Left ventricular concentric geometry predicts incident diabetes mellitus independent of established risk factors in the general population: the Copenhagen City Heart Study

The data, analytic methods, and study materials will not be made available to other researchers for purposes of reproducing the results or replicating the procedure. The data analysed in this paper is governed by the Danish Data Protection Agency and can only be made available for any additional researcher if a formal application is made to the Danish authorities.

The Copenhagen City Heart Study is a longitudinal cohort study designed to identify and characterize risk factors for CVD in the general population. The study population is based in and around the city of Copenhagen, Denmark. The population has been previously described in extensive detail [14‐17]. This echocardiographic sub study included all participants who had an echocardiographic exam including tissue Doppler imaging performed in the 4th round of examination, conducted from 2001 to 2003. Whether any specific participant underwent an echocardiographic exam was independent of both health status and risk factors. Figure 1 displays the flow diagram for the study sample selection process. Initially, 2221 participants underwent echocardiography including tissue Doppler echocardiography (Fig. 1). A total of 241 participants were excluded due to prevalent diabetes (Fig. 1). Then 270 participants had to be excluded due to missing LV chamber dimensions (Fig. 1). This left 1710 participants for inclusion into the study (Fig. 1). In the final cohort, no participants had significant valvular disease.

Written consent was collected from all participants. The study complies with the second declaration of Helsinki. The design of the study was approved by a regional scientific ethics committee.

All participants underwent a general health examination consisting of a questionnaire and a physical examination. Blood pressure readings were obtained by sphygmomanometer. Plasma cholesterol and plasma glucose levels were collected using non-fasting venous blood samples [18]. Plasma-pro-BNP (pro B-type natriuretic peptide) was measured by a processing-independent assay and abnormal values defined as pro-BNP > 150 pmol/L. Prevalent DM was defined as either a plasma glucose level ≥ 11.1 mmol/L or HbA1c ≥ 7.0% or the use of glucose-lowering medication [19]. Prevalent ischemic heart disease (IHD) was defined as either a history of admission for acute coronary artery occlusion, percutaneous coronary intervention, coronary artery bypass grafting or major electrocardiogram alterations per Minnesota codes 1.1 to 3.

Echocardiographic examinations were carried out by 3 experienced sonographers using Vivid 5 Ultrasound machines (GE Healthcare, Horten, Norway) with a 2.5 MHz transducer. All participants were subject to both conventional 2D-imaging and color tissue Doppler imaging (TDI). Echocardiograms were stored on magneto-optical disks and on an external FireWire harddrive (Lacie, France). All echocardiograms underwent offline analysis by experienced investigators blinded to all clinical data and information about outcomes using commercially available software (EchoPac, GE Healthcare, Horten, Norway).

Wall motion score indexing (WMSI) was determined using the 16-segment model as directed by the American Society of Echocardiography [20] with subsequent estimation of the left ventricular ejection fraction (LVEF) from the WMSI-score. An M-mode still frame perpendicular to the long axis of the LV at the level of the mitral valve leaflets was used to determine LV diastolic dimensions and LV wall dimensions. Left ventricular mass index (LVMI) was derived by indexation of LV mass with body surface area (BSA). Relative wall thickness (RWT) was calculated by indexation of the posterior wall thickness (PWT) with the left ventricular internal diameter at end diastole (LVIDd) (RWT = 2 * PWT/LVIDd) [20]. LV hypertrophy categories were defined from RWT and LVMI as directed in contemporary guidelines [20]. LV concentric geometry was defined as either LV concentric remodelling (RWT > 0.42 and LVMI ≤ 115 for men or RWT > 0.42 and LVMI ≤ 95 for women) or LV concentric hypertrophy (RWT > 0.42 and LVMI ≥ 115 for men or RWT > 0.42 and LVMI ≥ 95for women [20]. Average wall thickness was defined as the sum of interventricular septum diameter (IVSD) and PWT divided by 2 ((IVSD + PWT)/2). Mitral valve inflow patterns were determined using pulsed wave Doppler in the apical position between the mitral valve leaflet tips. From here, peak early (E) and late (A) inflow velocity was measured and E/A ratio was calculated. Furthermore, the deceleration time of the E wave (DT) was measured.

Color TDI velocity tracings were used to determine the peak longitudinal systolic velocity (s′), the early peak longitudinal diastolic velocity (e′) and the late longitudinal peak diastolic velocity (a′). In the 4-chamber view, the TDI range gate was placed in the septal and lateral position of the mitral annulus with subsequent determination of peak velocities from the TDI velocity tracings. Thus, s′, e′ and a′ were calculated as the averages of the septal and lateral values. Aortic valve timings were determined using a 2–4 cm straight M-mode line through the septal half of the mitral leaflet in the color TDI 4-chamber view [15].

Participants were followed from echocardiographic examination in 2001 to 2003 until time to event or, in the case of no event, until October 2014. The end-point of this study was incident DM. Follow-up data was collected through the Danish National Board of Health’s National Patient Registry using International Classification of Diseases, 10th revision codes (ICD). DM was defined as ICD-10 codes DE10-DE14.

All statistical analysis was done using STATA 13 for Mac OS. A p-value < 0.05 was considered statistically significant. In Table 1, continuous variables exhibiting Gaussian distribution were compared using the two-tailed students t-test. Continuous variables not displaying Gaussian distribution were compared using the Mann–Whitney U-test. Proportions were compared using the Chi squared test. In Table 2, univariable and multivariable Cox regression was used to assess prognostic strength of examined parameters. The extent of the multivariable analysis was limited by the relatively small number of events [21]. In Fig. 2, survival curves were constructed using the Kaplan–Meier method. In Fig. 3, Poisson cubic spline regression was used to estimate incidence rates of DM using RWT as a continuous measure of LV concentricity. In Fig. 4, Cox cubic spline regression was used to estimate the hazard ratio associated with increasing degree of LV concentricity evaluated as RWT. In Figs. 3 and 4 the number of knots were selected by calculating the Akaike Information Criterion (AIC) for each model with subsequent selection of the model displaying the lowest value. In Fig. 5 we studied the relationship between the degree of LV concentricity evaluated as RWT and diastolic function as quantified by E/e′ and e′ using restricted cubic spline regression. For the number of knots we chose the best fitting model as determined by the lowest AIC. We used Net Reclassification Analysis [22] and Integrated Discrimination Improvement Analysis [22] to evaluate the incremental prognostic value of LV concentric geometry and RWT in predicting DM over a model of established risk factors for DM. This model was limited by the number of events (n = 55) in the final multivariable model and hence the following risk factors were chosen: age, sex, systolic blood pressure, total cholesterol, triglycerides, HbA1c levels, and BMI. Finally, we assessed whether treating death as a competing event in competing risk Cox regressions changed our results (Additional file 1: Table S1).

Median follow-up time was 12.6 years (IQR: 12.0–12.8 years). Follow-up was a 100%. End-point was incident DM. A total of 55 participants (3.3%) developed DM during follow-up.

Participants who developed DM were significantly older and displayed higher values of systolic blood pressure (BP), diastolic BP and mean arterial pressure (Table 1). Also, they displayed higher values of body mass index and heart rate, and the prevalence of IHD was higher among participants who developed DM (Table 1). Furthermore, they displayed significantly higher values of blood glucose, HbA1c levels and blood triglycerides (Table 1).

Participants who developed DM displayed higher values of LVMI, PWT, average wall thickness, RWT, IVSD, A, E/e′, DT and a′ (Table 1). Individuals who developed DM displayed significantly lower values of E/A, s′ and e′ (Table 1). The prevalence of LV concentric remodelling, LV concentric hypertrophy and LV concentric geometry was significantly higher in participants who developed DM during follow-up (Table 1). Notably, at baseline, the prevalence of LV concentric geometry was twice as high (20% vs 42%, p < 0.001) in individuals who developed DM during follow-up.

Participants with concentric LV geometry were characterized by higher blood pressure, higher body mass index and higher heart rate (Table 2). Blood levels of glucose, triglycerides, total cholesterol and pro-BNP were also significantly higher in participants with concentric LV geometry (Table 2). They also had significantly higher LVMI, PWT, average wall thickness, RWT, IVSD, A, E/e′, DT and a′ (Table 2). Finally, they displayed significantly lower values of LVIDd/height, E, E/A, s′ and e′ (Table 2).

In univariable Cox regression, LV concentric geometry, LVMI, IVSD, PWT, average wall thickness, RWT, A, E/A E/e′, DT, e′ and a′ were significant predictors of incident DM (Table 3). In univariable Cox regression, individuals with LV concentric geometry exhibited an approx. 3 times greater risk of developing DM compared to participants without LV concentric geometry (Fig. 2). Both LV concentric remodelling and LV concentric hypertrophy were associated with an increased risk of developing DM, and no statistically significant difference in risk was present between the two groups (LV concentric remodelling: HR 3.22, 95% CI 1.74–5.98, p < 0.001) (LV concentric hypertrophy: HR 3.61, 95% CI 1.48-8.80, p = 0.005) (p for difference: p = 0.81). In univariable Cox regression interaction analysis, the association between LV concentric geometry or RWT and DM was not significantly modified by sex (LV concentric geometry, p for interaction: p = 0.19) (RWT, p for interaction: p = 0.88). Additionally, the risk of developing DM during follow-up increased exponentially with increasing degree of LV concentricity evaluated as RWT (Figs. 3 and 4).

In a multivariable model (Model 1, Table 3) adjusting for age, sex, hypertension, smoking status, total cholesterol levels, blood triglycerides, BMI, blood glucose, HbA1c levels, pro-BNP, prevalent IHD and prevalent HF, only A, RWT and LV concentric geometry remained independent predictors of DM (Table 3). In a final multivariable model (Model 2, Table 3) adjusting for the same variables as Model 1 with further adjustment for A, only RWT and LV concentric geometry remained independent predictors of DM (Table 3). Additional adjustment of our final multivariable model for either blood pressure lowering or lipid lowering medication did not attenuate the prognostic value of RWT and LV concentric geometry (Lipid lowering: Concentric geometry HR 1.83, 95%CI 1.01–3.30, p = 0.046; RWT HR 1.35, 95%CI 1.02–1.78, per 0.1 increase, p = 0.037) (blood pressure lowering: concentric geometry HR 1.98, 95%CI 1.10–3.56, p = 0.022; RWT HR 1.43, 95%CI 1.07–1.90, per 0.1 increase, p = 0.014). The results were similar in competing risk regression treating death as a competing event (Additional file 1: Table S1).

In restricted cubic spline regression LV concentricity as evaluated by RWT was significantly associated with diastolic function (Fig. 5). Diastolic function evaluated as e′ and E/e′ declined with increasing degree of LV concentricity (Fig. 5).

We assessed the incremental prognostic value of myocardial indices derived from echocardiography when added to already established risk factors for DM (age, sex, total cholesterol, triglycerides, haemoglobin A1C levels, systolic blood pressure, and BMI). In reclassification analysis, only RWT and LV concentric geometry provided incremental prognostic value when added to established risk factors (LV concentric geometry: Continuous NRI 0.429, 95% CI 0.023–0.680; IDI 0.010, 95%CI 0.000–0.033) (RWT: continuous NRI 0.431, 95% CI 0.040–0.760; IDI 0.008, 95%CI 0.000–0.037).

The notion of a true and distinct diabetic cardiomyopathy has been around for a long time, dating all the way back to the 1950s when Lundbæk, an internist in Denmark, observed frequent myocardial dysfunction in elderly patients with DM [23]. He suggested that DM patients may have heart disease in the absence of coronary blockage [23]. He became the first to suggest the existence of a specific diabetic form of cardiomyopathy. Now, two distinct phenotypes of diabetic cardiomyopathy have been suggested: an eccentric, dilated form with reduced systolic function (HFrEF phenotype), and a restricted, concentric form characterized by diastolic dysfunction (HFpEF phenotype) [13]. The HFpEF phenotype is the one most often encountered by clinicians [13]. Several cross-sectional studies have demonstrated impaired diastolic function and increased LV mass in DM individuals [4, 5, 7, 24]. In our study, diastolic parameters (A, DT, E/A, E/e′, e′, a′) were impaired and LVMI was increased in participants who developed DM during follow-up, supporting the findings of previous studies. However, only peak A-wave remained significant after adjustment for clinical characteristics, suggesting that the prognostic value of the other diastolic parameters was secondary to associations with clinical characteristics. In early diastolic dysfunction and with increasing age along with subtle wall hypertrophy, the A wave increases in magnitude due to increased atrial contribution to LV filling to compensate for the decreased inflow during early diastole (E wave), often referred to as “atrial kick” [25]. However, peak A-wave did not remain significant in a final model adjusting for clinical characteristics and LV concentric geometry. In this model, only LV concentric geometry and RWT were independent predictors of outcome, suggesting that the significance of A was secondary to clinical characteristics and the degree of LV concentricity. It is widely recognized that cardiac concentric remodelling can contribute to diastolic dysfunction, and thus our results support the hypothesis that diastolic parameters are associated with DM secondary to LV concentricity. A recent study by our group, examining cardiac function in a population of DM individuals, found that individuals with DM were mainly characterized by increased LV concentricity evaluated as RWT due to increased LV wall thickness and smaller LV cavities [7]. In the same study, we found that the severity of LV concentricity evaluated by RWT correlated significantly to the duration of DM [7]. Similar relationships between DM and LV concentricity have been found by other authors: In a recent cross-sectional study of type 2 DM patients, De Jong et al. [26] reported that LV concentricity and RWT were significantly increased in both obese and non-obese type 2 DM patients as compared to metabolically healthy obese patients, even in the absence of hypertension. Roberts et al. [27] compared the exercise capacity of type 2 DM patients to that of healthy age and sex-matched controls and found that increased sedentary behaviour and reduced LVIDd were significantly associated with reduced exercise capacity in patients with type 2 DM (in addition, type 2 DM patients also displayed increased RWT as compared to healthy controls). These findings concur with our results, since LV concentric geometry and RWT were the only echocardiographic predictors of incident DM in the present study, and since the risk of DM increased continuously with increasing degree of LV concentricity as evaluated by RWT, suggesting that LV concentricity is indeed correlated to abnormal glucose metabolism.

The finding that the prognostic value of diastolic parameters is secondary to indices of LV concentricity is supported by the suspected pathophysiological mechanisms underlying diabetic cardiomyopathy. It is thought, in the HFpEF phenotype, that hyperglycaemia, lipotoxicity and insulin resistance cause cardiomyocyte hypertrophy and increased resting tension, along with collagen deposition between cells [13]. This leads to LV wall thickening with a concomitant increase in stiffness and decrease in diastolic function. These pathophysiological considerations are reflected in our results, since participants with concentric LV geometry had significantly higher levels of blood glucose, blood triglycerides and cholesterol levels. They also had significantly higher body mass index and blood pressure and lower estimated glomerular filtration rates. These are all risk factors for CVD [28‐31], and they contribute significantly to the increased risk of CVD seen in DM [32]. Thus, the degree of LV concentricity appears to scale with the degree of dyslipidaemia, hyperglycaemia, hypertension, and body mass index even before development of DM, supporting its prognostic value in predicting DM.

Conventionally, it has been thought that a clear temporal relationship existed between cardiovascular risk factors and the development of end organ damage such as LV diastolic dysfunction, LV hypertrophy and LV systolic impairment. However, it has recently been shown that LV hypertrophy and arterial stiffness, markers of end organ damage, are predictors of incident hypertension in both normotensive and prehypertensive individuals [33, 34]. This raises questions about the temporal relationship between cardiovascular risk factors and end organ damage, indicating that end organ damage may contribute with novel prognostic value. Besides our study, one study has previously investigated the prognostic role of echocardiography with respect to development of DM in the general population. In this study, Park and Colleagues investigated the prognostic value of echocardiography in predicting incident type 2 DM in a cohort of 1817 non-diabetic individuals from Korea [9]. The age of their population was lower (mean 54 years vs mean 57.9 years) and the follow-up was shorter (6 years vs median 12.6 years) when compared to our study. Also, the prevalence of hypertension in their sample was much lower (22.1% vs 38.7%) than in our sample indicative of a younger and healthier population. During follow-up 273 individuals (15%) developed type 2 DM, which is higher than what we found in our study. This may be explained by the lower age of their study sample. Due to the higher age of participants in our study many had already developed DM at baseline and were therefore excluded. Park and Colleagues found that markers of diastolic dysfunction (e′ and E/e′) predicted incident DM. After adjustment for clinical risk factors, only e’ and the presence of diastolic dysfunction (yes/no) were independent predictors of DM. No results regarding the prognostic value of LV concentric geometry or RWT in predicting DM in the final multivariable model are shown, however, RWT was significantly higher in individuals who developed type 2 DM during follow-up (0.39 SD 0.07 vs 0.36 SD 0.06). This suggests that LV concentric geometry may have been an independent predictor of DM in their study as well, however no results regarding LV concentric geometry was reported. Nevertheless, it is possible that the prognostic value of echocardiography differs between ethnicities with regards to prediction of DM. Hence more research is needed to explain the differences between our two studies. However, altered LV concentricity seems to correspond more closely to the pathophysiological mechanisms suspected to underlie the cardiac alterations seen in the HFpEF phenotype of diabetic cardiomyopathy, and our results suggest that the predictive value of many diastolic parameters in predicting DM is secondary to LV concentric geometry. To our knowledge, ours is the second study to evaluate the prognostic value of echocardiography in predicting development of DM in the general population, and therefore, our results should be viewed as exploratory and hypothesis generating. More research, by independent groups in similar populations, is needed before any significant conclusions can be drawn. In summary, we show that end organ damage can contribute with prognostic value in predicting DM in the general population, and that this organ damage precedes the development of DM.

Several limitations to this study must be acknowledged. Firstly, mean age at baseline of participants included into this study was 56.6 years (SD: 16.3 years), and thus the prevalence of DM before exclusion of diabetic individuals was 10.9% (241 individuals). These were all excluded from the study. Therefore, due to the high age of this general population sample, a large proportion had already developed DM at baseline and were therefore excluded. Furthermore, we assessed the development of DM by ICD-10 codes, and therefore we do not know how rigorously or by what methods participants have been monitored for the development of DM. Due to unknown differences in the rigour of monitoring of participants, the time from echocardiographic examination to DM diagnosis may have been overestimated, since participants may not have sought out medical consultation at the onset of symptoms from DM. This could potentially alter the temporal relationship between RWT values and DM development observed in this study. Furthermore, in the Copenhagen City Heart study, prevalent DM was defined using the old HbA1c cut-off value (HbA1c ≥ 7.0%). Today a cut-off value of HBA1c ≥ 6.5% is used [35]. Therefore, our definition of prevalent DM may have underestimated the true prevalence. Also, the general population in Denmark is mainly of Caucasian ethnicity, and therefore we cannot extrapolate our results to other races—stressing this is particularly important when considering ethnicity in itself is a significant risk factor for DM, and that ethnicity modifies the contributions of many other risk factors for cardiovascular disease in diabetic individuals [11, 12]. We did not have information on the prevalence of rare cardiomyopathies such as hypertrophic cardiomyopathy, which may affect LV geometry. However, undiagnosed cases of rare cardiomyopathies are unlikely to have affected our results given their very low prevalence in the general population. Also, since rare cardiomyopathies do not cause DM, potential inclusion of such patients would only serve to weaken our results. Yet, we still found significant associations between LV concentric geometry and DM. A final limitation is related to the method of outcome assessment in the present study. In this study, DM outcomes were assessed through the Danish National Board of Health’s National Patient Registry using ICD 10 codes. However, this means that receiving a diagnosis of DM necessitates some type of hospital contact. This hospital contact can either be directly related to DM, or it can be unrelated to the presence of DM and the presence of DM may then be discovered through patient history or examination/assessment. Given the long follow-up of our study (median 12.6 years) and the high age of participants at baseline (mean 57 years) it is reasonable to assume that most patients will have at least one hospital contact not related to a potential DM diagnosis facilitating the discovery of DM if present. Yet, this does constitute a limitation of the present study and could have affected our results. In addition, it is also possible that the DM outcomes detected may represent sicker DM patients possibly requiring hospitalisation at DM debut since outcome assessment is related to hospital contact. Thus, although our study provides important hypothesis generating results, ideally, our findings should be evaluated in multiple ethnicities, in a younger study population and using a study design including closer, dedicated outcome assessment.