Echocardiographic phenotypes of Chinese patients with type 2 diabetes may indicate early diabetic myocardial disease

Abstract

Aim

Type 2 diabetes may impair cardiac structure and function at very early stage, other factors, for example, obesity and hypertension, can induce aforementioned abnormalities individually. This study aimed to explore precise prevention and treatment of diabetic cardiomyopathy (DCM) by using cluster analysis of echocardiographic variables.

Methods and results

A total of 66 536 inpatients with diabetes from 2013 to 2018 were investigated, and 7112 patients were available for analysis after nadir. The cluster analysis was performed on echocardiographic variables to assess the clinical profiles and risk factors of clusters. Two clusters were identified. Cluster 1 with 3576 patients (50.3%, including 62.5% female) had hypertension in 62.4%, while the lower rate of obesity (13.7%). Ultrasound findings showed that 79.9% of them had left ventricular diastolic dysfunction (LVDD), the most characteristic change in the early stages of DCM. Systolic blood pressure (SBP), uric acid and antithrombin III were independent risk factors for LVDD (P < 0.0001); 64.0% of the 3536 patients in the second group were male, with a high prevalence of obesity (30.1%) and a higher prevalence of hypertension (79.5%), In particular, decreased systolic function and a high rate of LV hypertrophy (46.8%) represented the progressive phase of DCM (P < 0.0001). SBP, diastolic blood pressure, BMI and creatinine were independent correlates of LV mass index (P < 0.05).

Conclusion

The cluster analysis of echocardiographic variables may improve the identification of groups of patients with similar risks and different disease courses and will facilitate the achievement of targeted early prevention and treatment of DCM.

Introduction

In China, the prevalence of type 2 diabetes mellitus (T2DM) is increasing yearly, reaching 11.2% of the global diabetic population, ¹ whereas the prevalence values of hypertension (HTN) and obesity increase yearly. ² , ³ HTN and obesity are common complications of T2DM. Furthermore, the abnormalities in cardiac geometry and function seen in patients with early diabetic cardiomyopathy (DCM) have many similarities to those seen in patients with HTN and obesity. ² , ⁴ , ⁵ , ⁶ In addition, age, a factor in T2DM, influences the early clinical features and functional changes of DCM. ⁷ , ⁸ These causative factors can lead to changes in the ultrasound structure of the patient's heart. The specific contribution of these causative factors and their synergistic contribution to cardiac dysfunction in T2DM are unclear. Classic statistical analyses are built on the a priori hypothesis, and cluster analysis may improve cardiac phenotyping and provide new insights into heterogeneous patient groups, such as those with T2DM, by providing an innovative exploratory analysis. ⁹ , ¹⁰ Therefore, we hypothesized that cluster analysis may allow the identification of patient groups (clusters) with T2DM and common cardiac phenotypes on the basis of ultrasound data and different clinical features, providing new insights and exploring strategies for the management of diabetic cardiac changes and cardiomyopathy. ¹⁰ , ¹¹

Methods

Study population

The study subjects were patients hospitalized in ZhongDa Hospital affiliated to Southeast University from July 2013 to the end of 2018. To avoid selection bias, we included all patients consecutively and enrolled 66 536 patients with diabetes based on primary discharge diagnosis. After excluding 5411 cases of type 1 diabetes, gestational diabetes, and other specific types of diabetes, there were 61 125 cases of T2DM. For patients hospitalized multiple times during this period, only the first hospitalization data were used (n = 31 112). After excluding 23 299 patients due to significant data gap (n = 419) and cardiac ultrasound data gap (n = 22 880), the remaining 7813 patients were included. After excluding obvious rheumatic heart disease, congenital heart disease, valvular disease, coronary atherosclerotic heart disease, pericardial disease, and hyperthyroidism, 7112 patients (3601 men and 3511 women) were enrolled (Supporting Information, Figure S1 ). The procedure of this study was approved by the Research Ethics Committee of ZhongDa Hospital affiliated to Southeast University (Approved No. of ethic committee: 2020ZDSYLL028‐P01).

Clinical and biological data

Relevant physical examination, laboratory biochemical examination and echocardiography were included on the same day. Age, sex, diabetes course, smoking history, alcohol drinking history, height, weight, and drug use were collected from clinical data. The patient is given a complete physical examination, blood pressure (systolic and diastolic) is measured, and a history of diabetes is collected. The diagnostic criteria for HTN are based on the recommendations of the American College of Cardiology (ACC)/American Heart Association (AHA) and the European Society of Cardiology (ESC)/European Society of HTN (ESH) blood pressure guidelines, ¹² , ¹³ and blood pressure >140/90 mmHg measured more than two times in resting state after admission, based on medical history. The diagnostic criteria of obesity are based on the recommendations of the World Health Organization (WHO) and the Working Group on Obesity in China. ¹⁴ , ¹⁵ Obesity is defined as body mass index (BMI) ≥ 28.0 kg/m², while overweight is defined as body mass index (BMI) ≥ 24.0 kg/m².

All patients had venous blood (approximately 2–4 mL of venous blood and 1–2 mL of serum extracted) collected and laboratory parameters measured on the second day of admission Fasting blood glucose was detected by hexokinase method. HbA1c was determined by cation exchange high performance liquid chromatography. Total cholesterol (T‐CHOL) was measured by cholesterol esterase/peroxidase enzyme method, and triglyceride by lipase glycerol kinase method. High density lipoprotein cholesterol (HDL‐C) and low density lipoprotein cholesterol (LDL‐C) were determined by direct antibody separation and elimination methods, respectively. The concentrations of apolipoprotein‐A1, apolipoprotein‐B and lipoprotein‐A in serum were determined by immunoturbidimetry. Serum urea nitrogen (BUN) was determined by glutamate dehydrogenase and urease kinetic methods. Serum creatinine (SCr) and uric acid (SUA) levels were determined by sarcosine oxidase method and uricase method respectively. These indicators were measured by an automatic biochemical analyser (AU5800, Beckman Coulter, California, USA). Prothrombin time (PT), activated partial thrombin time (APTT), thrombin time (TT), and fibrinogen level were measured by coagulation method. The concentration of antithrombin III was determined by chromogenic substrate method, and the levels of fibrinogen degradation product (FDP) and D dimer were determined by turbidimetric method. All fibrinolytic functions were measured using an automatic analyser (ACLTOP 700, Beckman Coulter, California, USA). The Laboratory Center of the University Hospital implements internal and external quality management procedures directed by the China Laboratory Quality Control Center. All blood samples were analysed by clinical laboratory professionals at ZhongDa Hospital.

Echocardiogram

All echocardiographic data were obtained using commercial echocardiographic systems (Vivid E9; Ge‐vingmed, Horten, Norway). The criteria for echocardiographic measurements are based on current guidelines of the American Society of Echocardiography (ASE)/European Association of Cardiovascular Imaging (EACVI). That is, a 2.5–3.5 mhz phased array probe is used by an experienced registered sonographer. Subjects were placed in partial left lateral decubitus, and images were obtained by the observer along the long and short axis of the prosternum and apical four compartment and two compartment long axis views. All records included at least five cardiac cycles and were stored digitally for offline analysis. Offline analysis Echocardiography was analysed without knowledge of the subjects' characteristics by a registered ultrasound physician with more than 10 years of experience. EchoPac software (GE Vingmed) is used for post‐processing of digitally stored images. Mean measurements were statistically analysed over at least three cardiac cycles. Measure and record the aortic root (AO) diameter, aorta ascendens (AAO) diameter, left atrium (LA) diameter, right ventricular (RV) end‐diastolic internal diameter, interventricular septum (IVS) thickness, left ventricular (LV) end‐diastolic internal diameter, left ventricular posterior wall (LVPW) thickness, diameter of pulmonary artery (PA), the right atrium (RA) diameter, mitral valve blood flow velocity (MV) tricuspid, and aortic blood flow velocity (AV), pressure gradient (PG), pulmonary valve (PV) opening velocity, and ejection fraction (EF). Left ventricular diastolic dysfunction (LVDD) was calculated and assessed according to expert consensus guidelines of the American Society of Echocardiography and the European Society of Cardiovascular Imaging. ¹⁶

Calculation of variable

Calculation of the glomerular filtration rate (GFR) was based on the guideline consensus, following the simplified MDRD formula: GFR = 186 × (SCr/88.402)^−1.154 × age^‐0.203 × (female × 0.742). ¹⁷ , ¹⁸ Left ventricular mass (LVM) was calculated by Devereaux's LVM correction formula. ¹⁹ LVM (g) = 0.8 × 1.04 × [(IVST + LVPWT + LVDD)³ − LVDD³] + 0.6. Formula for calculating body surface area (BSA): BSA (m²) = 0.0061 × height (cm) + 0.0128 × weight (kg) − 0.1529. Left ventricular mass index (LVMi) (g/m²) = LVM/BSA. The formula for calculating the relative wall thickness (RWT): RWT = (LVST + LVPWT)/LVDD. Echocardiographic criteria for the diagnosis of LV hypertrophy using the 2015 American and European Echocardiographic Association guidelines: LVMI ≥115 g/m² in men and LVMI ≥95 g/m² in women. ²⁰ Hospitalization rates for all cardiovascular diseases (including cardiovascular medicine and cardiovascular surgery) were obtained from medical records of ZhongDa Hospital.

Cluster analysis

The initial step analysis divided patients into four clinically defined groups based on admission records and diagnostic records for the presence of HTN and/or obesity. To assess differences in baseline data, laboratory biology, and echocardiographic features among patients with T2DM. Four clinically defined groups were thus constituted of (i) patients with T2DM alone; (ii) T2DM mellitus with obesity; (iii) patients with T2DM mellitus and HTN; and (iv) patients with T2DM mellitus with obesity and HTN.

In the second step of the analysis, we first analysed the included data by removing five variables with missing values greater than 25% (PA, MV, TR, PG, and e′/a′) from the 51 included variables and interpolating the remaining 46 variables. Parameter estimation using EM (Expectation Maximization) for great likelihood was used to reduce the loss due to missing data. These datasets were combined as complete data according to Rubin's standard rules. For cardiac ultrasound‐related metrics the raw data were transformed by Z‐score normalization to meet the data standardization requirements; data that conformed to a normal distribution were taken between −3 standard deviations and +3 standard deviations depending on the data type to exclude abnormally high and low values (n = 406); Euclidean distance was used as a similarity measure and the Ward method was used. All clusters were performed by one to two researchers with no knowledge of the clinical data.

Various unsupervised analyses were performed in turn in order to more accurately identify associations between echocardiographic phenotypic features and other clinical indicators and characteristics. Firstly, based on the collated echocardiographic data, a principal component analysis was performed in order to understand the correlation between the ultrasound data and to construct a three‐dimensional principal component analysis plot describing the categorical cardiac ultrasound characteristics of the patients. Based on the results of the subsequent cluster analysis, the clustering results were evaluated using principal component analysis double‐labelled plots by assigning different colours to different groups of patients.

The best number of clusters was obtained by sequentially analysing nine combinations of different numbers of clusters, based on the consistency matrix and cluster consistency scores, according to the echocardiographic characteristics of the patients. The conclusions suggest that the clustering analysis method divided into two classes has the best consistency matrix and the highest cluster consistency score. It is suggested that cluster analysis into two categories has the least intra‐group variability and the greatest inter‐group heterogeneity.

Statistical analysis

Numerical variables are expressed as mean ± standard deviation (SD), median (interquartile range), respectively, according to the presence or absence of a normal distribution. Qualitative variables are expressed as numbers (percentages). Numerical variables with and without normal distribution are expressed as mean ± standard deviation (SD), median (interquartile range), respectively. Qualitative variables are expressed as quantities (percentages). Comparisons between two groups were made using Student's t‐test for normally distributed quantitative variables, Mann–Whitney U‐test for asymmetrically distributed quantitative variables, and χ ² test for qualitative variables. Multiple comparisons of continuous data were performed using one‐way ANOVA or Kruskal–Wallis tests, with post hoc tests for paired comparisons if overall significance existed, and Sidak correction for multiplicity of tests. A threshold of statistical significance was set at a corrected two‐sided P < 0.05. Missing data were interpolated using the EM algorithm, standardized using the Z‐score method, analysed for extreme values and cleaned for data with ≥3 standard deviations and ≤3 standard deviations.

Data were processed, descriptively analysed and compared using SPSS 26.0 software (SPSS, Inc., Chicago, IL, USA), R statistical software (4.0.5) for cluster analysis and visualization (ConsensusClusterPlus, limma, pheatmap, ggradar, scatterplot3d, and other packages), Microsoft excel (16.43.1) for case screening and Prism (8.2.1) for plotting violin plots and forest plots. The new results after clustering were data‐completed using multiple interpolation. One‐way analysis of variance was used, followed by multi‐factor regression analysis using binary unconditional logistic regression. Prevalence 95% confidence intervals (CI) were calculated based on a normal approximation of the binomial distribution. A corrected two‐sided P < 0.05 was the threshold for a statistically significant difference.

Results

Traditional classification based on risk factors (presence or absence of hypertension and/or obesity)

A total of 7112 patients with T2DM were analysed. Baseline data were summarized for patients with type 2 diabetes only (n = 1788 (25.1%)), obese patients (n = 279 (3.9%)), hypertensive patients (n = 3769 (53.0%)), and obese patients with hypertension (n = 1276 (17.9%)), the characteristics of which are shown in Table 1 . Compared with the other groups: the T2DM alone group had the lowest systolic and diastolic blood pressure, lowest Apo‐B levels, lowest triglyceride levels, lowest creatinine levels and highest glycated haemoglobin levels; the obese group was younger, had a shorter duration of diabetes, had the highest Apo‐B levels and higher LDL‐C levels; the hypertensive group was older; the obese group with HTN had higher LDL‐C levels, the highest triglyceride levels and the highest creatinine levels. There were no significant between‐group differences in the use of antiplatelet agents, statin lipid‐regulating drugs, insulin drugs and metformin in the four groups. Little variability between groups can be seen for most variables, with similar patient characteristics between groups.

Table 1.

Clinical and biological characteristics

	N	All patients	Isolated T2DM	T2DM and obesity	T2DM and HTN	T2DM,Obesity and HTN	P‐value*
		(N = 7112)	(n = 1788)	(n = 279)	(n = 3769)	(n = 1276)
Clinical characteristics
Female, n (%)	7112	3511/7112 (49.4)	845/1788 (47.3)	138/279 (49.5)	1893/3769 (50.2)	635/1276 (49.8)	0.224
Age (year)	7112	67.5 ± 10.23, n = 7112	66.05 ± 10.54, n = 1788	61.76 ± 11.28†, n = 279	69.04 ± 9.53†‡, n = 3769	66.24 ± 10.65‡§, n = 1276	<0.0001
Diabetes duration (year)	6014	8 (6, 9), n = 6014	8 (7, 9), n = 1484	7 (6, 8)†, n = 218	8 (7, 9)†‡, n = 3238	8 (7, 9)‡§, n = 1074	<0.0001
Hypertension, n (%)	7112	5045/7112 (70.9)	‐	‐	3769/3769 (100.0)	1276/1276 (100.0)	‐
SBP (mmHg)	6382	136.3 ± 19.4, n = 6382	127.2 ± 16.1, n = 1633	129.9 ± 15.6†, n = 255	139.6 ± 19.4†‡, n = 3362	141.0 ± 19.6†‡, n = 1132	<0.0001
DBP (mmHg)	3880	78.0 ± 11.6, n = 3880	74.6 ± 10.1, n = 916	77.9 ± 10.2†, n = 150	78.6 ± 11.9†, n = 2145	80.5 ± 11.7†‡, n = 669	<0.0001
Smoking, n (%)	7028	1558/7028 (22.2)	390/1766 (22.1)	71/278 (25.5)	795/3720 (21.4)	302/1264 (23.9)	0.146
Drinking, n (%)	7026	922/7026 (13.1)	217/1765 (12.3)	32/278 (11.5)	478/3719 (12.9)	195/1264 (15.4)	0.05
BMI (kg/m2)	7112	25.24 ± 3.76, n = 7112	23.17 ± 2.77, n = 1788	30.00 ± 2.11, n = 279	24.09 ± 2.5, n = 3769	30.49 ± 2.49, n = 1276	‐
Obesity (≥28.0 kg/m2)	7112	1555/7112 (21.9)	‐	279/279 (100.0)	‐	1276/1276 (100.0)	‐
Metabolic factor
Apolipoprotein‐A1 (g/L)	6564	1.09 ± 0.28, n = 6564	1.1 ± 0.31, n = 1619	1.07 ± 0.26, n = 259	1.09 ± 0.28, n = 3501	1.09 ± 0.26, n = 1185	0.153
Apolipoprotein‐B (g/L)	6564	0.84 ± 0.25, n = 6564	0.82 ± 0.25, n = 1619	0.87 ± 0.25†, n = 259	0.83 ± 0.25, n = 3501	0.86 ± 0.25†§, n = 1185	<0.0001
T‐CHO (mmol/L)	6710	4.6 ± 1.27, n = 6710	4.54 ± 1.26, n = 1656	4.74 ± 1.18†, n = 262	4.58 ± 1.28, n = 3590	4.7 ± 1.28†§, n = 1202	0.001
HDL‐C (mmol/L)	6564	1.15 ± 0.3, n = 6564	1.17 ± 0.34, n = 1619	1.12 ± 0.26†, n = 259	1.15 ± 0.29†, n = 3501	1.14 ± 0.28†, n = 1185	0.003
LDL‐C (mmol/L)	6564	2.84 ± 0.96, n = 6564	2.79 ± 0.95, n = 1619	2.98 ± 0.94†, n = 259	2.82 ± 0.96, n = 3501	2.94 ± 0.97†§, n = 1185	<0.0001
Triglyceride (mmol/L)	6709	1.52 (1.06, 2.22), n = 6709	1.20 (0.82, 1.70), n = 1656	1.40 (1.00, 1.88), n = 262	1.39 (1.00, 2.01)†‡, n = 3589	1.60 (1.15, 2.29)†§, n = 1202	<0.0001
Lipoprotein‐a (mg/L)	6564	187 (101, 334), n = 6564	187 (194, 299), n = 1619	174 (122, 319), n = 259	210 (117, 352)†, n = 3501	183 (114, 330)§, n = 1185	0.006
BUN (mmol/L)	6940	5.6 (4.5, 7.1), n = 6940	5.5 (4.3, 6.9), n = 1734	5.2 (4.3, 6.4), n = 270	5.9 (4.6, 8.2)†‡, n = 3691	5.6 (4.5, 7.1)†‡, n = 1245	<0.0001
SCr (μmol/L)	7034	76 (63, 93), n = 7034	72 (58, 86), n = 1763	79 (69, 88), n = 275	79 (65, 104)†‡, n = 3735	83 (64, 105)†‡, n = 1261	<0.0001
SUA (μmol/L)	7034	309.92 ± 106.98, n = 7034	281.74 ± 99.88, n = 1763	303.76 ± 105.95†, n = 275	314.63 ± 107.09†, n = 3735	336.71 ± 107.7†‡§, n = 1261	<0.0001
FBG (mmol/L)	7034	8.25 (6.47, 11.49), n = 7034	8.23 (6.15, 11.10), n = 1763	9.60 (7.00, 11.44), n = 275	7.80 (6.08, 10.92)†‡, n = 3735	8.09 (6.28, 11.01), n = 1261	0.001
HbA1c (%)	5749	8.03 ± 2.07, n = 5749	8.42 ± 2.31, n = 1439	8.08 ± 1.85, n = 230	7.89 ± 2†, n = 3020	7.93 ± 1.9†, n = 1060	<0.0001
Blood clotting index
PT (s)	6415	11.1 (10.4, 11.9), n = 6415	11.2 (10.5, 12.1), n = 1610	11.3 (10.6, 12.2), n = 250	11.1 (10.4, 11.9)†, n = 3414	11.0 (10.2, 11.8)†, n = 1141	<0.0001
APTT (s)	6415	27.2 (29.7, 32.3), n = 6415	30.6 (28.3, 32.7), n = 1610	31.1 (27.5, 33.2), n = 250	30.3 (28.0, 32.9), n = 3414	29.8 (27, 32.6), n = 1141	0.341
TT (s)	6411	15.31 ± 3.17, n = 6411	15.49 ± 2.9, n = 1609	15.51 ± 3.96, n = 248	15.24 ± 3.31, n = 3413	15.22 ± 2.86, n = 1141	0.034
INR	6406	1.05 ± 0.19, n = 6406	1.06 ± 0.21, n = 1606	1.03 ± 0.11, n = 250	1.04 ± 0.17†, n = 3411	1.04 ± 0.23†, n = 1139	0.001
ATIII (%)	6411	96 (87, 110), n = 6411	97 (82, 109), n = 1608	89 (78, 109), n = 250	95 (84, 107)†, n = 3412	94 (83, 108), n = 1141	0.01
FDP (mg/L)	6392	1.6 (0.9, 2.97), n = 6392	1.5 (0.82, 3.1), n = 1603	1.39 (0.9, 2.85), n = 249	1.69 (0.95, 3.1)†, n = 3402	1.51 (0.88, 2.52)§, n = 1138	<0.0001
Fibrinogen (g/L)	6394	3.76 ± 0.92, n = 6394	3.69 ± 0.95, n = 1605	3.71 ± 0.93, n = 249	3.81 ± 0.91, n = 3402	3.74 ± 0.9, n = 1138	0.421
D dimer (μg/L)	6072	128 (38, 340), n = 6072	121 (11, 363), n = 1541	109 (11, 279), n = 237	138 (48, 352)†, n = 3220	119 (40, 283)§, n = 1074	0.001
Medications
Antiplatelet user, n (%)	6322	2599/6322 (41.1)	609/1594 (38.2)	94/245 (38.4)	1419/3345 (42.4)†	477/1138 (41.9)	0.029
Statin user, n (%)	6323	2237/6323 (35.4)	539/1592 (33.9)	94/247 (38.1)	1201/3346 (35.9)	403/1138 (35.4)	0.427
Insulin user, n (%)	6337	4291/6337 (67.7)	1126/1596 (70.6)	174/247 (70.4)	2247/3352 (67.0)†	744/1142 (65.1)†	0.012
Metformin user, n (%)	6329	1640/6329 (25.9)	418/1595 (26.2)	69/246 (28)	854/3348 (25.5)	299/1140 (26.2)	0.801

Values are mean ± SD, median (interquartile range), or n/N (% of non‐missing data). Bold results are statistically significant at the P < 0.05 level. Sidak correction: †P < 0.05 compared with T2DM; ‡P < 0.05 compared with T2DM and obesity; §P < 0.05 compared with T2DM and HTN.

APTT, activated partial thromboplastin time; ATIII, Antithrombin III; BUN, blood urea nitrogen; DBP, diastolic blood pressure; FBG, fasting blood‐glucose; FDP, fibrinogen degradation product; HbA1c, glycosylated haemoglobin; HDL‐C, high‐density lipoprotein cholesterol; HTN, hypertension; INR, international normalized ratio; LDL‐C, low‐density lipoprotein cholesterol; PT, prothrombin time; SBP, systolic blood pressure; SCr, serum creatinine; SUA, serum uric acid; T‐CHO, total cholesterol; T2DM, type 2 diabetes mellitus; TT, thrombin time.

Compares the echocardiographic data (Table 2 ) and calculated data (Table 3 ) among the four groups. Compared with the T2DM group alone and the obese with HTN group, the obese and hypertensive groups showed significant differences in aorta, aorta ascendens, left atrium, right ventricle, IVS, left ventricle, LVPW, pulmonary artery, right atrium mitral valve, TR, PG, pulmonary valve (m/s), E/A, e′/a′ were significantly different (P < 0.05). While the obese group and the hypertensive group, including aorta, aorta ascendens, left atrium, IVS, left ventricle, pulmonary artery, right atrium, aortic valve, PG, pulmonary valve, ejection fraction had significant overlap and were not significantly different (P > 0.05). LVMi and RWT levels were lower in T2DM mellitus group and obesity group, but higher in HTN group and obesity group with HTN, and the difference was statistically significant (P < 0.05). EGFR levels were higher in the simple T2DM group and the obesity group, but lower in the HTN group and the obesity group with HTN, and the difference was statistically significant (P < 0.05). The incidence of Left ventricular hypertrophy (LVH) was lower in T2DM mellitus group and obesity group, but higher in HTN group and obesity with HTN group (P < 0.05). Despite significant differences between mean values, an important overlap of individual values of echocardiographic parameters was observed among the four groups (Figure 1 ).

Table 2.

Echocardiographic variables

	N	All patients	Isolated T2DM	T2DM and obesity	T2DM and HTN	T2DM,Obesity and HTN	P‐value*
		(N = 7112)	(n = 1788)	(n = 279)	(n = 3769)	(n = 1276)
Aorta (cm)	6054	2.63 ± 0.48, n = 6054	2.57 ± 0.45, n = 1532	2.69 ± 0.43†, n = 239	2.63 ± 0.48†, n = 3188	2.69 ± 0.49†§, n = 1095	<0.0001
Aorta ascendens (cm)	5, 986	3.3 ± 0.38, n = 5986	3.21 ± 0.35, n = 1497	3.28 ± 0.42†, n = 225	3.31 ± 0.38†, n = 3176	3.4 ± 0.39†‡§, n = 1088	<0.0001
Left atrium (cm)	6126	3.72 ± 0.49, n = 6126	3.53 ± 0.47, n = 1532	3.75 ± 0.44†, n = 241	3.73 ± 0.48†, n = 3240	3.94 ± 0.47†‡§, n = 1113	<0.0001
Right ventricle (cm)	5980	2.36 ± 0.27, n = 5980	2.3 ± 0.27, n = 1492	2.41 ± 0.35†, n = 238	2.35 ± 0.27†‡, n = 3184	2.44 ± 0.26†§, n = 1066	<0.0001
IVS (cm)	6323	1.05 ± 0.16, n = 6323	0.98 ± 0.13, n = 1589	1.03 ± 0.23†, n = 244	1.06 ± 0.15†, n = 3353	1.1 ± 0.16†‡§, n = 1137	<0.0001
Left ventricle (cm)	6156	4.55 ± 0.48, n = 6156	4.44 ± 0.45, n = 1544	4.63 ± 0.43†, n = 235	4.55 ± 0.47†, n = 3291	4.71 ± 0.49†§, n = 1086	<0.0001
LVPW (cm)	6329	1.01 ± 0.14, n = 6329	0.96 ± 0.12, n = 1596	1.00 ± 0.12†, n = 247	1.02 ± 0.14†‡, n = 3343	1.07 ± 0.15†‡§, n = 1143	<0.0001
Pulmonary artery (cm)	4842	2.36 ± 0.33, n = 4842	2.29 ± 0.3, n = 1204	2.41 ± 0.31†, n = 177	2.36 ± 0.32†, n = 2576	2.47 ± 0.34†§, n = 885	<0.0001
Right atrium (cm)	6016	3.80 (3.53, 4.19), n = 6016	3.70 (3.47, 4.00), n = 1504	3.83 (3.49, 4.27)†, n = 246	3.80 (3.54, 4.20)†, n = 3186	3.95 (3.66, 4.25)†‡§, n = 1080	<0.0001
Mitral valve (m/s)	755	0.71 ± 0.26, n = 755	0.67 ± 0.2, n = 185	0.89 ± 0.54†, n = 21	0.71 ± 0.21‡, n = 418	0.73 ± 0.37‡, n = 131	0.001
Aortic valve (m/s)	6071	1.24 (1.10, 1.40), n = 6071	1.20 (1.07, 1.36), n = 1530	1.20 (1.08, 1.37), n = 230	1.27 (1.12, 1.43)†‡, n = 3209	1.28 (1.10, 1.43)†‡, n = 1102	<0.0001
TR	2104	2.60 (2.40, 2.85), n = 2104	2.55 (2.32, 2.80), n = 533	2.67 (2.38, 2.80), n = 70	2.60 (2.44, 2.90)†‡, n = 1156	2.61 (2.40, 2.84)†, n = 345	<0.0001
PG	2217	28.67 ± 9.2, n = 2217	27.8 ± 9.21, n = 556	26.91 ± 7.24, n = 70	29.3 ± 9.6†, n = 1230	28.21 ± 7.92, n = 361	0.003
Pulmonary valve (m/s)	6059	0.97 ± 0.34, n = 6059	0.93 ± 0.54, n = 1538	0.95 ± 0.16, n = 225	0.98 ± 0.26†, n = 3202	0.98 ± 0.21†, n = 1094	<0.0001
Ejection fraction (%)	5591	0.68 ± 0.07, n = 5591	0.68 ± 0.06, n = 1395	0.68 ± 0.06, n = 228	0.68 ± 0.07, n = 2946	0.68 ± 0.06, n = 1022	0.313
E/A ≤ 1, n (%)	5784	5319/5784 (92.0)	1207/1397 (86.4)	186/213 (87.3)	2957/3132 (94.4)†‡	969/1042 (93.0)†‡	<0.0001
e′/a′<1, n (%)	3749	3566/3749 (95.1)	843/917 (91.9)	116/135 (85.9)†	1965/2031 (96.8)†‡	642/666 (96.4)†‡	<0.0001

Values are mean ± SD, median (interquartile range), or n/N (% of non‐missing data). Bold results are statistically significant at the P < 0.05 level. Sidak correction: †P < 0.05 compared with T2DM; ‡P < 0.05 compared with T2DM and obesity; §P < 0.05 compared with T2DM and HTN.

A, peak late diastolic velocity; e′, mitral annular early diastolic velocity; E, peak early diastolic velocity; HTN, hypertension; IVS, interventricular septum; LVPW, left ventricular posterior wall; PG, pressure gradient; T2DM, type 2 diabetes mellitus; TR, tricuspid regurgitation.

Table 3.

Computed variables

	N	All patients	Isolated T2DM	T2DM and obesity	T2DM and HTN	T2DM, obesity, and HTN	P‐value*
		(N = 7112)	(n = 1788)	(n = 279)	(n = 3769)	(n = 1276)
LVMi (g/m2)	5467	96.67 ± 24.71, n = 5467	88.62 ± 19.74, n = 1359	89.34 ± 24.50, n = 212	99.85 ± 25.73†‡, n = 2915	99.94 ± 25.02‡, n = 981	<0.0001
RWT	5467	0.46 ± 0.11, n = 5467	0.44 ± 0.10, n = 1359	0.44 ± 0.08, n = 212	0.46 ± 0.13†‡, n = 2915	0.46 ± 0.07†‡, n = 981	<0.0001
eGFR (mL/min/1.73 m2)	7034	83.12 ± 33.77, n = 7034	91.83 ± 32.23, n = 1763	89.81 ± 31.65, n = 275	79.38 ± 33.76†‡, n = 3735	80.89 ± 33.99†‡, n = 1261	<0.0001
LVH, n (%)	5467	785/5467 (14.4)	91/1359 (6.7)	10/212 (4.7)	515/2915 (17.7)†‡	169/981 (17.2)†‡	<0.0001

Values are mean ± SD or n/N (% of non‐missing data). Bold results are statistically significant at the P < 0.05 level. Sidak correction: †P < 0.05 compared with T2DM; ‡P < 0.05 compared with T2DM and obesity; §P < 0.05 compared with T2DM and HTN.

eGFR, estimated glomerular filtration rate; HTN, hypertension; LVH, left ventricular hypertrophy; LVMi, left ventricular mass index; RWT, relative wall thickness; T2DM, type 2 diabetes mellitus.

Figure 1.

Echocardiographic morphological and functional parameters. The distribution of each index parameter in the echocardiogram shows significant overlap in the groups of values when plotted according to the presence or absence of obesity and/or hypertension in the patient. The violin plot shows the median (horizontal midline) and interquartile distance (upper and lower horizontal lines), with the width indicating the amount of data. †P < 0.05 compared with T2DM; ‡P < 0.05 compared with T2DM and obesity; §P < 0.05 compared with T2DM and HTN. DM, diabetes mellitus; HTN, hypertension; IVS, interventricular septum; LVMi, left ventricular mass index; LVPW, left ventricular posterior wall.

Cluster analysis of echocardiographic phenotypes

Echocardiographic variables identified three clusters of cardiac phenotypes (Figures 2 A , 2 B , 3 , and 4 ).

Figure 2.

(A) Consistency matrix plot for cluster analysis. Consistency matrix plots are shown for the cluster analysis using categories 1 to 9, respectively. A tree diagram is also used to illustrate the combinations between the different variables and the distances between the groups. The size of the matrix plot area represents the sample size. (B) Consistency scores for cluster analysis. The horizontal coordinate is the number of clusters and the vertical coordinate is the cluster consistency score.

Figure 3.

Heat map of a scheme based on echocardiographic variables for two classification. Red indicates an increase in standardized values expressed in SD; blue indicates a decrease in standardized values expressed in SD. The horizontal coordinates represent different patients and the vertical coordinates represent different echocardiographic variables. The bar attached at the top of the figure identifies and colours two clusters: Cluster 1 (blue), Cluster 2 (red). IVS, interventricular septum; LVPW, left ventricular posterior wall.

Figure 4.

Three‐dimensional principal component analysis graph. Visual representations were used to construct relationships between cluster variables while displaying patients (points) based on their individual echocardiographic characteristics. The results were projected onto a three‐dimensional principal component analysis plot. The colours observed correspond to the two‐group solutions from the cluster analysis, suggesting a very clear delineation between individuals of the clustering solutions based on the two classifications of the patient's echocardiographic characteristics.

Analysis of cardiac phenotypes and clinical phenotypes according to patient clusters

After clustering the two groups identified according to the echocardiographic phenotype data, the echocardiographic data (Table 4 ) and clinical characteristics data (Table 5 ) of the two clusters were further compared, and it was evident that there was significant heterogeneity between the echocardiographic and clinical characteristics groups of the two clustered patients (Figure 5 ).

Table 4.

Cardiac phenotypes according to patient clusters

	N	Cluster 1	Cluster 2	P‐value*
		(N = 3507)	(n = 3199)
Aorta (cm)	6054	2.51 ± 0.43, n = 3008	2.74 ± 0.45, n = 2690	<0.0001
Aorta ascendens (cm)	5986	3.16 ± 0.33, n = 2978	3.41 ± 0.33, n = 2653	<0.0001
Left atrium (cm)	6126	3.43 ± 0.34, n = 3036	3.98 ± 0.36, n = 2731	<0.0001
Right ventricle (cm)	5980	2.22 ± 0.21, n = 2977	2.48 ± 0.21, n = 2647	<0.0001
IVS (cm)	6323	0.97 ± 0.12, n = 3166	1.11 ± 0.12, n = 2789	<0.0001
Left ventricle (cm)	6156	4.31 ± 0.35, n = 3071	4.78 ± 0.39, n = 2722	<0.0001
LVPW (cm)	6329	0.94 ± 0.1, n = 3144	1.08 ± 0.11, n = 2821	<0.0001
Pulmonary artery (cm)	4842	2.25 ± 0.28, n = 2344	2.45 ± 0.31, n = 2186	<0.0001
Right atrium (cm)	6016	3.50 (3.30, 3.70), n = 2999	3.90 (3.70, 4.13), n = 2674	<0.0001
Aortic valve (m/s)	6071	1.22 (1.10, 1.40), n = 3001	1.27 (1.10,1.43), n = 2732	<0.0001
Pulmonary valve (m/s)	6059	0.95 ± 0.17, n = 2984	0.97 ± 0.18, n = 2731	<0.0001
Ejection fraction (%)	5591	0.69 ± 0.06, n = 2770	0.67 ± 0.06, n = 2494	<0.0001
E/A ≤ 1, n (%)	5784	2697/2942 (91.7)	2387/2580 (92.5)	0.245
LVMi (g/m2)	5467	82.93 ± 14.64, n = 2736	108.31 ± 20.56, n = 2406	<0.0001
RWT	5467	0.45 ± 0.06, n = 2736	0.46 ± 0.06, n = 2406	<0.0001

Values are mean ± SD, median (interquartile range). Bold results are statistically significant at the P < 0.05 level.

A, peak late diastolic velocity; E, peak early diastolic velocity; e′, mitral annular early diastolic velocity; eGFR, estimated glomerular filtration rate; HTN, hypertension; IVS, interventricular septum; LVH, left ventricular hypertrophy; LVMi, left ventricular mass index; LVPW, left ventricular posterior wall; PG, pressure gradient; RWT, relative wall thickness; T2DM, type 2 diabetes mellitus; TR, tricuspid regurgitation.

Table 5.

Clinical phenotypes according to patient clusters

	N	Cluster 1	Cluster 2	P‐value*
		(N = 3507)	(n = 3199)
Group
Isolated T2DM	1716	1186/3507 (33.8)	530/3199 (16.6)	<0.0001
T2DM + obesity	263	127/3507 (3.6)	136/3199 (4.3)	‐
T2DM + HTN	3547	1840/3507 (52.5)	1707/3199 (53.4)	‐
T2DM + obesity + HTN	1180	354/3507 (10.1)	826/3199 (25.8)	<0.0001
Clinical characteristics
Age (year)	6706	67.6 ± 10.03, n = 3507	67.24 ± 10.34, n = 3199	0.146
Female, n (%)	6706	2194/3507 (62.6)	1166/3199 (36.4)	<0.0001
Diabetes duration (year)	5673	8 (6, 9), n = 2958	8 (6, 9), n = 2715	0.007
BMI (kg/m2)	6706	24.26 ± 3.54, n = 3507	26.25 ± 3.66, n = 3199	<0.0001
Obesity (≥28.0 kg/m2)	6706	481/3507 (13.7)	962/3199 (30.1)	<0.0001
Smoking, n (%)	6625	592/3456 (17.1)	858/3169 (27.1)	<0.0001
Drinking, n (%)	6623	325/3455 (9.4)	530/3168 (16.7)	<0.0001
Blood pressure
Hypertension, n (%)	6706	2194/3507 (62.6)	2533/3199 (79.2)	<0.0001
SBP (mmHg)	6010	133.77 ± 18.71, n = 3149	138.75 ± 19.21, n = 2861	<0.0001
DBP (mmHg)	3645	76.6 ± 10.91, n = 1871	79.22 ± 11.66, n = 1774	<0.0001
Biochemistry
BUN (mmol/L)	6537	5.5 (4.4, 6.8), n = 3413	5.7 (4.6, 7.3), n = 3124	<0.0001
SCr (μmol/L)	6629	72 (60, 86), n = 3463	80 (66, 99), n = 3166	<0.0001
SUA (μmol/L)	6629	291.4 ± 99.48, n = 3463	326.13 ± 109.11, n = 3166	<0.0001
FBG (mmol/L)	6629	8.26 (6.46, 11.65), n = 3463	8.24 (6.49, 11.39), n = 3166	0.568
HbA1c (%)	5421	8.17 ± 2.15, n = 2790	7.91 ± 1.98, n = 2631	<0.0001
eGFR	6629	86.59 ± 30.94, n = 3463	80.78 ± 35.03, n = 3166	<0.0001
Lipid metabolism
Apolipoprotein‐A1 (g/L)	6192	1.12 ± 0.29, n = 3237	1.06 ± 0.28, n = 2955	<0.0001
Apolipoprotein‐B (g/L)	6192	0.86 ± 0.25, n = 3237	0.82 ± 0.25, n = 2955	<0.0001
T‐CHO (mmol/L)	6327	4.73 ± 1.27, n = 3302	4.5 ± 1.25, n = 3025	<0.0001
HDL‐C (mmol/L)	6192	1.19 ± 0.31, n = 3237	1.11 ± 0.28, n = 2955	<0.0001
LDL‐C (mmol/L)	6192	2.92 ± 0.96, n = 3237	2.79 ± 0.95, n = 2955	<0.0001
Triglyceride (mmol/L)	6326	1.5 (1.05, 2.19), n = 3301	1.57 (1.1, 2.31), n = 3025	<0.0001
Lipoprotein‐a (mg/L)	6192	190 (106, 341), n = 3237	181 (96, 322), n = 2955	0.26
Blood clotting index
PT (s)	6052	11.1 (10.4, 11.8), n = 3153	11.2 (10.5, 11.9), n = 2899	0.002
APTT (s)	6052	29.4 (26.9, 32.1), n = 3153	29.9 (27.4, 32.5), n = 2899	<0.0001
TT (s)	6048	15.3 ± 3.05, n = 3150	15.33 ± 3.28, n = 2898	0.71
INR	6043	1.04 ± 0.17, n = 3147	1.05 ± 0.19, n = 2896	0.003
ATIII (%)	6048	99 (88, 111), n = 3152	98 (86, 108), n = 2896	<0.0001
FDP (mg/L)	6029	1.51 (0.87, 2.77), n = 3140	1.64 (0.9, 3), n = 2889	0.005
Fibrinogen (g/L)	6031	3.73 ± 0.91, n = 3138	3.77 ± 0.93, n = 2893	0.121
D dimer (μg/L)	5723	118 (28, 304), n = 3009	135 (42, 342.25), n = 2714	0.002
Medications
Antiplatelet user, n (%)	5954	1246/3107 (40.1)	1205/2847 (42.3)	0.082
Statin user, n (%)	5956	1055/3110 (33.9)	1044/2846 (36.7)	0.026
Insulin user, n (%)	5969	2148/3114 (69.0)	1896/2855 (66.4)	0.034
Metformin user, n (%)	5961	809/3112 (26.0)	744/2849 (26.1)	0.917
Diagnostics and others
Hypercholesteraemia, n (%)	6327	327/3302 (9.9)	222/3025 (7.3)	<0.0001
Hypertriglyceridaemia, n (%)	6326	755/3301 (22.9)	764/3025 (25.3)	0.027
Low high‐density lipoproteinaemia, n (%)	6192	1035/3237 (32.0)	1202/2955 (40.7)	<0.0001
High low‐density lipoproteinaemia, n (%)	6192	308/3237 (9.5)	222/2955 (7.5)	0.005
Cardiovascular hospitalization rate, n (%)	6706	488/3507 (13.9)	501/3199 (15.7)	0.044
Left ventricular hypertrophy, n (%)	5142	376/2736 (13.7)	1127/2406 (46.8)	<0.0001
Left ventricular diastolic dysfunction, n (%)	6706	2802/3507 (79.9)	2313/3199 (72.3)	<0.0001

Values are mean ± SD, median (interquartile range), or n/N (% of non‐missing data). Bold results are statistically significant at the P < 0.05 level.

APTT, activated partial thromboplastin time; ATIII, Antithrombin III; BUN, blood urea nitrogen; DBP, diastolic blood pressure; FBG, fasting blood‐glucose; FDP, fibrinogen degradation product; HbA1c, glycosylated haemoglobin; HDL‐C, high‐density lipoprotein cholesterol; HTN, hypertension; INR, international normalized ratio; LDL‐C, low‐density lipoprotein cholesterol; PT, prothrombin time; SBP, systolic blood pressure; SCr, serum creatinine; SUA, serum uric acid; T‐CHO, total cholesterol; T2DM, type 2 diabetes mellitus; TT, thrombin time.

Figure 5.

Distribution of echocardiographic morphological and functional parameters after clustering.When patients are plotted according to the two‐category clustering scheme, there are significant differences in individual values. Violin plots show median (horizontal midline) and interquartile spacing (upper and lower horizontal lines), with widths indicating the amount of data. ****P < 0.0001 Comparison between the two groups. DM, diabetes mellitus; HTN, hypertension; IVS, interventricular septum; LVMi, left ventricular mass index; LVPW, left ventricular posterior wall.

Figure 6 shows and describes the profile of echocardiographic phenotypes and clinical indicators using radar plots.

Figure 6.

An overview of echocardiographic and clinical indicators is shown and depicted using radar plots. (A) Echocardiographic features in four clinical subgroups. (B) Radar plot of echocardiographic indices for two clusters. (C) Radar plot of clinical characteristics and indicators based on clinical subgroups. (D) Radar plot of clinical characteristics for two clusters. (E) Radar plot of differences in important clinical characteristics for two clusters. f: Radar plot of clinical diagnosis and medication use for two clusters. Panels (A) and (B) illustrate the superimposition of echocardiograms based on four clinical groups in and two clusters, respectively. Panels (C) and (D) illustrate the overlay of clinical indicators for the two classification methods described earlier, respectively. It is suggested that cluster analysis significantly reduced the overlap of patients. Panels (E) and (F) show the differences in important clinical characteristics and diagnoses between the two clusters (P < 0.05). The colours of the two clusters correspond to the two groups of solutions from the clustering analysis. T2DM, type 2 diabetes mellitus; HTN, hypertension; SBP, systolic blood pressure; DBP, diastolic blood pressure; T‐CHO, total cholesterol; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; BUN, blood urea nitrogen; SCr, serum creatinine; SUA, serum uric acid; FBG, fasting blood‐glucose; HbA1c, glycosylated haemoglobin; PT, prothrombin time; APTT, activated partial thromboplastin time; TT, thrombin time; INR, international normalized ratio; ATIII, Antithrombin III; FDP, fibrinogen degradation product.

Cluster 1 (female and diastolic dysfunction more) with smaller aortic internal diameter (2.51 ± 0.43), ascending aortic internal diameter (3.16 ± 0.33), pulmonary artery internal diameter (2.25 ± 0.28), left atrial internal diameter (3.43 ± 0.34), left intraventricular diameter (4.31 ± 0.35), right atrial internal diameter [3.50 (3.30, 3.70)], right intraventricular diameter (2.22 ± 0.21), lower aortic orifice flow velocity [1.22 (1.10, 1.40)], pulmonary orifice flow velocity (0.95 ± 0.17); there was a smaller septal thickness (0.97 ± 0.12), posterior left ventricular wall thickness (0.94 ± 0.10), left ventricular mass index (82.93 ± 14.64), relative ventricular wall thickness (0.45 ± 0.06), and a high ejection fraction (0.69 ± 0.06), presenting a ‘small, thin, slow heart’. The clinical features of T2DM mellitus without obesity and HTN are usually lean with a low prevalence of HTN, low prevalence of smoking and alcohol consumption, low levels of urea nitrogen, creatinine and uric acid, and high glomerular filtration rate; however, total cholesterol, HDL‐C, LDL‐C, Apo‐A1, and Apo‐B levels are high, and glycated haemoglobin levels are high (P < 0.05).

Cluster 2 (more men, higher LVMi) with larger aortic internal diameter (2.74 ± 0.45), ascending aortic internal diameter (3.41 ± 0.33), pulmonary artery internal diameter (2.45 ± 0.31), left atrial internal diameter (3.98 ± 0.36), left intraventricular diameter (4.78 ± 0.39), right atrial internal diameter [3.90(3.70, 4.13)], right intraventricular diameter (2.48 ± 0.21), higher aortic orifice flow velocity [1.27(1.10, 1.43)], and pulmonary orifice flow velocity (0.97 ± 0.18); there was greater septal thickness (1.11 ± 0.12), posterior left ventricular wall thickness (1.08 ± 0.11), left ventricular mass index (108.31 ± 20.56), and relative ventricular wall thickness (0.46 ± 0.06), and a low ejection fraction (0.67 ± 0.06), presenting a ‘large, thick, urgent’ heart. The clinical features of T2DM mellitus are more common in combination with obesity and HTN, resulting in a high prevalence of obesity and HTN, a high prevalence of smoking and alcohol consumption, high levels of urea nitrogen, creatinine and uric acid, and a low glomerular filtration rate, but low levels of total cholesterol, HDL‐C, LDL‐C, Apo‐A1, and Apo‐B, and low levels of glycated haemoglobin (P < 0.05).

Analysis of risk factors for each cluster

DCM is characterized by a decrease in myocardial compliance and diastolic function in the early stages of the disease, with further development of myocardial hypertrophy and even systolic dysfunction. ²¹ , ²² To further explore the risk factors in the two clustered groups of patients, we analysed patients in Cluster 1 with reduced left ventricular diastolic function as the independent variable and patients in Cluster 2 with LVMi as the independent variable.

Data from patients in Cluster 1 were further processed for analysis of covariance to remove variables with covariate associations. After the univariate analysis (Supporting Information, Table S1 ), the independent variable was taken as the test of <0.1 and the LVDD was used as the dependent variable to perform a binary logistic regression analysis. Figure 7 combined with forest plots yielded age, systolic blood pressure, uric acid level, and ATIII as risk factors in the Cluster 1 (female and diastolic dysfunction more) group (P < 0.05).

Figure 7.

Forest plot of risk factors for reduced left ventricular diastolic function in the cluster two groups of patients. All covariables listed were included in the model simultaneously. *Numbers for continuous variables are values of 1 standard deviation. T2DM, type 2 diabetes mellitus; HTN, hypertension; SBP, systolic blood pressure; DBP, diastolic blood pressure; T‐CHO, total cholesterol; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; BUN, blood urea nitrogen; SCr, serum creatinine; SUA, serum uric acid; FBG, fasting blood‐glucose; HbA1c, glycosylated haemoglobin; PT, prothrombin time; APTT, activated partial thromboplastin time; TT, thrombin time; INR, international normalized ratio; ATIII, Antithrombin III; FDP, fibrinogen degradation product.

Patient data from Cluster 2 were further processed for analysis of covariance to remove variables with covariate correlation. After conducting a univariate analysis, the independent variables were taken as the test of <0.1 and the LVMi was used as the dependent variable to perform a multiple linear regression analysis (Supporting Information, Table S2 ). Table 6 suggests that in the Cluster 2 (more men, higher LVMi) group, the factors associated with higher LVMi were: higher systolic blood pressure and creatinine levels, and lower diastolic blood pressure and BMI levels, respectively (P < 0.05).

Table 6.

Multiple linear regression for LVMi in cluster 2

Risk factors	β	95% CI		P
		Lower	Upper
Constant	99.999	84.500	115.498	0.000
Age (year)	−0.004	−0.104	0.097	0.944
Diabetes duration (year)	0.103	−0.229	0.435	0.541
SBP (mmHg)	0.246	0.192	0.301	0.000
DBP (mmHg)	−0.137	−0.248	−0.027	0.016
BMI (kg/m2)	−0.577	−0.788	−0.366	0.000
T‐CHO (mmol/L)	0.612	−1.283	2.507	0.524
HDL‐C (mmol/L)	−3.388	−7.490	0.714	0.104
LDL‐C (mmol/L)	−1.384	−3.397	0.629	0.177
Triglyceride (mmol/L)	−0.227	−0.781	0.327	0.421
Lipoprotein‐a (mg/L)	0.001	−0.002	0.005	0.397
BUN (mmol/L)	0.149	−0.114	0.411	0.267
SCr (μmol/L)	0.031	0.023	0.039	0.000
SUA (μmol/L)	−0.005	−0.013	0.004	0.300
FBG (mmol/L)	0.098	−0.073	0.268	0.259
TT (s)	0.131	−0.199	0.462	0.425
FDP (mg/L)	0.022	−0.067	0.112	0.618
D dimer (μg/L)	0.001	−0.001	0.002	0.556
Drinking	−0.892	−3.219	1.435	0.449

Bold results are statistically significant at the P < 0.05 level.

T2DM, type 2 diabetes mellitus; HTN, hypertension; SBP, systolic blood pressure; DBP, diastolic blood pressure; T‐CHO, total cholesterol; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; BUN, blood urea nitrogen; SCr, serum creatinine; SUA, serum uric acid; FBG, fasting blood‐glucose; HbA1c, glycosylated haemoglobin; PT, prothrombin time; APTT, activated partial thromboplastin time; TT, thrombin time; INR, international normalized ratio; ATIII, Antithrombin III; FDP, fibrinogen degradation product.

Discussion

Computer learning methods are divided into supervised and unsupervised learning methods. Supervised learning is the classification or prediction of a given learning or outcome. Another type of unsupervised learning analyses the internal structure and connections in the data. ²³ The cluster analysis is an unsupervised learning process. Different from common clinical classification methods, the cluster analysis does not rely on pre‐classification in accordance with clinically defined categories but requires machine learning algorithms to determine markers on the basis of non‐prior data sets and classify data into different clusters automatically. ²⁴ The cluster analysis based on machine learning has been validated in clinical practice and has profound importance in identifying clinical phenotypes and differentiating heterogeneity. ²⁴ , ²⁵ , ²⁶ Previous relevant studies have also suggested the clinical value of clustering studies of echocardiographic correlates, leading to different clustering results and clinical significance. ¹⁰ For example, unsupervised cluster analysis was used to propose a unique grouping model for evaluating left ventricular diastolic dysfunction. These cluster models can better identify patient groups with similar risk, and their inclusion in clinical practice may help eliminate uncertain results and improve clinical outcome prediction. ¹¹ Our study retrospectively investigated and collected clinical and cardiac ultrasound data of Chinese patients with T2DM in the last 5 years. It was also compared with clinical taxonomies to determine whether these clusters were associated with different clinical features and outcomes and explore echocardiographic parameters observed by unsupervised cluster analysis for grouping and identifying cardiac phenotypes in patients with similar morphology and function. The statistical analysis of risk factors of groups leads to the transcendental echocardiographic parameters and clinical data of the larger overlap, but the heart of the clustering analysis based on echocardiography‐related phenotype is divided into two groups. The two groups of clusters of echocardiography parameters have good distinction in clinical characteristics, good consistency within the group, and good heterogeneity between groups. The phenotypes of the two groups are associated with obesity and HTN and with diastolic function, systolic function, gender, lipid metabolism, renal function, coagulation index, and other important metabolic diseases. Compared with clinical classification, the clustering model has better clinical guidance.

DCM is the widespread focal necrosis of the myocardium caused by metabolic disorders and microvascular lesions, which gradually lead to diastolic and contractile dysfunction. This phenomenon eventually leads to heart failure or even sudden death. ²¹ Studies showed that DCM has diastolic function and structural abnormalities caused by metabolic disorders. ²² The earliest features of the disease are impaired cardiac compliance, early decreased diastolic function, and increased atrial filling. ²⁷ Diastolic function is further impaired by metabolic changes, neurohumoural activation, and development of myocardial fibrosis in the advanced and late stages of DCM. ²² , ²⁸ , ²⁹ Systolic dysfunction and heart failure with reduced ejection fraction are finally present. Thus, diastolic dysfunction, myocardial hypertrophy, and systolic dysfunction are successively established as pathological features of DCM. ²⁷ , ³⁰ , ³¹ However, heterogeneity is observed in related cardiovascular risk factors, such as age, duration of diabetes, obesity, and HTN, in patients with T2DM. ³² , ³³ These confounding factors influence the echocardiographic data in the daily clinical practice of patients with T2DM. T2DM with HTN, T2DM with obesity, or both with HTN further affect the occurrence and development of DCM, and whether these factors accelerate the occurrence and development of DCM due to the synergistic effect and interaction caused by the disorder of glucose and lipid metabolism. ³³ , ³⁴ Therefore, distinguishing the risk factors of DCM in the early and middle stages and carrying out clinical intervention strategies are challenging. This research is the first study to apply cluster analysis in the identification of different cardiac phenotypes and determine their correlation with clinical characteristics in a Chinese population with T2DM. Interestingly, the cardiac characteristics of the two groups of patients in our echocardiographic cluster analysis correspond to the early and middle changes in DCM. Guiding clinical identification and intervention strategies is important.

In our study, patients in Cluster 1 showed a high rate of LVDD and a low decline rate of hypertrophic systolic function, and these findings were consistent with the echocardiographic characteristics of the early identification of DCM. Moreover, the cardiac ultrasound indicators of these patients were all better than those in Cluster 2 and were in the low‐risk group of DCM. In addition, the clinical characteristics of this type of patients were evidently heterogeneous and were more common in simple patients with T2DM but no obesity and HTN. Gender analysis indicated that the majority of patients were female, and the BMI analysis indicated thin body; low rates of smoking and alcohol drinking; and low levels of urea nitrogen, creatinine, and uric acid. In addition, we found that Cluster 1 had inferior glucose and lipid metabolism control and higher levels of HbA1c, apolipoprotein‐A1, total cholesterol, and low‐density lipoprotein compared with Cluster 2, leading to increased prevalence of hypercholesterolemia, low‐density to high‐density lipoproteinaemia, and low‐density lipoproteinaemia. At the same time, the low utilization rate of statins and insulin in Cluster 1 might be related to the poor control of glucolipid metabolism indicators. Therefore, for this group of patients, our clinical treatment strategy should emphasize the early intervention of blood glucose and blood lipids. By further analysing the decline in diastolic function in this group of patients, we found that blood pressure (especially SBP), uric acid levels, and testing for antithrombin III might be risk factors for the decline in diastolic function.

Patients in Cluster 2 showed a high rate of hypertrophic systolic function decline. According to the occurrence and development of DCM, patients with reorganization might have been in the progression of DCM and were in the high‐risk group of DCM. In comparison with those of patients in Cluster 1, the clinical characteristics of patients in Cluster 2 were also evidently heterogeneous with more patients complicated with obesity and HTN. Gender analysis suggested that most patients were male, and the BMI analysis suggested that these patients were overweight and had high rates of smoking and alcohol drinking and high levels of urea nitrogen, creatinine, and uric acid. These findings indicated the necessity of lowering blood pressure, reducing weight, and improving bad living habits in such patients. In addition, we found that patients in Cluster 2 had lower levels of HbA1c, apolipoprotein‐A1, total cholesterol, and low‐density lipoprotein. Thus, the prevalence of hypercholesterolemia, low‐high‐density lipoproteinaemia, and low‐density lipoproteinaemia in patients in Cluster 2 was lower than that in Cluster 1, indicating that the control of glucose and lipid metabolism of patients in Cluster 2 was better than that in Cluster 1. Combined with our findings, this result might be related to the clinical use of statins, lipid‐lowering drugs, and insulin. Therefore, for this group of patients, our clinical treatment should emphasize intensive treatment of weight and blood pressure, smoking cessation, and alcohol restriction, and the control of urea nitrogen, creatinine, and uric acid levels might have some improvement in DCM with reduced hypertrophic systolic function. The multiple linear regression of the LVMi in this group of patients also suggested that blood pressure (SBP and DBP), BMI, and creatinine levels might be important in the course of DCM with hypertrophic systolic dysfunction. Thus, our data suggested that clinicians should pay attention to the treatment of female patients with decreased diastolic function in the early stages of DCM and treatment of male patients with HTN, obesity, and hypertrophic systolic dysfunction in the progressive stages of DCM (Figure 8).

Figure 8.

Cardiac phenotypes in T2DM by cluster analysis.

In conclusion, our findings suggested that the new clustering of altered cardiac function and structure in patients with T2DM is superior to traditional classification based on risk factors because the former allows for the improved identification of early clinical features in patients with DCM and provides information about risk factors. This study demonstrates the potential clinical value of a powerful algorithm incorporating artificial intelligence to improve the identification and use of alterations in the cardiac phenotype of patients with T2DM for the determination of different clinical features at different stages of DCM. Moreover, clustering can be easily applied to existing cohorts of patients with T2DM, and this new classification may ultimately help to bring benefits for targeted and tailored early treatment of cardiomyopathy in patients with T2DM, thus representing an important step towards precision medicine in diabetes.

Study limitations

First, this was a retrospective analysis with a statistical study of a large number of patients. There is a lack of further prospective studies on changes in different clinical indicators and cardiovascular endpoints in the two categories of patients screened by the cluster analysis, and the value of the cluster analysis in providing prognostic information lacks further validation. Secondly, this is a single‐centre study, with the majority of patients originating from a specific region, and information on patients from other regions needs to be added to further enrich the study area. Thirdly, during the 5‐year clinical data collection process, there was a lack of inclusion of cardiac‐related indicators such as patients' ambulatory ECG and BNP, making the results of the cluster analysis lacking clinical analysis for comparison of these two indicators on cardiac function. Fourth, the random absence of data during the clinical data collection process left a proportion of patients without sufficiently complete data. For example, the TR signal used to calculate TR velocity may not be available for many patients. For the small number of randomly missing data, we used the technique of interpolating the missing data. Finally, despite our strict enforcement of exclusion criteria, there are still many unknown confounders and comorbidities that may not have been followed up, and it is possible that future stratification could be further refined by including other clustering variables for cardiac ultrasound.

Funding

This work was supported by the Natural Science Foundation of China (No. 81870568, Shaohua Wang).

Conflict of interest

The authors declare that they have no competing interests.

Supporting information

Table S1. Cluster 1 univariate analysis.

Table S2. Cluster 2 univariate analysis.

Figure S1. Flowchart of the study population nadir.

Wang, Z. , Wang, C. , Xie, Z. , Huang, X. , ShangGuan, H. , Zhu, W. , and Wang, S. (2022) Echocardiographic phenotypes of Chinese patients with type 2 diabetes may indicate early diabetic myocardial disease. ESC Heart Failure, 9: 3327–3344. 10.1002/ehf2.14062.