Abstract
Aims
We aim to integrate the parameters of two‐dimensional (2D) echocardiography and identify the high‐risk population for all‐cause mortality in patients with acute ST‐segment elevation myocardial infarction (STEMI) undergoing percutaneous coronary intervention (PCI).
Methods
The study involved a retrospective cohort population with STEMI who were admitted to Yongchuan Hospital of Chongqing Medical University between January 2016 and January 2019. Baseline data were collected, including 2D echocardiography parameters and left ventricular ejection fraction (LVEF). The parameters of 2D echocardiography were subjected to cluster analysis. Logistic regression models were employed to assess univariate and multivariate adjusted odds ratios (ORs) of cluster information in relation to all‐cause mortality. Four logistic regression models were generated, utilizing cluster information, clinical variables, clinical variables in conjunction with LVEF, and clinical variables in conjunction with LVEF and cluster information as predictive variables, respectively. The area under the curve (AUC) were utilized to evaluate the incremental risk stratification value of cluster information.
Results
The study included 633 participants with 28.8% female, a mean age of 65.68 ± 11.98 years. Over the course of a 3‐year follow‐up period, 108 (17.1%) patients experienced all‐cause mortality. Utilizing cluster analysis of 2D echocardiography parameters, the patients were categorized into two distinct clusters, with statistically significant differences observed in most clinical variables, echocardiography, and survival outcomes between the clusters. Multivariate regression analysis revealed that cluster information was independently associated with the risk of all‐cause mortality with adjusted OR 7.33 (95% confidence interval [CI] 3.99–14.06, P < 0.001). The inclusion of LVEF enhanced the predictive capacity of the model utilized with clinical variables with AUC 0.848 (95% CI 0.809–0.888) versus AUC 0.872 (95% CI 0.836–0.908) (P < 0.001), and the addition of cluster information further improved its predictive performance with AUC 0.906 (95% CI 0.878–0.934, P < 0.001). This cluster analysis was translated into a free available online calculator (https://app‐for‐mortality‐prediction‐cluster.streamlit.app/).
Conclusions
The 2D echocardiographic diagnostic information based on cluster analysis had good prognostic value for STEMI population, which was helpful for risk stratification and individualized intervention.
Keywords: Cluster analysis, Echocardiography, Myocardial infarction, Risk stratification
Introduction
The advent of percutaneous coronary intervention (PCI) has substantially decreased the mortality of patients with acute ST‐segment elevated myocardial infarction (STEMI); however, the long‐term survival rate of this cohort remains considerably lower than that of the general population. 1 Precise evaluation of event risk and proactive intervention can enhance outcomes for cardiac disease. 2 , 3 Consequently, identifying populations with a high risk of mortality has become a crucial aspect of clinical practice.
Echocardiography, a non‐invasive imaging technology, has demonstrated a robust and repeatable capability for evaluating and stratifying cardiac ailments and has been recommended in the guidelines for precise risk stratification of STEMI patients. 4 However, it is important to acknowledge that medical professionals often prioritize a limited range of parameters, such as left ventricular ejection fraction (LVEF) and left ventricular diastolic diameter (LVDD), but neglect other potentially valuable image information, particularly two‐dimensional (2D) echocardiography parameters. The measurement of 2D echocardiographic parameters is comparatively straightforward, enabling its rapid adoption in underprivileged areas. However, it can be difficult to integrate so many variables.
Cluster analysis is an unsupervised machine learning technique that categorizes patients into groups or clusters based on various attributes, including demographics, medical history, and examination findings. By grouping patients with similar characteristics across multiple dimensions, it is possible to analyse and correlate the characteristics of these individuals with treatment responses or outcomes. Recent studies have demonstrated the effectiveness of unsupervised multi‐kernel learning in identifying similarities between patients in an ‘agnostic’ manner, using diverse and heterogeneous data sources, such as complex imaging‐based descriptions of ventricular structure and function. 5
Based on the above, in this study, the integration of 2D echocardiography into a novel parameter through cluster analysis was undertaken to enhance the optimization of the risk stratification system for STEMI patients.
Methods
Study design and participants
The study cohort was derived from a retrospective analysis conducted at Yongchuan Hospital, Chongqing Medical University, China. 6 , 7 The study encompassed STEMI patients who sought medical attention at Yongchuan Hospital of Chongqing Medical University between January 2016 and January 2019. Patients diagnosed with STEMI according to the guidelines. 4 All STEMI patients included in the study presented to the emergency department within 12 h of symptom onset and were administered 300 mg aspirin and 300 mg clopidogrel/180 mg ticagrelor with standard heparin. Patients with a high burden of thrombosis were treated with a glycoprotein IIb/IIIa inhibitor (uniform tirofiban) as determined by the interventionalist. The exclusion criteria employed in this analysis encompassed old myocardial infarction; conservative treatment without PCI; estimated life expectancy of less than 12 months; severe valvular heart disease; history of cerebrovascular disease or significant residual neurological deficit; history of chronic hepatitis or cirrhosis; severe renal insufficiency necessitating dialysis; known contraindications to statins, heparin, aspirin, clopidogrel, contrast media, or GPI; recent severe infection or connective tissue disease; malignant tumour; active severe bleeding and loss to follow‐up. All human participant research procedures adhered to the ethical standards of the Institution, the National Research Council, and the Declaration of Helsinki of 1964, along with its subsequent amendments or equivalent ethical standards. The retrospective design necessitated the removal of the informed consent requirement. The Ethics Committee of Yongchuan Hospital of Chongqing Medical University granted approval for this retrospective study.
Data collection and definitions
The study collected sociodemographic, lifestyle characteristics, past medical history, co‐morbidities, laboratory data, and imaging data upon admission.
Blood samples were obtained for assessment of liver function, lipids, and glucose, including alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma‐glutamyl transpeptidase, albumin, total bilirubin (TBIL), direct bilirubin (DBIL), total cholesterol (TC), hypertriglyceridemia, low‐density lipoprotein cholesterol (LDL‐C), high‐density lipoprotein cholesterol (HDL‐C), and fasting glucose levels. Other laboratory tests were obtained from specimens sent immediately after admission, including white blood cell count (WBC), C‐reactive protein (CRP), haemoglobin, platelets, uric acid, creatinine, glycosylated haemoglobin (HbA1c), cardiac troponin I (cTn I), creatine kinase‐MB values (CKMB), and N‐terminal B‐type natriuretic peptide (BNP). Cardiac ultrasound scans were performed within 3 days of PCI, and 2D echocardiographic parameters were measured using standard methods (Figure 1 ). In particular, diagnostic information such as left atrial anteroposterior diameter, LVDD, left ventricular systolic diameter (LVSD), right atrial anteroposterior diameter (RAD), right ventricular diameter (RVD), interventricular septum thickness (IVS), left ventricular posterior wall thickness (LVPW), ascend aorta diameter (AO), right ventricular outflow tract diameter (RVOT) and main pulmonary artery diameter were obtained based on 2D echocardiography. 8 , 9
Figure 1.
Schematic diagram of two‐dimensional parameters in echocardiography.
Following discharge, patients were subjected to annual telephone or clinic visits, with the latest follow‐up conducted in January 2020. The primary endpoint was defined as all‐cause mortality.
Statistical analysis
Continuous variables that exhibited normal distribution were presented as means (standard error), and intergroup comparisons were conducted using analysis of variance. Continuous variables that did not follow normal distribution were presented as medians (interquartile range), and intergroup comparisons were conducted using Kruskal–Wallis rank sum tests. Categorical variables were presented as frequencies and proportions, and intergroup comparisons were conducted using chi‐square test.
Utilizing all 10 items of 2D ultrasound diagnostic information, the partitioning around medoids (PAM) method was employed for cluster analysis. The optimal number of clusters was determined through the utilization of ‘average silhouette width’ and ‘total within sum of squared error’. The average silhouette width was calculated for varying numbers of clusters, and the cluster number with the highest average silhouette width was deemed optimal. The value of total within sum of squared error (WSS) decreased as the number of clusters increased. When the WSS exhibited a slow decrease with a low slope, it was concluded that further increases in cluster number would not enhance the cluster effect. The optimal cluster number served as the ‘turning point’, after which cluster information for each individual was obtained.
Then, logistic regression models were then constructed to estimate the odds ratio (OR), with four models fitted. The ‘model a’ solely included clusters, while ‘model b’ adjusted for age and gender, ‘model c’ adjusted for age, gender, diabetes, WBC, CRP, haemoglobin, albumin, creatinine, HbAIc, LDL‐C, CKMB, cTnI, BNP, and culprit vessel, and ‘model d’ adjusted for age, gender, diabetes, WBC, CRP, haemoglobin, albumin, creatinine, HbAIc, LDL‐C, CKMB, cTnI, BNP, culprit vessel, and LVEF. 10 , 11
Next, four models were developed to predict outcome events. The first model utilized clusters as the predictor to obtain ‘model A’, while the second model utilized clinical variables as predictors to obtain ‘model B’. The third model incorporated both clinical variables and LVEF as predictors to obtain ‘model C’, and the fourth model included clinical variables, LVEF, and clusters as predictors to obtain ‘model D’. ROC curves were plotted for each model, and the incremental risk stratification value of LVEF and cluster was evaluated by calculating the net reclassification improvement (NRI) and integrated discrimination improvement 12 of model C relative to model B, and model D relative to model C. Finally, the clinical effectiveness of the models was assessed using decision curve analysis.
Finally, we deployed the results of cluster analysis online to analyse the external data. R 4.2.2 was used for all statistical analysis with a significance level of P < 0.05 and Python was used for deployment.
Results
Cluster for population
The study included 633 participants with 28.8% female, a mean age of 65.68 ± 11.98 years. The results of cluster analysis indicated that the average silhouette width reached its maximum value at two clusters (Figure 2 A ), and the WSS value decreased rapidly from one cluster to two clusters and then decreased slowly (Figure 2 B ). Ultimately, the patients were categorized into two clusters, denoted as cluster A and cluster B (Figure 3 A ).
Figure 2.
The average silhouette width‐change (A) and total within sum of squared error‐change (B) diagram under different cluster numbers.
Figure 3.
Plot of the different clusters (A) and forest plot showing multivariate‐adjusted OR of the cluster (B).
Baseline characteristics of study participants
Based on the cluster information, the baseline characteristics of all patients were summarized in Table 1 and Table S1 . The cluster B populations exhibited lower levels of haemoglobin, LVDD, RAD, RVD, RVOT, higher levels of CRP, LDL‐C, LAD, MPA, and LVPW. The coordinates of the centres of cluster A and B were plotted (Figure S1 ), which reflected the relationship between 2D echocardiographic parameters and centres, and subjects would be clustered according to different distances from the two centres. Following a median follow‐up period of 3 years, 108 (17.1%) patients experienced all‐cause mortality. Patients with cluster B (27.8%) had higher mortality compared with cluster A (8.9%).
Table 1.
Baseline information of study population and comparison of different clusters
| Characteristics | Overall (n = 633) | Cluster A (n = 360) | Cluster B (n = 273) | P value |
|---|---|---|---|---|
| Male, n (%) | 451 (71.20) | 266 (73.90) | 185 (67.80) | 0.110 |
| Age (years), mean (SD) | 65.68 (11.98) | 65.48 (12.00) | 65.95 (11.97) | 0.629 |
| Hypertension, n (%) | 343 (54.20) | 217 (60.30) | 126 (46.20) | 0.001 |
| Diabetes, n (%) | 139 (22.00) | 86 (23.90) | 53 (19.40) | 0.211 |
| WBC (109/L), mean (SD) | 10.60 (4.42) | 10.71 (4.02) | 10.45 (4.89) | 0.457 |
| Haemoglobin (g/L), mean (SD) | 132.86 (20.31) | 135.12 (20.01) | 129.88 (20.35) | 0.001 |
| PLT (10^9/L), mean (SD) | 198.29 (67.83) | 204.95 (68.52) | 189.50 (66.02) | 0.004 |
| CRP (mg/L), mean (SD) | 20.95 (41.76) | 18.49 (37.75) | 24.19 (46.39) | 0.089 |
| ALT (U/L), mean (SD) | 63.98 (136.51) | 67.63 (164.69) | 59.17 (86.37) | 0.440 |
| AST (U/L), mean (SD) | 186.06 (248.76) | 194.26 (287.33) | 175.26 (186.06) | 0.342 |
| Albumin (g/L), mean (SD) | 38.97 (5.20) | 38.69 (5.42) | 39.33 (4.87) | 0.130 |
| UA (μmmol/L), mean (SD) | 347.95 (117.89) | 357.03 (112.57) | 335.97 (123.76) | 0.026 |
| Creatinine (μmmol/L), mean (SD) | 88.16 (51.83) | 86.89 (49.00) | 89.83 (55.39) | 0.481 |
| HDL‐C (mmmol/L), mean (SD) | 1.17 (0.43) | 1.18 (0.50) | 1.16 (0.33) | 0.584 |
| LDL‐C (mmmol/L), mean (SD) | 2.74 (0.95) | 2.68 (0.90) | 2.82 (1.02) | 0.067 |
| Glucose (mmmol/L), mean (SD) | 8.00 (4.98) | 8.23 (5.55) | 7.70 (4.09) | 0.183 |
| HbAIc (%), mean (SD) | 6.44 (1.59) | 6.49 (1.64) | 6.38 (1.52) | 0.385 |
| cTnI (ng/L), mean (SD) | 13.74 (20.74) | 16.48 (24.54) | 10.13 (13.50) | <0.001 |
| BNP (ng/L), mean (SD) | 2893.54 (4710.81) | 2891.47 (4838.68) | 2896.27 (4545.51) | 0.990 |
| LAD (mm), mean (SD) | 40.20 (10.73) | 32.40 (5.15) | 50.49 (6.77) | <0.001 |
| LVDD (mm), mean (SD) | 42.01 (9.31) | 48.54 (6.24) | 33.40 (4.36) | <0.001 |
| LVSD (mm), mean (SD) | 31.10 (6.09) | 33.61 (6.40) | 27.78 (3.58) | <0.001 |
| RAD (mm), mean (SD) | 27.62 (7.43) | 33.30 (3.81) | 20.12 (3.16) | <0.001 |
| RVD (mm), mean (SD) | 15.33 (5.32) | 19.60 (2.30) | 9.70 (1.71) | <0.001 |
| IVS (mm), mean (SD) | 9.77 (1.52) | 9.73 (1.61) | 9.84 (1.40) | 0.380 |
| LVPW (mm), mean (SD) | 10.97 (2.25) | 9.59 (1.18) | 12.78 (2.03) | <0.001 |
| AO (mm), mean (SD) | 27.97 (3.33) | 28.14 (3.52) | 27.75 (3.06) | 0.142 |
| RVOT (mm), mean (SD) | 24.51 (4.59) | 27.89 (2.78) | 20.05 (1.91) | <0.001 |
| MPA (mm), mean (SD) | 21.58 (5.30) | 20.16 (2.09) | 23.47 (7.30) | <0.001 |
| LVEF (%), mean (SD) | 56.85 (12.58) | 56.68 (11.38) | 57.08 (14.03) | 0.688 |
| All‐cause mortality, n (%) | 108 (17.1) | 32 (8.9) | 76 (27.8) | <0.001 |
ALT, alanine aminotransferase; AO, ascend aorta diameter; AST, aspartate aminotransferase; BNP, B‐type natriuretic peptide; CRP, C‐reactive protein; cTnI, cardiac troponin I; HDL‐C, high‐density lipoprotein cholesterol; IVS, interventricular septum thickness; LAD, left atrial anteroposterior diameter; LDL‐C, low‐density lipoprotein cholesterol; LVDD, left ventricular diastolic diameter; LVEF, left ventricular ejection fraction; LVPW, left ventricular posterior wall thickness; LVSD, left ventricular systolic diameter; MPA, main pulmonary artery diameter; PLT, platelets; RAD, right atrial anteroposterior diameter; RVD, right ventricular diameter; RVOT, right ventricular outflow tract diameter; UA, uric acid; WBC, white blood cell count.
Odd ratios for all‐cause mortality
After adjustment for age, gender, diabetes, WBC, CRP, haemoglobin, albumin, creatinine, HbAIc, LDL‐C, CKMB, cTnI, BNP, culprit vessel, and LVEF (model d), the multivariate odd ratio for all‐cause mortality was 7.33 (95% CI 3.99–14.06, P < 0.001) for cluster B compared with cluster A (Table 2 , Table S2 ). For each level of cluster, the risk of all‐cause mortality increased by 633% (Figure 3 B ).
Table 2.
Logistic models for the association between cluster and all‐cause mortality
| Case/total | Model a | Model b | Model c | Model d | |
|---|---|---|---|---|---|
| Cluster A | 360/633 | Ref | Ref | Ref | Ref |
| Cluster B | 273/633 | 3.95 (2.55,6.27) | 4.56 (2.84,7.50) | 5.29 (3.05,9.47) | 7.33 (3.99,14.06) |
Model a: cluster without adjust. Model b: adjusted for age and gender. Model c: adjusted for age, gender, diabetes, WBC, CRP, haemoglobin, albumin, creatinine, HbAIc, LDL‐C, CKMB, cTnI, BNP, and culprit vessel. Model d: adjusted for age, gender, diabetes, WBC, CRP, haemoglobin, albumin, creatinine, HbAIc, LDL‐C, CKMB, cTnI, BNP, culprit vessel, and LVEF.
BNP, B‐type natriuretic peptide; CKMB, creatine kinase‐MB values; CRP, C‐reactive protein; cTnI, cardiac troponin I; HbAIc, glycosylated haemoglobin; LDL‐C, low‐density lipoprotein cholesterol; LVEF, left ventricular ejection fraction; WBC, white blood cell count.
Incremental value of cluster
The analysis of ROC curves indicated that model D (AUC 0.906, 95% CI 0.878–0.934) exhibited superior predictive performance in comparison to model A (AUC 0.664, 95% CI 0.616–0.712, P < 0.001), model B (AUC 0.848, 95% CI 0.809–0.888, P < 0.001), and model C (AUC 0.872, 95% CI 0.836–0.908, P < 0.001) (Figure 4 A ), as evidenced by the calculated NRI and IDI values with 25% and 75% as the boundaries (Table 3 ). The decision curve analysis revealed that the net benefit level of model D surpassed that of ‘zero risk of mortality’ and ‘all mortality’, and outperformed the other three models across the threshold probability range of 0–100% (Figure 4 B ).
Figure 4.
Receiver operating characteristic curves (A) and decision analysis curves of four models (B) of the four models for predicting the endpoint event.
Table 3.
Net reclassification improvement and integrated discrimination improvement among models
| NRI [95% CI] | IDI [95% CI] | |
|---|---|---|
| Model C vs. Model B | 0.658 [0.459–0.857] | 0.056 [0.027–0.085] |
| Model D vs. Model C | 0.654 [0.461–0.846] | 0.072 [0.043–0.102] |
Model B: utilizing clinical variables (age, gender, diabetes, WBC, CRP, haemoglobin, albumin, creatinine, HbAIc, LDL‐C, CKMB, cTnI, BNP, and culprit vessel) as predictors. Model C: utilizing clinical variables and LVEsF as predictors. Model D: utilizing clinical variables, LVEF, and clusters as predictors.
IDI, integrated discrimination improvement; NRI, net reclassification improvement.
Cluster deployment
The results of the cluster analysis were deployed online to promote and use the prediction tool (https://app‐for‐mortality‐prediction‐cluster.streamlit.app/) (Figure 5 ). After the 10 variable values required for the cluster analysis were entered, the tool automatically clustered the patient into cluster A or B.
Figure 5.
The online calculator based on the cluster analysis. Input the patient information and click the ‘predict’ button to get the patient's mortality risk assessment results.
Discussion
This study introduced a novel parameter derived from 2D echocardiography based on cluster analysis. This novel parameter was an independent indicator for all‐cause mortality in the STEMI population. The inclusion of cluster information resulted in an additional prognostic stratification value in comparison to the utilization of solely clinical parameters or the combination of individual ultrasound parameters.
Echocardiography is the most commonly employed and easily accessible diagnostic tool for comprehensive evaluation of cardiac structure and function. 13 Despite 3D‐speckle tracking imaging or ultrasonic enhancing agents are recommended by guidelines, 14 their employment are limited in developing country or area. 15 2D echocardiography is still the most common tool in real world, and studies have confirmed that LAD, main pulmonary artery, 16 , 17 diastolic dysfunction, and LVEF are closely related to the prognosis of myocardial infarction. 18 , 19 Optimized 2D parameters selection and risk stratification algorithm have practical significance for those developing country or area. In our study, based on 10 2D parameters, cluster analysis categorized our patients into two groups. Multivariate analysis revealed that 2D echocardiography cluster information was an independent predictor of all‐cause mortality in STEMI patients undergoing PCI. Furthermore, the introduction of cluster information, in conjunction with clinical features, laboratory data, and LVEF, enhanced the model's risk prediction capabilities. Interestingly, in our cohort, the group of decreased patients had higher levels of LAD and lower levels of MPA and LVEF (Table S1 ). However, in cluster analysis, patients in cluster B had higher levels of LAD and MPA than patients in group A, and the same levels of LVEF (Table 1 ). From the perspective of groups data analysis, our findings seem to indicate that cluster analysis results identify high‐risk groups characterized by ventricular thickening and diastolic dysfunction, and the results of multivariate adjustment analysis indicate that the prognostic effects of cluster parameter is independent of LVEF. The cluster parameters representing ventricular thickening and diastolic dysfunction could be a powerful supplement to ventricular systolic dysfunction. 20 This suggests that cluster analysis, an advanced algorithm, captured both linear and nonlinear relationships between variables and outcomes, creatively identifying and classifying patients with similar characterizations. In this study, we have outlined a novel approach that we believe has made a valuable contribution to the rapid advancement of this field. By incorporating multiple variables into our analysis, our approach provided new insights that further the progress in this burgeoning area of research. This methodology will assume a more prominent role in the optimization of secondary prevention strategies of STEMI and in the identification of high‐risk populations.
Machine learning has demonstrated successful applications in the diagnosis, classification, and prognostication of myocardial infarction cohorts. 21 , 22 Contemporary data analysis techniques, such as deep learning, have demonstrated increased efficacy in solving intricate pattern recognition problems by using neural networks, and their reliability is contingent upon the substantial amounts of data acquisition. 23 As such, we opted for cluster analysis, a less complex and computationally intensive analytical approach. This method is designed to reduce the amount of data by identifying a subset of observations and grouping them into distinct clusters. A cluster is defined as a collection of multiple observations that exhibit higher similarity within the group than between groups. 24 This technique is a classic example of unsupervised learning in the field of machine learning, which enables us to investigate the inherent structure and distribution of the data. In both clinical practice and most prior research, medical professionals and researchers have tended to concentrate on a limited number of parameters in order to simplify their usage and avoid issues of multicollinearity. However, it should be noted that it is difficult for a relatively single indicator to provide effective and compelling explanations or predictions for complex diseases, which results in the neglect and loss of a significant amount of potentially valuable information. Cluster analysis, as a novel tool, allows researchers to integrate all the potentially valuable variables and convert multi‐dimensional variables into a single, more manageable variable, skilfully. This approach effectively addresses the aforementioned challenges. 25 This study utilized cluster analysis to incorporate all variable information presented in 2D echocardiography, without the need to discard any variables. In future, cluster analysis can be applied to various disease fields, facilitating clinical diagnosis, treatment, and prevention. It was noteworthy that the cluster information was derived from fundamental and uncomplicated two‐dimensional parameters, making the research outcome accessible and applicable to primary health‐care institutions.
Despite the widespread adoption of unsupervised machine learning models in numerous studies, they were rarely translated into practical solutions. In this study, cluster information was stored on cloud computing servers, and an application programming interface was developed to enable immediate and free access (https://app‐for‐mortality‐prediction‐cluster.streamlit.app/). This web‐based tool made the clinical use of 2D echocardiography parameters more comprehensive and accurate. These advancements were expected to significantly enhance the accessibility and practicality of our research. In the future, the cluster information may be incorporated into echocardiography reports to facilitate clinical decision‐making.
This study was not without limitations. First, this study constituted a post‐hoc analysis of a retrospective single‐centre cohort study, which may have been subject to population selection bias. Further population analyses are required to augment our findings. So we disseminated the study content online, which facilitated validation studies with external populations. Second, the study did not analyse major adverse cardiovascular events but all‐cause mortality as a hard endpoint due to the initial study design. 7
Conclusions
Our study demonstrated that a new 2D echocardiography cluster information was an independent risk factor for long‐term all‐cause mortality in patients with STEMI undergoing PCI. Furthermore, the risk warning strategy of echocardiography for patients was optimized based on the cluster analysis.
Funding
This work was support by Science and Technology Bureau of Yongchuan, Chongqing (2022yc‐jckx20005).
Conflict of interest
None declared.
Supporting information
Figure S1. Supporting Information
Table S1. Echocardiography information of study population.
Table S2. Odds ratios of logistic models for the association between cluster and all‐cause mortality.
Gao, H. , Wang, K. , Wang, X. , Zeng, D. , and Chen, Z. (2024) Integration of two‐dimensional echocardiography: A novel risk indicator for ST‐segment elevation myocardial infarction. ESC Heart Failure, 11: 3312–3321. 10.1002/ehf2.14939.
Hongli Gao and Kai Wang are first co‐authors.
Contributor Information
Deli Zeng, Email: chenzijun@cqmu.edu.cn, Email: 21258712@qq.com.
Zijun Chen, Email: chenzijun@cqmu.edu.cn.
References
- 1. Piironen M, Ukkola O, Huikuri H, Havulinna AS, Koukkunen H, Mustonen J, et al. Trends in long‐term prognosis after acute coronary syndrome. Eur J Prev Cardiol 2017;24:274‐280. doi: 10.1177/2047487316679522 [DOI] [PubMed] [Google Scholar]
- 2. Benjamin EJ, Muntner P, Alonso A, Bittencourt MS, Callaway CW, Carson AP, et al. Heart disease and stroke statistics—2019 update: A report from the American Heart Association. Circulation 2019;139:e56‐e528. doi: 10.1161/CIR.0000000000000659 [DOI] [PubMed] [Google Scholar]
- 3. Lee DS, Straus SE, Farkouh ME, Austin PC, Taljaard M, Chong A, et al. Trial of an intervention to improve acute heart failure outcomes. N Engl J Med 2023;388:22‐32. doi: 10.1056/NEJMoa2211680 [DOI] [PubMed] [Google Scholar]
- 4. Byrne RA, Rossello X, Coughlan J, Barbato E, Berry C, Chieffo A, et al. 2023 ESC guidelines for the management of acute coronary syndromes: developed by the task force on the management of acute coronary syndromes of the European Society of Cardiology (ESC). Eur Heart J Acute Cardiovasc Care 2024;13:55‐161. doi: 10.1093/ehjacc/zuad107 [DOI] [PubMed] [Google Scholar]
- 5. Shomorony I, Cirulli ET, Huang L, Napier LA, Heister RR, Hicks M, et al. An unsupervised learning approach to identify novel signatures of health and disease from multimodal data. Genome Med 2020;12:7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Wang K, Zeng D, Chen Z, Yu W. Impact of left atrial diameter on all‐cause mortality of patients with STEMI undergoing primary percutaneous coronary intervention. Saudi Med J 2023;44:1260‐1268. doi: 10.15537/smj.2023.44.12.20230235 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Cao M, Li T, Li Z, Gong F, Chen Z. A retrospective study on the relationship between fibrosis‐4 index and all‐cause mortality in patients with acute myocardial infarction. Exp Ther Med 2022;24:643. doi: 10.3892/etm.2022.11580 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Mitchell C, Rahko PS, Blauwet LA, Canaday B, Finstuen JA, Foster MC, et al. Guidelines for performing a comprehensive transthoracic echocardiographic examination in adults: recommendations from the American Society of Echocardiography. J Am Soc Echocardiogr 2019;32:1‐64. doi: 10.1016/j.echo.2018.06.004 [DOI] [PubMed] [Google Scholar]
- 9. Rudski LG, Lai WW, Afilalo J, Hua L, Handschumacher MD, Chandrasekaran K, et al. Guidelines for the echocardiographic assessment of the right heart in adults: a report from the American Society of Echocardiography. J Am Soc Echocardiogr 2010;23:685‐713. doi: 10.1016/j.echo.2010.05.010 [DOI] [PubMed] [Google Scholar]
- 10. Prasad SB, Camuglia A, Lo A, Holland DJ, Guppy‐Coles K, Whalley G, et al. Hemodynamic validation of the E/e' ratio as a measure of left ventricular filling pressure in patients with non‐ST elevation myocardial infarction. Am J Cardiol 2020;125:507‐512. doi: 10.1016/j.amjcard.2019.11.010 [DOI] [PubMed] [Google Scholar]
- 11. Tsega W, Awoke W, Sendekie AK, Dagnew EM, Bayih H. Electrocardiogram and echocardiography findings and the outcomes of patients with myocardial infarction: Retrospective study in tertiary care hospitals in Northwest Ethiopia. PLoS ONE 2023;18:e0288698. doi: 10.1371/journal.pone.0288698 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Mulatero P, Monticone S, Deinum J, Amar L, Prejbisz A, Zennaro MC, et al. Genetics, prevalence, screening and confirmation of primary aldosteronism: a position statement and consensus of the working group on endocrine hypertension of the European Society of Hypertension. J Hypertens 2020;38:1919‐1928. doi: 10.1097/HJH.0000000000002510 [DOI] [PubMed] [Google Scholar]
- 13. Seo JH, Park I, Park SJ, Jeong DS, Bak M, Kim J, et al. Association of longitudinal left atrial strain with mortality after tricuspid valve surgery. ESC Heart Fail 2022;9:3868‐3875. doi: 10.1002/ehf2.14057 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Porter TR, Mulvagh SL, Abdelmoneim SS, Becher H, Belcik JT, Bierig M, et al. Clinical applications of ultrasonic enhancing agents in echocardiography: 2018 American Society of Echocardiography guidelines update. J Am Soc Echocardiogr 2018;31:241‐274. doi: 10.1016/j.echo.2017.11.013 [DOI] [PubMed] [Google Scholar]
- 15. Chen X‐Y, Yang Q‐M, Fang J‐X, Guo H‐F. The value of three‐dimensional speckle‐tracking echocardiography in the prediction of cardiovascular events in patients with hypertension complicated by acute myocardial infarction: A long‐term follow‐up study. Int J Gen Med 2021;14:6703‐6712. doi: 10.2147/IJGM.S330789 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Chen Y, Tao Y, Zhang L, Xu W, Zhou X. Diagnostic and prognostic value of biomarkers in acute myocardial infarction. Postgrad Med J 2019;95:210‐216. doi: 10.1136/postgradmedj-2019-136409 [DOI] [PubMed] [Google Scholar]
- 17. Scarpato BM, Locke BW, Bledsoe J, Knox DB, Conner K, Stoddard GJ, et al. The association between pulmonary artery enlargement and mortality in an emergency department population undergoing computed tomography pulmonary angiography. Pulm Circ 2023;13:e12225. doi: 10.1002/pul2.12225 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Wei L, Dong JX, Jin LX, He J, Zhao CX, Kong LC, et al. Peak early diastolic strain rate improves prediction of adverse cardiovascular outcomes in patients with ST‐elevation myocardial infarction. Radiol Med 2023;128:1372‐1385. doi: 10.1007/s11547-023-01700-y [DOI] [PubMed] [Google Scholar]
- 19. Ng VG, Lansky AJ, Meller S, Witzenbichler B, Guagliumi G, Peruga JZ, et al. The prognostic importance of left ventricular function in patients with ST‐segment elevation myocardial infarction: The HORIZONS‐AMI trial. Eur Heart J Acute Cardiovasc Care 2014;3:67‐77. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. de Simone G, Greco R, Mureddu G, Romano C, Guida R, Celentano A, et al. Relation of left ventricular diastolic properties to systolic function in arterial hypertension. Circulation 2000;101:152‐157. doi: 10.1161/01.CIR.101.2.152 [DOI] [PubMed] [Google Scholar]
- 21. Kwiecinski J, Tzolos E, Meah MN, Cadet S, Adamson PD, Grodecki K, et al. Machine learning with 18F‐sodium fluoride PET and quantitative plaque analysis on CT angiography for the future risk of myocardial infarction. J Nucl Med 2022;63:158‐165. doi: 10.2967/jnumed.121.262283 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Khera R, Haimovich J, Hurley NC, McNamara R, Spertus JA, Desai N, et al. Use of machine learning models to predict death after acute myocardial infarction. JAMA Cardiol 2021;6:633‐641. doi: 10.1001/jamacardio.2021.0122 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Liu W‐C, Lin C‐S, Tsai C‐S, Tsao T‐P, Cheng C‐C, Liou J‐T, et al. A deep learning algorithm for detecting acute myocardial infarction: Deep learning model to detect AMI. EuroIntervention 2021;17:765‐773. doi: 10.4244/EIJ-D-20-01155 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Guo Q, Lu X, Gao Y, Zhang J, Yan B, Su D, et al. Cluster analysis: a new approach for identification of underlying risk factors for coronary artery disease in essential hypertensive patients. Sci Rep 2017;7:43965. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Fuest KE, Ulm B, Daum N, Lindholz M, Lorenz M, Blobner K, et al. Clustering of critically ill patients using an individualized learning approach enables dose optimization of mobilization in the ICU. Crit Care 2023;27:1. doi: 10.1186/s13054-022-04291-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Figure S1. Supporting Information
Table S1. Echocardiography information of study population.
Table S2. Odds ratios of logistic models for the association between cluster and all‐cause mortality.