| Research BackgroundChronic heart failure(CHF)is one of the diseases with the highest morbidity and mortality worldwide,and its major clinical manifestation is reduced exercise capacity,which is highly correlated with the risk of cardiovascular events in CHF patients.The cardiopulmonary exercise test(CPET)can measure various gas exchange parameters of the body during exercise and evaluate the exercise capacity of patients,including ventilation/carbon dioxide production(VE/VCO2slope)and maximal oxygen uptake(VO2max).VO2max is considered to be an important indicator for predicting the incidence of cardiovascular events in CHF patients.However,CPET has certain contraindications,such as acute myocardial infarction and physical disability,and is time-consuming and requires a large number of sites.Therefore,it is important to find non-invasive resting echocardiographic indicators that most accurately reflect the reduced exercise capacity of CHF patients,with clinical application value.However,some echocardiographic parameters may not show significant abnormalities at rest in patients with CHF and may change with exercise stress.To improve diagnostic accuracy,further assessment of cardiac function is often required in conjunction with exercise testing.In addition to CPET,exercise stress echocardiography(ESE)is a non-invasive test that can help accurately determine the degree of heart failure.Compared with CPET,ESE requires only a treadmill or running machine in addition to conventional echocardiography equipment,making it more suitable for promotion and application in small and medium-sized medical institutions.However,ESE has many parameters and its measurement steps are very complicated.Thus,optimizing the parameters and simplifying the parameter measurement steps is more conducive to the popularization and application of exercise echocardiography in patients with heart failure.In recent years,machine learning classifiers and mathematical statistics have been widely used in the study of clinical prediction models of cardiovascular diseases.However,there is no clinical prediction model established by ESE features to evaluate the exercise capacity of CHF patients.The use of extreme gradient boosting(XGBT),random forests(RF),classification and regression tree(CART),logistic regression combined with lasso regression(LR)is expected to discover new ultrasound features through feature importance ranking.Through the comparison or comprehensive use of these algorithms,the accuracy and practicability of diagnostic models can be improved.Therefore,establishing a diagnostic model of exercise capacity in patients with CHF based on ESE features and various algorithms,as well as verifying and comparing them,will help to innovatively discover important ESE features and optimize clinical prediction models.Furthermore,considering the diversity of clinical application scenarios,other models under different clinical application scenarios will be established and compared with the ESE model to further evaluate the diagnostic performance of the ESE model.In conclusion,this study aimed to predict the exercise capacity in patients with CHF using ESE features,explore the relationship between ESE features and the exercise capacity of CHF patients and obtain a clinical prediction model for assessing the exercise capacity in patients with CHF based on conventional ESE features or novel features.Additionally,the predictive ability of the model was verified and compared in different clinical application scenarios.Chapter 1 Predictive Value of Resting-Stage Echocardiography for Exercise Capacity in Patients with Chronic Heart FailureIntroduction:An important manifestation of chronic heart failure is reduced exercise capacity,which has been found to be correlated with the risk of cardiovascular events in patients with CHF.It has been previously reported that resting left atrial(LA)strain can predict VO2max in patients with CHF to assess exercise capacity.However,these existing studies have generally included a subset of patients not receiving beta-blockers.Since CHF patients receiving and not receiving beta-blockers have different VO2max reference values in cardiac rehabilitation guidelines,it is necessary to limit the enrolled population in order to exclude the influence of heterogeneity in the study.Objective:To evaluate the relationship between LA strain and exercise capacity only in CHF patients receiving β-blockers.Methods:The subjects of this study were CHF patients using beta blockers.All patients underwent complete resting echocardiography to obtain ultrasound parameters including left atrial strain.VO2max and other cardiopulmonary exercise parameters reflecting exercise capacity were obtained by cardiopulmonary exercise testing.Results:A total of 73 patients were enrolled with a mean age of 61±10 years,78%male,and a mean LVEF of 44 ± 10%.After adjusting for confounders such as age,sex,and body mass index(BMI),LA reservoir longitudinal strain(LALSr),maximal LA volume index(LAVImax),minimal LA volume index(LAVImin)(P<0.0001),and LA contractile longitudinal strain(LALSct)(P<0.01)were significantly correlated with VO2max.After adjusting for confounders such as left ventricular ejection fraction(LVEF),E/e’,and tricuspid annular plane systolic excursion(TAPSE),LALSr,LALSct(P<0.01),LAVImax and LAVImin(P<0.0001)were correlated with VO2max.LALSr with a cutoff value of 24.9%had 74%sensitivity and 63%specificity for identifying VO2max<16 mL/kg/min.In the subgroups aged≥or<60 years and in the subgroups of left ventricular ejection fraction(LVEF)≥50%and LVEF<50%,there was no difference in the ability of LALSr to discriminate between reduced VO2max.Conclusion:In CHF patients treated with β-blockers,resting LA strain was linearly and positively correlated with exercise capacity.LALSr and LALSct were independent predictors for detecting reduced exercise capacity.Chapter 2 Establishment of a Diagnostic Model for the Evaluation of Exercise Capacity in Chronic Heart Failure by Exercise Stress EchocardiographyIntroduction:Patients with heart failure often require further evaluation of heart function with an exercise stress test.Exercise stress echocardiography(ESE)is a noninvasive test that helps to accurately determine the degree of heart failure.Compared with CPET,ESE is relatively simple and more suitable for promotion and use in small and medium-sized medical facilities.However,ESE parameters to evaluate exercise capacity are usually numerous and complicated.Therefore,optimization of ESE parameters and a streamlining of its measuring processes are more conducive for its clinical application in heart failure patients..Objective:To develop a simple and efficient clinical diagnostic model for exercise capacity evaluation of CHF patients based on ESE by analyzing the relationship between each index of ESE and exercise capacity of CHF patients in resting,warm-up,peak and recovery periods.Methods:Eighty CHF patients were enrolled in this study.According to the Expert Consensus on Cardiac Rehabilitation and Exercise Physiology,the label for predicting reduced exercise capacity was defined as VE/VCO2slope≥30.0.Features analyzed included LV volume,LVEF,LV systolic reserve function,and other features reflecting LV systolic function in the four phases of ESE,as well as tissue or spectral Doppler functions such as E,e’,s’peak,and E/e’.Global longitudinal strain parameters for the left ventricle and left atrium were also included.A 10-fold crossvalidation was used to retrieve the training and internally validated test sets.To discriminate VC1(VE/VCO2slope<30.0),VC2(VE/VCO2slope ≥ 30.0),LR combined with Lasso regression model and three standard tree-based machine learning algorithms(XGBT,RF,CART)were used to established the model.The receiver operating characteristic curve and calibration curve were used to evaluate the performance of different models and select the model with the best performance.Results:The mean age of the subjects was 59.8±10.8 years and the mean LVEF was 44±10%.Thirty subjects were VC1 and 50 were VC2.The areas under curves of XGBT,RF,CART and LR models were 0.76,078,078 and 0.82,respectively.Among the four models,the LR model achieved the largest area under the curve(0.82,95%confidence interval:0.73 to 0.92)and the highest accuracy(0.78)and F1 score(0.75).Sex,difference between peak and resting E(Δ E),and s’ at peak exercise(s’peak)were identified as independent predictors of VC2 in the LR model.Conclusions:In this study,four clinical diagnostic models for evaluating exercise capacity of CHF patients were successfully established based on LR,XGBT,CART and RF algorithms,respectively.Furthermore,the LR model was verified to be the best in VE/VCO2slope classification for identifying CHF patients.Among them,gender,△E and s’peak are independent predictors of exercise capacity in CHF patients.This model is easy to operate and helpful for the accurate management and risk assessment of CHF patients.Chapter 3 Validation of the Stress Echocardiography Model and Comparison with Other ModelsIntroduction:At present,there are relatively few prediction models based on ultrasound parameters to evaluate the exercise capacity of patients with CHF.In Chapter 2,we successfully established for the first time a simple prediction model based on ESE for patients with CHF without laboratory tests.Due to various clinical applications,it would be of value to establish clinical prediction models using laboratory tests or other clinical data.Nevertheless,the diagnostic efficacy of such models and those compared to ESE models is yet to be determined.Objective:To establish Models 1 and 2 that are applicable for laboratory tests,and to compare the diagnostic efficacy of Models 1,2 and the ESE model constructed in Chapter 2 in the validation set.To further investigate the relationship between ESE characteristics and exercise capacity in the subgroups with reduced ejection fraction(HFrEF),moderate ejection fraction(HFmrEF),and preserved ejection fraction(HFpEF).Methods:(1)In a training set of 80 cases,Models 1 and 2 were developed,characterized by age,sex,BMI,and NT-proBNP,respectively.Model 2 was combined with ESE model characteristics(sex,△E,s’peak),and N-terminal pro brain natriuretic peptide(NT-proBNP).The diagnostic efficacy of Models 1,2,and the ESE model,was verified and compared in a validation set of 30 cases.(2)Further,subgroup analysis was conducted to compare the differences in ESE characteristics amongst subgroups with HFrEF,HFmrEF,and HFpEF,and correlation analysis was performed between ESE characteristics and VE/VCO2slope in each subgroup.Results:(1)Compared with Models 2 and ESE,the net reclassification index was<0(P=0.013).Compared with Model 1 and ESE model,the comprehensive discrimination improvement index was<0(P<0.001).Decision curve analysis showed that the ESE model had the best clinical application value.(2)In the HFpEF subgroup,most systolic functional characteristics of ESE(including LV end-systolic and end-diastolic volume indices and stroke volume at rest,warm-up and recovery,as well as LVEF at rest,warm-up,and recovery)were linearly correlated with VE/VCO2slope(P<0.05).Additionally,ΔE was a significant predictor of VE/VCO2slope in all three subgroups(HFpEF,HFmrEF,and HFrEF)and was linearly and independently correlated with VE/VCO2slope(P<0.05).Conclusion:The ESE model established in Chapter 2 has high diagnostic performance in the validation set,and the ESE model is superior to Models 1 and 2 in terms of discrimination and clinical practicability.Δ E was an independent predictor of exercise capacity in the HFpEF,HFmrEF,and HFrEF subgroups. |
