FCVM 09 1022658

TYPE Original Research
PUBLISHED 28 October 2022

DOI 10.3389/fcvm.2022.1022658
A prediction model for major

OPEN ACCESS adverse cardiovascular events in
EDITED BY
Masaki Izumo,
St. Marianna University School
patients with heart failure based
of Medicine, Japan
REVIEWED BY
on high-throughput
Kimi Sato,
University of Tsukuba, Japan
Toshinari Onishi,
echocardiographic data
Sakai City Medical Center, Japan
Norio Suzuki,
St. Marianna University School
Qinliang Sun1 , Shuangquan Jiang1 , Xudong Wang1 ,
of Medicine, Japan Jingchun Zhang2 , Yi Li1 , Jiawei Tian1* and Hairu Li1*
*CORRESPONDENCE 1
Department of Ultrasound Imaging, The Second Affiliated Hospital of Harbin Medical University,
Jiawei Tian
Harbin, China, 2 Department of Gastroenterology, Digestive Disease Hospital, Heilongjiang
jwtian2004@163.com
Provincial Hospital Affiliated to Harbin Institute of Technology, Harbin, China
Hairu Li
lihairu0419@126.com
SPECIALTY SECTION
This article was submitted to
Cardiovascular Imaging, Background: Heart failure (HF) is a serious end-stage condition of various
a section of the journal
heart diseases with increasing frequency. Few studies have combined clinical
Frontiers in Cardiovascular Medicine
features with high-throughput echocardiographic data to assess the risk of
RECEIVED 19August 2022
ACCEPTED 10 October 2022 major cardiovascular events (MACE) in patients with heart failure. In this
PUBLISHED 28 October 2022 study, we assessed the relationship between these factors and heart failure
CITATION to develop a practical and accurate prognostic dynamic nomogram model
Sun Q-L, Jiang S-Q, Wang X-D,
Zhang J-C, Li Y, Tian J-W and Li H-R
to identify high-risk groups of heart failure and ultimately provide tailored
(2022) A prediction model for major treatment options.
adverse cardiovascular events
in patients with heart failure based on Materials and methods: We conducted a prospective study of 468 patients
high-throughput echocardiographic
with heart failure and established a clinical predictive model. Modeling to
data.
Front. Cardiovasc. Med. 9:1022658. predict risk of MACE in heart failure patients within 6 months after discharge
doi: 10.3389/fcvm.2022.1022658 obtained 320 features including general clinical data, laboratory examination,
COPYRIGHT 2-dimensional and Doppler measurements, left ventricular (LV) and left atrial
© 2022 Sun, Jiang, Wang, Zhang, Li,
Tian and Li. This is an open-access (LA) speckle tracking echocardiography (STE), and left ventricular vector flow
article distributed under the terms of mapping (VFM) data, were obtained by building a model to predict the risk of
the Creative Commons Attribution
License (CC BY). The use, distribution
MACE within 6 months of discharge for patients with heart failure. In addition,
or reproduction in other forums is the addition of machine learning models also confirmed the necessity of
permitted, provided the original
increasing the STE and VFM parameters.
author(s) and the copyright owner(s)
are credited and that the original
Results: Through regular follow-up 6 months after discharge, MACE occurred
publication in this journal is cited, in
accordance with accepted academic in 156 patients (33.3%). The prediction model showed good discrimination
practice. No use, distribution or C-statistic value, 0.876 (p < 0.05), which indicated good identical calibration
reproduction is permitted which does
not comply with these terms. and clinical efficacy. In multiple datasets, through machine learning multi-
model comparison, we found that the area under curve (AUC) of the model
with VFM and STE parameters was higher, which was more significant with
the XGboost model.
Conclusion: In this study, we developed a prediction model and nomogram
to estimate the risk of MACE within 6 months of discharge among patients
with heart failure. The results of this study can provide a reference for
Frontiers in Cardiovascular Medicine 01 frontiersin.org

Sun et al. 10.3389/fcvm.2022.1022658
clinical physicians for detection of the risk of MACE in terms of clinical

characteristics, cardiac structure and function, hemodynamics, and enable its
prompt management, which is a convenient, practical and effective clinical
decision-making tool for providing accurate prognosis.
KEYWORDS
speckle tracking, vector flow mapping, heart failure, prediction model, nomogram
Introduction All efforts must be underway to examine clinical, laboratory,

and imaging data to better characterize heart failure phenotypes
Urbanization and the widespread use of cars have shifted (5, 6) and to develop cost-effective strategies to reliably
many people from active to sedentary lifestyles, increasing the identify at-risk populations at an early stage. Therefore,
incidence of chronic diseases such as obesity, hypertension, early identification of individuals with high-risk factors
diabetes, and coronary artery disease. Heart failure is a serious will provide an opportunity for the early intervention and
manifestation of the late stage of various heart diseases, prevention of MACE in these individuals. Stratifying patients
and its risk factors include coronary diseases, hypertension, according to risk of future outcomes and optimizing treatment
and diabetes, lifestyle factors such as smoking and alcohol strategies can help reduce their follow-up costs and mortality
consumption. Heart failure remains a serious clinical and public (7–9).
The 2019 American College of Cardiology expert consensus
health problem as the total number of patients living with heart
on heart failure suggests that assessing risk-increasing factors
failure increases, reflecting the chronic course of the disease as
can help inform decisions about preventive interventions
well as population growth and aging (1). With high readmission
(10). Many risk prediction models have been published
and mortality rate, heart failure places a huge financial burden
internationally (11–16). Predictors of hospitalization rates
on the healthcare system (2–4).
were age, history of hospitalization for heart failure, edema,
systolic blood pressure, and estimated glomerular filtration
Abbreviations: A, late diastolic transmitral flow velocity; a’, late diastolic rate (11). In 2019, the Korean Acute Heart Failure Registry
relaxation velocity at septal mitral annular position; ac, atrium establishes a risk score that predicts the risk of HF specific
contraction; ANOVA, analysis of variance; ALT, alanine transaminase;
ApoA, apolipoproteinA; ACEI, Angiotensin-converting enzyme inhibitors;
readmission or death at 30 days after discharge by using 12
ApoB, apolipoproteinB; AR, aortic regurgitation; ARBS, angiotensin predictors (12). In this study, Wang Lei developed a prediction
receptor blockers; AST, aspartate transaminase; AUC, area under curve; model and nomogram to estimate the risk of irreversible
BSA, body surface area; C2, 2-chamber; C3, 3-chamber; C4, 4-chamber;
CI, cardiac index; CV, cardiovascular; CW, clockwise; CCW, counter- worsening of cardiac function among acute decompensated
clockwise; DBP, diastolic blood pressure; DCM, dilated cardiomyopathy; HF patients, which provided a reference for clinicians to
ed, early diastole; e’, early diastolic relaxation velocity at septal mitral
detect and treat cardiac deterioration in a timely manner
annular position; E, early diastolic transmitral flow velocity; EL, energy
loss; ELC3P3, energy loss from 3-chamber images during isovolumic (13). Compared with traditional HF risk and non-race-specific
relaxation period; GDMT, Guideline Determined Medication Therapy; machine learning models, Race-specific and ML-based HF
GHB, glycatedhemoglobin; GLS, global longitudinal strain; GLSacC2LV,
left ventricular global longitudinal strain from 2-chamber images during
risk models that combine clinical, laboratory, and biomarker
atrial systolic period; GLSedC3LV, left ventricular global longitudinal data demonstrated superior performance which can identify
strain from 3-chamber images during early diastolic period; GLSacLV, distinct race-specific contributors of HF (14). A multivariate
average of left ventricular global longitudinal strain during atrial
systolic period; GLSR, global systolic strain rate; GLSRconduitC4LA,
left atrial global longitudinal strain rate during conduit period; HF,
heart failure; HCM, hypertrophic cardiomyopathy; HFpEF, heart failure index; MACE, major cardiovascular events; MEL, mean energy loss; MI,
with preserved ejection fraction; HFmrEF, heart failure with mildly myocardial infarction; MLP, Multi-LayerPerceptron; NA, not available;
reduced ejection fraction; HFrEF, heart failure with reduced ejection NT-pro BNP, N-terminal pro-brain natriuretic peptide; NYHA, New York
fraction; HFimpEF, heart failure with improved ejection fraction; HR, Heart Association; P, inorganicphosphorus; P1, isovolumic contraction
heart rate; HPLS, hyperlipidemia stability; ICM, ischemic cardiomyopathy; period; P2, ejection period; P3, isovolumetric relaxation period; P4,
INR, international normalized ratio; IQRs, interquartile ranges; IVPD, diastolic filling period; P5, atrial contraction period; PALS, peak atrial
intraventricular pressure difference; IVPDC2P4, intraventricular pressure longitudinal strain; PALSconduitLA, Peak LA longitudinal peak strain
differences from 2-chamber images during LV early filling period; during conduit period; s, systolic; S, septal side; SBP, systolic blood
IVPG, intraventricular pressure gradient; KNN, k-nearest neighbor pressure; SGLT-2, sodium-dependent glucose transporters 2; STE,
classifier; L, lateral side; LA, left atrial; LAD, left atrium diameter; speckle tracking echocardiography; SVM, support vector machine; TRV,
LASSO, Least Absolute Shrinkage and Selection Operator; LV, left tricuspid regurgitant flow velocity; TRPD, tricuspid regurgitation pressure
ventricular; LVEDd, left ventricular end diastolic diameter; LVESd, difference; VFM, vector flow mapping; VIF, the variance inflation factor;
left ventricular end systolic diameter; LVEF, left ventricular ejection VorAreaC4S, LV vortex area from 4-chamber images during systolic
fraction; LAVI, left atrial volume index; LVMI, left ventricular mass period; XGBoost, extreme gradient boosting classifier.

Sun et al. 10.3389/fcvm.2022.1022658
Cox regression model has been developed and validated to Standard echocardiography
predict long-term mortality and readmission risk of Chinese
patients with chronic heart failure (15). A convenient and All echocardiographic examinations follow the guidelines
accurate prognostic dynamic nomogram model for the risk of of the American Society of Echocardiography and use
all-cause death in acute heart failure patients was developed by commercially available ultrasound equipment (LISENDO 880,
Yin T, et al., which included N-terminal pro-brain natriuretic Hitachi Healthcare America, Twinsburg, Ohio, USA) (18, 19).
peptide (NT-pro BNP) and growth stimulation expresses gene 2
proteins (16).
However, most of these models are insufficient to reflect Speckle-tracking echocardiography
the overall situation of the patient. The aim of this study was to
develop a predictive model and a predictive nomogram model to Left atrial (LA) and LV endocardial boundaries were
estimate the risk of MACE within 6 months of discharge among manually determined using QRS early end-diastolic frames as
patients with HF taking into account clinical characteristics, a reference for image analysis (19). The trace can be adjusted
laboratory parameters of blood tests, speckle-tracking manually if necessary. The LV global longitudinal strain (GLS),
echocardiographic analysis, hemodynamic analysis, which the LV global systolic strain rate (GLSR) were calculated by
can better represent the structure and function of the heart. the software from 4-, 3-, and 2-chamber images, and averaged
for GLS and GLSR. We defined the following components of
LA strain: LA reservoir strain = peak (maximal) longitudinal
LA strain; LA pump strain = longitudinal LA strain measured
Materials and methods
between onset of the P wave and onset of the QRS complex; and
LA conduit strain = LA reservoir strain–LA pump strain (20).
Study cohort and study protocol Peak atrial longitudinal strain (PALS) represented LA reservoir
function, and peak atrial contraction strain represented LA
This was a single center study. Between July 2021
pump function, which were measured from the average of the
and February 2022, 505 consecutive patients who were
strain curves of all segments at the end of ventricular systole.
previously diagnosed with chronic heart failure referred to
Strain rate analysis was used to measure the peak LA strain rate
our institution for routine evaluation were screened for
during the same time-phase divisions described above. Oana
eligibility for this study. Criteria and definitions used in the
Mirea assess the level of agreement between non-dedicated
diagnosis of heart failure followed the 2021 heart failure
(left ventricular tracking software) and novel dedicated tracking
guidelines (17). Inclusion criteria were: (1) consecutive patients
software for RV and LA strain, and found left atrial mean values
who were previously diagnosed with chronic heart failure;
showed no statistical difference when obtained with the two
(2) pharmacologic therapy followed Guideline Determined
tracking tools (21; Figure 2).
Medication Therapy (GDMT) criteria before enrollment in
the study. Exclusion criteria were: (1) patients younger
than 18 years; (2) history of heart valve replacement, Vector flow mapping
congenital heart disease, severe heart Valve disease, malignant
tumors. (3) poor echocardiography windows, or suboptimal Images were analyzed using commercially available off-line
cardiac imaging. According to the criteria, 468 patients software (DASRS1, Hitachi Aloka Medical Ltd., Tokyo, Japan).
with HF were finally included (Figure 1). Heart failure Using the initial point of the QRS complex as a reference
patients were classified into the following four categories point for image analysis, the LV endocardial border was tracked
according to LVEF, including heart failure with preserved manually in end-diastolic frames and automatically by the
ejection fraction (HFpEF), heart failure with mildly reduced software. Previous studies have limited validation studies (22–
ejection fraction (HFmrEF), heart failure with reduced ejection 26), but the published data may contain useful clinical features.
fraction (HFrEF), heart failure with improved ejection fraction They included: (1) indexes of vortex from the flow-velocity
(HFimpEF). curve, such as vortex area, circulation, maximum vorticity, and
All scans were performed by a VFM imaging specialist, the vortex was automatically tracked and analyzed throughout
and all images were post-processing and analyzed by the cardiac cycle (27, 28); (2) energy loss (EL) and mean
two professional VFM researchers. Eligible patients were energy loss (MEL), which were calculated as peak values (23,
prospectively followed up at 28 days, 3 months, and 24); (3) the intraventricular pressure difference (IVPD) and
6 months after discharge. The study protocol followed the the intraventricular pressure gradient (IVPG), which were
Declaration of Helsinki and was approved by the Research measured on a line that went through the center of the LV
Ethics Committee. All subjects gave written informed consent from the base to the apex (28, 29). The STE and VFM analysis
for additional research tests and for the use of their data for were post-processed and analyzed by two professional VFM
research purposes. researchers (QL. SUN, Y. LI) (Figure 2).

Sun et al. 10.3389/fcvm.2022.1022658
FIGURE 1
Flowchart demonstrating the process of selection from a total of 504 patients.
FIGURE 2
Conventional echocardiography, speckle-tracking echocardiography, and vortex flow mapping echocardiography.
Follow-up and clinical outcome data cardiovascular (CV) death. If there are multiple events,
the first event is timed for analysis. Eligible patients were
Follow-up information was obtained during clinical prospectively followed up at 28 days, 3 months, and 6 months
consultations and patients who did not attend scheduled after discharge. Clinical and echocardiographic parameters were
consultations were contacted through telephone interviews tested for prediction of MACE in the study population. Clinical
with family members. The primary endpoint of the study was data, including clinical symptoms and signs, comorbidities,
the incidence of major adverse cardiac events (MACE). MACE laboratory test results, treatment during hospitalization, and
was defined as a complex of congestive HF hospitalization, clinical outcomes were obtained by reviewing each patient’s
non-fatal myocardial infarction (MI), non-fatal stroke, and medical records.

Sun et al. 10.3389/fcvm.2022.1022658
Statistical analysis prognostic variables were identified, including New York

Heart Association (NYHA), hyperlipidemia stability (HPLS),
We used R (R Foundation, Vienna, Austria) and Python ischemic cardiomyopathy (ICM), age, diastolic blood pressure
(Python Software Foundation, Beaverton, Oregon, USA) for (DBP), left ventricular end diastolic diameter (LVEDd), left
statistical analysis. Continuous variables are expressed as ventricular end systolic diameter (LVESd), peak mitral valve
mean ± SD, frequencies (percentages) for categorical variables blood flow during atrial contraction (A), LV ejection fraction
and medians (interquartile ranges, IQRs) for skewed variables. (LVEF), aortic regurgitation (AR), inorganicphosphorus (P),
The t-test was used to compare the means of continuous glycatedhemoglobin (GHB), apolipoproteinB (ApoB), sodium
variables when the data were normally distributed, else, the ion (Na), N-terminal brain natriuretic peptide (NTproBNP),
Mann-Whitney U test was used. One-way analysis of variance international normalized ratio (INR), fibrinogen, cardiac index
(ANOVA) was used for comparisons among multiple groups. (CI), LV vortex area from 4-chamber images during systolic
Categorical data comparisons between groups were performed period (VorAreaC4S), LV energy loss from 3-chamber images
using the χ2 test. The statistical significance level for each during isovolumic relaxation period (ELC3P3), intraventricular
test was set at α = 0.05; P < 0.05 (two-tailed) was considered pressure differences from 2-chamber images during LV
statistically significant. The Least Absolute Shrinkage and early filling period (IVPDC2P4), LA global longitudinal
Selection Operator (LASSO) regression algorithm and 10-fold strain rate during conduit period (GLSRconduitC4LA), LV
cross-validation were used to filter out the best variables most global longitudinal strain from 3-chamber images during
associated with 6-month MACE incidence. The importance of early diastolic period (GLSedC3LV), LV global longitudinal
variables is sorted by machine learning, a logistic regression strain from 2-chamber images during atrial systolic period
model is constructed after screening through Venn diagrams, (GLSacC2LV), average of LV global longitudinal strain during
and generates nomograms. We used the Kaplan-Meier method atrial systolic period (GLSacLV), peakLA longitudinal peak
to compare survival between groups. The concordance index strain during conduit period (PALSconduitLA) (Figures 3A,B).
(C-index) was used to measure the discriminative abilities Correlation heatmap showing associations between clinical
of the nomograms [Harrell et al. (30)]. Calibration was features and ultrasound parameters after lasso dimension
performed by examining the survival probability plot predicted reduction analysis (Figure 4A). Here, we used three different
by the nomogram. popular machine-learning methods including extreme gradient
boosting classifier (XGBoost), random forest classifier, k-nearest
neighbor classifier (KNN), which were widely applied in
Results bioinformatics in order to calculate the importance of each
influencing factor to the classification model and rank it,
Study population the top 20 parameters were selected (Figures 4B–D). R
(Venn Diagram package) was employed to generate the Venn
A total of 468 patients were included, with an average age of diagram, then 13 variables were recognized through the Venn
62 years, including 321 males (47.30%). There were 156 cases of diagram (Figure 4E), including NTproBNP, NYHA, GLSLV,
MACE, the incidence rate was 33.3%. Patients are divided into LVEDd, LVESd, A, Na, PALSconduitLA, VorAreaC4S, ApoB,
four categories based on LVEF: including HFpEF (147 patients, IVPDC2P4, GLSacC2LV, DBP. In this study, LVEF with high
31.4%), HFmrEF (95 patients, 20.29%), HFrEF (136 patients, clinical acceptance was included, and the variance inflation
29.06%), HFimpEF (90 patients, 19.23%). Demographic and factor (VIF) was calculated by collinearity analysis. LVESd was
baseline criteria are detailed by MACE (Table 1) and by LVEF excluded because of the higher VIF, and A was excluded because
classification (Table 2). Echocardiographic characteristics of of the higher missing. Finally, 12 candidate predictors were
STE parameters and VFM parameters are detailed by MACE selected to build the prediction model, including NYHA, DBP,
(Table 3) and by LVEF classification (Table 4). NTproBNP, Na,ApoB, LVEF, LVEDd, GLSLV, PALSconduitLA,
GLSacC2LV, VorAreaC4S, IVPDC2P4, and built a nomogram
based on the logistic regression model (Table 5).
Select optimal prognostic variables
and model development
Nomogram interpretation and model
Feature selection was performed using the LASSO validation
regression algorithm via the R package glmnet and 10-
fold cross-validation. This is consistent with the glmnet Nomogram for predicting MACE risk, and the point was
package recommendation for choosing λ, as either λ min the selected scoring standard or scale. For each independent
(minimum mean square error) or this value plus one standard variable, by drawing a line (through the ruler) perpendicular to
error. When lambda equaled 0.035, twenty-six optimal the point axis, the intersection points represent the values of the

Sun et al. 10.3389/fcvm.2022.1022658
TABLE 1 Baseline characteristics for heart failure (HF) patients during follow-up divided by major cardiovascular events (MACE).
Variables Overall No event at follow-up Event at follow-up Statistics P value

(n = 468) (n = 312) (n = 156)
Patient characteristics
Sex, male, n (%) 321 (68.59) 212 (67.95) 109 (69.87) 0.18 0.673
Age (years), median [IQR] 62.00 [53.00, 69.00] 62.00 [53.00, 69.00] 62.00 [53.00, 68.00] −0.02 0.982
BSA (m2 ), median [IQR] 1.79 [1.68, 1.93] 1.79 [1.68, 1.93] 1.78 [1.66, 1.95] 0.28 0.783
DBP (mmHg), median [IQR] 90.00 [80.00, 100.00] 90.00 [80.00, 101.00] 90.00 [80.00, 100.00] 1.85 0.063
SBP (mmHg), median [IQR] 150.00 [120.00, 170.00] 150.00 [126.00, 170.00] 140.00 [120.00, 170.00] 2.65 0.008
Smoke, n (%) 213 (45.51) 136 (43.59) 77 (49.36) 1.4 0.237
Alcohol, n (%) 134 (28.63) 85 (27.24) 49 (31.41) 0.88 0.347
HR (bpm), median [IQR] 79.00 [70.00, 90.00] 78.00 [69.00, 88.00] 81.00 [71.00, 92.00] −1.88 0.059
NYHA class, n (%) 63.19 < 0.001
I 145 (30.98) 127 (40.71) 18 (11.54)
II 96 (20.51) 70 (22.44) 26 (16.67)
III 114 (24.36) 67 (21.47) 47 (30.13)
IV 113 (24.15) 48 (15.38) 65 (41.67)
LVEF classification, n (%) 129.42 < 0.001
HFpEF, n (%) 147 (31.41) 127 (40.71) 20 (12.82)
HFmrEF, n (%) 95 (20.29) 63 (20.19) 32 (20.51)
HFrEF, n (%) 136 (29.06) 41 (13.14) 95 (60.90)
HFimpEF, n (%) 90 (19.23) 81 (25.96) 9 (5.77)
Medical history
Hypertension, n (%) 332 (70.94) 231 (74.04) 101 (64.74) 4.36 0.037
Hyperlipidemia, n (%) 233 (49.79) 171 (54.81) 62 (39.74) 9.44 0.002
Diabetes mellitus, n (%) 161 (34.40) 114 (36.54) 47 (30.13) 1.89 0.169
HCM, n (%) 16 (3.42) 15 (4.81) 1 (0.64) 5.47 0.019
DCM, n (%) 68 (14.53) 32 (10.26) 36 (23.08) 13.76 < 0.001
Ischemic cardiomyopathy, n (%) 200 (42.74) 123 (39.42) 77 (49.36) 4.2 0.041
Atrial fibrillation, n (%) 105 (22.44) 60 (19.23) 45 (28.85) 5.53 0.019
Medications
ACEI/ARBS, n (%) 249 (53.21) 167 (53.53) 82 (52.56) 0.04 0.844
ARNI, n (%) 230 (49.15) 154 (49.36) 76 (48.72) 0.02 0.896
β-blockers, n (%) 440 (94.02) 292 (93.59) 148 (94.87) 0.3 0.581
Aldosterone antagonists, n (%) 431 (92.09) 289 (92.63) 142 (91.03) 0.37 0.545
SGLT-2 inhibitors, n (%) 204 (43.59) 128 (41.03) 76 (48.72) 2.5 0.114
diuretics, n (%) 270 (57.69) 148 (47.44) 122 (78.21) 40.34 < 0.001
Antiplatelets, n (%) 305 (65.17) 190 (60.90) 115 (73.72) 7.53 0.006
Anticoagulants, n (%) 268 (57.26) 173 (55.45) 95 (60.90) 1.26 0.261
Statins, n (%) 144 (30.77) 102 (32.69) 42 (26.92) 1.63 0.202
Laboratory examinations
Glycatedhemoglobin (%), median [IQR] 6.00 [5.60, 7.20] 6.10 [5.50, 7.10] 6.00 [5.60, 7.50] 0.54 0.587
ApolipoproteinA (g/L), median [IQR] 1.15 [1.00, 1.32] 1.18 [1.05, 1.35] 1.07 [0.92, 1.23] 4.39 < 0.001
ApolipoproteinB (g/L), median [IQR] 0.92 [0.72, 1.10] 0.90 [0.72, 1.10] 0.96 [0.73, 1.11] −0.7 0.483
Inorganicphosphorus (mmol/L), median [IQR] 1.07 [0.93, 1.19] 1.04 [0.90, 1.17] 1.11 [0.98, 1.22] −3.29 < 0.001
Na (mmol/L), median [IQR] 139.00 [137.00, 141.40] 139.00 [137.00, 141.20] 138.60 [136.00, 141.90] 1.68 0.092
INR, median [IQR] 1.03 [0.97, 1.12] 1.02 [0.96, 1.09] 1.06 [1.00, 1.20] −4.51 < 0.001
AST (U/L), median [IQR] 22.00 [17.00, 31.00] 21.00 [16.00, 29.00] 25.00 [19.00, 35.00] −3.48 < 0.001
ALT (U/L), median [IQR] 24.00 [16.00, 35.00] 24.00 [17.00, 35.00] 23.00 [15.00, 36.00] 0.71 0.478
Creatin (umol/L), median [IQR] 84.00 [69.00, 102.00] 82.00 [69.00, 98.00] 87.00 [72.00, 116.00] −2.58 0.01
(Continued)

Sun et al. 10.3389/fcvm.2022.1022658
TABLE 1 (Continued)

(n = 468) (n = 312) (n = 156)
Urea/Crea, median [IQR] 83.96 [71.56, 100.66] 83.14 [70.79, 100.66] 86.91 [73.03, 100.61] −0.73 0.468
Plasma D-dimer (ng/mL), median [IQR] 125.00 [67.00, 263.00] 119.00 [61.00, 235.00] 155.00 [84.00, 323.00] −3.08 0.002
Fibrinogenr (g/L), median [IQR] 3.01 [2.59, 3.49] 2.98 [2.59, 3.41] 3.11 [2.62, 3.63] −1.67 0.095
SerumcalciumproteinI (ug/L), median [IQR] 0.11 [0.08, 0.14] 0.11 [0.08, 0.14] 0.11 [0.08, 0.13] −0.33 0.74
NTproBNP (pg/mL), median [IQR] 971.00 [303.0, 2383.00] 626.00 [223.0, 1565.00] 2372.00 [902.0, 6245.0] −8.77 < 0.001
General echocardiographic data
LVESd (mm), median [IQR] 44.40 [31.30, 53.30] 39.70 [29.10, 50.00] 51.20 [42.10, 59.20] −7.59 < 0.001
LVEDd (mm), median [IQR] 55.80 [48.30, 64.20] 53.10 [47.30, 60.80] 61.80 [54.40, 69.40] −7.05 < 0.001
LAD (mm), median [IQR] 40.50 [37.00, 44.30] 39.80 [36.30, 42.80] 42.70 [39.00, 46.60] −5.18 < 0.001
LAVI (ml/m2), median [IQR] 35.68 [27.89, 47.82] 33.36 [26.22, 44.71] 40.42 [33.39, 51.36] −4.62 < 0.001
LVMI (g/m2), median [IQR] 134.25 [107.54, 164.68] 129.41 [99.62, 159.00] 142.01 [122.18, 174.90] −4.45 < 0.001
E (m/s), median [IQR] 0.80 [0.62, 1.01] 0.78 [0.60, 1.00] 0.88 [0.67, 1.11] −3.09 0.002
A (m/s), median [IQR] 0.78 [0.33, 0.98] 0.81 [0.51, 1.00] 0.55 [0.00, 0.90] 3.83 < 0.001
E/A, mean (± SD) 1.13 ± 0.81 1.00 ± 0.63 1.42 ± 1.05 −3.89 < 0.001
e’-S (m/s), median [IQR] 4.90 [3.70, 6.20] 5.10 [4.00, 6.40] 4.30 [3.20, 5.70] 4.42 < 0.001
a’-S (m/s), mean (± SD) 7.48 ± 2.54 7.90 ± 2.42 6.53 ± 2.56 4.92 < 0.001
e’-L (m/s), median [IQR] 6.80 [4.90, 9.20] 7.30 [5.20, 9.40] 6.20 [4.20, 8.60] 3.39 < 0.001
a’-L (m/s), mean (± SD) 9.53 ± 3.29 9.88 ± 3.08 8.73 ± 3.58 2.96 0.003
e’/a’-S, mean (± SD) 0.70 ± 0.29 0.69 ± 0.28 0.73 ± 0.32 −1.31 0.191
e’/a’-L, mean (± SD) 0.80 ± 0.44 0.79 ± 0.42 0.84 ± 0.50 −1.02 0.31
TRV (m/s), mean (± SD) 2.90 ± 0.60 2.85 ± 0.59 2.96 ± 0.61 −1.41 0.159
TRPD (mmHg), mean (± SD) 34.96 ± 14.64 33.73 ± 14.50 36.43 ± 14.66 −1.34 0.181
tei-RV, median [IQR] 0.37 [0.29, 0.45] 0.35 [0.27, 0.44] 0.40 [0.33, 0.47] −4.25 < 0.001
tei-LV, median [IQR] 0.38 [0.31, 0.44] 0.36 [0.29, 0.43] 0.41 [0.37, 0.47] −5.96 < 0.001
E/e’-RVav, median [IQR] 6.84 [5.39, 8.40] 6.55 [5.16, 8.09] 7.38 [5.85, 9.22] −4.28 < 0.001
E/e’-LVav, median [IQR] 14.16 [10.50, 18.29] 13.16 [9.66, 16.49] 16.13 [13.20, 21.76] −6.29 < 0.001
LVEF, median [IQR] 41.80 [32.40, 61.00] 50.20 [36.00, 62.00] 34.70 [27.50, 43.70] 8.34 < 0.001
p < 0.05 indicates statistical significance. BSA, body surface area; DBP, diastolic blood pressure; SBP, systolic blood pressure; HR, heart rate; NYHA, New York Heart Association;
DCM, Dilated cardiomyopathy; HCM, hypertrophic cardiomyopathy; ACEI, Angiotensin-converting enzyme inhibitors; ARBS, angiotensin receptor blockers; SGLT-2, sodium-dependent
glucose transporters 2; INR, international normalized ratio; ALT, alanine transaminase; AST, aspartate transaminase; LVEDd, left ventricular end diastolic diameter; LVESd, left ventricular
end systolic diameter; LAD, left atrium diameter; LAVI, left atrial volume index; LVMI, left ventricular mass index; A, late diastolic transmitral flow velocity; E, early diastolic transmitral
flow velocity; e’, early diastolic relaxation velocity at septal mitral annular position; L, lateral side; S, septal side; TRV, tricuspid regurgitant flow velocity; TRPD, tricuspid regurgitation
pressure difference; LVEF, left ventricular ejection fraction; NA, not available.
independent variables. The value of each variable is scored on a MACE, and solid horizontal line is patients without MACE. The
scale of 0 to 100, and the scores for each variable are summed graph shows the expected net benefit per patient in relation to
to estimate the position perpendicular to the axis. That sum the nomogram MACE risk prediction. Decision curve analysis
enables us to predict the probability of MACE risk in patient indicated that the clinical validity of the model was moderate.
with HF (Figure 5A). Estimated odds ratios determined in a
logistic regression model as shown in the forest plot (Figure 5B).
Receiver operating characteristic curve for the nomogram Survival curves based on nomogram
generated using bootstrap resampling, which showed a good scores and predictive risk
discriminative ability for the prediction model (C-statistics:
0.876 [95% CI, 0.844– 0.907]) (Figure 5C). The Nomogram Patients were divided into four groups according to the
calibration plots of the model based on the bootstrap method quartile total nomogram scores (nomgroup = 0, 1, 2, 3),
showed good performance (Figure 5D). When the solid line and each group had 117 patients. A prediction model with
(performance nomogram) was closer to the dotted line (ideal a prediction probability of less than 0.5 is considered a low
model), the prediction accuracy of the nomogram was better. risk group, and patients were divided into high-risk and low-
Decision curve analysis for predictive models (Figure 5E). Solid risk groups (low-risk group = 1, high-risk group = 2). Of
red line is predictive models, solid blue line is patients with these, 342 patients (73.08%) belonged to the low-risk group.

Sun et al. 10.3389/fcvm.2022.1022658
TABLE 2 Baseline characteristics for heart failure (HF) patients divided by left ventricular end fraction (LVEF) classification.
Variables Overall HFpEF HFmrEF HFrEF HFimpEF Statistics P value

(n = 468) (n = 147) (n = 95) (n = 136) (n = 90)
Patient characteristics
Sex, male, n (%) 321 (68.59) 96 (65.31) 57 (60.00) 102 (75.00) 66 (73.33) 7.52 0.057
Age (years), median [IQR] 62.00 [53.00, 61.00 [53.00, 64.00 [55.00, 63.00 [53.00, 58.00 [52.00, 5.59 0.133
69.00] 68.00] 71.00] 69.00] 67.00]
BSA (m2 ), median [IQR] 1.79 [1.68, 1.93] 1.80 [1.68, 1.95] 1.79 [1.69, 1.92] 1.78 [1.66, 1.92] 1.80 [1.66, 1.93] 0.94 0.816
DBP (mmHg), median [IQR] 90.00 [80.00, 98.00 [90.00, 88.00 [78.00, 85.00 [75.00, 90.00 [80.00, 34.84 < 0.001
100.00] 110.00] 100.00] 93.00] 100.00]
SBP (mmHg), median [IQR] 150.00 [120.00, 160.00 [140.00, 146.00 [121.00, 130.00 [115.00, 150.00 [120.00, 54.36 < 0.001
170.00] 180.00] 166.00] 150.00] 170.00]
Smoke, n (%) 213 (45.51) 63 (42.86) 46 (48.42) 65 (47.79) 39 (43.33) 1.2 0.753
Alcohol, n (%) 134 (28.63) 36 (24.49) 25 (26.32) 49 (36.03) 24 (26.67) 5.30 0.151
HR (bpm), median [IQR] 79.00 [70.00, 77.00 [67.00, 77.00 [70.00, 85.00 [75.00, 78.00 [69.00, 26.40 < 0.001
90.00] 85.00] 86.00] 96.00] 89.00]
NYHA class, n (%) NA NA
I 145 (30.98) 109 (74.15) 10 (10.52) 0 (0.00) 26 (28.89)
II 96 (20.51) 25 (17.01) 51 (53.68) 10 (7.35) 10 (11.11)
III 114 (24.36) 12 (8.16) 22 (23.15) 49 (36.03) 31 (34.44)
IV 113 (24.15) 1 (0.68) 12 (12.63) 77 (56.62) 23 (25.56)
Medical history
Hypertension, n (%) 332 (70.94) 125 (85.03) 67 (70.53) 77 (56.62) 63 (70.00) 27.74 < 0.001
Hyperlipidemia, n (%) 233 (49.79) 95 (64.63) 47 (49.47) 47 (34.56) 44 (48.89) 25.60 < 0.001
Diabetes mellitus, n (%) 161 (34.40) 52 (35.37) 37 (38.95) 36 (26.47) 36 (40.00) 5.97 0.113
HCM, n (%) 16 (3.42) 16 (10.88) 0 (0.00) 0 (0.00) 0 (0.00) NA NA
DCM, n (%) 68 (14.53) 0 (0.00) 5 (5.26) 51 (37.50) 12 (13.33) NA NA
Ischemic cardiomyopathy, n (%) 200 (42.74) 25 (17.01) 60 (63.16) 63 (46.32) 52 (57.78) 64.99 < 0.001
Atrial fibrillation, n (%) 105 (22.44) 23 (15.65) 14 (14.74) 48 (35.29) 20 (22.22) 20.05 < 0.001
Medications
ACEI/ARBS, n (%) 249 (53.21) 12 7 (86.40) 36 (37.90) 52 (38.24) 34 (37.78) 94.83 < 0.001
ARNI, n (%) 230 (49.15) 26 (17.69) 61 (64.21) 81 (59.56) 62 (68.89) 86.77 < 0.001
β-blockers, n (%) 440 (94.02) 130 (88.44) 92 (96.84) 133 (97.79) 85 (94.44) 12.97 0.005
Aldosterone antagonists, n (%) 431 (92.09) 136 (92.51) 85 (89.47) 126 (92.64) 84 (93.33) 1.18 0.76
SGLT-2 inhibitors, n (%) 204 (43.59) 47 (31.97) 36 (37.90) 78 (57.35) 43 (47.78) 20.44 < 0.001
diuretics, n (%) 270 (57.69) 30 (20.41) 72 (75.79) 116 (85.29) 52 (57.78) 138.92 < 0.001
Antiplatelets, n (%) 305 (65.17) 77 (52.38) 72 (75.79) 92 (67.65) 64 (71.11) 17.08 < 0.001
Anticoagulants, n (%) 268 (57.27) 72 (48.98) 59 (62.11) 85 (62.50) 52 (57.78) 6.57 0.087
Statins, n (%) 144 (30.77) 59 (40.14) 29 (30.53) 30 (22.06) 26 (28.89) 11.05 0.011
Laboratory examinations
Glycatedhemoglobin (%), median [IQR] 6.00 [5.60, 7.20] 6.00 [5.50, 6.80] 6.10 [5.60, 8.00] 5.90 [5.60, 7.00] 6.00 [5.80, 7.80] 12.61 0.006
ApolipoproteinA (g/L), median [IQR] 1.15 [1.00, 1.32] 1.20 [1.09, 1.38] 1.12 [0.99, 1.35] 1.07 [0.92, 1.21] 1.17 [1.03, 1.31] 27.39 < 0.001
ApolipoproteinB (g/L), median [IQR] 0.92 [0.72, 1.10] 0.94 [0.72, 1.15] 0.90 [0.71, 1.12] 0.93 [0.73, 1.09] 0.87 [0.69, 1.02] 4.71 0.195
Inorganicphosphorus (mmol/L), median [IQR] 1.07 [0.93, 1.19] 1.02 [0.89, 1.15] 1.08 [0.98, 1.18] 1.13 [0.98, 1.22] 1.04 [0.89, 1.19] 16.50 < 0.001
Na (mmol/L), median [IQR] 139.00 [137.00, 139.00 [137.00, 138.60 [137.00, 138.50 [136.30, 139.00 [137.30, 4.23 0.238
141.40] 142.00] 141.40] 141.40] 141.00]
INR, median [IQR] 1.03 [0.97, 1.12] 1.00 [0.96, 1.06] 1.02 [0.97, 1.09] 1.09 [1.02, 1.22] 1.02 [0.96, 1.08] 46.31 < 0.001
AST (U/L), median [IQR] 22.00 [17.00, 19.00 [16.00, 23.00 [16.00, 27.00 [20.00, 21.00 [17.00, 26.04 < 0.001
31.00] 26.00] 34.00] 36.00] 29.00]
ALT (U/L), median [IQR] 24.00 [16.00, 21.00 [15.00, 26.00 [17.00, 24.00 [15.00, 25.00 [19.00, 6.14 0.11
35.00] 30.00] 38.00] 35.00] 37.00]
Creatin (umol/L), median [IQR] 84.00 [69.00, 79.00 [66.00, 84.00 [68.00, 91.00 [75.00, 81.00 [71.00, 17.69 < 0.001
102.00] 93.00] 105.00] 116.00] 95.00]
(Continued)

Sun et al. 10.3389/fcvm.2022.1022658
TABLE 2 (Continued)

(n = 468) (n = 147) (n = 95) (n = 136) (n = 90)
Urea/Crea, median [IQR] 83.96 [71.56, 83.26 [71.45, 85.19 [72.39, 84.17 [73.03, 83.13 [69.46, 0.6 0.89
100.66] 105.80] 99.56] 99.12] 102.71]
Plasma D-dimer (ng/mL), median [IQR] 125.00 [67.00, 87.00 [56.00, 142.00 [82.00, 186.00 [93.00, 110.00 [58.00, 33.00 < 0.001
263.00] 158.00] 263.00] 344.00] 242.00]
Fibrinogenr (g/L), median [IQR] 3.01 [2.59, 3.49] 2.99 [2.59, 3.38] 3.03 [2.57, 3.58] 3.02 [2.50, 3.63] 2.99 [2.67, 3.49] 0.49 0.92
SerumcalciumproteinI (ug/L), median [IQR] 0.11 [0.08, 0.14] 0.11 [0.08, 0.14] 0.10 [0.08, 0.14] 0.11 [0.08, 0.14] 0.10 [0.07, 0.14] 3.36 0.339
NTproBNP (pg/mL), median [IQR] 971.00 [303.00, 283.00 [52.00, 833.00 [452.00, 2959.00 1030.00 [304.00, 180.37 < 0.001
2383.00] 799.00] 1980.000] [1551.00, 1788.00]
7144.00]
General echocardiographic data
LVESd (mm), median [IQR] 44.40 [31.30, 28.60 [24.70, 43.10 [40.00, 56.20 [51.90, 47.90 [33.70, 303.31 < 0.001
53.30] 32.30] 46.30] 61.10] 55.00]
LVEDd (mm), median [IQR] 55.80 [48.30, 47.30 [44.20, 55.80 [52.800, 65.80 [60.60, 58.70 [50.50, 245.00 < 0.001
64.20] 49.70] 6.10] 71.70] 65.20]
LAD (mm), median [IQR] 40.50 [37.00, 37.70 [35.00, 40.00 [37.30, 44.20 [40.30, 41.00 [36.50, 75.81 < 0.001
44.30] 41.30] 42.00] 48.70] 44.00]
LAVI (g/m2 ), median [IQR] 35.69 [27.89, 30.65 [25.11, 34.20 [26.88, 42.07 [34.80, 35.15 [27.32, 50.32 < 0.001
47.82] 41.12] 44.32] 56.68] 49.87]
LVMI (g/m2 ), median [IQR] 134.25 [107.54, 108.33 [92.91, 127.38 [108.26, 158.73 [134.46, 138.60 [118.58, 96.57 < 0.001
164.68] 136.72] 153.64] 187.16] 166.09]
E (m/s), median [IQR] 0.80 [0.62, 1.01] 0.75 [0.60, 0.90] 0.70 [0.60, 0.83] 1.00 [0.80, 1.24] 0.80 [0.61, 1.00] 48.36 < 0.001
A (m/s), median [IQR] 0.78 [0.33, 0.98] 0.84 [0.64, 1.00] 0.84 [0.62, 1.00] 0.42 [0.00, 0.80] 0.78 [0.30, 1.00] 47.41 < 0.001
E/A ratio, mean (± SD) 0.79 [0.64, 1.32] 0.77 [0.64, 0.90] 0.75 [0.60, 0.90] 1.44 [0.73, 2.52] 0.80 [0.65, 1.29] 25.67 < 0.001
e’-S (m/s), median [IQR] 4.90 [3.70, 6.20] 5.60 [4.50, 6.70] 4.50 [3.60, 5.90] 4.50 [3.10, 5.50] 4.80 [3.50, 6.00] 40.54 < 0.001
a’-S (m/s), mean (± SD) 7.50 [5.70, 9.10] 8.7 [7.40, 10.20] 7.00 [5.90, 8.50] 5.50 [4.10, 7.60] 7.40 [5.30, 9.00] 70.43 < 0.001
e’-L (m/s), median [IQR] 6.80 [4.90, 9.20] 8.7 [6.10, 10.30] 6.20 [4.30, 8.40] 6.30 [4.50, 8.70] 6.50 [4.60, 8.80] 35.78 < 0.001
a’-L (m/s), mean (± SD) 9.4 [7.50, 11.60] 10.2 [9.00, 12.10] 9.0 [7.60, 11.40] 8.0 [5.50, 11.10] 8.7 [7.20, 11.10] 28.40 < 0.001
e’/a’-S, mean (± SD) 0.66 [0.52, 0.81] 0.62 [0.52, 0.73] 0.65 [0.46, 0.83] 0.71 [0.59, 0.88] 0.66 [0.46, 0.83] 11.60 0.009
e’/a’-L, mean (± SD) 0.73 [0.52, 0.90] 0.76 [0.59, 0.93] 0.65 [0.47, 0.85] 0.77 [0.54, 0.95] 0.66 [0.49, 0.86] 9.49 0.023
TRV (m/s), mean (± SD) 2.80 [2.40, 3.22] 2.82 [2.42, 3.10] 2.70 [2.400, 2.90] 2.81 [2.52, 3.31] 2.85 [2.40, 3.40] 2.96 0.397
TRP (mmHg), mean (± SD) 31.00 [24.00, 31.00 [24.00, 29.00 [23.00, 32.00 [25.00, 32.00 [24.00, 4.02 0.26
42.00] 38.00] 33.00] 44.00] 46.00]
tei-RV, median [IQR] 0.37 [0.29, 0.45] 0.30 [0.25, 0.38] 0.37 [0.31, 0.43] 0.42 [0.37, 0.49] 0.38 [0.29, 0.44] 77.33 < 0.001
tei-LV, median [IQR] 0.38 [0.31, 0.44] 0.31 [0.26, 0.38] 0.38 [0.34, 0.43] 0.44 [0.39, 0.50] 0.37 [0.32, 0.47] 114.76 < 0.001
E/e’-RVav, median [IQR] 6.84 [5.39, 8.40] 6.19 [5.12, 7.57] 6.71 [5.16, 7.73] 7.43 [5.82, 9.45] 6.98 [5.59, 8.37] 21.51 < 0.001
E/e’-LVav, median [IQR] 14.16 [10.51, 10.08 [8.30, 14.82 [12.25, 17.61 [13.46, 15.31 [11.57, 113.05 < 0.001
18.29] 13.62] 17.09] 23.88] 20.62]
LVEF, median [IQR] 41.80 [32.40, 62.00 [61.00, 43.40 [41.30, 29.30 [26.60, 35.50 [30.80, 350.79 < 0.001
61.00] 64.00] 46.60] 33.50] 57.90]
p < 0.05 indicates statistical significance. BSA, body surface area; DBP, diastolic blood pressure; SBP, systolic blood pressure; HR, heart rate; NYHA, New York Heart Association;
DCM, Dilated cardiomyopathy; HCM, hypertrophic cardiomyopathy; ACEI, angiotensin-converting enzyme inhibitors; ARBS, angiotensin receptor blockers; SGLT-2, sodium-dependent
glucose transporters 2; INR, international normalized ratio; ALT, alanine transaminase; AST, aspartate transaminase; LVEDd, left ventricular end diastolic diameter; LVESd, left ventricular
end systolic diameter; LAD, left atrium diameter; LAVI, left atrial volume index; LVMI, left ventricular mass index; A, late diastolic transmitral flow velocity; E, early diastolic transmitral
flow velocity; e’, early diastolic relaxation velocity at septal mitral annular position; L, lateral side; S, septal side; TRV, tricuspid regurgitant flow velocity; TRPD, tricuspid regurgitation
pressure difference; LVEF, left ventricular ejection fraction; NA, not available.
The KM survival curves and cumulative survival curves were Performance of machine learning
used to compare survival times among different groups was algorithms among the groups
made by log-rank test. The median survival time (LT50) for
different groups. The horizontal axis of the KM survival curve We divided patients into training and validation sets using
represents time, and the vertical axis represents probability MACE as the categorical outcome variable. Parameters include
(Figure 6). general parameters, STE-related parameters and VFM-related

Sun et al. 10.3389/fcvm.2022.1022658
TABLE 3 Echocardiographic characteristics of speckle tracking echocardiography (STE) parameters and vector flow mapping (VFM) parameters
divided by major cardiovascular events (MACE).

(n = 468) (n = 312) (n = 156)
STE parameters
GLS-LV (%), median [IQR] −10.81 [−14.90, −8.22] −12.04 [−16.30, −9.47] −8.64 [−11.28, −6.47] −8.05 < 0.001
GLSR-sLV (s−1 ), median [IQR] −0.60 [−0.76, −0.46] −0.64 [−0.83, −0.49] −0.50 [−0.62, −0.36] −6.76 < 0.001
GLSR-edLV (s−1 ), median [IQR] 0.56 [0.40, 0.76] 0.61 [0.44, 0.81] 0.49 [0.35, 0.64] 4.54 < 0.001
GLSR-acLV (s−1 ), median [IQR] 0.49 [0.32, 0.68] 0.51 [0.38, 0.71] 0.42 [0.27, 0.58] 4.13 < 0.001
PALS-reservoir LA (%),median [IQR] 21.14 [12.96, 30.29] 24.08 [15.69, 32.48] 17.11 [10.36, 23.34] 6.16 < 0.001
PALS-conduit LA (%),median [IQR] 9.53 [6.58, 13.45] 10.66 [7.47, 14.37] 7.60 [5.50, 10.30] 6.5 < 0.001
PALS-pump LA (%),median [IQR] 11.42 [4.98, 17.06] 13.39 [5.89, 18.02] 8.81 [4.00, 14.04] 4.65 < 0.001
GLSR-reservoir LA (s−1 ), median [IQR] 0.95 [0.66, 1.37] 1.09 [0.73, 1.46] 0.83 [0.62, 1.08] 5.44 < 0.001
GLSR-conduit LA (s−1 ), median [IQR] −0.93 [−1.30, −0.69] −0.99 [−1.38, −0.73] −0.81 [−1.06, −0.63] −4.6 < 0.001
GLSR-pump LA (s−1 ), median [IQR] −1.21 [−1.83, −0.68] −1.33 [−1.96, −0.78] −0.99 [−1.61, −0.58] −4.01 < 0.001
VFM parameters
MeanELP1 [J/(mˆ3 s)], median [IQR] 3.05 [1.84, 4.89] 3.16 [2.01, 4.94] 2.90 [1.40, 4.84] 1.91 0.057
EnergyLossP1 [J/(m s eˆ3)], median [IQR] 20.44 [13.13, 33.81] 20.31 [13.13, 32.68] 20.54 [13.19, 37.08] 0.16 0.872
EnergyLossP3 [J/(m s eˆ3)], median [IQR] 14.69 [8.76, 24.14] 14.15 [7.67, 23.70] 16.15 [9.87, 25.07] −1.82 0.068
EnergyLossP4 [J/(m s eˆ3)], median [IQR] 27.11 [16.70, 46.62] 26.46 [16.03, 45.89] 28.40 [18.08, 48.99] −0.66 0.509
IVPDP1 (mmHg), median [IQR] 0.78 [0.47, 1.22] 0.86 [0.52, 1.26] 0.67 [0.40, 1.00] 3.72 < 0.001
IVPDP2 (mmHg), median [IQR] 1.02 [0.71, 1.41] 1.07 [0.71, 1.45] 0.91 [0.71, 1.29] 2.09 0.036
IVPDP3 (mmHg), median [IQR] −0.36 [−0.60, −0.25] −0.38 [−0.63, −0.28] −0.32 [−0.52, −0.20] −3.28 0.001
IVPDP4 (mmHg), median [IQR] −0.97 [−1.39, −0.65] −1.03 [−1.42, −0.70] −0.86 [−1.17, −0.56] −3.23 0.001
IVPDP5 (mmHg), median [IQR] 0.52 [0.29, 0.89] 0.53 [0.27, 0.91] 0.51 [0.30, 0.78] 0.46 0.649
IVPGP1 (mmHg/mmeˆ3), median [IQR] −8.96 [−15.22, −5.53] −9.91 [−15.97, −6.02] −7.74 [−12.11, −4.37] −4.07 < 0.001
IVPGP2 (mmHg/mmeˆ3), median [IQR] −12.12 [−16.90, −8.30] −12.84 [−17.95, −8.57] −10.42 [−15.23, −8.15] −2.59 0.01
IVPGP3 (mmHg/mmeˆ3), median [IQR] 4.45 [3.00, 7.30] 4.89 [3.37, 7.69] 3.76 [2.47, 6.32] 3.89 < 0.001
IVPGP4 (mmHg/mmeˆ3), median [IQR] 11.98 [7.98, 17.47] 13.25 [8.80, 18.25] 10.15 [6.57, 14.56] 3.97 < 0.001
IVPGP5 (mmHg/mmeˆ3), median [IQR] −6.46 [−10.95, −3.31] −6.62 [−11.55, −3.34] −6.06 [−9.53, −3.30] −1.02 0.309
VorArea-s (mmˆ2), median [IQR] 454.22 [354.72, 574.46] 456.49 [344.67, 568.96] 452.18 [367.50, 577.34] −0.14 0.888
Circulation-s (mˆ2/s eˆ3), median [IQR] 16.90 [11.77, 22.03] 17.42 [12.43, 22.27] 16.10 [10.37, 21.37] 1.95 0.051
VorArea-ed (mmˆ2), median [IQR] 359.00 [257.14, 476.48] 356.72 [248.46, 468.67] 362.95 [291.99, 493.80] −1.96 0.05
Circulation-ed (mˆ2/s eˆ3), median [IQR] 13.13 [7.57, 21.80] 13.03 [7.40, 21.90] 13.90 [7.60, 21.77] −0.35 0.727
VorArea-ac (mmˆ2), median [IQR] 428.02 [323.87, 531.24] 425.12 [324.89, 523.87] 441.81 [319.24, 552.93] −0.41 0.682
Circulation-ac (mˆ2/s eˆ3), median [IQR] 18.03 [12.10, 25.53] 18.97 [12.77, 26.57] 15.77 [10.93, 23.57] 2.73 0.006
vorticity-ac-CCW (1/s), median [IQR] −105.8 [−142.94, −80.24] −107.2 [−144.99, −84.58] −103.6 [−140.37, −75.86] −1.2 0.231
vorticity-ac-CW (1/s), median [IQR] 114.97 [85.83, 149.68] 119.39 [89.86, 153.89] 108.21 [78.40, 145.27] 2 0.045
vorticity-ed-CCW (1/s), median [IQR] −99.07 [−135.19, −76.44] −97.37 [−134.90, −74.64] −101.4 [−136.47, −80.00] 0.58 0.564
vorticity-ed-CW (1/s), median [IQR] 101.54 [72.03, 140.46] 99.37 [71.15, 140.68] 107.40 [76.73, 139.43] −1.05 0.292
vorticity-s-CCW (1/s), median [IQR] −100.7 [−130.23, −75.93] −102.9 [−132.92, −78.65] −97.6 [−124.15, −68.81] −1.97 0.049
vorticity-s-CW (1/s), median [IQR] 110.59 [82.79, 149.78] 115.23 [86.19, 152.53] 103.63 [74.75, 144.74] 2.31 0.021
p < 0.05 indicates statistical significance. STE, speckle tracking echocardiography; GLS, global longitudinal strain; GLSR, global systolic strain rate; PALS, peak atrial longitudinal strain;
VFM, vector flow mapping; MeanEL, mean energy loss; IVPD, intraventricular pressure difference; IVPG, intraventricular pressure gradient; s, systolic; ed, early diastolic; ac, atrium
contraction; P1-P5, isovolumic contraction period, ejection period, isovolumetric relaxation period, diastolic filling period, atrium contraction period; CW, clockwise; CCW, counter-
clockwise.

Sun et al.
TABLE 4 Echocardiographic characteristics of speckle tracking echocardiography (STE) parameters and vector flow mapping (VFM) parameters divided by left ventricular end fraction
(LVEF) classification.

(n = 468) (n = 147) (n = 95) (n = 136) (n = 90)
STE parameters
GLS-LV (%), median [IQR] −10.81 [−14.90, −8.22] −16.25 [−18.74, −13.16] −11.06 [−12.81, −9.30] −7.29 [−9.11, −5.99] −10.07 [−12.64, −8.05] 246.93 < 0.001
GLSR-sLV (s−1 ), median [IQR] −0.60 [−0.76, −0.46] −0.83 [−0.96, −0.66] −0.60 [−0.68, −0.49] −0.43 [−0.53, −0.35] −0.55 [−0.70, −0.46] 189.24 < 0.001
GLSR-edLV (s−1 ), median [IQR] 0.56 [0.40, 0.76] 0.76 [0.54, 0.98] 0.55 [0.41, 0.70] 0.46 [0.32, 0.59] 0.49 [0.38, 0.66] 99.02 < 0.001
GLSR-acLV (s−1 ), median [IQR] 0.49 [0.32, 0.68] 0.56 [0.44, 0.76] 0.57 [0.46, 0.72] 0.34 [0.25, 0.49] 0.45 [0.33, 0.61] 77.00 < 0.001
PALS-reservoir LA (%),median [IQR] 21.14 [12.96, 30.29] 30.86 [21.19, 37.80] 23.32 [18.88, 29.11] 12.70 [9.11, 19.07] 18.72 [11.47, 26.94] 151.51 < 0.001
PALS-conduit LA (%), median [IQR] 9.53 [6.58, 13.45] 13.78 [10.92, 18.03] 9.53 [7.25, 12.35] 6.87 [5.01, 9.56] 8.49 [6.48, 10.66] 137.72 < 0.001
PALS-pump LA (%), median [IQR] 11.42 [4.98, 17.06] 16.64 [11.14, 21.79] 13.17 [9.05, 17.06] 5.07 [3.49, 10.55] 9.72 [4.52, 15.41] 103.30 < 0.001
GLSR-reservoir LA (s−1 ), median [IQR] 0.95 [0.66, 1.37] 1.42 [0.96, 1.83] 1.08 [0.82, 1.35] 0.69 [0.55, 0.91] 0.89 [0.62, 1.25] 133.14 < 0.001
GLSR-conduit LA (s−1 ), median [IQR] −0.93 [−1.30, −0.69] −1.24 [−1.66, −0.88] −0.95 [−1.24, −0.73] −0.72 [−0.96, −0.58] −0.83 [−1.13, −0.65] 77.08 < 0.001
GLSR-pump LA (s−1 ), median [IQR] −1.21 [−1.83, −0.68] −1.60 [−2.28, −1.14] −1.36 [−1.89, −1.04] −0.68 [−1.28, −0.48] −1.00 [−1.61, −0.60] 89.56 < 0.001
VFM parameters
MeanELP1 [J/(mˆ3 s)], median [IQR] 3.05 [1.84, 4.89] 3.56 [2.27, 5.68] 3.10 [2.09, 4.24] 2.51 [1.30, 4.30] 2.88 [1.73, 5.22] 18.06 < 0.001
11
MeanELP4 [J/(mˆ3 s)], median [IQR] 4.36 [2.54, 7.49] 4.90 [3.20, 8.04] 4.19 [2.32, 7.10] 3.95 [2.16, 7.28] 4.18 [2.39, 7.47] 10.93 0.012
EnergyLossP1 [J/(m s eˆ3)], median [IQR] 20.45 [13.13, 33.81] 20.20 [13.29, 30.70] 20.86 [14.55, 32.68] 21.61 [12.68, 36.57] 18.96 [12.61, 36.64] 0.40 0.941
IVPDP1 (mmHg), median [IQR] 0.78 [0.47, 1.22] 1.13 [0.71, 1.46] 0.71 [0.43, 1.15] 0.62 [0.378, 0.96] 0.71 [0.43, 1.00] 49.71 < 0.001
IVPDP2 (mmHg), median [IQR] 1.02 [0.71, 1.41] 1.12 [0.74, 1.56] 1.05 [0.72, 1.39] 0.90 [0.70, 1.28] 1.020 [0.657, 1.33] 8.35 0.039
IVPDP3 (mmHg), median [IQR] −0.36 [−0.60, −0.25] −0.51 [−0.74, −0.33] −0.31 [−0.43, −0.24] −0.34 [−0.57, −0.16] −0.34 [−0.52, −0.24] 37.65 < 0.001
IVPDP4 (mmHg), median [IQR] −0.97 [−1.39, −0.65] −1.22 [−1.71, v0.88] −0.88 [−1.22, −0.63] −0.81 [−1.16, −0.52] −0.97 [−1.32, −0.61] 45.79 < 0.001
10.3389/fcvm.2022.1022658
IVPDP5 (mmHg), median [IQR] 0.52 [0.29, 0.89] 0.67 [0.33, 1.13] 0.61 [0.32, 0.91] 0.45 [0.27, 0.70] 0.45 [0.21, 0.78] 13.80 0.003
IVPGP1 (mmHg/mmeˆ3), median [IQR] −8.96 [−15.22, −5.53] −14.52 [−19.48, −8.47] −8.45 [−13.37, −5.40] −6.96 [−10.34, −3.93] −8.35 [−12.47, −5.18] 67.63 < 0.001
IVPGP2 (mmHg/mmeˆ3), median [IQR] −12.12 [−16.90, −8.30] −14.45 [−20.48, −9.29] −12.58 [−16.63, −8.44] −10.15 [−14.84, −7.49] −12.21 [−16.32,- −8.01] 22.76 < 0.001
< 0.001
frontiersin.org
IVPGP3 (mmHg/mmeˆ3), median [IQR] 4.45 [3.00, 7.30] 6.56 [4.38, 10.20] 3.78 [2.96, 5.18] 3.73 [1.99, 6.74] 4.06 [2.95, 6.51] 58.77
(Continued)
Sun et al.
TABLE 4 Continued

(n = 468) (n = 147) (n = 95) (n = 136) (n = 90)
IVPGP4 (mmHg/mmeˆ3), median [IQR] 11.98 [7.98, 17.47] 15.80 [11.76, 23.72] 11.01 [7.73, 15.08] 9.23 [5.65, 13.70] 11.46 [7.26, 16.12] 75.75 < 0.001
IVPGP5 (mmHg/mmeˆ3), median [IQR] −6.46 [−10.95, −3.31] −9.11 [−14.72, −4.37] −7.71 [−10.88, −4.05] −5.04 [−7.80, −2.87] −5.76 [−9.65, −2.50] 24.53 < 0.001
VorArea-s (mmˆ2), median [IQR] 454.22 [354.72, 574.46] 423.08 [337.60, 551.43] 460.09 [384.78, 576.38] 461.12 [360.21, 579.91] 465.20 [350.05, 587.72] 3.07 0.381
Circulation-s (mˆ2/s eˆ3), median [IQR] 16.90 [11.77, 22.03] 18.57 [14.27, 23.33] 17.13 [12.63, 23.23] 15.03 [8.50, 19.30] 16.10 [10.80, 21.07] 19.85 < 0.001
VorArea-ed (mmˆ2), median [IQR] 359.00 [257.14, 476.48] 337.17 [225.00, 471.63] 345.53 [263.20, 448.11] 386.44 [284.88, 509.13] 359.00 [289.74, 470.27] 7.96 0.047
Circulation-ed (mˆ2/s eˆ3), median [IQR] 13.13 [7.57, 21.80] 14.63 [9.03, 22.30] 12.57 [7.33, 20.87] 12.87 [7.07, 21.17] 13.10 [7.13, 21.73] 2.85 0.416
VorArea-ac (mmˆ2), median [IQR] 428.02 [323.87, 531.24] 404.47 [312.14, 504.52] 447.10 [338.74, 556.72] 457.42 [346.38, 561.19] 417.28 [324.89, 517.32] 7.85 0.049
Circulation-ac (mˆ2/s eˆ3), median [IQR] 18.03 [12.10, 25.53] 19.70 [13.30, 27.10] 19.00 [13.40, 27.13] 15.47 [10.83, 21.93] 17.90 [11.73, 24.73] 10.65 0.014
12
vorticity-ac-CCW (1/s), median [IQR] −105.79 [−142.94, −113.56 [−152.41, −103.55 [−148.56, −94.59 [−131.58, −106.5 [−135.9, −86.7] 10.76 0.013
−80.24] −89.29] −76.79] −73.93]
vorticity-ac-CW (1/s), median [IQR] 114.97 [85.83, 149.68] 128.97 [95.50, 165.98] 117.48 [91.04, 151.08] 107.01 [73.69, 133.29] 105.04 [88.66, 137.78] 22.55 < 0.001
vorticity-ed-CCW (1/s), median [IQR] −99.07 [−135.19, −96.51 [−127.33, −92.34 [−124.11, −107.40 [−151.76, −99.77 [−131.18, 7.01 0.072
−76.44] −78.97] −69.19] −81.92] −74.86]
vorticity-ed-CW (1/s), median [IQR] 101.54 [72.03, 140.46] 98.34 [71.22, 140.46] 98.33 [71.15, 143.29] 115.89 [74.59, 140.68] 95.48 [72.29, 140.17] 2.94 0.401
vorticity-s-CCW (1/s), median [IQR] −100.69 [−130.23, −112.44 [−145.15, −97.57 [−124.52, −90.01 [−122.77, −104.3 [−134.5, −85.9] 17.34 < 0.001
−75.93] −83.37] −77.99] −67.34]
vorticity-s-CW (1/s), median [IQR] 110.59 [82.79, 149.78] 132.47 [99.22, 173.11] 105.17 [80.06, 147.14] 97.38 [69.16, 131.18] 105.57 [75.87, 144.28] 37.41 < 0.001
p < 0.05 indicates statistical significance. STE, speckle tracking echocardiography; GLS, global longitudinal strain; GLSR, global systolic strain rate; PALS, peak atrial longitudinal strain; VFM, vector flow mapping; MeanEL, mean energy loss; IVPD,
intraventricular pressure difference; IVPG, intraventricular pressure gradient; s, systolic; ed, early diastolic; ac, atrium contraction; P1-P5, isovolumic contraction period, ejection period, isovolumetric relaxation period, diastolic filling period, atrium
contraction period; CW, clockwise; CCW, counter-clockwise.
10.3389/fcvm.2022.1022658
frontiersin.org
Sun et al. 10.3389/fcvm.2022.1022658
FIGURE 3
Feature selection using the least absolute shrinkage and selection operator (LASSO) binary logistic regression model. (A) LASSO coefficient
profiles of the 320 features. Vertical line was drawn at the value selected using 10-fold cross-validation, where optimal resulted in 26 non-zero
coefficients. (B) Tuning parameter (λ) selection in the LASSO model used 10-fold cross-validation via minimum criteria. A coefficient profile plot
was produced against the log (λ) sequence. The area under the receiver operating characteristic (AUC) curve was plotted versus log(λ). This is
consistent with the glmnet package recommendation for choosing λ, as either λ min (minimum mean square error) or this value plus one
standard error.
parameters. General parameters include basic characterization STE and VFM parameters in the XGBoost classifiers model
parameters, clinical parameters, blood laboratory test results, showed an improved ability to classify patients with MACE
and conventional ultrasound parameters. Four machine correctly compared with the other learning algorithms, which
learning models were used to complete the data sample is more beneficial to predicting the occurrence of MACE. The
classification task, including: XGBoost Classifier, Random performances (AUC and accuracy) of the model in classifying
Forest Classifier, Multi-LayerPerceptron (MLP) Classifier, patients correctly in each group using learning algorithms
support vector machine (SVM) Classifier. By repeated sampling are shown (Figures 7D,E). Calibration plot of four learning
10 times and each resampling training validation set of algorithms (Figure 7F). When the solid line was closer to the
20.0% of the total sample, training set of 80.0%, the model dotted line (perfectly calibrated), the prediction accuracy was
classification was successively performed and drawed the ROC better. Among the comparison of Brier scores for the above four
curve. General Parameters group (AUC: 0.79, accuracy: 0.75, learning algorithms, XGBoost is the lowest, and his prediction
AUC: 0.76, accuracy: 0.74, AUC: 0.68, accuracy: 0.69, AUC: 0.70, calibration is the best (Brier scores = 0.07).
accuracy: 0.72, respectively), General Parameters and STE group
(AUC: 0.81, accuracy: 0.75; AUC: 0.80, accuracy: 0.76, AUC:
Discussion
0.67, accuracy: 0.73, AUC: 0.70, accuracy: 0.74, respectively),
General Parameter, STE and VFM group (AUC: 0.84, accuracy:
The primary objective of this study was to develop a
0.77, AUC: 0.82, accuracy: 0.79, AUC: 0.71, accuracy: 0.72, AUC:
scoring system to predict MACE risk in patients with chronic
0.74, accuracy: 0.76, respectively) were classified in the test set heart failure. This study used clinical trial data from patients
performance (Figures 7A–C). with CHF, and age was not included in the risk assessment
Comparing AUC and accuracy, the XGBoost classifier model due to the short follow-up time. In previous models,
performed the best among all four learning algorithms. In the most common parameters used to predict major adverse
each group, we found that AUC and accuracy of four cardiovascular events were general clinical data, blood test
learning algorithms were higher when the parameters included parameters, and conventional ultrasound parameters (11–16).
general parameters, STE-related parameters and VFM-related There are also articles that use LA and LV strain parameters
parameters, and were more pronounced when using the to create models. LA function assessed by speckle tracking
XGBoost classifier. The inclusion of STE and VFM parameters echocardiography is an independent prognostic marker in heart
in the XGBoost classifier model improved its ability to failure patients with reduced ejection fraction (31). These results
correctly classify MACE patients compared to other learning suggest that STE variable information corresponding to two-
algorithms (AUC: 0.84, accuracy: 0.77). The inclusion of the dimensional and doppler analysis can provide independent

Sun et al. 10.3389/fcvm.2022.1022658
FIGURE 4
Select optimal prognostic variables. (A) Correlation heatmap showing associations between clinical features and ultrasound parameters. (B–D)
XGBoost, Random forest, KNN calculate the importance of each influencing factor to the classification model and rank it, the top 20 parameters
were selected. (E) Venn diagram, 13 variables were recognized.
assessments of diastolic function and LV filling pressures, which explored using other imaging techniques. Decreased LA ejection
are increasingly moving toward precision medicine (32). fraction measured by cardiac magnetic resonance imaging was
Left atrial (LA) function is closely related to LV function associated with a higher incidence of AF and poorer prognosis
and plays a key role in maintaining optimal cardiac functional in patients with HF (37). Previous studies have shown that
performance. The left atrium, through its reservoir, conduit, LV diastolic dysfunction is related to LA volume (38, 39).
and booster pump stages, regulates filling of the left ventricle, Moreover, LV diastolic function is closely related to atrial
while the LV influences LA function throughout the cardiac function; Thus, the transmitral flow parameters of LV must be
cycle. Long-term exposure to high LV filling pressures leads interpreted in terms of LV relaxation, LA function, and loading
to an increase in LA volume, presumably reduced LA parameters, each of which affects LV filling pressure. Due to
function is only a marker of LV deterioration (33, 34). the high dimensionality and complexity of variables affecting
In patients with HFrEF, LA function index measured at LV diastolic function and filling pressures, echocardiographic
first admission was associated with cardiovascular outcomes assessment requires a multiparametric approach (19, 40).
during the first 6 months of follow-up (35), and LA systolic Heart failure patients are divided into four categories based
function measured by LA strain rate was shown to provide on LVEF: HFpEF, HFmrEF, HFrEF HFimpEF (17). Different
prognostic stratification for outpatients with new-onset HF subtypes have different clinical features and disease courses,
(36). The prognostic value of LA function has also been and different conditions can lead to misjudgment of heart

Sun et al. 10.3389/fcvm.2022.1022658
TABLE 5 Prognostic model for major cardiovascular events (MACE) based on logistic regression model.
Predictor Estimate SE Z p Odds ratio Lower (95% CI) Upper (95% CI)
IVPDC2P4 (mmHg) 0.345 0.163 2.116 0.034 1.411 1.039 1.978

VorAreaC4S (mmˆ2) 0.001 0.001 1.664 0.046 1.001 1 1.002
GLSacC2LV (%) 0.155 0.079 1.967 0.049 1.168 1.003 1.367
PALSconduitLA (%) −0.06 0.039 −1.539 0.024 0.942 0.87 1.014
GLSLV (%) −0.041 0.056 −0.726 0.048 0.96 0.86 1.072
LVEDd (mm) 0.05 0.021 2.379 0.017 1.051 1.009 1.096
LVEF (%) 0.033 0.022 1.495 0.035 1.033 0.99 1.079
ApoB (g/L) 1.145 0.458 2.501 0.012 3.143 1.288 7.821
Na (mmol/L) −0.056 0.034 −1.674 0.044 0.945 0.885 1.009
NTproBNP (mmol/L) 0.001 0.001 4.745 0.001 1 1 1.001
DBP (mmHg) −0.014 0.007 −1.972 0.049 0.986 0.971 1
NYHA II 2.263 0.797 2.841 0.005 9.611 2.439 64.577
NYHA III 2.918 0.857 3.406 0.001 18.503 4.052 134.706
NYHA IV 3.367 0.896 3.758 0.001 28.979 5.799 222.979
p < 0.05 indicates statistical significance. IVPDC2P4, intraventricular pressure differences from 2-chamber images during LV early filling period; VorAreaC4S, LV vortex area from 4-
chamber images during systolic period; GLSacC2LV, left ventricular global longitudinal strain from 2-chamber images during atrial systolic period; PALSconduitLA, peakLA longitudinal
peak strain during conduit period; GLSLV, left ventricular global longitudinal strain; LVEDd, left ventricular end diastolic diameter; LVEF, left ventricular ejection fraction; ApoB,
apolipoproteinB; NT-pro BNP, N-terminal pro-brain natriuretic peptide; DBP, diastolic blood pressure; NYHA, New York Heart Association.
failure and bring difficulties to the diagnosis of heart failure. stress (24, 44, 45). Automated phenotype calculation is an
It is well-known that HFpEF differs in clinical features efficient strategy to fuse multidimensional parameters of LV
and disease course (including treatment strategies), but all- structure and function to collect STE- and CFM-related
cause readmission rates remained the highest in HFpEF vs. parameters (46).
HFrEF and HFmrEF (41). This study hopes to establish a In this study, the left ventricular high-throughput
predictive model suitable for different heart failure populations, parameters were constructed by collecting mechanics and
and only LVEF analysis variables are included, not LVEF hydrodynamic parameters, including STE parameters of LV
classification. We will continue to expand the sample size and LA, VFM parameters of LV flow. For the first time,
and construct respective prediction models for different types we integrate multiple echocardiographic parameters to
of heart failure patients, so as to improve the prediction meaningfully reflect changes in left ventricular structure and
effect of the model. function. Through screening, key indicators are incorporated
Currently, left chamber fluid mechanics is lacking to into the construction of the model. Based on selecting
participate in model construction. There is a growing interest optimal prognostic variables, a logistic regression model
in the imaging and visualization of intracardiac blood flow and nomogram is constructed, which can well-predict the
(42). VFM is a two-dimensional cross-sectional image acquired possibility of MACE within 6 months. Here, we evaluated
based on b-type color Doppler echocardiography, which can the relationship between these factors and HF to create a
enables visualize the cross-sectional image with blood flow practical and accurate prognostic dynamic nomogram model
as a velocity vector (43). As mentioned earlier, calculate the to identify high-risk groups of heart failure and ultimately
flow vector using the continuity equation and decompose develop targeted treatment options. From Figure 5C, we
the vector into vertical and parallel velocity components, can find the performance effect of constructing the new
determined from wall motion spot tracking and color doppler model is good [C-statistics: 0.876 (95% CI, 0.844– 0.907)].
images (25). The advantages of VFM are that it is relatively From Figure 6, survival curves were generated using the
inexpensive, less time-consuming, easy to use at the bedside, Kaplan–Meier method, and log-rank tests were used to
and does not require the use of contrast agents. This compare survival curves, which based on nomogram scores
model adds the VFM parameters compared with the previous and predictive risk. For survival analysis, significant differences
models, which makes the significance of the model more among the groups were seen according to Kaplan–Meier
explanatory and the evaluation more comprehensive. The survival curves. We can find patients with high nomogram
potential application of VFM to quantify aortic regurgitation has scores and predictive risk had the shortest survival time.
been reported. Compared with other conventional measures, Although the number of high-risk group was significantly
energy loss can more clearly quantify patients, subjective fewer than low risk group, the number of patients with
symptoms and help to assess disease severity based on cardiac MACE was significantly more than the low risk group,

Sun et al. 10.3389/fcvm.2022.1022658
FIGURE 5
Nomogram interpretation and model validation. (A) Nomogram for predicting major cardiovascular events (MACE) risk, and the point was the
selected scoring standard or scale. (B) Odds ratios determined in a logistic regression model as shown in the forest plot. (C) Receiver operating
characteristic curve for the nomogram. (D) The Nomogram calibration plots of the model. (E) Decision curve analysis for predictive models.
better demonstrating the clinical significance of the model. Furthermore, analyzing and extracting data from STE and
In each group, we found that AUC and accuracy of four VFM techniques requires a certain amount of time and effort,
learning algorithms were higher when the parameters included which is often difficult to perform in routine clinical practice.
general parameters, STE-related parameters and VFM-related However, with continuous attempts to explore and screen
parameters, and were more pronounced when using the more valuable parameters, the most valuable indicators can
XGBoost classifier. The inclusion of general parameters and be easily extracted from data, which is beneficial for disease
STE-related parameters and VFM-related parameters in the assessment and clinical decision-making in patients with heart
XGBoost classifiers model showed an improved ability to failure.
classify patients with MACE correctly compared with the
other learning algorithms (AUC: 0.84, accuracy: 0.77), which
is more beneficial to predicting the occurrence of MACE. Study limitations
Calibration plot of four learning algorithms, we can find among
the comparison of Brier scores, XGBoost is the lowest, and There were several limitations to this study that require
his prediction calibration is the best (Brier scores = 0.07). comment. First, this was a single-center study with a limited
By machine learning, we found that the AUC of the models sample size. A larger study sample from various centers
generated by general parameters, STE-related parameters may be helpful to further confirm these observations. The
and VFM-related parameters was higher than that of the findings of the present study, although showing statistical
other models, but the increase in AUC was not significant. significance, were based on a relatively small number of

Sun et al. 10.3389/fcvm.2022.1022658
FIGURE 6
Survival curves based on nomogram scores and predictive risk. (A,B) Patients were divided into four groups according to the quartile total
nomogram scores (nomgroup = 0, 1, 2, 3), the KM survival curves and cumulative survival curves were used to compare survival times among
different groups was made by log-rank test. (C,D) Patients were divided into high-‘and low-risk groups (low-risk group = 1, high–risk group = 2)
according to the predicted probabilities of the prediction model. The median survival time (LT50) for different groups. The horizontal axis of the
KM survival curve represents time, and the vertical axis represents probability.
patients and are indeed not suitable for data from countries this time, only to predict the possibility of MACE within
that respond to an aging population. The results should 6 months after discharge. In future studies, the continuous
therefore be regarded as preliminary, and further enlarge follow-up will be summarized to establish an effective cox
studies adequately powered for clinical outcomes are warranted prediction model. Third, VFM technology relies on strict
to confirm our results. Second, a relatively short follow-up contour condition provided by the endocardial-blood interface
duration was used to detect cardiac events in this study, the and is limited by the frame rate applied and the Nyquist
effective time follow-up within 6 months is relatively short, limit. Aliasing may result in errors in the VFM calculated
the specific time of MACE cannot be clearly defined by blood flow data, mild aliasing was manually corrected on
patients, therefore logist regression is used to draw nomgraph the selected frames. However, severe aliasing could not be

Sun et al. 10.3389/fcvm.2022.1022658
FIGURE 7
Performance of machine learning algorithms among the groups. (A–C) The ROC curve of the model performance in classifying patients
correctly in each group using learning algorithms. (D,E) The performances [area under curve (AUC) and accuracy] of the model in classifying
patients correctly in each group using learning algorithms. (F) The calibration plots of learning algorithms in group [General Parameters, speckle
tracking echocardiography (STE) and vector flow mapping (VFM) group].
corrected can underestimate the true velocity, we should its prompt management, which is a convenient, practical
minimize this phenomenon by optimizing the Nyquist limit and effective clinical decision making tool for providing
and ensuring appropriate patient selection Patients with accurate prognosis.
valcular heart diseases were not included in the analysis
due to severe aliasing produced by the fast valve flow.
Fourth, we did not include additional measurements for right
ventricular function, which would need to be addressed in Data availability statement
future investigations.
The original contributions presented in the study are
included in the article/supplementary material, further inquiries
can be directed to the corresponding authors.
Conclusion
In this study, we developed a prediction model and
nomogram to estimate the risk of MACE within 6 months Ethics statement
of discharge among patients with heart failure. The findings
may provide a reference for clinical physicians for detection Written informed consent was obtained from the
of the risk of MACE in terms of clinical characteristics, individual(s) for the publication of any potentially identifiable
cardiac structure and function, hemodynamics, and enable images or data included in this article.

Sun et al. 10.3389/fcvm.2022.1022658
Author contributions Conflict of interest

QL-S proposed the hypothesis, designed the experiments, The authors declare that the research was conducted in the
and drafted the manuscript. QL-S, SQ-J, XD-W, and YL absence of any commercial or financial relationships that could
made contributions to data collection. QL-S and JC-Z made be construed as a potential conflict of interest.
contributions to the analysis and interpretation of the data.
JW-T and HR-L conceived the study, participated in its design,
and helped edit the manuscript. All authors contributed to the
article and approved the submitted version. Publisher’s note
All claims expressed in this article are solely those of the
Funding authors and do not necessarily represent those of their affiliated
organizations, or those of the publisher, the editors and the
This study was funded by China Postdoctoral Science reviewers. Any product that may be evaluated in this article, or
Foundation (2021MD703828) and Heilongjiang Postdoctoral claim that may be made by its manufacturer, is not guaranteed
Fund (LBH-Z20172). or endorsed by the publisher.
References
1. Groenewegen A, Rutten FH, Mosterd A, Hoes AW. Epidemiology of heart 13. Wang L, Zhao YT. Development and validation of a prediction model for
failure. Eur J Heart Fail. (2020) 22:1342–56. doi: 10.1002/ejhf.1858 irreversible worsened cardiac function in patients with acute decompensated heart
failure. Front Cardiovasc Med. (2021) 8:785587. doi: 10.3389/fcvm.2021.785587
2. Ambrosy AP, Parikh RV, Sung SH, Tan TC, Narayanan A, Masson R, et al.
Analysis of worsening heart failure events in an integrated health care system. J Am 14. Segar MW, Jaeger BC, Patel KV, Nambi V, Ndumele CE, Correa A, et al.
Coll Cardiol. (2022) 80:111–22. doi: 10.1016/j.jacc.2022.04.045 Development and validation of machine learning-based race-specific models to
predict 10-year risk of heart failure: a multicohort analysis. Circulation. (2021)
3. Lesyuk W, Kriza C, Kolominsky-Rabas P. Cost-of-illness studies in heart
143:2370–83. doi: 10.1161/CIRCULATIONAHA.120.053134
failure: a systematic review 2004-2016. BMC Cardiovasc Disord. (2018) 18:74. doi:
10.1186/s12872-018-0815-3 15. Zhuang B, Shen T, Li D, Jiang Y, Li G, Luo Q, et al. A model for the prediction
of mortality and hospitalization in chinese heart failure patients. Front Cardiovasc
4. Urbich M, Globe G, Pantiri K, Heisen M, Bennison C, Wirtz HS, et al.
Med. (2021) 8:761605. doi: 10.3389/fcvm.2021.761605
A systematic review of medical costs associated with Heart Failure in the USA
(2014-2020). Pharmacoeconomics. (2020) 38:1219–36. doi: 10.1007/s40273-020-0 16. Yin T, Shi S, Zhu X, Cheang I, Lu X, Gao R, et al. A survival prediction
0952-0 for acute heart failure patients via web-based dynamic nomogram with internal
validation: a prospective cohort study. J Inflamm Res. (2022) 15:1953–67. doi:
5. Thenappan T, Prins KW, Cogswell R, Shah SJ. Pulmonary hypertension
10.2147/JIR.S348139
secondary to heart failure with preserved ejection fraction. Can J Cardiol. (2015)
31:430–9. doi: 10.1016/j.cjca.2014.12.028 17. Bozkurt B, Coats Andrew JS, Tsutsui H, Abdelhamid Ca M, Adamopoulos S,
Albert N, et al. Universal definition and classification of heart failure: a report of
6. Francis GS, Cogswell R, Thenappan T. The heterogeneity of heart failure: will
the Heart Failure Society of America, Heart Failure Association of the European
enhanced phenotyping be necessary for future clinical trial success? J Am Coll
Society of Cardiology, Japanese Heart Failure Society and Writing Committee
Cardiol. (2014) 64:1775–6. doi: 10.1016/j.jacc.2014.07.978
of the Universal Definition of Heart Failure: Endorsed by the Canadian Heart
7. Hsu B, Korda RJ, Lindley RI, Douglas KA, Naganathan V, Jorm LR. Use Failure Society, Heart Failure Association of India, Cardiac Society of Australia
of health and aged care services in Australia following hospital admission for and New Zealand, and Chinese Heart Failure Association. Eur J Heart Fail. (2021)
myocardial infarction, stroke or heart failure. BMC Geriatr. (2021) 21:538. doi: 23:352–80. doi: 10.1002/ejhf.2115
10.1186/s12877-021-02519-w
18. Lang RM, Badano LP, Mor-Avi V, Afilalo J, Armstrong A, Ernande L,
8. Valley TS, Sjoding MW, Ryan AM, Iwashyna TJ, Cooke CR. Intensive care unit et al. Recommendations for cardiac chamber quantification by echocardiography
admission and survival among older patients with chronic obstructive pulmonary in adults: an update from the American Society of Echocardiography and the
disease, Heart Failure, or Myocardial Infarction. Ann Am Thorac Soc. (2017) European Association of Cardiovascular Imaging. J Am Soc Echocardiogr. (2015)
14:943–51. doi: 10.1513/AnnalsATS.201611-847OC 28:1–39.e14. doi: 10.1016/j.echo.2014.10.003
9. Hsu B, Korda R, Naganathan V, Lewis P, Ooi SY, Brieger D, et al. Burden of 19. Nagueh SF, Smiseth OA, Appleton CP, Byrd BF, Dokainish H, Edvardsen T,
cardiovascular diseases in older adults using aged care services. Age Ageing. (2021) et al. Recommendations for the evaluation of left ventricular diastolic function
50:1845–9. doi: 10.1093/ageing/afab083 by echocardiography: an update from the american society of echocardiography
and the european association of cardiovascular imaging. Eur Heart J Cardiovasc
10. Arnett DK, Blumenthal RS, Albert MA, Buroker AB, Goldberger ZD,
Imaging. (2016) 17:1321–60. doi: 10.1093/ehjci/jew082
Hahn EJ, et al. 2019 ACC/AHA guideline on the primary prevention of
cardiovascular disease: a report of the American College of Cardiology/American 20. Unger ED, Dubin RF, Deo R, Daruwalla V, Friedman JL, Medina C, et al.
Heart Association task force on clinical practice guidelines. J Am Coll Cardiol. Association of chronic kidney disease with abnormal cardiac mechanics and
(2019) 74:e177–232. adverse outcomes in patients with heart failure and preserved ejection fraction. Eur
J Heart Fail. (2016) 18:103–12. doi: 10.1002/ejhf.445
11. Voors AA, Ouwerkerk W, Zannad F, van Veldhuisen DJ, Samani NJ,
Ponikowski P, et al. Development and validation of multivariable models to predict 21. Mirea O, Duchenne J, Voigt JU. Comparison between nondedicated and
mortality and hospitalization in patients with heart failure. Eur J Heart Fail. (2017) novel dedicated tracking tool for right ventricular and left atrial strain. J Am Soc
19:627–34. doi: 10.1002/ejhf.785 Echocardiogr. (2022) 35:419–25. doi: 10.1016/j.echo.2021.11.011
12. Lim NK, Lee SE, Lee HY, Cho HJ, Choe WS, Kim H, et al. Risk prediction 22. Asami R, Tanaka T, Shimizu M, Seki Y, Nishiyama T, Sakashita H, et al.
for 30-day heart failure-specific readmission or death after discharge: data from the Ultrasonic vascular vector flow mapping for 2-D flow estimation. Ultrasound Med
Korean Acute Heart Failure (KorAHF) registry. J Cardiol. (2019) 73:108–13. Biol. (2019) 45:1663–74. doi: 10.1016/j.ultrasmedbio.2019.02.014

Sun et al. 10.3389/fcvm.2022.1022658
23. Garcia D, Del Alamo JC, Tanne D, Yotti R, Cortina C, Bertrand E, 35. Chrysohoou C, Kotroyiannis I, Antoniou CC, Brili S, Vaina S, Latsios G, et al.
et al. Two-dimensional intraventricular flow mapping by digital processing Left atrial function predicts heart failure events in patients with newly diagnosed
conventional color-Doppler echocardiography images. IEEE Trans Med Imaging. left ventricular systolic heart failure during short-term follow-up. Angiology. (2014)
(2010) 29:1701–13. doi: 10.1109/TMI.2010.2049656 65:817–23. doi: 10.1177/0003319713506109
24. Stugaard M, Koriyama H, Katsuki K, Masuda K, Asanuma T, Takeda Y, 36. Sanchis L, Andrea R, Falces C, Lopez-Sobrino T, Montserrat S, Perez-Villa F,
et al. Energy loss in the left ventricle obtained by vector flow mapping as a new et al. Prognostic value of left atrial strain in outpatients with de novo heart failure.
quantitative measure of severity of aortic regurgitation: a combined experimental J Am Soc Echocardiogr. (2016) 29:1035–42.e1. doi: 10.1016/j.echo.2016.07.012
and clinical study. Eur Heart J Cardiovasc Imaging. (2015) 16:723–30. doi: 10.1093/
37. Pellicori P, Zhang J, Lukaschuk E, Joseph AC, Bourantas CV, Loh H, et al.
ehjci/jev035
Left atrial function measured by cardiac magnetic resonance imaging in patients
25. Asami R, Tanaka T, Kawabata KI, Hashiba K, Okada T, Nishiyama T. with heart failure: clinical associations and prognostic value. Eur Heart J. (2015)
Accuracy and limitations of vector flow mapping: left ventricular phantom 36:733–42. doi: 10.1093/eurheartj/ehu405
validation using stereo particle image velocimetory. J Echocardiogr. (2017) 15:57–
38. Tsang TS, Barnes ME, Gersh BJ, Bailey KR, Seward JB. Left atrial volume as a
66. doi: 10.1007/s12574-016-0321-5
morphophysiologic expression of left ventricular diastolic dysfunction and relation
26. Du Y, Goddi A, Bortolotto C, Shen Y, Dell’Era A, Calliada F, et al. Wall shear to cardiovascular risk burden. Am J Cardiol. (2002) 90:1284–9. doi: 10.1016/s0002-
stress measurements based on ultrasound vector flow imaging: theoretical studies 9149(02)02864-3
and clinical examples. J Ultrasound Med. (2020) 39:1649–64. doi: 10.1002/jum.
39. Hung MJ, Cherng WJ. Analysis of left atrial volume change rate for evaluation
15253
of left ventricular diastolic function. Echocardiography. (2004) 21:593–601. doi:
27. Rodríguez Muñoz D, Moya Mur JL, Fernández-Golfín C, Becker Filho 10.1111/j.0742-2822.2004.03154.x
DC, González Gómez A, Fernández Santos S, et al. Left ventricular vortices
40. Mottram PM, Marwick TH. Assessment of diastolic function: what the
as observed by vector flow mapping: main determinants and their relation to
general cardiologist needs to know. Heart. (2005) 91:681–95. doi: 10.1136/hrt.2003.
left ventricular filling. Echocardiography. (2015) 32:96–105. doi: 10.1111/echo.1
029413
2584
41. Cui X, Thunström E, Dahlström U, Zhou J, Ge J, Fu M. Trends in
28. Ji L, Hu W, Yong Y, Wu H, Zhou L, Xu D. Left ventricular energy loss and
cause-specific readmissions in heart failure with preserved vs. reduced and mid-
wall shear stress assessed by vector flow mapping in patients with hypertrophic
range ejection fraction. ESC Heart Fail. (2020) 7:2894–903. doi: 10.1002/ehf2.
cardiomyopathy. Int J Cardiovasc Imaging. (2018) 34:1383–91. doi: 10.1007/
12899
s10554-018-1348-7
42. Rodriguez Muñoz D, Markl M, Moya Mur JL, Barker A, Fernández-Golfín
29. Chen M, Jin JM, Zhang Y, Gao Y, Liu SL. Assessment of left ventricular
C, Lancellotti P, et al. Intracardiac flow visualization: current status and future
diastolic dysfunction based on the intraventricular velocity difference by vector flow
directions. Eur Heart J Cardiovasc Imaging. (2013) 14:1029–38. doi: 10.1093/ehjci/
mapping. J Ultrasound Med. (2013) 32:2063–71. doi: 10.7863/ultra.32.12.2063
jet086
30. Harrell FE, Lee KL, Mark DB. Multivariable prognostic models: issues
43. Uejima T, Koike A, Sawada H, Aizawa T, Ohtsuki S, Tanaka M, et al. A
in developing models, evaluating assumptions and adequacy, and measuring
new echocardiographic method for identifying vortex flow in the left ventricle:
and reducing errors. Stat Med. (1996) 15:361–87. doi: 10.1002/(SICI)1097-
numerical validation. Ultrasound Med Biol. (2010) 36:772–88. doi: 10.1016/j.
0258(19960229)15:4<361::AID-SIM168>3.0.CO;2-4
ultrasmedbio.2010.02.017
31. Malagoli A, Rossi L, Bursi F, Zanni A, Sticozzi C, Piepoli MF, et al. Left atrial
44. Li C, Zhang J, Li X, Zhou C, Li H, Tang H, et al. Quantification of
function predicts cardiovascular events in patients with chronic heart failure with
chronic aortic regurgitation by vector flow mapping: a novel echocardiographic
reduced ejection fraction. J Am Soc Echocardiogr. (2019) 32:248–56. doi: 10.1016/j.
method. Eur J Echocardiogr. (2010) 11:119–24. doi: 10.1093/ejechocard/
echo.2018.08.012
jep175
32. Collins FS, Varmus H. A new initiative on precision medicine. N Engl J Med.
45. Kainuma A, Itatani K, Akiyama K, Naito Y, Ishii M, Shimizu M, et al.
(2015) 372:793–5. doi: 10.1056/NEJMp1500523
Preoperative left ventricular energy loss in the operating theater reflects subjective
33. Matsuda Y, Toma Y, Ogawa H, Matsuzaki M, Katayama K, Fujii T, symptoms in chronic aortic regurgitation. Front Surg. (2022) 9:739743. doi: 10.
et al. Importance of left atrial function in patients with myocardial infarction. 3389/fsurg.2022.739743
Circulation. (1983) 67:566–71. doi: 10.1161/01.cir.67.3.566
46. Cho JS, Shrestha S, Kagiyama N, Hu L, Ghaffar YA, Casaclang-Verzosa G,
34. Rosca M, Lancellotti P, Popescu BA, Piérard LA. Left atrial function: et al. A network-based “Phenomics” approach for discovering patient subtypes
pathophysiology, echocardiographic assessment, and clinical applications. Heart. from high-throughput cardiac imaging data. JACC Cardiovasc Imaging. (2020)
(2011) 97:1982–9. doi: 10.1136/heartjnl-2011-300069 13:1655–70. doi: 10.1016/j.jcmg.2020.02.008

FCVM 09 1022658

Uploaded by

Copyright:

Available Formats

FCVM 09 1022658

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

FCVM 09 1022658

Uploaded by

Copyright:

Available Formats

TYPE Original Research

PUBLISHED 28 October 2022

A prediction model for major

Frontiers in Cardiovascular Medicine 01 frontiersin.org

clinical physicians for detection of the risk of MACE in terms of clinical

Introduction All efforts must be underway to examine clinical, laboratory,

Frontiers in Cardiovascular Medicine 02 frontiersin.org

Frontiers in Cardiovascular Medicine 03 frontiersin.org

Frontiers in Cardiovascular Medicine 04 frontiersin.org

Statistical analysis prognostic variables were identified, including New York

Frontiers in Cardiovascular Medicine 05 frontiersin.org

Variables Overall No event at follow-up Event at follow-up Statistics P value

Frontiers in Cardiovascular Medicine 06 frontiersin.org

Variables Overall No event at follow-up Event at follow-up Statistics P value

Frontiers in Cardiovascular Medicine 07 frontiersin.org

Variables Overall HFpEF HFmrEF HFrEF HFimpEF Statistics P value

Frontiers in Cardiovascular Medicine 08 frontiersin.org

Variables Overall HFpEF HFmrEF HFrEF HFimpEF Statistics P value

Frontiers in Cardiovascular Medicine 09 frontiersin.org

Variables Overall No event at follow-up Event at follow-up Statistics P value

Frontiers in Cardiovascular Medicine 10 frontiersin.org

Variables Overall HFpEF HFmrEF HFrEF HFimpEF Statistics P value

Variables Overall HFpEF HFmrEF HFrEF HFimpEF Statistics P value

Frontiers in Cardiovascular Medicine 13 frontiersin.org

Frontiers in Cardiovascular Medicine 14 frontiersin.org

IVPDC2P4 (mmHg) 0.345 0.163 2.116 0.034 1.411 1.039 1.978

Frontiers in Cardiovascular Medicine 15 frontiersin.org

Frontiers in Cardiovascular Medicine 16 frontiersin.org

Frontiers in Cardiovascular Medicine 17 frontiersin.org

Frontiers in Cardiovascular Medicine 18 frontiersin.org

Author contributions Conflict of interest

Frontiers in Cardiovascular Medicine 19 frontiersin.org

Frontiers in Cardiovascular Medicine 20 frontiersin.org

You might also like