Exercise capacity in diabetes mellitus is predicted by activity status and cardiac size rather than cardiac function: a case control study

Subjects with T1DM and T2DM attending specialist hospital diabetes outpatient clinics were recruited via promotional leaflets and approval of their treating doctor. An equal number of those with- and without microvascular complications formed each group, defined by diabetic retinopathy, nephropathy and/or neuropathy according to current guidelines [7]. Inclusion criteria consisted of confirmed diabetes diagnosis, age 18–70 years, and sufficient physical capacity to perform low-intensity exercise.

A group of healthy controls were age- and sex- matched to each diabetes subgroup in a 2:1 ratio to balance demographic differences between the DM subgroups and allow inferences to be made when considering results for T1DM and T2DM groups. For both diabetes and control subjects, exclusion criteria included known coronary artery disease, resting LV systolic dysfunction (defined as LV ejection fraction < 40%), significant nephropathy (eGFR < 30 ml/min/1.73 m²), and chronic obstructive airways disease. Subjects were also excluded if echocardiographic images were non-diagnostic.

A detailed baseline assessment of microvascular and cardiovascular disease was performed in all subjects, including serum creatinine, urinalysis, retinal imaging for diabetic retinopathy grading, 24-h blood pressure monitoring, and echocardiography. History of diabetic neuropathy was assessed by review of each participant’s hospital medical record and diabetes outpatient clinic charts. A detailed diary of exercise habits over the 4 weeks prior to the study were recorded and the product of weekly exercise (hours) and intensity were used to quantify the metabolic equivalent (MET) hours.

Cardiopulmonary exercise testing was performed on an upright bicycle (Excaliber Sport, Lode, The Netherlands) using an individualized continuous incremental ramp protocol until exhaustion. Peak exercise capacity (\(\dot{V}O_{2}\)peak) was aimed to be reached within 10 min of exercise. Continuous 12-lead electrocardiography (ECG) monitored heart rate (HR), ST-segment changes and arrhythmia throughout exercise (Norav Medical, Israel) whilst blood pressure was recorded using an automated ECG-gated auscultatory device (Tango M2, SunTech Medical, USA). Breath-by-breath analysis of oxygen consumption and carbon dioxide production (JLab, CareFusion, Germany) was averaged over five breaths. Respiratory exchange ratio (RER), ventilatory threshold (VT) and ventilatory efficiency (\(\dot{V}{\text{E}}/\dot{V}{\text{CO}}_{2}\)) were calculated by standard measures [8].

Exercise echocardiography was completed on a semi-supine bicycle ergometer with lateral tilt (Lode, The Netherlands) over four exercise stages at increasing power calculated according to an individual’s \(\dot{V}O_{2}\)peak. We have demonstrated previously that 66% of maximal power obtained on an upright ergometer is equivalent to near-maximal intensity on a semi-supine ergometer [9]. Thus low, medium, high and maximal intensity exercise was prescribed as 15, 25, 50 and 66% of \(\dot{V}O_{2}\)peak respectively. After 1 min of commencing each exercise stage, image acquisition began with the aim to collect all images within 3 min. Breath holding was required in only a minority of studies predominantly in the latter stages of exercise to improve image quality, although a preference to extend the number of cardiac cycles recorded was favored. At rest and during each stage of exercise, apical four- and two-chamber, and parasternal long- and short-axis transthoracic images were collected to calculate comprehensive measures of biventricular systolic function according to guideline recommendations for performing exercise echocardiography [10]. Non-invasively derived central hemodynamic parameters including stroke volume (SV), cardiac output (CO), and pulmonary artery systolic pressure (PASP) were also calculated by standard measures [11]. The Doppler envelope of tricuspid regurgitation (TR) peak velocity was optimized by injection of 1–2 ml of an agitated colloid contrast to calculate PASP. Measurements of chamber volume, SV and CO were indexed to body surface area (BSA). Three to five beat loops were recorded for each 2D window while continuous wave (CW) and pulse wave (PW) Doppler recordings were extended to allow measurement and averaging of three beats. The highest recorded TR peak velocity was used for PASP at each level of exercise. Heart rate reserve was the difference of peak exercise and resting heart rate. All other measures of cardiac reserve were defined as the difference between peak exercise and resting measures as a percentage. Global longitudinal strain (GLS) was measured at rest using 2-dimensional speckle-tracking methods as described in detail previously [12]. Diastolic function was assessed at rest, and in early recovery after sufficient separation of the early (E) and late (A) diastolic mitral inflow waves. A single experienced cardiac sonographer conducted all studies using a Vivid E9 cardiac ultrasound machine (GE Healthcare), while one cardiologist with expertise in echocardiography analyzed all images offline using EchoPAC software (Version 113, GE Healthcare).

Data was analyzed using IBM SPSS Version 22 (SPSS Inc., Chicago, USA). Analysis of data normality was determined by the Shapiro–Wilk test. Continuous data was expressed as the mean value ± standard deviation (SD). A P < 0.05 was considered statistically significant. Differences between groups were assessed using unpaired t tests and Chi square Fisher’s exact test. Pearson’s coefficient evaluated univariate correlations. Two-way repeated measure ANOVA was used to assess the effect of diabetes and exercise stage on echocardiographic measures with changes during exercise analyzed as a within subjects factor and comparisons between groups as between subjects factors. A forward stepwise multiple linear regression was used to assess predictors of exercise capacity (\(\dot{V}O_{2}\)peak). Age, gender and BSA were included in the multiple regression as they are well described determinants of \(\dot{V}O_{2}\)peak. In addition, the strongest univariate predictors of \(\dot{V}O_{2}\)peak were included in the regression after excluding those variables demonstrating significant collinearity. Collinearity was considered significant when two factors were closely associated (R > 0.7). The number of variables entered into the model was restricted to 7, selected as a balance between including variables relevant to the study hypothesis whilst maintaining a reasonable balance for the cohort size.

\(\dot{V}O_{2}\)peak was significantly lower in T2DM subjects compared to their controls (Table 1 and Fig. 1). In contrast, those with T1DM had above-average fitness (power: 120 ± 33% predicted) and no significant difference in exercise capacity compared to their controls.

Both DM groups had normal resting measures of biventricular systolic function and pulmonary pressures (Table 2). There was greater concentric remodeling amongst the T2DM cohort relative to their controls (55% vs. 10%; χ² P = 0.024), because of lower LV end-diastolic volume (LVEDV) and a tendency to greater LV mass. There were no differences in cardiac structure (LV relative wall thickness, LV mass, LVEDV and left atrial volume) in T1DM subjects and their controls.

The diastolic E/A ratio was lower and E/e’ ratio higher in T2DM subjects compared to their controls, while counter-intuitively left atrial volume index (LAVI) was lower. The prevalence of LV diastolic dysfunction according to current diagnostic guidelines was not different (20% of T2DM and control subjects alike; χ² P = 1.0). No subject with T1DM had evidence of diastolic dysfunction.

All measures of biventricular systolic function and hemodynamic parameters augmented significantly from resting to maximal exercise intensity in both DM groups (Table 3a, b and Fig. 2). No subject had evidence of inducible regional wall motion abnormalities on echocardiography or significant ischemic ECG changes.

E/e′ post exercise was surprisingly normal in all four subjects identified to have LV diastolic dysfunction at rest. Only one subject in the study cohort—with T2DM—met criteria (E/e′ > 15) for LV diastolic dysfunction at peak exercise.

Significant univariate correlates of \(\dot{V}O_{2}\)peak are presented in Table 4. The strongest univariate associations with \(\dot{V}O_{2}\)peak were LVEDV (accounting for 44% of variance, see Fig. 4), RV end-diastolic area (RVEDA), MET-hours of physical activity and cardiac index reserve.

A multivariate analysis was performed to predict \(\dot{V}O_{2}\)peak from a selection of the significant univariate correlates. Age, sex and BSA accounted for 49% of the variance in \(\dot{V}O_{2}\)peak (Table 5a). After adjustment for the above three factors, significant positive predictors of \(\dot{V}O_{2}\)peak were LVEDV and estimated MET-hours of physical activity, whilst T2DM was a negative predictor (Table 5b). The combined model accounted for 80% of the variance in \(\dot{V}O_{2}\)peak (R² = 0.80; P < 0.0001).

Chronic hyperglycemia is associated with endothelial dysfunction and microvascular disease, and a causal link with congestive heart failure (CHF) and reduced exercise capacity [2] has been proposed. In contrast to previous investigations [13, 14], we observed a significant inverse correlation between HbA1c and \(\dot{V}O_{2}\)peak in our cohort. However, this association was largely abolished when other confounding factors such as diabetic status, age and sex were included in the multivariate model.

It may be argued that the associations between hyperglycemia and exercise capacity are best explored in T1DM rather than T2DM subjects given the lower prevalence of other confounding factors such as obesity, additional cardiovascular risk factors and lower exercise participation. Most studies have failed to identify a clear link between glycemic control and exercise capacity [15, 16], with the exception of Baldi et al. [17]. Baldi compared a group of 12 T1DM endurance triathletes with matched controls and whilst there was no difference in exercise capacity, the 6 T1DM athletes with worse glycemic control (mean HbA1c = 62 mmol/mol) had 15% lower \(\dot{V}O_{2}\)peak than the six athletes with better control (mean HbA1c = 48 mmol/mol). The degree to which these differences can be accounted for by hyperglycemia as opposed to confounding behavioral and health factors is difficult to quantify. The data presented here suggests that once multiple factors are considered, the influence of glycemic control on exercise capacity is, at most, modest.

A higher incidence of congestive heart failure (CHF) has been observed in diabetes subjects [18, 19], leading to the concept of a diabetes-specific cardiomyopathy [2]. Diabetic cardiomyopathy has evolved into a specific clinical entity with two distinct phenotypes proposed: heart failure with preserved ejection fraction (HFpEF) and heart failure with reduced ejection fraction (HFrEF) [2]. Furthermore, Widya et al. [20] have reported changes in the right ventricle that parallel those seen in the left ventricle in T2DM subjects. On the other hand, the entity of diabetic cardiomyopathy remains contentious amongst some given the dependence of small animal and molecular models rather than prospective human data [6, 21].

A high prevalence of LV diastolic dysfunction has been observed in T2DM [22, 23] and its association with impaired exercise capacity has been suggested to represent a subclinical phase of diabetic cardiomyopathy [24, 25]. In the present study, we found no difference in the prevalence of diastolic dysfunction between the T2DM group and their controls. Interestingly, the four T2DM subjects identified to have diastolic dysfunction at rest had normal diastolic stress test results. Similarly, there were no differences in the change of LVEDV during exercise when DM subjects were compared with matched controls (Fig. 3). These findings strengthen the argument that diastolic dysfunction was not the cause for the reduced \(\dot{V}O_{2}\)peak.

There were also no significant differences in LV or RV systolic function to explain the lower \(\dot{V}O_{2}\)peak. Subjects with reduced LVEF were excluded from participating in our study, and the mean LVEF in the total cohort was 60 ± 5%. Reductions in newer echocardiographic measures of LV systolic function such as global longitudinal strain (GLS) [26‐28] and tissue Doppler myocardial velocity [29], however, can identify subclinical LV systolic dysfunction in the presence of a normal ejection fraction. In our cohort of T2DM subjects there was a non-significant trend to reduced resting LV GLS. However, LV GLS did not correlate with \(\dot{V}O_{2}\)peak amongst T2DM subjects (R = − 0.08, P = 0.78), and other echocardiography measures of LV function had similarly poor associations with exercise capacity. During exercise, contractile reserve, as quantified by LVEF and LVs’ augmentation, was similar and normal in both the T2DM group and controls.

It has been demonstrated previously that resting cardiac index is similar in healthy individuals regardless of fitness level [30], and thus the similar resting cardiac index across the groups in our cohort was expected. However, in T2DM subjects the augmentation of cardiac index during exercise was reduced, explained by significant differences in heart rate but not stroke volume.

Heart rate reserve was significantly reduced in those with T2DM compared to controls (mean 52 ± 13 bpm vs. 64 ± 11 bpm; P = 0.018) due to both a higher resting heart rate and a lower peak exercise heart rate.

Possible explanations for the difference in heart rate reserve include diabetic cardiovascular autonomic neuropathy (CAN), reduced β-adrenoreceptor sensitivity or remodeling of the sinoatrial node. It is possible that the reduced HR reserve could also reflect a relative lack of physical exercise conditioning. Only one of our T2DM participants had a diagnosis of diabetic neuropathy, which is perhaps less than may be expected [31, 32]. There is a large variability in the reported prevalence of diabetic CAN [33], although it appears to affect a similar proportion of people with type 1 and T2DM [34]. We note that our T1DM cohort displayed several risk factors for CAN including lengthy duration of DM, suboptimal glycemic control and additional microvascular disease. We therefore contend that should we have underestimated the number of subjects with CAN, both T1DM and T2DM groups should have been affected equally. This is not supported by our findings of reduced exercise capacity only in the T2DM group and not in those with T1DM.

Wilson et al. [35, 36] recently reported an almost identical reduction in HR reserve in T2DM subjects as compared with control subjects. They investigated whether this may be attributable to reduced β-adrenoreceptor sensitivity. However, the β-adrenoreceptor agonist dobutamine was associated with a greater relative increase in HR in T2DM subjects as compared with controls suggesting that factors other than β-adrenoreceptor dysfunction were responsible. It is possible that diabetes is associated with remodeling of ion-channels within the sino-atrial node that affect rate control, but this has not previously been investigated.

On the other hand, exercise conditioning may explain the differences in heart rate reserve. Habitual exercise is associated with greater heart rate reserve, mainly because of lower resting heart rate [30] consistent with the observations in this study. Regular exercise programs, including high intensity interval training (HIIT), have been repeatedly shown to improve a range of cardiovascular measures and outcomes in people with DM, including lower resting heart rates and improved heart rate reserve [37‐44]. Thus, it is reasonable to argue that the observation in our study that heart rate reserve was reduced in T2DM, but not in T1DM, may be explained by prior exercise conditioning rather than by diabetes itself.

The most definitive demonstration that cardiac function did not differ between DM subjects and matched controls was the demonstration that cardiac volumes changed similarly during exercise. There were no differences in SVI, LVEDVI or LVESVI during incremental exercise between T2DM or T1DM and their respective control groups. Resting LVEDVI was lower in T2DM subjects and remained so throughout incremental exercise (16% overall, P = 0.023). As illustrated in Fig. 3, the LVEDVI change mirrored that of controls (interaction of exercise and group, P = 0.96) suggesting that there was no impairment in diastolic filling. This is consistent with a recent study by Wilson et al. [35] who also found no difference between changes in LVEDV during exercise in T2DM as compared with controls. Furthermore, the pattern of change in cardiac volumes during exercise in our study are the same as those described for normal physiology using exercise echocardiography [45] and exercise cardiac magnetic resonance imaging [9, 46]. We contend that the anatomy of the LV itself (smaller LVEDV) may be the major factor contributing to attenuated cardiac reserve during exercise, rather than impaired diastolic filling of a small LV cavity.

It has been well demonstrated that exercise capacity correlates strongly with cardiac size (LV volumes, LV mass or whole heart volumes), both in athletic and non-athletic populations [30, 47, 48]. We previously demonstrated that in the absence of overt contractile dysfunction, differences in cardiac function during exercise were modest and thus the major determinant of stroke volume during intense exercise was stroke volume at rest [30]. In the current study, we found no evidence of LV systolic or diastolic dysfunction and, like the experience in athletes, found strong correlations between resting cardiac geometry measures (LV mass, LVEDV and RVEDA) and \(\dot{V}O_{2}\)peak. LVEDV was most strongly associated with \(\dot{V}O_{2}\)peak in the entire cohort and remained a significant independent predictor on multivariate analysis after adjusting for age, sex and BSA. Most literature to date has focused on measures of cardiac function to explain exercise capacity in diabetes. We could not identify differences in function despite comprehensive resting and exercise measures. On the other hand, we identified differences in cardiac size (LVEDV) and we found a robust correlation between LVEDV and \(\dot{V}O_{2}\)peak. This is perhaps not surprising given that cardiac output is a major determinant of exercise capacity and, in the absence of differences in cardiac function, the major determinant of stroke volume is cardiac size.

It is notable, however, that we did not find a strong relationship between LVEDV and \(\dot{V}O_{2}\)peak in all sub-groups. Whilst the association in the combined group of T1DM, T2DM and controls was strong, the correlation was diminished when comparing T2DM and controls or T1DM and controls (R = 0.51, P = 0.004 and R = 0.79, P < 0.0001, respectively). These comparisons are challenged by the limitations of sample size and also by the influence of covariates. Simple correlations do not consider the important confounders of age, gender and BSA, all of which influence \(\dot{V}O_{2}\)peak. Thus, the more robust analysis is the multivariate analysis in which LVEDV remained a significant independent predictor of \(\dot{V}O_{2}\)peak.

It is generally reported that levels of physical activity in people with T2DM are less than that of the general non-diabetic population [49, 50] and this was also true in our comparison between T2DM subjects and controls according to estimated MET-hours. More importantly, weekly exercise participation was a significant independent predictor of \(\dot{V}O_{2}\)peak on multivariate analysis.

The question arises as to how \(\dot{V}O_{2}\)peak could diminish in the setting of preserved cardiac function and how this could relate to reduced physical activity? Haykowsky et al. elegantly demonstrated that sedentary behavior was associated with reductions in \(\dot{V}O_{2}\)peak that was related to fatty infiltration and atrophy of peripheral muscle, thereby compromising peripheral O₂ diffusive metabolism [51]. Similarly, Russell et al. [52] attributed differences in exercise capacity between T2DM and controls to changes in the peripheral vasculature and muscle interface. We can only hypothesize by inference from our data as we did not assess the quality of peripheral muscle gas exchange.

T2DM is most prevalent in older adults, and aging is associated with a linear decline in exercise capacity, accelerating after age 50 [53]. Physical inactivity compounds this age-related decline in \(\dot{V}O_{2}\)peak and increases the risk of cardiovascular mortality in people with T2DM [54]. The Dallas Bed Rest and Exercise Study [55] eloquently described a loss of cardiorespiratory fitness, cardiac mass and ventricular volumes that occurred following 3 weeks of bed rest. This extreme intervention resulted in cardiovascular changes equivalent to 30 years of aging. It has also been reported that one of the strongest predictors of heart failure in older adults is their level of fitness two decades earlier [56], and thus chronic inactivity could result in cardiac atrophy and reductions in functional capacity whilst also increasing insulin resistance and diabetes risk. As summarized in Fig. 5, our observed association between less physical activity, smaller cardiac volumes and T2DM, lends weight to the premise that inactivity could be the common underlying factor causing both diabetes and exercise intolerance.