This retrospective, observational, cohort study was conducted at a single tertiary cardiac center (the Affiliated Hospital of Xuzhou Medical University, China) and was approved by the Institutional Review Board (Approval number: XYFY2021-KL116-01; May 25, 2021). The patients were recruited between July 2016 and September 2021, and they provided written informed consent during follow-up. The inclusion criteria for the HFpEF group and exclusion criteria for the study were consistent with those in our previous study (11). The inclusion criteria for the HFrEF and HFmEF group were as follows: (1) patients having typical HF signs and symptoms; (2) those with an N-terminal pro-brain natriuretic peptide level of >125 pg/ml; (3) those with LVEF between 41% and 49% were included in HFmEF group and those with LVEF of <40% were included in HFrEF group (12). Controls were recruited based on the following criteria: (1) be free of HF symptoms (2) LVEF ≥50% and be free of echocardiographic signs of severe diastolic dysfunction. Participants were excluded if they have: (1) Severe liver and kidney dysfunction (2). Severe heart valve disease (severe mitral stenosis, severe main stenosis) or known infiltrative or hypertrophic cardiomyopathy (3). Malignant tumors and severe hematological diseases (4). Severe infection (5). Autoimmune system diseases. All participants underwent a review of medical records and clinical data (general condition, medication, and blood sample analysis); transthoracic echocardiography and CMR were performed during the same visit.

All participants underwent transthoracic echocardiography (iE33, Philips Healthcare, Best, The Netherlands), and cardiac systolic and diastolic function indices were obtained using a similar procedure used in our previous study (11).

A retrospectively gated cine-CMR method was utilized to capture cine images in cardiac short-axis, vertical long-axis, and horizontal long-axis orientations, while a steady-state free precession sequence was employed for volumetric analysis. The steady-state free precession images were utilized for cine imaging with specific parameters (repetition time msec/echo time msec, 3.2/1.2; flip angle, 64°; voxel size, 1.41.46 mm; matrix, 180 × 256 pixels).

CVI42 software (v5.13.5, Circle Vascular Imaging, Canada) was used by a single observer with over 3 years of experience to evaluate the CMR images, maintaining blindness to all data. Ventricular volumes, ejection fraction (EF), and left ventricular (LV) mass (excluding papillary muscles) were determined using balanced steady-state cine imaging and calculated from the short-axis cine stack. Endocardial borders were traced in end-systole and end-diastole, and volumes were averaged for calculation.

All volumetric and mass data were indexed to body surface area to derive ventricular volumetric index and mass index. The biplane approach was employed to calculate left atrial volumes, excluding the appendage and pulmonary veins, with maximum and minimum volumes determined. End-diastolic volume (EDV) and end-systolic volume (ESV) of the ventricular volumes were assessed from the volume-time curve for the maximal and minimal values to calculate EF.

Endocardial and epicardial borders of the LV, right ventricle (RV), and left atrium (LA) were delineated to enable semi-automatic tracking of the myocardium throughout the cardiac cycle. After automatic profile delineation, tracking performance was verified and manually adjusted if necessary to ensure accurate strain analysis. LA strain analysis utilized one two-chamber view and one four-chamber view. LV and RV strain were assessed using a stack of short-axis views and three long-axis views (two-chamber, three-chamber, and four-chamber views). GCS, GLS, and GRS values (GLS and GRS were obtained from LA strain analysis) were averaged from peak values of all 16 American Heart Association segments, excluding the most basal and apical sections. Tracking was repeated three times, and the average of these repetitions was used for further analysis, with repetitions conducted in separate sessions (13). The cardiologists who assessed outcomes of clinical outcomes and CMR parameters were all blinded to the group of patients.

Strain analysis was performed on LV, LA, and RV in the previous study, and in this study, we added the right atrium (RA). The delineation of endocardial and epicardial borders in RA was similar to that in LA; thus, only GLS and GRS were acquired from RA strain analysis.

All patients were followed up through telephone calls and interviewed by a single cardiologist in February 2022. The clinical endpoint was a composite of mortality or HF readmission (12). Assuming that recall may be inaccurate in some elderly patients, we conducted a secondary validation using hospital databases.

In total, 354 patients, including 143 with HFpEF, 43 with HFmEF, 82 with HFrEF, and 86 participants without HF, were enrolled. Table 1 shows the baseline characteristics of the study population, including demographics, clinical findings, functional status, medical history, medication history, and biochemistry results. Patients with HFpEF (58 ± 11 years) and HFmEF (56 ± 14 years) were older than the controls (44 ± 17 years). Patients with HFrEF had higher body mass index levels (26.5 ± 4.5 kg/m²) than the other patients. From the perspective of heart rate, it can be observed that as LVEF decreases, the heart rate of the subjects gradually increases. As LVEF decreases, blood pressure, especially SBP, tends to decrease. Compared to HFpEF subjects, HFmrEF and HFrEF subjects have a higher proportion of NYHA grade III or IV evaluations. Compared with the normal population, patients with HF have a higher proportion of previous coronary heart disease, myocardial infarction, hypertension, and cardiomyopathy. Moreover, they used more heart failure medication and anticoagulant drugs for maintaining or restoring cardiac function.

The baseline imaging characteristics, including echocardiography and CMR measurements, are shown in Table 2. The ejection fraction of patients with different phenotypes of HF decreased to varying degrees (55.3 ± 5.6 for HFpEF, 46.5 ± 2.6 for HFmEF, 31.1 ± 5.3 for HFrEF, 62.5 ± 4.4 for controls), met the criteria. The CMR analysis of LV, RV, LA, and RA revealed the impairment of chambers in chamber volume and mass. Patients with HF had enlarged chamber volumes and larger chamber weights. These results remained unchanged after controlling for body mass index each strain was altered in the three groups compared with the control group. The mean values for left atrial GRS, left ventricular GRS, and left ventricular global circumferential strain (LVGCS) in the HFpEF group were different from that in the HFmEF group (P < 0.001). Moreover, all eight strains were altered in these groups compared with the HFrEF group (P < 0.001).

The follow-up was completed for all participants during a median follow-up of 999 days (616–1,334); 20.6% (73/354) of participants experienced adverse events, including HF readmission or cardiovascular death. Due to the serious confounding bias caused by the small sample size of HFmEF and HFrEF, regression analysis could not be implemented, but difference analysis was performed. The results are shown in Figures 1, 2. We performed COX regression analysis on HFpEF and obtained the results.

For 82 patients with HFrEF, LAGLS showed variation in any adverse event (t = 2.79, P = 0.007), similarly to LVGCS (t = 5.74, P < 0.001).Among the 43 patients with HFmEF in this study, we observed differences in LVGLS (t = 5.34, P < 0.001), LVGCS (t = 5.28, P < 0.001), and LAGRS (t = −4.41, P < 0.001) in relation to the endpoint events. The results are shown in Figures 1, 2.

In our study, the myocardial strain in the left ventricle and left atrium was associated with cardiovascular adverse events. After adjusting for other prognostic indicators, such as diabetes mellitus and smoking history, we found that the LAGLS (HR, 1.46; 95% CI: 1.07–1.46; P = 0.015), LVGRS (HR, 1.34; 95% CI: 1.07–1.68; P = 0.009), LVGCS (HR, 1.44; 95% CI: 1.06–1.94; P = 0.017), LVGLS (HR, 2.30; 95% CI: 1.513.49; P < 0.001) were all significantly associated with the occurrence of adverse cardiovascular events. The results are shown in Table 3.

The univariate Cox regression analysis revealed that a 1% increase in left atrial global longitudinal strain (LAGLS) was associated with a hazard ratio (HR) of 1.21 [95% confidence interval (CI): 1.15–1.28; P < 0.001]. LVGCS (HR, 1.18; 95% CI: 1.12–1.24; P < 0.001) and LVGLS (HR, 1.27; 95% CI: 1.20–1.36; P < 0.001) were also associated with adverse events. Among clinical variables, hypertension (HR, 2.11; 95% CI: 1.33–13.36; P = 0.002) and cardiomyopathy (HR, 2.26; 95% CI: 1.42–3.60; P < 0.001) were associated with adverse outcomes.

In multivariable analyses, LAGLS (95% CI: 1.08–1.29; P < 0.001) and LVGCS (95% CI: 1.19–1.51; P < 0.001) were significantly associated with outcomes; the clinical variable hypertension (95% CI: 1.09–3.73; P < 0.025) remained a risk factor. Tables 3, 4 shows the results of univariate and multivariable Cox regression analyses. The results are shown in Table 4.

Further receiver operating characteristic curve (ROC) analysis of LAGLS, LVGCS, and LVGLS showed that the critical value of LAGLS was −9.72 [area under the curve (AUC) 0.858, sensitivity 86.3%, specificity 78.6%], that of LVGCS was −12.43 (AUC 0.833, sensitivity 91.7%, specificity 61.2%), and that of LVGLS was −9.56 (AUC 0.866, sensitivity 93.1%, specificity 71.5%).

The patients were divided into groups according to the critical value. We drew the Kaplan–Meier survival curve of all patients in the group with a LAGLS of ≧9.72 and LAGLS of ≤9.72. The median survival time of patients with a LAGLS of ≧9.72 was 1,736 days (95% CI: 1,659–1,812) and that of patients with a LAGLS of ≤9.72 was 1,371 days (95% CI: 1,163–1,578). Significant differences were observed between the two groups (LogRank P < 0.001). Similarly, we drew a Kaplan–Meier survival curve with an LVGCS of 12.43 and an LVGLS of 9.56 as the critical value. The median survival time of the groups divided by the critical value was significantly different, and significant differences were observed between the groups (LogRank P < 0.001). The results are shown in Figures 3–5.