In our analyses we found inverse associations of FG, FI, HOMA-IR, 2HG and 2HI with LVEDVI in individuals without known diabetes. FG mainly represents nocturnal hepatic gluconeogenesis dependent on hepatic insulin sensitivity and 2HG mainly reflects postprandial hyperglycemia. Conversely, we found positive linear associations of these glycemic variables with LVWTI. The net result of the associations of higher values of glycemic indicators on the left ventricular geometry was a greater left ventricular concentricity. This concentric remodeling was independent of other determinants, such as hypertension.
Noteworthy, the higher values of the glycemic variables were also accompanied by greater values of ASI and lower values of LVSI. The higher values of HR observed in association with the greater values of the glycemic variables might be a compensatory mechanism, for the lower LVSI, trying to avoid any deleterious consequence on the LVCI which was, in reality, not affected as well as the LVEF. This means that subjects with higher values of the glycemic variables would have a smaller “reserve”, in this circumstance heart rate, to utilize under stress when compared with individuals with lower levels.
In summary, our findings showed a direct relation between higher glucose and/or insulin levels and greater arterial stiffness, smaller LV chamber size and higher LV thickness with resultant LV concentric remodeling and lower LV stroke volume. These changes in heart geometry and function may be related to the development of heart failure with preserved ejection fraction (HFpEF).
In the context of the published literature
A previous analysis of the Multiethnic Study of Atherosclerosis (MESA) [
4] showed that individuals with IFG had smaller LVEDV and LVSV and no significant difference regarding LVM, LVCO and LVEF, when compared with normoglycemic subjects. When diabetic subjects were compared with persons with normoglycemia, besides the smaller LVEDV and LVSV, the LVEF was lower and the LVM was higher. A more recent analysis of the MESA study [
5] demonstrated that IFG and HOMA-IR were positively associated with LVC. Moreover, in univariate analyses stratified by BMI, subjects with IFG had a higher LVC and LVMI and a lower LVEDVI when compared with individuals with normal fasting glucose.
An investigation of 1603 individuals with a mean age of 64 years from the Framingham Heart Study [
6], demonstrated positive associations in age-adjusted models of HOMA-IR with LVMI, LVC, relative wall thickness, LVCO and LVEF for men and women. Noteworthy, after further adjustments, that included BMI, just LVC remained positively associated with HOMA-IR, while the relation with LVMI became even an inverse one.
A previous investigation of the Atherosclerosis Risk in Communities (ARIC) study [
7] showed that prediabetes and type 2 diabetes were associated with higher arterial stiffness when compared with subjects with normal glucose levels. A recent study [
32] suggests that insulin resistance might be an early marker of arterial stiffness in healthy and active young to middle-age men. Triglyceride glucose index is the product of fasting plasma glucose and triglycerides and is a strong surrogate for insulin resistance [
33]. Previous studies found that the triglyceride glucose index was positive associated with the risk for incident type 2 diabetes [
34] and with arterial stiffness in a relatively healthy Korean population [
35] and in lean postmenopausal women [
33]. Another study [
36] indicated that lipopolysaccharide-binding protein levels, a surrogate of inflammation immune responses, were associated with arterial stiffness among male patients with type 2 diabetes independently of obesity and traditional cardiovascular risk factors. Interestingly, treatment with liraglutide in patients with recently diagnosed type 2 diabetes reduced oxidative stress resulting in an improvement of arterial stiffness and left ventricular myocardial strain [
37]. Moreover, in an experimental model with induced type 2 diabetes in female mice, empagliflozin improved kidney injury by promoting glycosuria, and probably by reducing systemic and renal artery stiffness [
38].
Finally, our group has previously published findings from two independent analyses of KORA FF4 samples. The first one [
15] which used multivariate models without adjustment for height, weight or BMI, showed that individuals with prediabetes and type 2 diabetes had lower LVEDV, LVESV, LVSV and higher LVM and LVEF (just individuals with prediabetes) when compared with subjects with normal glucose metabolism. The second one [
16], which further adjusted for BMI, showed that individuals with prediabetes and type 2 diabetes had higher LVWT when compared with subjects with normal glucose metabolism. We did not find any other MRI study regarding LVWT.
All the previously cited studies included MRI determined heart parameters. The results of our current analyses are, in general, in line with these studies regarding the findings of associations of higher glycemic and insulin levels with lower LV cavity size and greater wall thickness, concentricity and arterial stiffness without effect on LVCO and LVEF. On the other hand, compared with previous studies, we did not find any association of the OGTT parameters with LVMI. We believe that the main reason for this specific finding was the use of body fat-free and fat mass, instead of BMI, as covariate (after the initial normalization of cardiac parameters to height
2.7). As far as we know, almost all previous studies that analyzed the associations of glucose and/or insulin levels with cardiac structure and function by echocardiography or MRI utilized BMI as a covariate. BIA assess body composition which is considered a better measurement of obesity than BMI as it allows to differentiate between fat-free and fat mass which both contribute to BMI [
39]. In sensitivity analyses (Additional file
1: Table S4), we compared three regression models of the associations of HOMA-IR with LVMI. In the first model, besides the adjustment for age, sex, systolic blood pressure, use of antihypertensive medication, smoking status, alcohol consumption, sedentarism, estimated glomerular filtration rate, fasting time and study sample we further adjusted for body fat-free and fat mass (our original model). In the second model, we did not adjust for fat mass and fat-free-mass, but for BMI. Finally, in the third model we further adjusted for weight and height. Remarkably, while in our original model we had no significant association (p = 0.602), it was highly significant, when adjusted for BMI and the association was an inverse one (Additional file
1: Table S4). This seems to be the same phenomenon that has been observed in the above-discussed analysis of the Framingham Heart Study [
6]. Actually, we believe that adjustment for BMI (which includes height in its calculation) in a model that included a variable already normalized to height
2.7, might represent an over adjustment and thus, potentially, might lead to a misleading result. The reason for our choice of body fat-free and fat mass, instead of BMI, as cofounders in the multivariate models was the potential diverse effects of the different components of the body composition. Body fat-free and fat mass might have diverse effects on the OGTT parameters and, at the same time, the LV structure and function explainable by their different structural composition, metabolic demands and functional manifestations. Body fat-free mass is responsible for almost all of the body’s metabolic requirements while body fat mass may be accountable for the release of numerous biomarkers and inflammatory cytokines. Moreover, body fat-free mass might have a metabolic protective effect that mitigates the excess of the previously mentioned markers. We believe that our approach to normalize the cardiac parameters to height
2.7 and subsequent adjust for body fat-free and fat mass was the most feasible strategy to analyze the effects of OGTT parameters on the heart independently of obesity which is highly correlated with both exposures and outcomes.
Potential mechanisms for the observed associations
Glycemic disorders and insulin resistance are usually accompanied by several cardiovascular and metabolic risk factors and co-morbidities like older age, obesity, hypercholesterolemia, hypertension and coronary heart disease. Otherwise, the involved pathologic mechanisms that might explain these associations are still not completely clarified [
2]. Our study protocol was not designed to elucidate possible pathophysiological mechanisms that might be involved in the associations described in our analyses and the cross-sectional design of our study restricts the evaluation of causal relationships. Besides that, there is still no prospective clinical trial that had undoubtedly demonstrated that higher glucose and insulin levels might have a causal association with changes in the heart [
2]. Nevertheless, we have integrated various risk factors in our multivariable regression models. Our results might support a direct relation between higher glucose and/or insulin levels and greater arterial stiffness, smaller LV chamber size and higher LV thickness with resultant LV concentric remodeling and lower LV stroke volume. The LV cardiac output would be kept due to higher heart rate.
Dose–response modeling results suggest that fasting glycemic levels do not have a clearly defined threshold in their relations with cardiac parameters, as with retinopathy, but rather a continuous association. Hyperglycemia and hyperinsulinemia are usually accompanied by increased free fatty acid levels, systemic and tissue inflammation, oxidative stress, and activation of the renin–angiotensin–aldosterone system and the sympathetic nervous system. Otherwise, the effects of increased glycemic and insulinemic levels on the cardiac structure are, in the beginning, clinically asymptomatic [
2]. These initial effects are characterized by higher stiffness (mainly by increased intracellular Ca
2+), increased collagen and advanced glycation end products, fibrosis and cellular hypertrophy (mainly by expression of hypertrophic genes) [
2]. Noteworthy, this process should be considered as not specific of the heart, but rather of the entire cardio-vascular system. Because of this initial process, the LV chamber size decreases and the LV wall hypertrophies leading to cardiac remodeling, cardiac diastolic dysfunction and eventually systolic dysfunction. Our analyses are in line with the subclinical presentation of this process.
Finally, one of the possible complications that may affect individuals with type 2 diabetes and even with prediabetes is a cardiovascular autonomic neuropathy. This condition is characterized by sinus tachycardia, exercise intolerance, and orthostatic hypotension [
40]. Moreover, it may also be associated with left ventricular systolic and diastolic dysfunction independent of any other cardiac disease [
41]. Interestingly, while the cardiovascular autonomic neuropathy might result in a decrease in the left ventricular filing volume, leading to a lower stroke volume, it also causes sinus tachycardia that will result in a normal cardiac output. Unfortunately, our study protocol did not include measurements, such as heart rate variation, the Valsalva maneuver and postural changes in blood pressure to evaluate in more detail the cardiac autonomic function.
Study limitations
We need to mention some limitations of our analyses. First, our study sample consisted of European Caucasians; therefore, further analyses of samples with different ethnicity and age groups would be desirable to investigate the strength of this association across those groups. Second, the cross-sectional design is a limitation, which means that relationships between cause and effect might be not recognized. Third, our sample consisted of data from two separate studies (SHIP-TREND-0 and KORA FF4) with some minor methodological variances including difference in sample ages (21 to 81 in SHIP-TREND-0 and 39 to 73 in KORA FF4). While we consider that this did not have an effect on our results, we cannot conclusively exclude it (even after supplementary adjustment for study sample in our regression models). Finally, although we have incorporated numerous confounders into our multivariable regression models, we cannot exclude residual confounding due to unmeasured conditions.
However, our analyses also have some noteworthy strengths, including the large number of subjects based on two cohorts of the general population, the standardized evaluation of OGTT data after an overnight fast, and the availability of lifestyle data and multiple metabolic risk factors.