile range for continuous variables. We calculated incidence rate according to sex‐specific quartiles of BPA and PAT ratio. The association of endothelial function sex‐specific quartiles (PAT ratio and BPA) with the time until the first diagnosis of diabetes was evaluated using Cox proportional hazard models. We demonstrated this association adjusted for sex and age (Model 1) and then further adjusted for education (Model 2), physical activity, smoking and alcohol consumption (Model 3) and finally by BMI (or WC, as a sensitivity analysis) (Model 4). Proportional hazards assumptions were evaluated using Schoenfeld residuals, and violations of proportionality were not identified. Given the minimal amount of missing data, analyses were conducted using complete‐case data. To investigate whether the effect of the BPA/PAT ratio varied according to obesity, we included multiplicative interaction terms in the final models, considering BMI or WC [28], both as continuous variables. We also conducted a sensitivity analysis excluding participants with prevalent CVD and additional analyses adjusting for baseline fasting glucose. Analysis was performed using Stata software version 14.00 (Stata Corporation, College Station, United States). Data analytic methods and study materials will be made available to other researchers for reproducing the results or replicating the procedure from the corresponding author upon reasonable request. Ethics Statement This study was conducted in accordance with the principles of the Declaration of Helsinki. The protocol was approved by the Institutional Review Board of UFMG ETIC 186/06. Written informed consent was obtained from all participants before any study procedure. Results Participants and Descriptive Data Of the 1535 participants with valid PAT at baseline, we excluded 214 with prevalent diabetes. Among the remaining participants, we excluded six participants who died between baseline and the first follow‐up visit, as they did not contribute longitudinal data and therefore were not eligible for inclusion in the time‐to‐event analysis. In addition, 47 were lost to follow‐up during follow‐up, resulting in a final sample of 1268 participants (Figure 1). Among the 1268 participants included in the study, a total of 28 deaths (2.21%) occurred during follow‐up. Participants who died after contributing follow‐up data were retained in the analyses and censored at the time of death, in accordance with standard survival analysis procedures. Mean age was 51 ± 8 years, with 45% women. A high proportion of sedentary (73%) and highly educated participants were observed, the latter reflecting the work‐based origin of the sample. Mean BMI was 26.4 ± 4.4 kg/m2. There were no statistically significant differences in baseline characteristics according to BPA's quartiles (Table 1) or PAT ratio (Table S1), except for BMI, which was higher in individuals with more impaired endothelial function (p < 0.001). Flowchart of the study sample in present analysis. (N = 1268). TABLE 1Characteristics of study participants at baseline (2008–2010), according to sex‐specific quartiles a of mean basal pulse amplitude (BPA). The Brazilian Longitudinal Study of Adult Health (ELSA‐Brasil), N = 1268.CharacteristicsMean (SD) or N (%)Total populationQ1Q2Q3Q4 N = 1268 N = 318 N = 316 N = 318 N = 316Age (years)51 ± 851 ± 951 ± 851 ± 852 ± 8Sex, women %566 (44.6)142 (44.7)141 (44.6)142 (44.7)141 (44.7)Educational level, %Incomplete elementary school44 (3.5)13 (4.1)9 (2.9)10 (3.1)12 (3.8)Complete elementary school49 (3.9)15 (4.7)13 (4.1)8 (2.5)13 (4.1)Complete secondary school330 (26.0)81 (25.5)81 (25.6)82 (25.8)86 (27.2)University degree845 (66.6)209 (65.7)213 (67.4)218 (68.6)205 (64.9)Physical activity, %Sedentary928 (73.2)232 (73.0)228 (72.2)234 (73.6)234 (74.1)Active340 (26.8)86 (27.0)88 (27.9)84 (26.4)82 (26.0)Smoking, %144 (11.4)39 (12.3)30 (9.5)35 (11.0)40 (12.7)Excessive alcohol consumption, %122 (9.6)23 (7.2)37 (11.6)35 (11.0)27 (8.5)Body mass index (kg/m2)26.4 ± 4.424.8 ± 3.526.2 ± 4.326.8 ± 4.227.8 ± 4.9BPA6.0 ± 0.84.9 ± 0.65.8 ± 0.46.3 ± 0.46.8 ± 0.3 a The BPA quartiles were sex‐specific: for men, the cutoffs were Q1 (≤ 5.90), Q2 (> 5.90 to 6.42), Q3 (> 6.42 to 6.79) and Q4 (> 6.79); for women, the cutoffs were Q1 (≤ 5.04), Q2 (> 5.04 to 5.72), Q3 (> 5.72 to 6.22) and Q4 (> 6.22). Characteristics of study participants at baseline (2008–2010), according to sex‐specific quartiles a of mean basal pulse amplitude (BPA). The Brazilian Longitudinal Study of Adult Health (ELSA‐Brasil), N = 1268. The BPA quartiles were sex‐specific: for men, the cutoffs were Q1 (≤ 5.90), Q2 (> 5.90 to 6.42), Q3 (> 6.42 to 6.79) and Q4 (> 6.79); for women, the cutoffs were Q1 (≤ 5.04), Q2 (> 5.04 to 5.72), Q3 (> 5.72 to 6.22) and Q4 (> 6.22). Outcome Data The mean follow‐up time was 8.8 years (SD = 2.1), ranging from 0.2 to 10.2 years. We identified 159 new cases of diabetes. The incidence rates per 1000 persons‐years for diabetes varied from 6.5 (95% CI 4.0–10.2) in the first quartile to 19.7 (95% CI 15.1–25.8) in the fourth quartile for BPA and from 7.2 (95% CI 4.7–11.1) in the fourth quartile to 15.3 (95% CI 11.4–20.6) in the first quartile of PAT ratio. In model 1, we found that the incidence of diabetes increased as BPA levels rise, showing a clear dose–response gradient, except in the fourth quartile for the fully‐adjusted model (Table 2). Nevertheless, the association remained statistically significant. Compared to the first quartile of BPA (better endothelial function), individuals in the second quartile were 1.68 times (95% CI 0.96–2.95, p = 0.07) more likely to experience incident diabetes. This increased in subsequent quartiles, with 2.42 times (95% CI 1.42–4.12, p < 0.05) higher likelihood in third quartile, and 2.13 times (95% CI 1.24–3.66, p < 0.05) higher in the fourth quartile (Table 3). Similar results were found in the final model for PAT ratio, although higher risk estimates were observed across lower PAT ratio quartiles, a clear dose–response pattern was not evident. Compared to the fourth quartile (better endothelial function), the risk of developing diabetes increased in 2.20 (95% CI 1.32–3.68, p < 0.05) in third quartile. In the second and first quartiles a risk of 2.04 (95% CI 1.22–3.42, p < 0.05) and 1.54 (95% CI 0.90–2.60, p = 0.11), was observed, respectively. Across sequential models, the association between BPA and incident diabetes remained consistent, although the magnitude was attenuated after adjustment for behavioural factors and further reduced after inclusion of BMI (Model 4). A similar pattern was observed for PAT ratio, although associations were weaker and less consistent across quartiles. The variables were also evaluated together in the same model, with no changes in the results. When BMI was replaced by WC in the fully adjusted models, the results were similar (Table S2). Figure 2 shows the unadjusted cumulative incidence of diabetes according to BPA and PAT ratio sex‐specific quartiles, revealing greater difference between the quartiles that represent better endothelial function versus the other three quartiles, for both BPA and PAT ratio. Additional adjustment for baseline fasting glucose was performed (Table S3) and the magnitude of the association was substantially attenuated or it lost statistical significance. TABLE 2Incidence rate of type II diabetes according to sex‐specific quartiles of mean basal pulse amplitude and peripheral arterial tonometry (PAT) ratio. The Brazilian Longitudinal Study of Adult Health (ELSA‐Brasil, 2008–2019).Person‐time at riskNumber eventsIncidence rate, 95% CI (per 1000 person‐year)Mean basal pulse amplitude (sex specific quartiles)First (best endothelial function)2923.5196.5 (4.0–10.2)Second2832.23512.4 (8.9–17.2)Third2759.35118.5 (14.0–24.3)Fourth (worse endothelial function)2731.25419.7 (15.1–25.8)Peripheral arterial tonometry ratio (sex specific quartiles)Fourth (best endothelial function)2908.8217.2 (4.7–11.1)Third2766.74917.7 (13.4–23.4)Second2768.44616.6 (12.4–22.2)First (worse endothelial function)2808.44315.3 (11.4–20.6) Incidence rate of type II diabetes according to sex‐specific quartiles of mean basal pulse amplitude and peripheral arterial tonometry (PAT) ratio. The Brazilian Longitudinal Study of Adult Health (ELSA‐Brasil, 2008–2019). TABLE 3Adjusted cox regression results for mean basal pulse amplitude (BPA) and peripheral arterial tonometry (PAT) ratio in sex‐specific quartiles. The Brazilian Longitudinal Study of Adult Health (ELSA‐Brasil, 2008–2019).Model 1Model 2Model 3Model 4HR (95% CI)HR (95% CI)HR (95% CI)HR (95% CI)Mean basal pulse amplitude (sex specific quartiles)First (better endothelial function)Ref.Ref.Ref.Ref.Second1.90 (1.09–3.33)* 2.29 (1.10–3.45)* 1.99 (1.14–3.49)* 1.68 (0.96–2.95)Third2.79 (1.65–4.73)** 3.86 (1.67–4.78)** 2.88 (1.70–4.88)** 2.42 (1.42–4.12)* Fourth (worse endothelial function)2.93 (1.74–4.95)** 4.05 (1.75–4.98)** 2.98 (1.76–5.02)** 2.13 (1.24–3.66)* Peripheral arterial tonometry ratio (sex specific quartiles)Fourth (better endothelial function)Ref.Ref.Ref.Ref.Third2.44 (1.46–4.07)** 2.45 (1.47–4.09)** 2.46 (1.45–4.10)** 2.21 (1.30–3.68)* Second2.27 (1.35–3.80)* 2.29 (1.36–3.83)* 2.28 (1.36–3.81)* 2.04 (1.22–3.42)* First (worse endothelial function)2.06 (1.22–3.47)* 2.08 (1.23–3.50)* 2.04 (1.21–3.45)* 1.54 (0.90–2.61) Note: Model 1: sex and age; Model 2: Model 1+ education; Model 3: Model 2+ physical activity, smoking and excessive alcohol consumption; Model 4: Model 3+ body mass index.* p < 0.05.** p ≤ 0.001. Adjusted cox regression results for mean basal pulse amplitude (BPA) and peripheral arterial tonometry (PAT) ratio in sex‐specific quartiles. The Brazilian Longitudinal Study of Adult Health (ELSA‐Brasil, 2008–2019). Note: Model 1: sex and age; Model 2: Model 1+ education; Model 3: Model 2+ physical activity, smoking and excessive alcohol consumption; Model 4: Model 3+ body mass index. p < 0.05. p ≤ 0.001. Cumulative incidence of type II diabetes according to mean basal pulse amplitude—BPA (A) and peripheral arterial tonometry—PAT ratio (B) sex‐specific quartiles. The Brazilian Longitudinal Study of Adult Health (ELSA‐Brasil, 2008–2019). N = 1268. In our analysis, we tested for interaction between endothelial function and BMI by including a multiplicative interaction term, and found no statistical evidence of effect modification [29]. Sensitivity analysis excluding participants with prevalent CVD was performed, and the results were similar as shown in Table S4. Discussion In the present study, we evaluated the longitudinal association between endothelial function and incident diabetes in a large cohort of Brazilian adults. Prior studies have described the association of digital vascular function with incident diabetes in individuals of European descent [7, 8, 30], who have different clinical and epidemiological characteristics compared to the Brazilian population [9]. We herein demonstrate similar associations of more impaired endothelial function with incident diabetes after 8.8 years of follow‐up in individuals from Brazil, a middle‐income country, where demographic and nutritional shifts are increasing the prevalence of metabolic risk factors, particularly diabetes and obesity, which threaten to raise the incidence of CVD [4, 5]. We also observed that there was no effect modification by BMI or WC in the association between endothelial function and incident diabetes, suggesting that this association does not differ across levels of BMI or WC. After adjustments, we found that the risk of diabetes increased significantly in those with more impaired endothelial function, and individuals in the fourth quartile of BPA had a 113% higher hazard of diabetes compared to those in the first quartile of BPA. When baseline glycemia was included in the adjustment model, the magnitude of the association was substantially attenuated or, in some cases, lost statistical significance. Nevertheless, the results of this model should be interpreted with caution, as baseline glycemia may act as a mediator rather than solely as a confounder in the relationship between endothelial function and diabetes incidence [30]. Our findings are consistent with previous results from the Gutenberg Heart Study (GHS), a cohort study conducted in Europeans [14], which demonstrated that the reactive hyperemia index (RHI) was independently associated with an increased risk of type 2 diabetes by 16% (95% CI, 1.01–1.34; p = 0.041) per 1 SD. Furthermore, BPA was independently associated with incident diabetes (RR, 1.17; 95% CI, 1.02–1.33; p = 0.022). These results corroborate what we observed in direction and significance, even though there were methodological differences in our studies, precluding direct comparison of effect sizes: we used PAT ratio instead of RHI [7]; we used different distribution of data and statistical model, as the associations of our explanatory variables with diabetes were non‐linear; we adjusted for a smaller number of covariables that were considered confounding factors in our conceptual model, including BMI/WC. Some studies have also evaluated whether endothelial dysfunction can predict diabetes in different scenarios [31, 32, 33, 34, 35, 36]. To compile these results, in 2012, Muris et al. published a meta‐analysis of 13 studies demonstrating that various estimates of microvascular dysfunction were associated with incident diabetes, including plasma samples of E‐selectin, vascular cell adhesion molecule‐1 (VCAM‐1), intercellular adhesion molecule‐1 (ICAM‐1) and Von Willebrand factor (vWF) analysed by use of ELISAs; peripheral vascular reactivity assessed by a dose–response curve to intra‐arterial infusion of acetylcholine; retinal diameters evaluated by the use of fundus photos and microalbuminuria. However, only one of these studies evaluated vascular function with an invasive method, and in large conduction vessels [13]. Our study brings novelty by investigating a non‐invasive method in the community‐based and racially diverse ELSA‐Brasil study. We also observed that BPA was more consistently associated with incident diabetes than PAT ratio. This finding has also been previously reported with other cardiovascular risk factors [7, 8, 9] suggesting that these risk factors may be more associated with structural changes than physiological responses, reflected by the higher basal vascular tone, expressed by BPA. A question already raised was that, given the strong correlation between the PAT ratio and BPA, part of the PAT ratio values could be a result of an impaired BPA and not a response to hyperemia. The influence of the resting pulse amplitude in PAT ratio could be a physiological effect, given that the already higher resting pulse amplitude could preclude vasodilation, or a mathematical calculation effect, because PAT ratio is calculated using BPA in the denominator [6]. One possible explanation for the weaker performance of the PAT ratio compared to BPA is the presence of a ceiling effect. Individuals with higher BPA may have a limited capacity for further vasodilatory response, which could attenuate the variability captured by the PAT ratio and reduce its ability to discriminate differences in risk across individuals. Taken together, these findings suggest that basal vascular tone, as reflected by BPA, may capture more sustained or structural alterations in the microvasculature. In contrast, PAT ratio reflects reactive vasodilatory capacity and may be influenced by baseline vascular tone. Consequently, in the presence of higher BPA, the capacity for further vasodilation may be reduced, which may partially explain the weaker and less consistent associations observed for the PAT ratio. We did not confirm our hypothesis that overweight/obesity or abdominal obesity would amplify the effect of endothelial dysfunction on diabetes risk, as we found that BMI and WC did not modify the relationship between endothelial dysfunction and diabetes risk. However, we observed that after adjusting the models for BMI and WC, the association between BPA and diabetes risk reduced, suggesting that part of the effect of endothelial dysfunction on diabetes risk is explained by obesity/abdominal obesity. This finding may reflect either confounding or mediation mechanisms involving adiposity, which could be explained by two possible pathways: (1) Obesity may act as a common cause of both endothelial dysfunction and diabetes, thereby confounding the observed relationship between endothelial function and diabetes [37, 38]; (2) endothelial dysfunction could increase the risk of obesity, which in turn could contribute to the development of diabetes, with BMI and WC potentially serving as mediators in the endothelial dysfunction diabetes association [39, 40]. Longitudinal studies have shown that obesity is an important predictor of endothelial dysfunction [37, 38], supporting the first hypothesis. On the other hand, previous studies hypothesise that alterations in endothelial function could accelerate obesity [39, 40]. Some molecules that are expressed in a dysfunctional endothelium are considered key and capable of modulating systemic metabolism, including transcription factors (P53, PGC1a, FOXOs and NF‐kB), pro‐angiogenic signals (VEGF, VEGFR2, ANG2/TIE2, DLL4/NOTCH1), components of the insulin cascade (IRS‐2, insulin receptor) and mediators of fatty acid transport (NOTCH1, VEGF‐B/NRP1 and CD36) [41, 42, 43, 44, 45, 46, 47, 48]. The interaction between glucose metabolism and endothelial dysfunction is more complex and involves several feedback mechanisms. In the arterial microcirculation, endothelial dysfunction can impair insulin's function and redirect blood flow from non‐nutritive to nutritive capillaries, reducing muscle glucose uptake [11, 49]. Pancreatic microvascular dysfunction can cause apoptosis of β‐cells in the pancreas, reducing insulin secretion from the pancreas and thus leading to hyperglycemia [50] which has been associated with endothelial dysfunction [51]. Furthermore, microvascular dysfunction may act as an intermediate step linking obesity, low physical activity and chronic low‐grade inflammation to insulin resistance [52], although the effect modification by obesity status was not observed in this analysis. Our study has limitations. First, because the endothelial function evaluation was a substudy of ELSA‐MG, our sample size was smaller than that of the ELSA‐Brasil study, precluding stratified analysis for incident pre‐diabetes and diabetes. Second, this study was conducted in a cohort of civil servants holding stable positions at two universities, with an average education level higher than that of the general Brazilian population. However, the ELSA‐Brasil cohort comprises participants from diverse socioeconomic backgrounds and includes a racially heterogeneous population [18]. Although representativeness is essential when estimating population‐level measures such as prevalence or incidence, it is not a strict requirement for the validity of exposure–outcome associations in well‐conducted epidemiological studies. Therefore, diabetes incidence estimates from this cohort may not be readily generalizable to the broader population, whereas the observed associations are likely to be more broadly applicable [53, 54, 55]. Third, since ELSA‐Brasil is an observational study, we cannot rule out the possibility of residual confounding by the variables included in the adjustment model, nor can we completely exclude the possibility of omitted confounding variables. Furthermore, although our study employs a prospective cohort design that enables the assessment of temporal relationships, conclusions regarding causality should be interpreted with caution due to its observational nature. Despite rigorous adjustments for known confounders, the possibility of unmeasured or residual confounding factors influencing the results cannot be entirely ruled out. Fourth, regarding PAT exam, it evaluates microvascular endothelial function, which has the advantage of being noninvasive and operator independent, but may not be used interchangeably with other measures of vascular function [9, 10]. Fifth, that BPA and PAT ratio do not have cutoffs that could be used to categorise the endothelial function variables. To overcome this issue, we studied the variables continuously and in sex‐specific quartiles before deciding how to analyse them. We recognise that the absence of a dose–response relationship between PAT ratio and incident diabetes is also a limitation to the interpretation of results, but the significant difference between the quartile of better endothelial function versus the others corroborates the findings. Lastly, we acknowledge that BMI alone is not the ideal measure of excess adipose tissue [28], and that WC may be more related to metabolic risk [56, 57]. We chose to use BMI as it is more frequently used in clinical practice. To overcome this limitation, we performed a sensitivity analysis using WC instead of BMI and the results were largely unchanged. Strengths of the present study include the use of a robust statistical model demonstrating the association between endothelial dysfunction and incident diabetes in a diverse population with a high prevalence and incidence of diabetes, in which this association had not yet been demonstrated, and the comprehensive and rigorous assessment of covariables and ascertainment of outcomes. Conclusion Our findings support the hypothesis that impaired endothelial function in the microvasculature occurs before the onset of type 2 diabetes, corroborating data observed in the European population, and advancing the field by showing that this association is not modified by obesity status. Our findings may reflect that endothelial dysfunction could be part of a cyclical mechanism that will feedback metabolic endocrine changes, predisposing to diseases such as diabetes and subsequently, CVDs. These findings may have implications for the understanding of cardiometabolic diseases in the context of pathophysiology, prognosis and treatment. Funding The ELSA‐Brasil cohort study was funded by the Brazilian Ministry of Health (Department of Science and Technology) and by the Ministry of Science, Technology, and Innovation (FINEP—Funding Authority for Studies and Projects—and CNPq—National Council for Scientific and Technological Development), under grant numbers 01 06 0010.00, 01 06 0212.00, 01 06 0300.00, 01 06 0278.00, 01 06 0115.00, and 01 06 0071.00. The Endothelial function subproject at ELSA‐Brasil was funded by FAPEMIG. SMB receives a CNPq Research Productivity Fellowship (Grant number: 303656/2021‐2). ALPR LCCB is supported in part by CNPq (307329/2022‐4, 408659/2024‐6, 445011/2023‐8, and 406938/2024‐5) and FAPEMIG (RED‐00192‐23), CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superiorl). Conflicts of Interest The authors declare no conflicts of interest. Supporting information Table S1: Characteristics of study participants at baseline (2008–2010), according to peripheral arterial tonometry ratio (PAT‐ratio) sex‐specific quartiles. The Brazilian Longitudinal Study of Adult Health (ELSA‐Brasil), N = 1268. Table S2: Adjusted cox regression results for mean basal pulse amplitude (BPA) and peripheral arterial tonometry (PAT) ratio in sex‐specific quartiles using waist circumference instead of body mass index. The Brazilian Longitudinal Study of Adult Health (ELSA‐Brasil, 2008–2019). Table S3: Adjusted cox regression results for mean basal pulse amplitude (BPA) and peripheral arterial tonometry (PAT) ratio in sex‐specific quartiles including Adjustment for Baseline Fasting Glucose. The Brazilian Longitudinal Study of Adult Health (ELSA‐Brasil, 2008–2019). Table S4: Adjusted Cox regression results for mean basal pulse amplitude (BPA) and peripheral arterial tonometry (PAT) ratio in sex‐specific quartiles excluding 39 participants with prevalent cardiovascular disease. The Brazilian Longitudinal Study of Adult Health (ELSA‐Brasil, 2008–2019). Figure S1: Distribution of mean basal pulse amplitude (BPA) and peripheral arterial tonometry (PAT) ratio by sex: The Brazilian Longitudinal Study of Adult Health (ELSA‐Brasil, 2008–2010). N = 1268.